Volume 451 Issue 7174



News Features

News & Views


A receptor that mediates the post-mating switch in reproductive behaviour p.33

Mating in many species induces a dramatic switch in female reproductive behaviour. In most insects, this switch is triggered by factors present in the male’s seminal fluid. How these factors exert such profound effects in females is unknown. Here we identify a receptor for the Drosophila melanogaster sex peptide (SP, also known as Acp70A), the primary trigger of post-mating responses in this species. Females that lack the sex peptide receptor (SPR, also known as CG16752), either entirely or only in the nervous system, fail to respond to SP and continue to show virgin behaviours even after mating. SPR is expressed in the female’s reproductive tract and central nervous system. The behavioural functions of SPR map to the subset of neurons that also express the fruitless gene, a key determinant of sex-specific reproductive behaviour. SPR is highly conserved across insects, opening up the prospect of new strategies to control the reproductive and host-seeking behaviours of agricultural pests and human disease vectors.

doi: 10.1038/nature06483


A young massive planet in a star–disk system p.38

version="1.0" encoding="utf-8"?> ]>
Nature 451 7174 20080103 38414 0028-0836 1476-4687 2007Nature Publishing Group Supplementary Information

The file contains Supplementary Tables 1-3; Supplementary Figures 1-3 with Legends; Supplementary Discussion; Supplementary Notes; Supplementary Methods and additional references.

A young massive planet in a star–disk system J.SetiawanJ Th.HenningT R.LaunhardtR A.MüllerA P.WeiseP M.KürsterM Max-Planck-Institut für Astronomie, Heidelberg, D-69117, Germany Correspondence and requests for materials should be addressed to J.S. (setiawan@mpia.de) or R.L. (rlau@mpia.de). &e080103-9; &nature06426-s1;

There is a general consensus that planets form within disks of dust and gas around newly born stars. Details of their formation process, however, are still a matter of ongoing debate. The timescale of planet formation remains unclear, so the detection of planets around young stars with protoplanetary disks is potentially of great interest. Hitherto, no such planet has been found. Here we report the detection of a planet of mass (9.8±3.3)MJupiter around TW Hydrae (TW Hya), a nearby young star with an age of only 8–10 Myr that is surrounded by a well-studied circumstellar disk. It orbits the star with a period of 3.56 days at 0.04 au, inside the inner rim of the disk. This demonstrates that planets can form within 10 Myr, before the disk has been dissipated by stellar winds and radiation.

With the discovery of the first planet orbiting another Sun-like star, our understanding of planet formation has experienced a renaissance. The vast majority of exoplanets have been discovered with the radial velocity (RV) technique. This method is most sensitive to giant planets on short-period orbits and it works best with non-active solar-like stars. Other techniques with different detection biases are currently emerging (for example, transit photometry, astrometry, direct imaging), but the RV method still remains the most successful technique for detecting exoplanets. Correspondingly, our picture of extrasolar planetary systems is still dominated by the detection biases of this method.

Planets form from dust and gas in circumstellar disks around young stars. The growth from micrometre-sized dust grains to planetary embryos through collisions is believed to be the key mechanism leading to the formation of planetary cores. As these cores grow, they eventually become massive enough to accrete gas from the disk. Alternatively, giant planets are also proposed to form directly via gravitational instabilities in the disk.

One of the most important timescales for planet formation is the disk dispersal time. From observations of near-infrared excess and millimetre-wavelength emission of young stars it has been concluded that circumstellar disks dissipate within about 10 Myr after the star formation. The formation of planets from disk material must occur in this time window. Until now, no planet has been detected by RV surveys around a star younger than 100 Myr old. The only known young planet was detected by direct imaging at 55 au from the brown dwarf 2MASS1207 (ref. 9). The main reason for this lack of detections is that young stars were systematically excluded from large RV surveys because they usually exhibit high levels of stellar activity. However, when carefully analysing and characterizing all activity effects, it is also possible to detect planets around young stars with the RV technique. The young star TW Hya (see Supplementary Table 1) is surrounded by a circumstellar disk that has been studied by using a wide variety of different observing techniques and analysis methods (Fig. 1). From Hubble Space Telescope observations the disk was found to be oriented almost face-on.

We acquired high-resolution spectroscopic measurements of TW Hya with the Fibre-fed Extended Range Optical Spectrograph (FEROS) at the 2.2 m Max-Planck-Gesellschaft/European Southern Observatory (MPG/ESO) telescope (La Silla Observatory). Data reduction and RV determination procedures for FEROS data are described in ref. 21. For the RV computations we excluded spectral regions that contain strong emission lines (such as Ca ii H&K, H&bgr;, He i, Na i and H&agr;). The RV measurements are shown in Fig. 2 and listed in Supplementary Table 2. The sine-fitting periodogram of the entire RV data set shows three distinct significant periods (Fig. 3). The first and most pronounced signal at P = 3.56 days has a false alarm probability (FAP) of 10-14. This period corresponds to the regular sinusoidal variation shown in Fig. 2. Further signals are found at 0.78 and 1.39 days. However, after subtracting the 3.56-day period from the data and recomputing the periodogram for the residual RVs, these two peaks disappear. Hence, they are aliases. Other peaks that appear in the periodogram of the residual RVs have high FAPs and are not significant.

RV variations can be induced either by an orbiting companion, rotational modulation due to starspots, or nonradial pulsations. In the case of a companion, all spectral lines move simultaneously without affecting the line profile. In the case of rotational modulation and nonradial pulsations, the integral line shapes will vary and cause changes in the measured effective RV. To verify the nature of the observed RV variations, it is therefore mandatory to analyse the stellar activity indicators.

The 3.56-day RV variation appears to be regular during both observing periods. The phase-folded RV curve reveals very clearly the nearly sinusoidal variation. We find that this period is neither correlated with photometric variations nor with any stellar activity indicators. The bisector analysis of the line profile asymmetries confirms that there is no significant correlation between the 3.56-day period and the stellar activity (Fig. 4). The most probable explanation of the 3.56-day RV variation is therefore the presence of a companion orbiting TW Hya. We calculated an orbital solution using a keplerian fit to the RV variation (Table 1) and derived a minimum companion mass of (1.2 ± 0.4)MJupiter. Assuming that the companion orbits the star in the plane of the disk (i = 7° ± 1°) and also taking into account all other uncertainties, we computed a true companion mass of (9.8 ± 3.3)MJupiter. This mass range clearly qualifies the companion as a giant planet. The orbit of the planet is almost circular and has a semimajor axis of (0.041 ± 0.002) au, placing it inside the inner hole of the disk.

Several authors have tried to derive the rotation period of TW Hya from photometric observations (Supplementary Table 3). However, classical T Tauri stars do not easily reveal their rotation periods, because random variations due to accretion and related short-lived hot spots tend to mask their periodic behaviour. Phase-coherent periodic modulations may only persist for a few rotation cycles.

We investigated all available activity indicators in the spectra, including the H&agr; emission line (Supplementary Fig. 2). We selected several short time intervals during which obvious periodic variations are persistent (Supplementary Fig. 3). We then computed the periodicities independently for each subset. For the residual RVs and the Ca ii activity index we found periods of 1.4, 1.7 and 2.3 days with typical uncertainties of ±0.3 day. Other activity indicators (for example, H&bgr;, He i and effective temperature (Teff) variation) yielded similar results. Assuming that all individual values represent uncertain measurements of the same underlying period, we attribute the weighted mean to the stellar rotation period, that is Prot = 1.6 ± 0.3 days. This period is in agreement with the photometric variability reported by other authors. Combining Prot with the stellar radius and rotation velocity of TW Hya, we derived an inclination angle of the stellar rotation axis of 14° ± 4°, comparable with that of ref. 24. When we analysed the bisector velocity spans and bisector curvatures we found no significant correlation with the RVs (Fig. 4), confirming that the 3.56-day RV period is not caused by activity-related lineshape variations.

There is one activity indicator that varies with another significant period, at 9.05 days: the equivalent width of the H&agr;. The interpretation of this variability is still inconclusive, but it is probably related to a phenomenon in the disk. The corresponding keplerian radius would be ∼0.07 au, which is very close to the inner edge of the disk.

The detection of a young (8–10 Myr) and massive planet with 9.8MJupiter on a 0.04 au orbit around TW Hya provides important constraints on theories of planet formation and migration. It gives a real upper limit to the timescales of planet formation and migration that is not based on purely statistical arguments. Furthermore, this is the first direct observational link (to our knowledge) between a circumstellar disk and a newly formed planet, thus justifying calling such disks ‘protoplanetary’.

In the period–mass distribution of known exoplanets, massive planets with masses >2MJupiter seem to be rare at semi-major axes of less than 0.6 au. However, such statistics are based mostly on RV data, which provide only lower limits to the planet masses. When we use msini (1.2MJupiter) instead of the true mass (where m is the mass and i is the the inclination angle), TW Hya b falls in the middle of the period–mass distribution for RV planets. These facts show that our current understanding of the population of exoplanets is still biased by the detection method.

Models that assume the formation of planets beyond the snowline and include migration and accretion allow the formation of massive planets. However, it is still not clear whether a planet as massive as 9.8MJupiter could have formed through core accretion, or whether gravitational instabilities in the disk must have been involved.

A massive planet like TW Hya b would open a gap in the disk and undergo type II migration on a typical timescale of 105 years. TW Hya b may have formed and started its migration path close to the present inner edge of the optically thick disk at 1–4 au. We can even speculate that the planet is responsible for the clearing of the inner disk through the accretion of the gas. Its inward migration came to a halt when the planet entered the gas-free zone at the inner edge of the optically thin disk at ∼0.06 au. This inner hole can have formed, for example, by disk dispersal processes or by the stellar magnetosphere.

The detection of TW Hya b opens up the possibility of directly connecting the disk evolution and planet formation processes. It is the ideal system to test numerical simulations of planet core formation, migration and accretion.

A pictographic sketch of the TW Hya system.

The star, the disk with a central hole, and the newly discovered planet are shown. The circumstellar disk around TW Hya is almost face-on. Recent measurements yielded a disk inclination of 7° ± 1° (ref. 12). Ref. 13 concluded that an optically thin inner-disk zone void of large dust particles would model the complete spectral energy distribution well. The relatively sharp transition to the outer optically thick and geometrically flared disk was predicted to be located at ∼4 au. They speculated about the existence of a giant planet in this region that could have caused the clearing. The existence of such a transition at ∼4 au was confirmed by millimetre interferometric observations from 7 mm Very Large Array observations. Mid-infrared interferometric observations showed that the transition between cleared inner and optically thick outer disk must occur at radii between 0.5 to 0.8 au. Mid-infrared observations are more sensitive to smaller dust particles compared to the millimetre observations, which may explain the difference. On the other hand, CO gas emission from a region inside 1.0 au has been detected, showing that this region is not completely void of material. From Keck near-infrared interferometric observations it was concluded that the inner edge of the disk is located at about 0.06 ± 0.01 au from the star. This is somewhat larger than the dust sublimation radius, but the precise value depends critically on the assumed dust properties. The authors speculated that magnetospheric accretion may be responsible for this truncation. From the H&agr; line and short-wavelength continuum excess, ref. 19 derived a mass accretion rate from the disk onto the star of (∼5 × 10-10)MSun per year and concluded that TW Hya is at the end of its accretion stage. The planet at 0.04 au marked here is the result of our RV monitoring survey.

Radial velocity variation of TW Hya.

a, The RVs were obtained during two observing runs with 12 consecutive nights (between 28 February and 12 March 2007) and 20 consecutive nights (24 April to 13 May 2007). With typically three spectra per night, we sampled possible variability periods from about 1 to 12 days. For the RV calculations we used a cross-correlation technique, in which about 1,300 spectral lines were cross-correlated with a numerical template. The error bars are standard errors of the mean RV value. The typical accuracy of the individual RV is about 40 m s-1, which is mostly due to the rapid rotation and activity of TW Hya. For comparison, the typical accuracy achieved with FEROS for quiet and slow rotating solar-type stars is about 5 m s-1. The solid line shows a keplerian fit with a period of 3.56 days. The scatter of the data points around this curve (residuals) is probably due to stellar activity. b, The phase-folded (with P = 3.56 days) RV curve (blue line) of the planet around TW Hya. This periodic variation is stable within the observation time window. The amplitude of RV variation is 196 ± 61 m s-1. The black lines represent the uncertainty of ±61 m s-1 below and above the blue curve.

Sine-fitting periodogram of RV variation.

A period analysis was performed using both a sine-fitting routine minimizing &khgr;2 (ref. 29), and the Lomb–Scargle periodogram (Supplementary Fig. 1). The window function is displayed in the inset and has only a single peak at P = 1 day. For the sine-fitting we used the bootstrap randomization method to determine FAPs—that is, the probability that a value of &khgr;2 as small as (or smaller than) the optimum value found for the data was obtained purely by chance. We found three significant &khgr;2 minima at distinct periods. The first and most pronounced &khgr;2 minimum at P = 3.56 days has a FAP = 10-14. Further &khgr;2 minima are seen at 0.78 and 1.39 days, both with FAP = 10-5. We calculated the residual RVs by subtracting the 3.56-day periodicity. In the sine-fitting periodogram of the residual RVs, the 0.78 and 1.39-day periods are no longer seen in the periodogram. Thus, they are aliases.

Bisector analysis of line profile asymmetry.

We used a cross-correlation technique, using several hundred spectral lines of TW Hya. We measured the bisector velocity spans (a) and bisector curvatures (b), which are well known as excellent stellar activity indicators. a, Bisector velocity span versus RV for the entire data set. There is no significant correlation (correlation coefficient ∼0.2), indicating that the 3.56-day RV variation is not caused by the line profile changes. b, The bisector curvature does not show a significant correlation with the RV (correlation coefficient ∼0.3), confirming that stellar activity is not responsible for the observed 3.56-day RV variation. The error bars are the standard mean errors of the mean bisector velocity span/curvature, computed from the bisectors of each echelle order.

Orbital parameters for TW Hya b Primary mass (0.7 ± 0.1)MSun Orbital period (3.56 ± 0.02) days Offset RV (12,420.7 ± 4.1) m s-1 RV semi–amplitude (196 ± 61) m s-1 Inclination angle (7 ± 1)° Eccentricity 0.04 ± 0.03 Periastron longitude (105 ± 27)° Reduced &khgr;2 3.32 Minimum companion mass (1.2 ± 0.4)MJupiter True companion mass (9.8 ± 3.3)MJupiter Orbital semi–major axis (0.041 ± 0.002) au

We thank the 2.2 m MPG/ESO La Silla team, especially P. Francois, B. Conn, M. Stefanon, O. Schütz, M. Morell and A. Gonzales for their help during the observations. We thank W. Herbst for constructive discussion and providing the supporting data.

Author Contributions The observations were carried out by J.S. and A.M.; T.H. and R.L. were responsible for the project planning. The data analysis was done by J.S., A.M., P.W. and M.K.

Safronov, V. S. The protoplanetary cloud and its evolution. Sov. Astron. 10, 650658 (1967) Pollack, J. B. et al. Formation of the giant planets by concurrent accretion of solids and gas. Icarus 124, 6285 (1996) Mayor, M. & Queloz, D. A. Jupiter-mass companion to a solar-type star. Nature 378, 355359 (1995) The Extrasolar Planet Encyclopediawww.exoplanet.eu Boss, A. P. Gas giant protoplanet formation. disk instability models with thermodynamics and radiative transfer. Astrophys. J. 563, 367373 (2001) Haisch, K. E., Lada, E. A. & Lada, C. J. Disk frequencies and lifetimes in young clusters. Astrophys. J. 553, L153L156 (2001) Cieza, L. et al. The Spitzer c2d Survey of weak-line T-Tauri stars. II. New constraints on the timescale for planet building. Astrophys. J. 667, 308328 (2007) Carpenter, J. M. et al. Evolution of cold circumstellar dust around solar-type stars. Astron. J. 129, 10491062 (2005) Chauvin, G. et al. Giant planet companion to 2MASSW J1207334–393254. Astron. Astrophys. 438, L25L28 (2005) Setiawan, J. et al. Evidence for a planetary companion around a nearby young star. Astrophys. J. 660, L145L148 (2007) Krist, J. E. et al. WFPC2 images of a face-on disk surrounding TW Hydrae. Astrophys. J. 538, 793800 (2000) Qi, C. et al. Imaging the disk around TW Hydrae with the submillimeter array. Astrophys. J. 616, L11L14 (2004) Calvet, N. et al. Evidence for a developing gap in a 10 Myr old protoplanetary disk. Astrophys. J. 568, 10081016 (2002) Hughes, A. M. et al. An inner hole in the disk around TW Hydrae resolved in 7 mm dust emission. Astrophys. J. 664, 536542 (2007) Ratzka, T. et al. High spatial resolution mid-infrared observations of the low-mass young star TW Hya. Astron. Astrophys. 471, 173185 (2007) Rettig, T. W. et al. Discovery of CO gas in the inner disk of TW Hydrae. Astrophys. J. 616, L163L166 (2004) Salyk, C., Blake, G. A., Boogert, A. C. A. & Brown, J. M. Molecular gas in the inner 1 AU of the TW Hya and GM Aur transitional disks. Astrophys. J. 655, L105L108 (2007) Eisner, J. A., Chiang, E. I. & Hillenbrand, L. A. Spatially resolving the inner disk of TW Hydrae. Astrophys. J. 637, L133L136 (2006) Muzerolle, J. et al. Disk accretion in the 10 Myr old T Tauri stars TW Hydrae and Hen 3–600A. Astrophys. J. 535, L47L50 (2000) Kaufer, A. et al. Performance report on FEROS, the new fiber-linked echelle spectrograph at the ESO 1.52-m telescope. In Proc. SPIE Optical and IR Telescope Instrumentation and Detectors. 4008, 459466 (2000) Setiawan, J. et al. Precise radial velocity measurements of G and K giants. First results. Astron. Astrophys. 397, 11511159 (2003) Herbst, W. & Koret, D. L. Rotation periods of four T Tauri stars. Astron. J. 96, 19491955 (1988) Mekkaden, M. V. Photometric and polarimetric variability of the isolated T Tauri star TW Hydrae. Astron. Astrophys. 340, 135142 (1998) Alencar, S. H. P. & Batalha, C. Variability of southern T Tauri stars. II. The spectral variability of the classical T Tauri star TW Hydrae. Astrophys. J. 571, 378393 (2002) Alibert, Y., Mordasini, C., Benz, W. & Winisdoerffer, C. Models of giant planet formation with migration and disc evolution. Astron. Astrophys. 434, 343353 (2005) Matsuo, T., Shibai, H., Ootsubo, T. & Tamura, M. Planet formation scenarios revisited: core accretion versus disk instability. Astrophys. J. 662, 12821292 (2007) Boss, A. Evolution of the solar nebula. VII. Planet formation and survival of protoplanets formed by disk instability. Astrophys. J. 629, 535548 (2005) Papaloizou, J. C. B. Disk-planet interaction during planet formation. Protostars Planets 5, 655668 (2007) Cumming, A., Marcy, G. W. & Butler, R. P. The Lick planet search: detectability and mass thresholds. Astrophys. J. 526, 890915 (1999) Kürster, M., Schmitt, J. H. M. M., Cutispoto, G. & Dennerl, K. ROSAT and AB Doradus: the first five years. Astron. Astrophys. 320, 831839 (1997)
doi: 10.1038/nature06426

Magnetic monopoles in spin ice p.42

version="1.0" encoding="utf-8"?> ]>
Nature 451 7174 20080103 42454 0028-0836 1476-4687 2007Nature Publishing Group Supplementary Information

The file contains Supplementary Notes with Supplementary Figures 1-2

Magnetic monopoles in spin ice C.CastelnovoC R.MoessnerR S. L.SondhiS L Rudolf Peierls Centre for Theoretical Physics, Oxford University, Oxford OX1 3NP, UK Max-Planck-Institut für Physik komplexer Systeme, 01187 Dresden, Germany PCTP and Department of Physics, Princeton University, Princeton, New Jersey 08544, USA Correspondence and requests for materials should be addressed to C.C. (castel@physics.ox.ac.uk). &e080103-1; &nature06433-s1;

Electrically charged particles, such as the electron, are ubiquitous. In contrast, no elementary particles with a net magnetic charge have ever been observed, despite intensive and prolonged searches (see ref. 1 for example). We pursue an alternative strategy, namely that of realizing them not as elementary but rather as emergent particles—that is, as manifestations of the correlations present in a strongly interacting many-body system. The most prominent examples of emergent quasiparticles are the ones with fractional electric charge e/3 in quantum Hall physics. Here we propose that magnetic monopoles emerge in a class of exotic magnets known collectively as spin ice: the dipole moment of the underlying electronic degrees of freedom fractionalises into monopoles. This would account for a mysterious phase transition observed experimentally in spin ice in a magnetic field, which is a liquid–gas transition of the magnetic monopoles. These monopoles can also be detected by other means, for example, in an experiment modelled after the Stanford magnetic monopole search.

Spin-ice materials are characterized by the presence of magnetic moments &mgr;i residing on the sites i of a pyrochlore lattice (depicted in Fig. 1). These moments are constrained to point along their respective local Ising axes (the diamond lattice bonds in Fig. 1), and they can be modelled as Ising spins &mgr;i = &mgr;Si, where Si = ±1 and . For the spin-ice compounds discussed here, Dy2Ti2O7 and Ho2Ti2O7, (where Dy is dysprosium and Ho is holmium) the magnitude µ of the magnetic moments equals approximately ten Bohr magnetons (µ ≈ 10µB). The thermodynamic properties of these compounds are known to be described with good accuracy by an energy term that accounts for the nearest-neighbour exchange and the long-range dipolar interactions (for a review of spin ice, see ref. 4):

The distance between spins is rij, and a ≈ 3.54 Å is the pyrochlore nearest-neighbour distance. D = µ0µ2/(4&pgr;a3) = 1.41 K is the coupling constant of the dipolar interaction.

Spin ice was identified as a very unusual magnet when it was noted that it does not order to the lowest temperatures T even though it appeared to have ferromagnetic interactions. Indeed, spin ice was found to have a residual entropy at low T (ref. 5), which is well-approximated by the Pauling entropy for water ice, S ≈ SP = (1/2)log(3/2) per spin. Pauling’s entropy measures the huge ground-state degeneracy arising from the so-called ice rules. In the context of spin ice, its observation implies a macroscopically degenerate ground state manifold obeying the ‘ice rule’ that two spins point into each vertex of the diamond lattice, and two out.

We contend that excitations above this ground-state manifold—that is, defects that locally violate the ice rule—are magnetic monopoles with the necessary long-distance properties. From the perspective of the seemingly local physics of the ice rule, the emergence of monopoles at first seems rather surprising. We will probe deeper into how the long-range magnetic interactions contained in equation (1) give rise to the ice rule in the first place. We then incorporate insights from recent progress in understanding the entropic physics of spin ice, and the physics of fractionalization in high dimensions, of which our monopoles will prove to be a classical instance.

We consider a modest deformation of equation (1), loosely inspired by Nagle’s work on the ‘unit model’ description of water ice: we replace the interaction energy of the magnetic dipoles living on pyrochlore sites with the interaction energy of dumbbells consisting of equal and opposite magnetic charges that live at the ends of the diamond bonds (see Fig. 2). The two ways of assigning charges on each diamond bond reproduce the two orientations of the original dipole. Demanding that the dipole moment of the spin be reproduced quantitatively fixes the value of the charge at ±µ/ad, where the diamond lattice bond length .

The energy of a configuration of dipoles is computed as the pairwise interaction energy of magnetic charges, given by the magnetic Coulomb law:where Q&agr; denotes the total magnetic charge at site &agr; in the diamond lattice, and r&agr;&bgr; is the distance between two sites. The finite ‘self-energy’ &ugr;0/2 is required to reproduce the net nearest-neighbour interaction correctly. Equation (2)—which is derived in detail in the Supplementary Information—is equivalent to the dipolar energy equation (1), up to corrections that are small everywhere, and vanish with distance at least as fast as 1/r5.

We consider first the ground states of the system. The total energy is minimized if each diamond lattice site is net neutral, that is, we must orient the dumbbells so that Q&agr; = 0 on each site. But this is just the above-mentioned ice rule, as illustrated in Fig. 2. Thus, one of the most remarkable features of spin ice follows directly from the dumbbell model: the measured low-T entropy agrees with the Pauling entropy (which follows from the short-distance ice rules), even though the dipolar interactions are long-range.

We now turn to the excited states. Naively, the most elementary excitation involves inverting a single dipole / dumbbell to generate a local net dipole moment 2µ. However, this is misleading in a crucial sense. The inverted dumbbell in fact corresponds to two adjacent sites with net magnetic charge Q&agr; = ±qm = ±2µ/ad—a nearest-neighbour monopole–antimonopole pair. As shown in Fig. 2e, the monopoles can be separated from one another without further violations of local neutrality by flipping a chain of adjacent dumbbells. A pair of monopoles separated by a distance r experiences a Coulombic interaction, , mediated by monopolar magnetic fields, see Fig. 3.

This interaction is indeed magnetic, hence the presence of the vacuum permeability µ0, and not 1/ε0, the inverse of the vacuum permittivity. It takes only a finite energy to separate the monopoles to infinity (that is, they are deconfined), and so they are the true elementary excitations of the system: the local dipolar excitation fractionalizes.

By taking the pictures from the dumbbell representation seriously, we may be thought somehow to be introducing monopoles where there were none to begin with. In general, it is of course well known that a string of dipoles arranged head to tail realizes a monopole–antimonopole pair at its ends. However, to obtain deconfined monopoles, it is essential that the cost of creating such a string of dipoles remain bounded as its length grows, that is, the relevant string tension should vanish. This is evidently not true in a vacuum (such as that of the Universe) where the growth of the string can only come at the cost of creating additional dipoles. Magnetic materials, which come equipped with vacua (ground states) filled with magnetic dipoles, are more promising. However, even here a dipole string is not always a natural excitation, and when it is—for example, in an ordered ferromagnet – a string of inverted dipoles is accompanied by costly domain walls along its length (except, as usual, for one-dimensional systems), causing the incipient monopoles to remain confined.

The unusual properties of spin ice arise from its exotic ground states. The ice rule can be viewed as requiring that two dipole strings enter and exit each site of the diamond lattice. In a typical spin-ice ground state, there is a ‘soup’ of such strings: many dipole strings of arbitrary size and shape can be identified that connect a given pair of sites. Inverting the dipoles along any one such string creates a monopole–antimonopole pair on the sites at its ends. The associated energy cost does not diverge with the length of the string, unlike in the case of an ordered ferromagnet, because no domain walls are created along the string, and the monopoles are thus deconfined.

We did not make reference to the Dirac condition that the fundamental electric charge e and any magnetic charge q must exhibit the relationship eq = nh/µ0 whence any monopoles in our universe must be quantized in units of qD = h/µ0e. This follows from the monopole being attached to a Dirac string, which has to be unobservable. By contrast, the string soup characteristic of spin ice at low temperature makes the strings energetically unimportant, although they are observable and are therefore not quantized.

Indeed, the monopoles in spin ice have a magnitude qm = 2µ/ad = 2(µ/µB)(&agr;&lgr;C/2&pgr;ad)qD ≈ qD/8,000, where &lgr;C is the Compton wavelength for an electron, and &agr; is the fine-structure constant. The charge of a monopole in spin ice can even be tuned continuously by applying pressure, because this changes the value of µ/ad.

The monopoles are sources and sinks of the magnetic field H, as is appropriate to the condensed matter setting. More precisely, as in other instances of fractionalization, we can define a ‘smeared’ magnetic charge , where ∇ · H is the divergence of the magnetic field. For a monopole at the origin, separated by L ≫ &xgr; ≫ a from any other monopoles, this gives &rgr;m(0) = ±qm. The form of the magnetic induction B is also monopolar, but with the important difference that a compensating flux travels along the (unquantized) ‘Dirac string’ of flipped dipoles created along with the monopole (see Fig. 2).

Our magnetic monopoles would in principle show up in one of the best-known monopole searches, the Stanford experiment to detect fundamental magnetic monopoles from cosmic radiation. This experiment is based on the fact that a long-lived current is induced in a superconducting ring when a monopole passes through it. We can easily check that the presence of the Dirac string of flipped dipoles is immaterial to the establishment of a current.

The above observations are the central qualitative results of our work: ice-rule-violating defects are deconfined monopoles of H, they exhibit a genuine magnetic Coulomb interaction (see equation (2)), and they produce Faraday electromotive forces in the same way as elementary monopoles would.

We re-emphasize that the ice rule alone does not permit a consistent treatment of the excited states of the physical problem: crucially, the energetic interaction between our defects is absent altogether. Also, in previous discussions of the purely ice-rule problem and related short-range problems it has been noted that the defects do acquire a purely entropic Coulomb (that is, 1/r) interaction, which has a strength that vanishes proportionally to T at low temperatures. This interaction will be present in addition to the magnetic Coulomb interaction discussed in this paper, and is clearly much smaller as T → 0. Also, it will not be accompanied by a magnetic field, it will not renormalize the monopole charge, and it will not be felt by a stationary magnetic test particle that is embedded in the lattice but is not attached to a lattice site.

The most satisfactory way to demonstrate the presence of a monopole would be to measure the force on magnetic test particles, say by a Rutherford scattering experiment or by clever nanotechnological means. Unfortunately, given the lack of elementary magnetic monopoles, we would have to use dipoles as test particles, which significantly weakens such signatures.

An alternative strategy is to look for consequences of the presence of magnetic monopoles in the collective behaviour of spin ice. This is most elegantly achieved by applying a magnetic field in the [111] crystallographic direction. Such a field acts as a (staggered) chemical potential (see Fig. 2), favouring the creation of monopoles of a given sign on either sublattice of the diamond lattice.

We thus have a tuneable lattice gas of magnetic monopoles on the diamond lattice. The basic structure of the phase diagram as a function of magnetic field and temperature can be inferred from work by Fisher and collaborators in the context of ionic lattice gases and Coulombic criticality. At high T, there is no phase transition but a continuous crossover between the high- and low-density phases as the chemical potential is varied. At low T, a first-order phase transition separates the two regimes. This transition terminates in a critical point at (hc, Tc), not unlike the liquid–gas transition of water. This serves as a useful diagnostic, because the liquid–gas transition is absent for a nearest-neighbour spin-ice model, in which defects interact only entropically. In that case, it is known that there cannot be a first-order transition in the limit of low T (ref. 22).

To confirm this scenario, we have demonstrated by Monte Carlo simulations that the actual phase diagram of dipolar spin-ice model has precisely this structure. To rule out the appearance of the liquid–gas transition being due to effects introduced by the approximations leading to equation (2), we simulated directly the original dipolar spin-ice model, equation (1). The resulting phase diagram is depicted in Fig. 4. The critical endpoint is located around (Tc, hc) = (0.57 ± 0.06 K, 0.86 ± 0.03 T). The error bars are mainly due to finite-size effects, as the intensive nature of the simulations of long-range dipolar interactions prohibits simulating very large systems.

This scenario is indeed observed experimentally in spin-ice materials, and our results provide a natural explanation. Spin ice in a [111] magnetic field is a problem that has already attracted considerable attention. The low-density phase of monopoles is known as kagome ice, a quasi-two-dimensional phase with algebraic correlations and a reduced residual entropy. The high-density phase is an ordered state with maximal polarization along the field direction. Experimental results on the liquid–gas transition and its endpoint are also displayed in Fig. 4 for comparison. Our numerical results are in good qualitative agreement with both experiment and the analytic calculations of ref. 21. Our value of the critical field agrees with ref. 6 to within a few per cent, which is less than the uncertainty due to demagnetization effects. However, the experimental value of Tc is about a third lower than the numerical one, most probably due to farther-neighbour (exchange) interaction terms, which—although small—can shift the location of a transition temperature considerably.

The presence of a liquid–gas transition was noted to be very remarkable because there are few, if any, other experimentally known instances in localized spin systems. No mechanism was known to account for this phenomenon, and our theory of magnetic monopoles fills this gap.

The existence of magnetic monopoles in a condensed matter system is exciting in itself. (The monopoles appearing in the interesting work on the anomalous Hall effect are not excitations and do not involve the physical magnetic field.) Moreover, these monopoles are a rare instance of high-dimensional fractionalization, of interest in fields as diverse as correlated electrons and topological quantum computing. We hope our analysis will encourage experiments aimed at directly detecting these monopoles. There are many avenues to explore in search of useful signatures, among them scattering, transport and noise measurements, and flux detection.

The pyrochlore and diamond lattices.

The magnetic moments in spin ice reside on the sites of the pyrochlore lattice, which consists of corner-sharing tetrahedra. These are at the same time the midpoints of the bonds of the diamond lattice (black) formed by the centres of the tetrahedra. The ratio of the lattice constant of the diamond and pyrochlore lattices is . The Ising axes are the local [111] directions, which point along the respective diamond lattice bonds.

Mapping from dipoles to dumbbells.

The dumbbell picture (c, d) is obtained by replacing each spin in a and b by a pair of opposite magnetic charges placed on the adjacent sites of the diamond lattice. In the left panels (a, c), two neighbouring tetrahedra obey the ice rule, with two spins pointing in and two out, giving zero net charge on each site. In the right panels (b, d), inverting the shared spin generates a pair of magnetic monopoles (diamond sites with net magnetic charge). This configuration has a higher net magnetic moment and it is favoured by an applied magnetic field oriented upward (corresponding to a [111] direction). e, A pair of separated monopoles (large red and blue spheres). A chain of inverted dipoles (‘Dirac string’) between them is highlighted in white, and the magnetic field lines are sketched.

Monopole interaction.

Comparison of the magnetic Coulomb energy (equation (2); solid line) with a direct numerical evaluation of the monopole interaction energy in dipolar spin ice (equation (1); open circles), for a given spin-ice configuration (Fig. 2e), as a function of monopole separation.

Phase diagram of spin ice in a [111] field.

The location of the monopole liquid–gas transition from numerics (blue line) compared to experiment (black line; ref. 6). The solid lines are first-order transitions terminated by critical endpoints (filled circles). The dashed lines are crossovers. The inset shows magnetization curves showing the onset of first-order behaviour as the temperature is lowered. Our simulations cover the range 0.335 K < T < 0.8 K for 1,024 spins. At the lowest temperatures, the parallel tempering code we use in our simulations of the Ewald-summed dipolar interaction no longer completely suppresses the hysteresis, and we have extended the first-order transition line using Clausius–Clapeyron.

We thank S. Bramwell, J. Chalker, C. Chamon and S. Kivelson (especially for pointing out ref. 8) for discussions. This work is supported in part by EPSRC (CC), and NSF (SLS). We also thank A. Canossa for support with the graphics.

Author Contributions All authors contributed equally to the manuscript.

Milton, K. A. Theoretical and experimental status of magnetic monopoles. Rep. Prog. Phys. 69, 16371712 (2006) Laughlin, R. B. Nobel lecture: Fractional quantization. Rev. Mod. Phys. 71, 863874 (1999) Harris, M. J., Bramwell, S. T., McMorrow, D. F., Zeiske, T. & Godfrey, K. W. Geometrical frustration in the ferromagnetic pyrochlore Ho2Ti2O7. Phys. Rev. Lett. 79, 25542557 (1997) Bramwell, S. T. & Gingras, M. J. P. Spin ice state in frustrated magnetic pyrochlore materials. Science 294, 14951501 (2001) Ramirez, A. P., Hayashi, A., Cava, R. J., Siddharthan, R. B. & Shastry, S. Zero-point entropy in ‘spin ice’. Nature 399, 333336 (1999) Aoki, H., Sakakibara, T., Matsuhira, K. & Hiroi, Z. Magnetocaloric effect study on the pyrochlore spin ice compound Dy2Ti2O7 in a [111] magnetic field. J. Phys. Soc. Jpn 73, 28512856 (2004) Higashinaka, R., Fukazawa, H., Deguchi, K. & Maeno, Y. Low temperature specific heat of Dy2Ti2O7 in the kagome ice state. J. Phys. Soc. Jpn 73, 28452850 (2004) Cabrera, B. First results from a superconductive detector for moving magnetic monopoles. Phys. Rev. Lett. 48, 13781381 (1982) Siddharthan, R. et al. Ising pyrochlore magnets: low-temperature properties, ice rules, and beyond. Phys. Rev. Lett. 83, 18541857 (1999) Ruff, J. P. C., Melko, R. G. & Gingras, M. J. P. Finite-temperature transitions in dipolar spin ice in a large magnetic field. Phys. Rev. Lett. 95, 097202 (2005) Huse, D. A., Krauth, W., Moessner, R. & Sondhi, S. L. Coulomb and liquid dimer models in three dimensions. Phys. Rev. Lett. 91, 167004 (2003) Hermele, M., Fisher, M. P. A. & Balents, L. Pyrochlore photons: The U(1) spin liquid in a S = 1/2 three-dimensional frustrated magnet. Phys. Rev. B 69, 064404 (2004) Henley, C. L. Power-law spin correlations in pyrochlore antiferromagnets. Phys. Rev. B 71, 014424 (2005) Fulde, P., Penc, K. & Shannon, N. Fractional charges in pyrochlore lattices. Ann. Phys. 11, 892900 (2002) Levin, M. & Wen, X. G. Quantum ether: Photons and electrons from a rotor model. Phys. Rev. B 73, 035122 (2006) Nagle, J. F. Theory of the dielectric constant of ice. Chem. Phys. 43, 317328 (1979) Jackson, J. D. Classical Electrodynamics 251–260, Ch. 6.12–6.13 (Wiley, New York, 1975) Saitoh, E., Miyajima, H., Yamaoka, T. & Tatara, G. Current-induced resonance and mass determination of a single magnetic domain wall. Nature 432, 203206 (2004) Dirac, P. A. M. Quantised singularities in the electromagnetic field. Proc. R. Soc. A 133, 6072 (1931) Rajaraman, R. Fractional charge. Preprint at 〈http://arxiv.org/abs/cond-mat/0103366v1〉 (2001) Kobelev, V., Kolomeisky, A. B. & Fisher, M. E. Lattice models of ionic systems. J. Chem. Phys. 116, 75897598 (2002) Heilmann, O. J. & Lieb, E. H. Theory of monomer-dimer systems. Commun. Math. Phys. 25, 190232 (1972) Isakov, S. V., Raman, K. S., Moessner, R. & Sondhi, S. L. Magnetization curve of spin ice in a [111] magnetic field. Phys. Rev. B 70, 104418 (2004) Fang, Z. et al. The anomalous Hall effect and magnetic monopoles in momentum space. Science 302, 9295 (2003) Kitaev, A. Fault tolerant quantum computing. Ann. Phys. 303, 230 (2003)
doi: 10.1038/nature06433

Three-dimensional atomic-scale structure of size-selected gold nanoclusters p.46

version="1.0" encoding="utf-8"?> ]>
Nature 451 7174 20080103 46480 0028-0836 1476-4687 2007Nature Publishing Group Supplementary Information

The file contains Supplementary Notes and Figures.

Three-dimensional atomic-scale structure of size-selected gold nanoclusters Z. Y.LiZ Y N. P.YoungN P M.Di VeceM S.PalombaS R. E.PalmerR E A. L.BlelochA L B. C.CurleyB C R. L.JohnstonR L J.JiangJ J.YuanJ Nanoscale Physics Research Laboratory, School of Physics and Astronomy, School of Chemistry, University of Birmingham, Birmingham B15 2TT, UK UK SuperSTEM Laboratory, Daresbury Laboratory, Daresbury WA4 4AD, UK Beijing Electron Microscopy Centre; Laboratory of Advanced Materials and Department of Materials Science and Engineering, Tsinghua University, Beijing 100084, China Correspondence and requests for materials should be addressed to Z.Y.L. (ziyouli@nprl.ph.bham.ac.uk). &e080103-10; &nature06470-s1;

An unambiguous determination of the three-dimensional structure of nanoparticles is challenging. Electron tomography requires a series of images taken for many different specimen orientations. This approach is ideal for stable and stationary structures. But ultrasmall nanoparticles are intrinsically structurally unstable and may interact with the incident electron beam, constraining the electron beam density that can be used and the duration of the observation. Here we use aberration-corrected scanning transmission electron microscopy, coupled with simple imaging simulation, to determine with atomic resolution the size, three-dimensional shape, orientation and atomic arrangement of size-selected gold nanoclusters that are preformed in the gas phase and soft-landed on an amorphous carbon substrate. The structures of gold nanoclusters containing 309±6 atoms can be identified with either Ino-decahedral, cuboctahedral or icosahedral geometries. Comparison with theoretical modelling of the system suggests that the structures are consistent with energetic considerations. The discovery that nanoscale gold particles function as active and selective catalysts for a variety of important chemical reactions has provoked much research interest in recent years. We believe that the detailed structure information we provide will help to unravel the role of these nanoclusters in size- and structure-specific catalytic reactions. We note that the technique will be of use in investigations of other supported ultrasmall metal cluster systems.

Scanning transmission electron microscopy (STEM), in the mode where incoherently scattered electrons are collected by a high-angle annular dark field (HAADF) detector, is appealing as a method of probing three-dimensional structure of nanoparticles (via an analysis of the intensity map from a single HAADF-STEM image) because its intensity is strongly dependent not only on the atomic number Z of the observed atoms but also on the number of atoms in a column. The recent successful implementation of spherical aberration (Cs) correction in STEM enables us to achieve the same analysis, but now at the atomic scale. We show that, by combining quantitative HAADF-STEM analysis with molecular-dynamics-based model structure search procedures and realistic image contrast simulations, it is possible to identify not only the size and shape but also the structure and orientation of soft-landed Au nanoclusters.

We demonstrate this for size-selected AuN (where N = 309 ± 6) clusters, where Au309 is known to be a possible ‘magic number’ nanocluster (see Supplementary Information). Figure 1a–c displays three high-resolution HAADF images taken from a Au309 nanoparticle. The images show outline shapes, which are approximately pentagonal, square and hexagonal, respectively. Close inspection of the intensity variation within the individual cluster images further reveals that the arrangement of the atomic columns varies from one cluster morphology to the other. Single Au atoms are also visible in the vicinity of the clusters or occasionally some distance away (two such atoms are marked by the circles in Fig. 1a and c). We have imaged the same clusters in several successive frames. The clusters sometimes move/rotate by small amounts and sometimes their structures change too. The individual atoms imaged on the carbon surface around the cluster also move from one frame to the next. The images shown in Fig. 1 are the first-pass images in each case.

To establish a foundation on which to analyse quantitatively the structure of the Au309 clusters, we carried out a series of integrated HAADF intensity measurements on size-selected Au clusters in the size range N = 55–1,500 atoms. For each sample, the HAADF intensity integrated over each cluster shows a narrow distribution (see Fig. 2 inset for N = 309). The mean intensity value taken for each cluster size is plotted as a function of the selected-size N in Fig. 2, where the standard deviation is used for estimating the error bars. The finite width of the distribution can be attributed partly to the resolution of the mass-selector (±2%) and suggests that clusters soft-landing on the surface do not suffer significant fragmentation or coalescence. This is consistent with the results of our detailed image analysis, which reveals no extensive rafts of single atom layers on the amorphous carbon support, apart from a few ‘shake-off’ atoms. Figure 2 shows that the integrated HAADF intensity from the clusters increases linearly with the number of constituent atoms up to about N = 1,500. This linearity implies that, at small cluster size, atoms within the cluster contribute equally to the total scattered electron signal detected by the HAADF detector.

The linearity shown in Fig. 2 also suggests that the HAADF intensities for the individual atomic columns in Fig. 1 can be directly associated with the number of atoms in each column. The clear five-fold symmetry in the atomic column arrangement in Fig. 1a, for example, suggests that the cluster has Ino-decahedral geometry and is oriented on the substrate such that the five-fold axis is parallel to the electron beam, as shown in the hard-sphere representation in Fig. 3a. Figure 3b displays an illustrative line intensity profile from the centre of this cluster to one of the corners, averaging over three experimental pixels (equal to 0.9 Å). Five peaks and a shoulder (marked by the arrow) are apparent, with the peak intensity decreasing gradually towards the corner. Using a simple kinematical approach, the simulated HAADF-STEM image of the decahedral Au309 cluster is shown in Fig. 3c, together with the intensity profile (the solid red curve). The correspondence between the simulated profile (Fig. 3c), and the experimental profile (Fig. 3b), with respect to both the peak positions and the relative peak intensities, is remarkable, indicating the correct identification of the atomic column structure. An icosahedron also has a five-fold rotational symmetry axis; however, the rotation-reflection symmetry of this structure results in a STEM image having ten-fold symmetry, and the technique described has the potential to discriminate between these two possible structures (see Supplementary Information). We have also conducted a full dynamical calculation using the multislice method. The corresponding line profile is shown by the dashed line in Fig. 3c. The similarity between the two simulated line profiles confirms the validity of the simple kinematical approximation for HAADF-STEM image simulation of the Au309 clusters and that the quantization of the HAADF intensity correlates directly with the quantization of the number of the atoms.

Close comparison between Fig. 3b and c also highlights a discrepancy in the atom columns at the edge of the cluster. The experimental intensity of the outermost atomic column is significantly lower than those predicted by either simulation. In addition, an extra shoulder appears in the experimental profile, as indicated by the arrow, with a peak intensity lower than that of the isolated single Au atom observed on the same sample (Fig. 1a). This shoulder exists for all the clusters inspected, irrespective of their shape. The discrepancy cannot be wholly explained by effects such as the rocking movement of the cluster under the electron beam, electron-beam scan instability or noise from other sources, because all these effects would result in the smearing out of the overall image. We take it as evidence that atoms in the surface layer of the clusters fluctuate significantly, on a timescale shorter than the period for the data acquisition. This is similar to the dynamic motion of surface atoms previously observed for larger nanoparticles.

The observed structures of the Au clusters can be understood from their calculated total potential energies for different polyhedral geometries (icosahedral, Ino-decahedral and cuboctahedral) as a function of the number of constituent atoms. After local energy minimization, it was found that, for very small Au clusters (N < 100), the icosahedral structure is much more stable than the Ino-decahedral and the cuboctahedral structures. The total potential energy is in the order of: icosahedral < Ino-decahedral < cuboctahedral. However, for the larger clusters (N ≈ 500–1,000), the order of stability begins to change, with the Ino-decahedral structure becoming more stable than the icosahedral geometry: Ino-decahedral < icosahedral < cuboctahedral. Further increasing the cluster size results in the order of stability changing to: Ino-decahedral < cuboctahedral < icosahedral. For Au309, the difference in total energy between different geometries was less than 1.2 eV (that is, less than 3.88 meV per atom) from the most stable to the least stable. Moreover, there are many local energy minima and the energy barriers between these structures are small. These results support our experimental findings that no one structure dominates. For Au309, we see a similar proportion of clusters with Ino-decahedral (32%) and cuboctahedral (25%) structures and a much lower population of icosahedral structures (8%). In the remaining population, some clusters show irregular facets and some do not show any ordered geometry, possibly because of significant rearrangement of the outer-shell atoms, akin to the solid–liquid phase coexistence predicted for other systems. Given the narrow size distribution of the deposited clusters, our results may shed light on the relative structural stabilities of the various cluster isomers in the gas phase, information that has solicited many theoretical investigations but little hard experimental evidence.

In conclusion, we have demonstrated the suitability of high-angle annular dark-field imaging in the aberration-corrected STEM for detailed structural and stability analysis of size-selected metallic clusters on solid supports at atomic resolution. The multiplicity of cluster geometries revealed by our detailed study of the atomic arrangement of soft-landed Au309 clusters on amorphous carbon supports is consistent with many local energy minima predicted for clusters of this size by cluster simulations. Evidence for increased fluctuations and motion of cluster surface atoms relative to the core atoms within the Au309 clusters may reflect an inherent property of the nanometre-sized gold clusters that could be related to their enhanced catalytic properties through much reduced coordination. Vertical depth information can be extracted from a single projection, with single-atom sensitivity, opening up the possibility to use the technique as a routine three-dimensional structural characterization tool for small nanoparticles at the atomic-scale level, with the help of image simulation based on ab initio cluster modelling including dynamical effects. The experimental approach has practical advantages, such as the more relaxed constraints on cluster stability. When combined with time-lapsed imaging techniques, our approach could provide the dynamical insight into the atomistic structural changes of nanoparticles that commonly occur in some catalytic reactions.

Methods Summary

The gold clusters are formed by gas-phase condensation of sputtered atoms in a rare-gas atmosphere, size-selected by a lateral time-of-flight mass spectrometer and soft-landed on an amorphous carbon support for examination by high-angle annular dark-field STEM, using a spherical-aberration-corrected machine for the atomically resolved imaging. The three-dimensional atomic structures of the size-selected clusters are obtained by comparison of the experimental results with both kinematic and dynamical image simulations, based on structural models optimized by a realistic many-body potential.

Cluster formation and deposition

Gold cluster beams were produced by a source based on radio-frequency magnetron plasma sputtering and gas aggregation. The positively ionized clusters were extracted and focused by ion optics and mass-selected (resolution, ±2%) by a lateral time-of-flight mass filter. The clusters were deposited on a transmission electron microscope grid coated with an amorphous carbon film with a deposition energy of 500 eV, which is in the soft-landing regime, to ensure the maximum probability of retaining the preformed cluster structure upon deposition.

Electron microscope imaging

Systematic measurements of integrated HAADF intensities for a wide range of size-selected Au clusters were carried out using a Tecnai F20 200 kV STEM at the Nanoscale Physics Research Laboratory, University of Birmingham. High-resolution imaging of Au309 clusters was performed using a dedicated VG HB501 STEM at the UK SuperSTEM facility at Daresbury. This was fitted with a second-generation spherical-aberration corrector from Nion Inc. and operated at 100 kV. These images were captured by a high-angle annular dark-field detector with a probe convergence angle of 24 mrad and collection angle from 70 to 200 mrad. The beam current was typically 70 pA. All images were recorded at a rate of 19 &mgr;s per pixel. The images in Fig. 1 were extracted from a 1,024 × 1,024 pixel image and are 170 × 170 pixels. They have been low-pass filtered (with a 3 × 3 kernel and a weight of 1.2) using the SPIP program (version, Image Metrology).

Structure modelling

To simulate HAADF images of the Au309 clusters for comparison with experiments, we first generated idealized icosahedral, Ino-dechedral and cuboctahedral geometries with the magic number N = 309. The structures, which were modelled by the Gupta many-body potential for Au, were then locally relaxed. This was followed by a detailed genetic algorithm search for Au309 cluster structures.

Kinematical STEM simulation

This was done by locating each trial cluster within a two-dimensional grid of dimensions 50 Å× 50 Å, with an interval spacing of 0.25 Å in both the x and y coordinates. A probe of diameter 1 Å visited each grid point and checked for atoms within the probe radius of this grid point in the two-dimensional x–y plane. The quantitative contribution of an atom to the intensity was related to the actual distance between the probe centre and the centre of the atom by a gaussian distribution. All the atomic contributions within the probe radius were then summed and used to calculate the total intensity. Finally, we examined the calculated image for various alignments of each cluster relative to the electron beam probe. The above procedure was repeated for high-symmetry clusters with the magic number N = 309, including the icosahedron, the cuboctahedron and the Ino-decahedron structures.

Multislice STEM simulation

We have used an approach described by Kirkland that allows the atomic potential at each layer to be different. Therefore, the image of a particle can be calculated by putting the appropriate number of atoms in each layer of the multislice calculation in accordance with the cluster geometry. To simulate the HAADF image of the cluster, a supercell was built by putting a single cluster into a box of dimension 60 × 60 × 25 Å3. The same relaxed atomic coordinates as in the kinematical simulation were used for the Au309 clusters. In the beam direction, the sample is sliced into 17 layers. The sampling is 1,024 × 1,024 for the supercell and this configuration covers the scattering angle up to about 210 mrad. The frozen phonon method was used to include the thermal diffuse scattering with a constant Debye–Waller factor of 0.0057 nm2. The experimental probe parameters we used were C5 = 70 mm, C3 = -0.026 mm and C1 = 5 nm with convergence angle aperture of 24 mrad. The resulting probe profile has a full-width at half-maximum (FWHM) of about 0.08 nm. The scattering factor for gold was taken from ref. 16.

High-resolution HAADF-STEM images of Au309 clusters on a carbon film.

Typical images show various outline shapes, that is, cluster projections: pentagon (a), square (b) and hexagon (c). The intensity variation within the clusters clearly demonstrates atomic column resolution. Single atoms can be seen in the vicinity of the clusters, as indicated by the circle in c, and occasionally some distance away, as indicated by the circle in a. Resolution of the mass selector is ±2%.

Relationship between integrated HAADF intensity and size of gold clusters.

Integrated HAADF intensity of size-selected Au clusters on amorphous carbon film plotted as a function of the number of atoms they contain, showing a linear relationship. The line is drawn as a guide to the eye. Each data point is obtained from a statistical intensity distribution analysis over a large number of clusters with a given number of atoms. The standard deviation is used for estimating the error bars. An example of such a distribution for Au309 is shown in the inset.

Three-dimensional atomic structure of a gold cluster (N = 309 ± 6).

a, Three-dimensional atom density profile of Au309, derived from Fig. 1a. A hard-sphere model for an Ino-decahedral structure is shown with the electron beam (arrow) parallel to the five-fold axis. b, Experimental intensity line profile taken from the central atom column of the cluster to one of the corners (indicated in inset with red line. c, Simulated HAADF-STEM image (inset), obtained with a simple kinematical approach, of an Au309 cluster with Ino-decahedral geometry. An intensity profile (solid curve) across one ridge (indicated in inset with red line) is compared with the result from a full dynamical multislice calculation (dashed line).

We thank Y. Chen for assistance with the electron microscopy work in Birmingham. We gratefully acknowledge the UK Engineering and Physical Science Research Council (EPSRC) and the EU for their financial support of the cluster work. The EPSRC funded the UK SuperSTEM facility at Daresbury Laboratory. N.P.Y. and B.C.C. acknowledge the EPSRC and the University of Birmingham for PhD funding, respectively. The work at Tsinghua University was supported by the Ministry of Science and Technology and Ministry of Education in China.

Koga, K., Ikeshoji, T. & Sugawara, K. Size- and temperature-dependent structural transitions in gold nanoparticles. Phys. Rev. Lett. 92, 115504115507 (2004) Midgley, P. A. & Weyland, M. 3D electron microscopy in the physical sciences: the development of Z-contrast and EFTEM tomography. Ultramicroscopy 96, 413431 (2003) Arslan, I., Yates, T. J. V., Browning, N. D. & Midgley, P. A. Embedded nanostructures revealed in three dimensions. Science 309, 21952198 (2005) Iijima, S. & Ichihashi, T. Structural instability of ultrafine particles of metals. Phys. Rev. Lett. 56, 616619 (1986) Ajayan, P. M. & Marks, L. D. Experimental evidence for quasimelting in small particles. Phys. Rev. Lett. 63, 279282 (1989) Marks, L. D. Experimental studies of small particles structures. Rep. Prog. Phys. 57, 603649 (1994) Batson, P. E., Dellby, N. & Krivanek, O. L. Sub-angstrom resolution using aberration corrected electron optics. Nature 418, 617620 (2002) Haruta, M. Size- and support-dependency in the catalysis of gold. Catal. Today 36, 153166 (1997) Sanchez, A. et al. When gold is not noble: nanoscale gold catalysts. J. Phys. Chem. A 103, 95739578 (1999) Hughes, M. D. et al. Tunable gold catalysts for selective hydrocarbon oxidation under mild conditions. Nature 437, 11321135 (2005) Tsunoyama, H., Sakurai, H., Negishi, Y. & Tsukuda, T. Size-specific catalytic activity of polymer-stabilized gold nanoclusters for aerobic alcohol oxidation in water. J. Am. Chem. Soc. 127, 93749375 (2005) Chen, M. S. & Goodman, D. W. Structure-activity relationships in supported Au catalysts. Catal. Today 111, 2233 (2006) Li, Z. Y., Yuan, J., Chen, Y., Palmer, R. E. & Wilcoxon, J. P. Direct imaging of core-shell structure in silver-gold bimetallic nanoparticles. Appl. Phys. Lett. 87, 243103 (2005) Li, Z. Y., Yuan, J., Chen, Y., Palmer, R. E. & Wilcoxon, J. P. Local three-dimensional visualization of nanoparticle assemblies. Adv. Mater. 17, 28852888 (2005) Krivanek, O. L., Delby, N. & Lupini, A. R. Towards sub-angstrom electron beams. Ultramicroscopy 78, 111 (1999) Kirkland, E. J. Advanced Computing in Electron Microscopy (Plenum Press, New York/London, 1998) Smith, D. J., Petford-Long, A. K., Wallenberg, L. R. & Bovin, J.-O. Dynamic atomic-level rearrangements in small gold particles. Science 233, 872875 (1986) Bovin, J.-O., Wallenberg, R. & Smith, D. J. Imaging of atomic clouds outside the surfaces of gold crystals by electron microscopy. Nature 317, 4749 (1985) Schebarchov, D. & Hendy, S. C. Transition from icosahedral to decahedral structure in a coexisting solid-liquid nickel cluster. Phys. Rev. Lett. 95, 116101116104 (2005) Hansen, P. L. et al. Atom-resolved imaging of dynamic shape changes in supported copper nanocrystals. Science 295, 20532055 (2003) Pratontep, S., Carroll, S. J., Xirouchaki, C., Streun, M. & Palmer, R. E. Size-selected cluster beam source based on radio frequency magnetron plasma sputtering and gas condensation. Rev. Sci. Instrum. 76, 045103045109 (2005) von Issendorff, B. & Palmer, R. E. A new high transmission infinite range mass selector for cluster and nanoparticle beams. Rev. Sci. Instrum. 70, 44974501 (1999) Palmer, R. E., Pratontep, S. & Boyen, H.-G. Nanostructured surfaces from size-selected clusters. Nature Mater. 2, 443448 (2003) Bleloch, A. L. & Lupini, A. Imaging at the picoscale. Mater. Today 7, 4248 (2004) Jiang, J., Yuan, J. & Bleloch, A. L. Cluster-scale composition determination in a boron-rich compound. Appl. Phys. Lett. 91, 113107 (2007) Cleri, F. & Rosato, V. Tight-binding potentials for transition metals and alloys. Phys. Rev. B 48, 2233 (1993) Johnston, R. L. Evolving better nanoparticles: Genetic algorithms for optimising cluster geometries. Dalton Trans. 41934207 (2003)
doi: 10.1038/nature06470

Net carbon dioxide losses of northern ecosystems in response to autumn warming p.49

version="1.0" encoding="utf-8"?> ]>
Nature 451 7174 20080103 49524 0028-0836 1476-4687 2007Nature Publishing Group Supplementary Information

The file contains Supplementary Notes, Supplementary Tables S1-S4 and a Supplementary Figure S1 with Legend.

Net carbon dioxide losses of northern ecosystems in response to autumn warming ShilongPiaoS PhilippeCiaisP PierreFriedlingsteinP PhilippePeylinP MarkusReichsteinM SebastiaanLuyssaertS HankMargolisH JingyunFangJ AlanBarrA AnpingChenA AchimGrelleA David Y.HollingerD Y TuomasLaurilaT AndersLindrothA Andrew D.RichardsonA D TimoVesalaT LSCE, UMR CEA-CNRS, Bâtiment 709, CE, L’Orme des Merisiers, F-91191 Gif-sur-Yvette, France Laboratoire de Biogéochimie Isotopique, LBI, Bâtiment EGER, F-78026 Thiverval-Grignon, France Max Planck Institute for Biogeochemistry, PO Box 100164, 07701 Jena, Germany Department of Biology, University of Antwerp, Universiteitsplein 1, 2610 Wilrijk, Belgium Faculté de foresterie et de géomatique, Université Laval, Sainte-Foy, Quebec G1K 7P4, Canada Department of Ecology, Peking University, Beijing 100871, China Climate Research Division, Environment Canada, 11 Innovation Boulevard, Saskatoon, Saskatchewan S7N 3H5, Canada Department of Ecology and Evolutionary Biology, Princeton University, Princeton, New Jersey 08544, USA Department of Ecology, Swedish University of Agricultural Sciences, SE-750 07 Uppsala, Sweden USDA Forest Service Northern Research Station, 271 Mast Road, Durham, New Hampshire 03824, USA Finnish Meteorological Institute, FIN-00101 Helsinki, Finland Department of Physical Geography and Ecosystems Analysis, Lund University, SE-22362 Lund, Sweden Complex Systems Research Center, University of New Hampshire, Durham, New Hampshire 03824, USA Department of Physical Sciences, University of Helsinki, PO Box 64, FIN-00014 Helsinki, Finland Correspondence and requests for materials should be addressed to S.L.P. (slpiao@lsce.ipsl.fr) or P.C. (philippe.ciais@lsce.ipsl.fr). &e080103-11;

The carbon balance of terrestrial ecosystems is particularly sensitive to climatic changes in autumn and spring, with spring and autumn temperatures over northern latitudes having risen by about 1.1 °C and 0.8 °C, respectively, over the past two decades. A simultaneous greening trend has also been observed, characterized by a longer growing season and greater photosynthetic activity. These observations have led to speculation that spring and autumn warming could enhance carbon sequestration and extend the period of net carbon uptake in the future. Here we analyse interannual variations in atmospheric carbon dioxide concentration data and ecosystem carbon dioxide fluxes. We find that atmospheric records from the past 20 years show a trend towards an earlier autumn-to-winter carbon dioxide build-up, suggesting a shorter net carbon uptake period. This trend cannot be explained by changes in atmospheric transport alone and, together with the ecosystem flux data, suggest increasing carbon losses in autumn. We use a process-based terrestrial biosphere model and satellite vegetation greenness index observations to investigate further the observed seasonal response of northern ecosystems to autumnal warming. We find that both photosynthesis and respiration increase during autumn warming, but the increase in respiration is greater. In contrast, warming increases photosynthesis more than respiration in spring. Our simulations and observations indicate that northern terrestrial ecosystems may currently lose carbon dioxide in response to autumn warming, with a sensitivity of about 0.2 PgC °C-1, offsetting 90% of the increased carbon dioxide uptake during spring. If future autumn warming occurs at a faster rate than in spring, the ability of northern ecosystems to sequester carbon may be diminished earlier than previously suggested.

The carbon balance of terrestrial ecosystems is highly sensitive to climate changes at the edges of the growing season. In response to warmer springs, for example, several field studies have shown that boreal forests absorb more carbon as a result of an earlier beginning of the growing season. A strong autumn warming is currently occurring in eastern Asia and eastern North America. However, little attention has been given to the impacts of this forcing on the terrestrial carbon cycle. We have analysed how interannual variations and trends in autumn temperatures have recently affected atmospheric CO2 concentrations, ecosystem CO2 fluxes measured by eddy covariance, and remotely sensed vegetation greenness values. A process-oriented terrestrial biosphere model (ORCHIDEE) is combined with an atmospheric transport model (LMDZt) to quantify the processes through which autumn warming controls the carbon balance of ecosystems (see Methods).

The seasonal cycle of atmospheric CO2 concentrations provides an integrated measure of the net land–atmosphere carbon exchange (net ecosystem productivity; NEP) and its temporal characteristics. We analysed the ten atmospheric CO2 measurement records from the NOAA–ESRL air-sampling network, which cover at least 15 years of data in the Northern Hemisphere (Fig. 1 and Supplementary Table 1). The upward zero-crossing date of CO2 was determined as the day when the de-trended atmospheric CO2 seasonal cycle crosses the zero line from negative to positive values (see Methods). This date occurs in autumn at northern high-latitude stations and in early winter at northern tropical stations (Supplementary Table 1). We found that variations in the CO2 zero-crossing date are negatively correlated with anomalies in autumn air temperatures over a broad region surrounding each station by ±20° of latitude. All CO2 records show a negative correlation, with four out of ten sites having statistically significant correlations (Supplementary Table 2). The probability that this occurs purely by chance is estimated to be about 10-5 if all station records are assumed to be independent (see Supplementary Information). The striking anticorrelation between autumnal temperature and CO2 zero-crossing date is illustrated in Fig. 1a for the 23-year-long atmospheric measurement record of Point Barrow in northern Alaska (R = -0.61, P = 0.002). In contrast with the widespread influence of temperature, the upward CO2 zero-crossing date shows no significant correlation with precipitation anomalies (Supplementary Table 2). If soil moisture calculated by the ORCHIDEE model (see Methods) is used instead of precipitation as a predictor of CO2 upward zero-crossing dates, then only six of the ten sites show a positive correlation, and only three of the ten sites show a higher correlation with soil moisture than with temperature. Similar results are also inferred from a partial correlation analysis in which the controlling effects of other variables on temperature were removed (Supplementary Table 2).

We verified that the strong negative correlation between upward CO2 zero-crossing date and temperature predominantly reflects climate-driven fluctuations in NEP, rather than interannual fluctuations in atmospheric transport. To do so, we prescribed either variable NEP or climatological NEP fluxes from ORCHIDEE to the global transport model LMDZt driven by variable wind fields (see Methods). With the exception of the Mt Cimone (CMN) and Cape Kumukahi (KUM) stations, we found that the fluctuations in upward zero-crossing dates are driven mainly by changes in NEP, and only partly by interannual wind changes (see Methods, Supplementary Table 3 and Supplementary Fig. 1). We also verified that accounting for increasing ocean uptake and fossil fuel emissions in the LMDZt transport model did not significantly affect the zero-crossing dates because these two fluxes contribute less than 4% of the variation for all sites (except for station KUM). The possible changes in seasonal fossil fuel emissions over time may only marginally impact the upward CO2 zero-crossing date changes (see Methods).

There is also a long-term trend in the autumn upward zero-crossing date of atmospheric CO2 superimposed on interannual fluctuations. At Point Barrow, for instance, we determined a systematic advance of -0.40 day yr-1 (Fig. 1b), which was not primarily caused by changes in atmospheric transport, because the trends in zero-crossing date simulated with climatological ORCHIDEE fluxes and interannual transport are only about -0.12 day yr-1. Overall, eight of ten sites show an earlier trend in upward zero-crossing date, with four sites being statistically significant (Fig. 1b). This trend towards earlier or increased ecosystem losses of CO2 in autumn becomes apparent when analysing CO2 data from the past decade, whereas it was non-existent in the CO2 data from 1970 to 1994 (ref. 18) as a result of the time-frame of their analysis. This trend towards larger autumn CO2 losses is not a legacy from drier summers, because atmospheric CO2 data show that weaker summer CO2 minima are not significantly associated with an advanced upward zero-crossing date at all sites. The advance in autumnal atmospheric CO2 zero-crossing date clearly exceeds that of the spring zero-crossing date (Supplementary Table 3). Thus, the duration of the net carbon uptake period (CUP), defined as the difference between autumn upward and spring downward CO2 zero-crossing dates, has on average decreased at nearly all Northern Hemisphere atmospheric CO2 stations (Fig. 1b).

Next, we analysed 108 site-years of eddy-covariance CO2 measurement data from 24 northern ecosystem sites (Supplementary Table 4) to quantify the response of the CUP ending date to interannual variations in autumn temperature (see Methods). All sites combined show that the CUP terminates systematically earlier when autumn conditions are warmer, and vice versa (Fig. 2). Further, stronger temperature anomalies seem to have stronger effects on ecosystem carbon balance than weak anomalies (P < 0.05). Hence, despite a large scatter in the individual yearly eddy-covariance CUP dates (see insets to Fig. 2), these micrometeorological observations corroborate the atmospheric concentration records.

The large-scale atmospheric concentration records, taken together with the ecosystem-scale eddy-covariance flux measurements (about 1 km2) suggest that warmer temperatures in autumn increase ecosystem CO2 losses by shortening the net CUP. This finding stands in apparent contradiction of the autumn ‘greening’ and longer-lasting vegetation activity detected at mid-to-high northern latitudes by remote sensing and by numerous in situ phenological indicators. However, the underlying mechanisms and processes are yet to be explained. NEP results from the balance between gross primary photosynthesis (GPP) and total ecosystem respiration (TER), necessitating separate investigations into the response of each gross flux to temperature changes. We provide some indication of possible controlling mechanisms by using the ORCHIDEE terrestrial biosphere simulation model forced by variable climate fields over the period 1980–2002 (see Methods). The model’s ability to capture the timing of the CUP and the length of the growing season successfully was verified by using the following: first, eddy-covariance CO2 flux measurements, second, satellite-derived observations of global leaf area index, and third, interannual and seasonal variations in atmospheric CO2 (see Methods and Supplementary Fig. 1). Results from these studies suggest that it is possible to use this model tool to help in disentangling the response of photosynthesis, respiration and NEP to climate variability.

Simulated September to November NEP shows a trend towards increasing carbon losses in the Northern Hemisphere (north of 25° N) at a rate of 13 Tg C yr-1 (P = 0.01) during 1980–2002. In the ORCHIDEE model long-term simulation, the increasing autumn source of carbon to the atmosphere offsets about 90% of the increasing carbon sink in spring. This result is consistent with the atmospheric concentration analysis (Supplementary Table 3). We attribute the trend in net carbon loss during autumn to increases in TER (21 Tg C yr-1) dominating over increasing GPP (8 Tg C yr-1 owing to delayed leaf senescence). In autumn, both modelled GPP and TER increase with increasing temperature, but the temperature sensitivity of TER (5.0 g C m-2 °C-1) exceeds that of GPP (2.5 g C m-2 °C-1). This is due to limitations of radiation and temperature on GPP during the autumn, and to soil desiccation carried over from the summer dry period. As a result, autumn NEP is simulated to be an increasing source of CO2 in response to warming, with a mean sensitivity of -2.5 g C m-2 °C-1 (or about -0.2 Pg C °C-1 north of 25° N), which is comparable to that derived from eddy-covariance measurements (-3.2 g C m-2 °C-1; Fig. 2).

Our results suggest that net carbon uptake of northern ecosystems is being decreased in response to autumnal warming. The spatial distribution of the response of carbon flux to temperature, as projected by the ORCHIDEE model, is shown in Fig. 3. Warmer autumns coincide with greater than normal GPP (Fig. 3a). However, because of a concurrent stimulation of plant respiration, the geographical area where autumn NPP increases with temperature (slope > 5 g C m-2 °C-1) is much less extensive than the area where GPP increases (Fig. 3b). The spatial pattern of the autumn increase in NPP in response to warming is remarkably similar to that of the NOAA/AVHRR vegetation index (NDVI) data (Fig. 3d), suggesting that results from the ORCHIDEE model for NPP are realistic. However, this ‘extra’ autumn NPP is being accompanied by even more respiration in response to warming, so that the modelled NEP response shows systematic anomalous carbon losses during warmer autumns, in particular over North America and Europe (Fig. 3c).

Observed historical climate data reveal that Eurasia experienced a stronger warming in spring (0.06 °C yr-1, P = 0.001) than in autumn (0.02 °C yr-1, P = 0.15) over the past two decades. In contrast, North America has experienced a larger warming in autumn (0.05 °C yr-1, P = 0.03) than in spring (0.02 °C yr-1, P = 0.36). In addition, a more significant and coherent greening pattern in Eurasia than in North America has been detected in the remote sensing data. This suggests that the processes and the magnitude of seasonal changes in NEP in Eurasia and North America are different, which may control the annual carbon balance of their ecosystems. Further constraints on the spatial and temporal patterns of large-scale ecosystem fluxes will be delivered in the future from atmospheric inversions constrained with longer-term ecosystem flux data.

Applying the future Northern Hemisphere warming of 3.8–6.6 °C predicted by a climate model to the sensitivity of the autumn zero-crossing date of atmospheric CO2 at Point Barrow (about 5 days °C-1) gives a projected advance of 19–33 days by the end of the twenty-first century. Previous model assessments of the response of land ecosystems to climate change concluded that terrestrial carbon sinks should peak by about the year 2050 and then diminish towards the end of the twenty-first century. The asymmetrical impact of autumn versus spring warming on ecosystem carbon exchange contributes significant uncertainty to future projections. If warming in autumn occurs at a faster rate than in spring, the ability of northern ecosystems to sequester carbon may diminish in the future. Acquiring a greater understanding of responses of terrestrial ecosystems to climate trends at the edges of the growing season, including potential acclimation processes, is clearly a priority, and should come from controlled ecosystem experiments and long-term eddy-covariance data sets.

Methods Summary

We analysed the effects of autumn temperature on the carbon balance of northern ecosystems at different scales, using three different methods.

First, we used smoothed flask CO2 data from the NOAA/ESRL network to characterize changes in the seasonal CO2 zero-crossing dates for ten stations over the Northern Hemisphere (Fig. 1). We correlated each zero-crossing date with the corresponding observed temperature or precipitation in spring (March to May) and autumn (September to November), and with the ORCHIDEE-modelled soil moisture content. The trends in CO2 zero-crossing dates and their correlation with climate factors were computed by using linear least-squares regression. The significance of statistical analyses in this study were assessed on the basis of two-tailed significance tests. To isolate further the contribution of fluxes and transport to the year-to-year atmospheric CO2 signal, we performed factorial simulation experiments in which NEP from the ORCHIDEE vegetation model forced by varying climate fields provided surface boundary conditions for simulated CO2 in the atmospheric transport model LMDZt driven by interannual winds.

Second, we analysed the net CO2 flux data measured by the eddy-covariance technique from 24 northern ecosystem sites (Supplementary Table 4). The end of the CUP is defined as the last day in a year when the NEP five-day running means exceed zero. Autumn is defined as the interval of ±30 days around the average CUP ending date at each site. We grouped the 108 year-site data into distinct 0.5 °C bins of autumn temperature anomaly. For each autumn temperature bin we calculated the median and mean anomaly of the ending date of the CUP.

Third, hints on the processes that control the integrated autumn NEP response to temperature, through the individual sensitivity of photosynthesis and respiration, were provided by integrating the ORCHIDEE vegetation model forced by historic climate data during the period 1980–2002.

Atmospheric CO2 data

We used flask data from the NOAA/ESRL network to characterize trends in the CO2 zero-crossing dates (spring downward and autumn CO2 upward) that correspond roughly to the time of maximum NEP uptake in spring and maximum release in autumn. Following the approach described in ref. 27, we first removed the interannual trend in the atmospheric CO2 concentration for each site with a polynomial curve of degree 2, four harmonic seasonal function, and time-filtered residuals. We then used the harmonics plus the residuals (detrended CO2 seasonal cycle) to define the downward and upward CO2 zero-crossing dates as the day on which the detrended curve crossed the zero line from positive to negative and from negative to positive, respectively. We considered only northern stations for which at least 15 years of data were available during the period 1980–2002 (Supplementary Table 1).

The downward and upward CO2 zero-crossing dates were correlated with spring (March to May) and autumn (September to November) air temperature over a broad region surrounding each station by ±20° of latitude, respectively (Supplementary Table 1). Using a similar method, we also evaluated the correlation with precipitation and modelled change in soil moisture.

Eddy-covariance data

The eddy-covariance CO2 flux observations were performed in accordance with the routine procedures established by regional networks (for example, Fluxnet-Canada and CARBOEUROPE). Half-hourly data were quality-controlled, filtered against low turbulence by using friction velocity as a heuristic criterion and gap-filled by the method developed in ref. 29. Data were aggregated to daily flux integrals and were only used for the analyses if more than 80% of the half-hourly values were either direct measurements or gap-filled with high confidence. The end of the CUP was calculated as the last day in a year when NEP five-day running means exceeded zero; that is, when the ecosystems became a source of CO2 to the atmosphere. Autumn was defined as the period during ±30 days of the average ending date of CUP for each site. Because the flux records are not long enough to assess the long-term impact of autumn temperature trends on the ending date of CUP and net CO2 exchange for each site, we grouped the 108 year-site data into different 0.5 °C bins of autumn temperature anomaly. For each bin we calculated the median and average anomaly of the ending date of CUP. To ensure the reliability of the statistical analysis of median and average calculation, we used only the data in temperature classes with a sample size greater than 3.

Global vegetation model

The global vegetation model called ORCHIDEE (‘ORganizing Carbon and Hydrology In Dynamic Ecosystems’) was used to simulate the terrestrial biogeochemical processes. ORCHIDEE describes the turbulent surface fluxes of CO2, water and energy (transpiration, photosynthesis and respiration), the dynamics of water and carbon pools (soil moisture budget and allocation, growth, mortality, and soil carbon decomposition) and longer-term ecosystem dynamics (fire, sapling establishment and light competition). Fluxes were calculated each hour, and carbon pools were updated each day. Onset and senescence of foliage development depend on a critical leaf age, water and temperature stresses.

With the use of 1901 climate data and the 1860 atmospheric CO2 concentration of 286 p.p.m., a first model spin-up was performed to bring carbon pools to equilibrium. A second spin-up was performed with interannually variable climate data over 1901–1910 to define the initial condition of a run covering 1901–2002. The monthly climate data sets were supplied by the Climatic Research Unit, University of East Anglia, UK. These data were transformed to half-hourly weather variables by using a weather generator.

The modelled NEP over 1980–2002, prescribed in an atmospheric transport model (LMDzt), was found to faithfully reproduce the interannual variations in the spring drawdown date and autumn build-up date at high-latitude (north of 50° N) stations that are predominantly affected by the fluxes of the Northern Hemisphere (Supplementary Table 3 and Supplementary Fig. 1). Furthermore, the modelled upward zero crossing at high-latitude (north of 50° N) stations has advanced by an average of -0.19 ± 0.05 days yr-1, which is comparable to that estimated from atmospheric CO2 concentration data (-0.22 ± 0.23 days yr-1).

Atmospheric transport model

We used the three-dimensional eulerian transport model LMDzT derived from the general circulation model of the Laboratoire de Météorologie Dynamique, LMDz, to compute the daily CO2 concentration at each station driven by daily NEP variations from ORCHIDEE during 1980–2002. The model has a horizontal resolution of 3.75° × 2.5° and 19 vertical levels. The simulated winds are relaxed towards the analysed field of ECMWF (‘nudging’ mode) and therefore vary from year to year according to the observations. Advection, deep convection and turbulent mixing of tracer are calculated by following the schemes proposed in refs 31, 32 and 17, respectively.

To separate the effects of transport and terrestrial carbon fluxes on the zero-crossing date signal, we performed two simulations. The first one used interannual daily NEP fluxes calculated during the period 1980–2002 by ORCHIDEE. The second simulation (referred to as ‘transport only’) used climatological but daily variable NEP. The contribution of interannually varying fluxes to the variability in zero-crossing date is assessed by the difference in simulated atmospheric CO2 between the first and the second simulations, which is referred to as ‘flux only’. In addition, we computed the contribution to CO2 concentrations from air–sea exchange and fossil fuel emissions and their increase (annual increase per group of countries) by following estimates from refs 33 and 34, respectively. Because of the lack of information on seasonal variations of fossil fuel emissions, we tested the impact of possible changes in fossil seasonality using the method of ref. 35 to construct seasonally varying fossil fuel emission. We modelled the impact on atmospheric CO2 in the LMDZt transport model of using a modified fossil fuel source with seasonal amplitude of 40%, 20%, 10%, 5% and 0%. The results showed that the change in zero-crossing date is less than 1.3 days, even when the seasonal amplitude of fossil fuel emissions changed by 40%.

Atmospheric CO2 concentration data analysis from long-term records of the global NOAA-ERSL air-sampling network.

a, Interannual variability in anomaly of upward zero-crossing date (red) observed at Point Barrow, Alaska, and the corresponding autumn (September to November) temperature (black) over the region between 51° and 90° N over the past two decades. Upward zero-crossing date is strongly anti-correlated with autumn temperature (slope = -5.4 days °C-1; R = -0.61, P = 0.002). The vertical dotted line indicates the time of the eruption of Mount Pinatubo. b, Trends in upward zero-crossing date (red) and length of the net CUP (green) from long-term Northern Hemisphere atmospheric observations during at least the past 15 years (see Methods). The differences in the trends between autumn upward zero-crossing date and CUP reflects changes in the spring downward zero crossing. As a result of the earlier autumn upward zero-crossing date, CUP has persistently decreased by an average of 0.36 ± 0.38 days per year since 1980. The inset shows the distribution of the stations used in this study. Station abbreviations are defined in Supplementary Table 1.

Eddy-covariance flux data analysis from boreal sites in North America and Eurasia.

A total of 108 site-years have been aggregated in this figure. The average (blue) and median (green) anomaly of ending date of net CUP is shown for different autumn temperature anomalies binned into 0.5 °C intervals. The top horizontal axis labels correspond to the number of site-years and sites (in parenthesis) in each temperature bin. The bottom left inset shows the relationships between ending date of CUP and temperature anomalies. There is a marginally negative correlation between autumn CUP ending date and temperature anomalies (y = -1.7x - 0.0087, P = 0.07). If we exclude the four site-years with the most extreme cold anomalies (&Dgr;T < -2 °C), the negative correlation between CUP ending date and temperature becomes highly significant (P = 0.03) and the slope is steeper (y = -2.4x + 0.3007), suggesting that below a certain threshold of cold anomaly there is no further decrease in respiration. The top right inset shows the relationships between autumn NEP and temperature anomalies. A positive NEP value indicates an increased carbon uptake. Autumn was defined as the 60-day interval around the average CUP ending date for each site. Eddy-covariance data show increased carbon losses under warmer conditions, with a temperature sensitivity of NEP of -3.2 g C m-2 °C-1 (y = -3.17x - 5 × 10-6, P = 0.04).

A model view of the spatial distribution of the effects of autumn (September to November) temperature warming on gross and net carbon fluxes, obtained with the ORCHIDEE model.

a, ORCHIDEE model-derived autumn GPP. b, ORCHIDEE model-derived autumn NPP. c, ORCHIDEE model-derived autumn NEP. d, Sum of satellite-derived autumn normalized difference vegetation index (NDVI). The sensitivity is expressed as the linearly regressed slope of autumn carbon flux or of NDVI against autumn temperature for each pixel over the past two decades. A positive slope of NEP indicates that terrestrial carbon uptake is increasing with warmer temperatures, and vice versa. Areas with a low sensitivity or insignificant (P > 0.05) relationships between the variables are coloured in grey.

We thank all the people and their respective funding agencies who worked to provide data for this study, and specifically B. Amiro, M. A. Arain, T. A. Black, C. Bourque, L. Flanagan, J. H. McCaughey and S. Wofsy for providing some of the flux data from the Canadian sites, and M.-A. Giasson and C. Coursolle for their help in compiling the data. We also thank A. Friend, P. Rayner and N. Viovy for helpful comments and discussions. This study was supported by European Community-funded projects ENSEMBLES and CARBOEUROPE IP, and by the National Natural Science Foundation of China as well as by Fluxnet-Canada, which was supported by CFCAS, NSERC, BIOCAP, MSC and NRCan. The computer time was provided by CEA. We thank the NOAA-ERSL global air sampling program for collecting and analysing the long-term CO2 flask data, and K. Masarie at NOAA-ERSL for generating each year the GLOBALVIEW-CO2 collaborative data product, which formed the basis of our atmospheric data analysis. The ongoing exchange of ideas, data and model results in the international research community on the carbon cycle is facilitated by the Global Carbon Project.

Author Contributions P.C., S.P., P.F. and P.P. designed the research. S.P., P.C. and P.F. performed ORCHIDEE modelling analysis. P.P. and S.P. performed transport analysis. S.P., S.L., M.R., H.M. and P.C. performed eddy-covariance data analysis. S.P., P.C. and J.F. performed satellite data analysis. All authors contributed to the interpretation and writing.

Goulden, M. L. et al. Sensitivity of boreal forest carbon balance to soil thaw. Science 279, 214217 (1998) Randerson, J. T., Field, C. B., Fung, I. Y. & Tans, P. P. Increases in early season ecosystem uptake explain recent changes in the seasonal cycle of atmospheric CO2 at high northern latitudes. Geophys. Res. Lett. 26, 27652768 (1999) Morgenstern, K. et al. Sensitivity and uncertainty of the carbon balance of a Pacific Northwest Douglas-fir forest during an El Niño La Niña cycle. Agric. For. Meteorol. 123, 201219 (2004) Bergeron, O. et al. Comparison of carbon dioxide fluxes over three boreal black spruce forests in Canada. Glob. Change Biol. 13, 89107 (2007) Mitchell, T. D. & Jones, P. D. An improved method of constructing a database of monthly climate observations and associated high-resolution grids. Int. J. Climatol. 25, 693712 (2005) Myneni, R. B., Keeling, C. D., Tucker, C. J., Asrar, G. & Nemani, R. R. Increased plant growth in the northern high latitudes from 1981 to 1991. Nature 386, 698702 (1997) Zhou, L. M. et al. Variations in northern vegetation activity inferred from satellite data of vegetation index during 1981 to 1999. J. Geophys. Res. 106, 2006920083 (2001) Churkina, G., Schimel, D., Braswell, B. H. & Xiao, X. M. Spatial analysis of growing season length control over net ecosystem exchange. Glob. Change Biol. 11, 17771787 (2005) Cramer, W. et al. Global response of terrestrial ecosystem structure and function to CO2 and climate change: results from six dynamic global vegetation models. Glob. Change Biol. 7, 357373 (2001) Sitch, S. et al. Evaluation of the terrestrial carbon cycle, future plant geography and climate–carbon cycle feedbacks using 5 Dynamic Global Vegetation Models (DGVMs). Glob. Change Biol. (in the press) Black, T. A. et al. Increased carbon sequestration by a boreal deciduous forest in years with a warm spring. Geophys. Res. Lett. 27, 12711274 (2000) Tanja, S. et al. Air temperature triggers the recovery of evergreen boreal forest photosynthesis in spring. Glob. Change Biol. 9, 14101426 (2003) Menzel, A. et al. European phenological response to climate change matches the warming pattern. Glob. Change Biol. 12, 19691976 (2006) Schwartz, M. D., Ahas, R. & Aasa, A. Onset of spring starting earlier across the Northern Hemisphere. Glob. Change Biol. 12, 343351 (2006) IPCC. Climate Change 2007: The Physical Sciences Basis: Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change (Cambridge Univ. Press, Cambridge, 2007) Krinner, G. et al. A dynamic global vegetation model for studies of the coupled atmosphere–biosphere system. Glob. Biogeochem. Cycles 19 doi: 10.1029/2003GB002199 (2005) Hourdin, F. & Armengaud, A. The use of finite-volume methods for atmospheric advection of trace species. Part I: Test of various formulations in a general circulation model. Mon. Weath. Rev. 127, 822837 (1999) Keeling, C. D., Chin, J. F. S. & Whorf, T. P. Increased activity of northern vegetation inferred from atmospheric CO2 measurements. Nature 382, 146149 (1996) Heimann, M. et al. Evaluation of terrestrial carbon cycle models through simulations of the seasonal cycle of atmospheric CO2: First results of a model intercomparison study. Glob. Biogeochem. Cycles 12, 124 (1998) GLOBALVIEW-CO2. Cooperative Atmospheric Data Integration Project Carbon Dioxide, 〈ftp://ftp.cmdl.noaa.gov/ccg/co2/GLOBALVIEW〉 (2004) Angert, A. et al. Drier summers cancel out the CO2 uptake enhancement induced by warmer springs. Proc. Natl Acad. Sci. USA 102, 1082310827 (2005) Linderholm, H. W. Growing season changes in the last century. Agric. For. Meteorol. 137, 114 (2006) Ciais, P. et al. Europe-wide reduction in primary productivity caused by the heat and drought in 2003. Nature 437, 529533 (2005) Piao, S. L., Friedlingstein, P., Ciais, P., Zhou, L. M. & Chen, A. P. Effect of climate and CO2 changes on the greening of the Northern Hemisphere over the past two decades. Geophys. Res. Lett. 33 doi: 10.1029/2006GL028205 (2006) Myneni, R. B. & Song, X. NDVI Version 3 based on Pathfinder, 〈ftp://primavera.bu.edu/pub/datasets/AVHRR_DATASETS/PATHFINDER/VERSION3_DATA/〉 (2003) Marti, O. et al. The New IPSL Climate System Model: IPSL-CM4 (Institut Pierre Simon Laplace, Paris, 2006) Thoning, K. W., Tans, P. P. & Komhyr, W. D. Atmospheric carbon-dioxide at Mauna Loa Observatory. 2. Analysis of the NOAA GMCC data, 1974–1985. J. Geophys. Res. 94, 85498565 (1989) Aubinet, M. et al. Estimates of the annual net carbon and water exchange of forest: the EUROFLUX methodology. Adv. Ecol. Res. 30, 113175 (2000) Papale, D. et al. Towards a standardized processing of Net Ecosystem Exchange measured with eddy covariance technique: algorithms and uncertainty estimation. Biogeosciences 3, 571583 (2006) Richardson, C. W. & Wright, D. A. A Model for Generating Daily Weather Variables, Technical Report (US Department of Agriculture, Agricultural Research Service, Washington DC, 1984) van Leer, B. Towards the ultimate conservative difference scheme: IV. A new approach to numerical convection. J. Comput. Phys. 23, 276299 (1977) Tiedtke, M. A comprehensive mass flux scheme for cumulus parameterization in large-scale models. Mon. Weath. Rev. 117, 17791800 (1989) Takahashi, T. et al. in Extended Abstracts of the 2nd International CO in the Oceans Symposium 18–01 (Tsukuba, Japan, 1999) Marland, G., Boden, T. A. & Andres, R. J. Global, Regional, and National CO2 Emissions. Trends: A Compendium of Data on Global Change.. 〈http://cdiac.esd.ornl.gov/trends/emis/tre_glob.htm〉 (2007) Gurney, K. R. et al. Sensitivity of atmospheric CO2 inversions to seasonal and interannual variations in fossil fuel emissions. J. Geophys. Res. 110 doi: 10.1029/2004JD005373 (2005)
doi: 10.1038/nature06444

Vertical structure of recent Arctic warming p.53

version="1.0" encoding="utf-8"?> ]>
Nature 451 7174 20080103 53564 0028-0836 1476-4687 2007Nature Publishing Group Supplementary Information

The file contains Supplementary Discussion, Supplementary Figures 1-3 with Legends and additional references.

Vertical structure of recent Arctic warming Rune G.GraversenR G ThorstenMauritsenT MichaelTjernströmM ErlandKällénE GunillaSvenssonG Department of Meteorology, Stockholm University, S-106 91 Stockholm, Sweden Correspondence and requests for materials should be addressed to R.G.G. (rune@misu.su.se). &e080103-6; &nature06502-s1;

Near-surface warming in the Arctic has been almost twice as large as the global average over recent decades—a phenomenon that is known as the ‘Arctic amplification’. The underlying causes of this temperature amplification remain uncertain. The reduction in snow and ice cover that has occurred over recent decades may have played a role. Climate model experiments indicate that when global temperature rises, Arctic snow and ice cover retreats, causing excessive polar warming. Reduction of the snow and ice cover causes albedo changes, and increased refreezing of sea ice during the cold season and decreases in sea-ice thickness both increase heat flux from the ocean to the atmosphere. Changes in oceanic and atmospheric circulation, as well as cloud cover, have also been proposed to cause Arctic temperature amplification. Here we examine the vertical structure of temperature change in the Arctic during the late twentieth century using reanalysis data. We find evidence for temperature amplification well above the surface. Snow and ice feedbacks cannot be the main cause of the warming aloft during the greater part of the year, because these feedbacks are expected to primarily affect temperatures in the lowermost part of the atmosphere, resulting in a pattern of warming that we only observe in spring. A significant proportion of the observed temperature amplification must therefore be explained by mechanisms that induce warming above the lowermost part of the atmosphere. We regress the Arctic temperature field on the atmospheric energy transport into the Arctic and find that, in the summer half-year, a significant proportion of the vertical structure of warming can be explained by changes in this variable. We conclude that changes in atmospheric heat transport may be an important cause of the recent Arctic temperature amplification.

The recent warming of the Earth’s surface is most probably due to an increase of atmospheric greenhouse-gas concentrations. Although most greenhouse gases are fairly uniformly distributed around the globe, the temperature response to greenhouse-gas forcing is thought to be larger in polar than equatorial regions. The response depends on various feedbacks within the climate system. In addition to snow and ice processes, the strength of the atmospheric stratification constitutes such a feedback. The troposphere is more stably stratified in the polar regions than closer to the Equator. An increase in downwelling long-wave radiation at the surface (for example, due to an altered atmospheric CO2 level) causes warming, which at high latitudes is confined to the lower troposphere. In the tropics, in contrast, the warming is distributed vertically by deep convection. It has also been proposed that the increase of polluting materials (such as black carbon) on Arctic ice and snow have caused albedo changes and added to the Arctic warming. Common to all these processes is that they are expected to induce the largest warming in the lowermost part of the atmosphere.

The Arctic amplification can also be caused by other processes. Idealized experiments with models that have no surface-albedo feedback also reveal a polar-temperature-amplification response to a doubling of CO2 concentration. It is found that the excessive Arctic warming is due to an increase of the atmospheric northward transport of heat and moisture. These results are supported by observational studies, which suggest that changes of the heat transport have added to the recent Arctic surface warming.

The linkage between Arctic warming and changes of atmospheric circulation has been investigated by studying various Northern Hemisphere circulation indices, such as that associated with the Arctic Oscillation. Generally, different phases of these indices are associated with linear temperature responses characterized by east–west heat redistribution between the mid-latitude ocean and land, whereas the high latitudes are less affected. However, in the winter season, high phases of the circulation indices are associated with a warmer Arctic. This warming is particularly pronounced over the northern rims of the continents. From the 1970s through to the mid-1990s, the indices were in their high phases, while since then, they have relaxed towards neutral values. The Arctic warming, on the other hand, has shown a persistently positive trend over the past 30 years. It is therefore difficult to associate changes in these indices, such as the Arctic Oscillation index, with the recent Arctic warming trend.

The vertical structure of the Arctic warming during the 1980s and 1990s, based on the ERA-40 reanalysis (see Methods), exhibits trends throughout large parts of the troposphere that are comparable in magnitude to those at the surface (Fig. 1). In fact, the Arctic warming in the reanalysis data shows clear maxima well above the surface in winter and in summer, and the trends are almost equal at heights below the 400 hPa level in the atmosphere during autumn. This vertical structure is not consistent with the hypothesis that retreating snow and ice cover is the main cause of the amplification. Retreating snow and ice are associated with energy input at the surface, which—along with the stable stratification conditions often prevailing in the Arctic—means that this process would be expected to induce the largest temperature response in the lowermost part of the troposphere. But we only observe this vertical structure of warming in spring. It is worth noting that this is when the trends above the boundary layer are of comparable magnitude to those at the surface. We note that the lack of amplification near the surface in summer is consistent with expectations because surface air temperatures over the Arctic Ocean are constrained to be close to the freezing point owing to the melting of sea ice, but that the amplification aloft cannot be explained by surface feedbacks.

It is also notable that during the same period observations solely from Arctic land stations reveal an amplification of the temperature trend during the dark months, November–February (Fig. 2). This amplification cannot be explained by snow-cover changes, as the albedo effect is practically absent during this dark period. Moreover, the heat flux from the ground is very small. This is contrary to conditions in the ice-free parts of the Arctic oceans during winter, where convection ensures that cold water at the surface is replaced by warmer water from below; this process maintains a large vertical temperature gradient between the ocean surface and the cold atmosphere. In addition, reduction of sea-ice cover during summer results in increased sea-ice formation during autumn and winter, such that latent heat is stored during the warm season and released into the atmosphere during the subsequent months.

So what are the mechanisms giving rise to the vertical structure of the Arctic warming? Changes in the advection of atmospheric energy into the Arctic region might imply Arctic warming with a maximum not necessarily located at the surface. Mid-tropospheric temperatures in the Arctic are sensitive to advection of energy across the Arctic boundary: this is evident from linear regressions of the Arctic 500 hPa temperature field on the atmospheric northward energy transport (ANET) across 60° N (Fig. 3). Positive (negative) anomalies of the ANET at 60° N are followed by positive (negative) temperature anomalies over the Arctic area, where the anomalies of the temperatures lag those of the ANET by about 5 days (Fig. 3a). Positive regressions are found for positive time lag north of 60° N and for negative time lag south of 60° N, whereas the opposite distribution is found for negative regressions. In a statistical sense, this indicates that large energy transport follows conditions where a larger-than-usual north–south temperature gradient at 60° N has prevailed. This transport, in turn, is succeeded by warming of areas north of 60° N and cooling south of this latitude—a signature of energy convergence and divergence north and south of 60° N, respectively. The warming evaluated at five-day lag (Fig. 3b) is distributed over the major part of the Arctic area, whereas the cooling in the mid-latitudes is found over large parts of the continents. This linkage between the ANET across 60° N and the temperature field is found through the entire vertical extent of the troposphere (not shown), and is similar to that found from regressions of the surface temperatures on the ANET.

The ANET across 60° N has increased during recent decades, except in January and February. For the summer half-year, April through to October, the ANET can explain a substantial part of the Arctic temperature trends (Fig. 4). The part of the temperature trends that can be linked to the ANET (Fig. 4b) shows roughly the same vertical distribution as the total temperature trends (Fig. 4a), with a maximum at around 700 hPa. At 60° N, the ANET is mainly accomplished by atmospheric waves, such as Rossby waves and cyclone systems. Hence, the ANET at mid-latitudes can be viewed as an index of atmospheric circulation patterns. This index appears to be an efficient indicator of a linkage between circulation changes and Arctic temperature trends.

Other processes that might be important contributors to the warming above the surface include changes in cloud cover and the atmospheric water vapour content. In the Arctic, except possibly for a short summer period, persistent low clouds are believed to induce surface warming. Often the greenhouse effect of clouds dominates over the albedo effect, as the clouds cover an already highly reflecting surface. By absorbing radiation, clouds may furthermore warm the atmosphere at the height where they are present. As a result, it is possible that an increase in cloud cover at a given atmospheric height may cause warming there. Observations from satellites indeed suggest an increase of Arctic cloud fraction in summer during the 1980s and 1990s. Warming of the Arctic middle troposphere might also partly be an effect of changes in the atmospheric radiative properties. These changes could be associated with the above-mentioned increase in advection of energy, which is basically a transport of warm and/or moist air into the Arctic. The advection in itself accounts for a considerable part of the maximum warming at 700 hPa, but additional warming at this height would occur if the advected air is more humid than the ambient air and hence absorbs long-wave radiation more efficiently; water vapour is an efficient greenhouse gas.

Our results do not imply that studies based on models forced by anticipated future CO2 levels are misleading when they point to the importance of the snow and ice feedbacks. It is likely that a further substantial reduction of the summer ice-cover would strengthen these feedbacks and they could become the dominant mechanism underlying a future Arctic temperature amplification. Much of the present warming, however, appears to be linked to other processes, such as atmospheric energy transports.

Methods Summary

The ERA-40 reanalysis data are used for Figs 1, 3 and 4. A discussion of the data quality and a comparison with two other reanalysis data sets are given in the Supplementary Discussion. The ANET at a given latitude is defined as the total energy flux across this particular latitude. Hence, this quantity constitutes one time series. Using daily data, the temperature field has been regressed on the ANET at 60° N for different time lags of the temperature field relative to the ANET (Fig. 3). When these regressions are multiplied by the ANET time series, projections of the temperature field on the ANET are obtained. On the basis of monthly mean data, linear trends of these projections have been estimated (Fig. 4) using a least squares fit. A Monte Carlo approach with a large number of artificial ANET time series has been used in order to estimate the significance of the results in Figs 3 and 4.

Energy transport

The atmospheric energy can be divided into four components: potential, gz; internal, cvT; kinetic, ; and latent, Lq. Here g is gravity, z is height, cv is specific heat capacity for constant volume, T is absolute temperature, u is the three-dimensional wind vector, L is the specific heat of condensation or sublimation, and q is specific humidity. The ANET across latitude &phgr;o is defined by:where &phgr; is latitude, cp is specific heat capacity for constant pressure, v is the northward wind component, p is pressure, &eegr; is the vertical hybrid coordinate used within the ERA-40 framework, and x is the east–west coordinate.

Linear regression

The sensitivity of the Arctic temperature field to the ANET across 60° N is estimated as follows: first, the ANET across 60° N based on 6-hourly data is determined using equation (1). Then a barotropic mass correction is applied to daily averages of the ANET. At each grid point and based on the daily data, the temperature time series have been regressed on the mass-corrected ANET time series for different time lags of the temperature relative to the ANET. The annual cycle was removed and a 7-day running mean applied to all data before performing the regressions. Normalized regressions at the 500 hPa level are shown in Fig. 3.

For estimations of the Arctic temperature trends, which are linked to the ANET across 60° N, monthly mean rather than daily data are used as the warming response lags the ANET by around 5 days (Fig. 4b). The regressions multiplied by the ANET time series constitute the projection of the temperature field on the ANET. Linear trends of these projections are shown in Fig. 4b.

Significance test

The shading in Fig. 3 and the white contours in Fig. 4b show the results of a statistical significance test based on a Monte Carlo approach. The temperature field is regressed on artificial time series with the same power spectrum as the ANET, but with arbitrary phases of the modes. These regressions are compared to the original regression at each spatial grid point. At a given grid point, the temperature trend linked to the ANET (Fig. 4b) is taken to differ significantly from zero on, say, a 99% level, if less than 1% of the artificial regressions show trends that are numerically larger than the one from the original regression. A corresponding procedure is used when estimating significance of the lagged regressions (Fig. 3). We have compared with at least 10,000 artificial projections.

Averaged temperature trends around latitude circles for 1979–2001 plotted versus latitude and height for the four seasons.

Trends are shown for winter (a, December–February), spring (b, March–May), summer (c, June–August) and autumn (d, September–November). The linear trends are estimated from monthly mean data using a least-squares fit.

Dark-month (November–February) anomalies of mean temperature relative to the 1850–1900 average as function of year.

Data were obtained from land-station observations. In the main figure, the symbols represent means from individual years, whereas the lines show the temporal evolution when variability over timescales smaller than 20 years has been removed using a wavelet filter. Solid line and open circles are based on observations north of 65° N, while the dashed line and dots are for the entire Northern Hemisphere. Inset shows the smoothed temperature time series for the full instrumental period. The data were provided by the Climate Research Unit (CRU) as a 5° × 5° gridded data set.

Regressions of the 500 hPa temperature field on the atmospheric northward energy transport (ANET) across 60° N.

a, Regressions averaged around latitude circles as a function of latitude and time lag; b, regressions for 5-day lag as a function of longitude and latitude. Solid and dotted contours indicate positive and negative regressions, respectively. In each point the regression has been scaled by the spatial standard deviation of all regressions. Light- and dark-grey shading shows areas where regressions differ significantly from zero at the 99% and 99.9% level, respectively. The regressions indicate temperature anomalies associated with an ANET anomaly at lag zero. For instance, a positive ANET anomaly is followed 5 days later by warming and cooling north and south of 60° N, respectively.

Averaged temperature trends around latitude circles for 1979–2001 plotted versus latitude and height for April–October.

a, Total trends; b, trends that are linked to the ANET across 60° N. The shadings indicate trends, and the white contours indicate areas where trends differ significantly from zero at the 99% and 99.9% level, respectively.

We thank P. Lundberg for comments on the manuscript. The ERA-40 data were obtained from the European Centre for Medium-Range Weather Forecasts (ECMWF) data server, whereas the Climate Research Unit (CRU) at the University of East Anglia provided the observational data used for Fig. 2.

Author Contributions The analysis was performed and the manuscript written by R.G.G., and to some extent T.M. The original idea to use ERA-40 data to study Arctic warming was due to R.G.G., M.T. and E.K. All authors contributed with ideas, discussions and text.

Simon, C., Arris, L. & Heal, B. Arctic Climate Impact Assessment (Cambridge Univ. Press, New York, 2005) Johannessen, O. M. et al. Arctic climate change: observed and modeled temperature and sea-ice variability. Tellus A 56, 328341 (2004) Polyakov, I. V. et al. Observationally based assessment of polar amplification of global warming. Geophys. Res. Lett. 29 doi: 1029/2001GL011111 (2002) Rigor, I. G., Colony, R. L. & Martin, S. Variations in surface air temperature observations in the Arctic, 1979–97. J. Clim. 13, 896914 (2000) Serreze, M. C. & Francis, J. A. The Arctic amplification debate. Clim. Change 76, 241264 (2006) Cavalieri, D. J., Gloersen, P., Parkinson, C. L., Comiso, J. C. & Zwally, H. J. Observed hemispheric asymmetry in global sea ice changes. Science 278, 11041106 (1997) Stroeve, J. C. et al. Tracking the Arctic’s shrinking ice cover: Another extreme September minimum in 2004. Geophys. Res. Lett. 32 doi: 10.1029/2004GL021810 (2005) Solomon, S. et al. (eds) Climate Change 2007: The Physical Science Basis (Cambridge Univ. Press, Cambridge, UK, 2007) Holland, M. M. & Bitz, C. M. Polar amplification of climate in coupled models. Clim. Dyn. 21, 221232 (2003) Hansen, J. et al. Efficacy of climate forcing. J. Geophys. Res. 110 doi: 10.1029/2005JD005776 (2005) Chapman, W. L. & Walsh, J. E. Simulation of Arctic temperature and pressure by global coupled models. J. Clim. 20, 609632 (2007) Alexeev, V. A., Langen, P. L. & Bates, J. R. Polar amplification of surface warming on an aquaplanet in “ghost forcing” experiments without sea ice feedbacks. Clim. Dyn. 24, 655666 (2005) Thompson, D. W. J. & Wallace, J. M. Regional climate impacts of the Northern Hemisphere annular mode and associated climate trends. Science 293, 8589 (2001) Moritz, R. E., Bitz, C. M. & Steig, E. J. Dynamics of recent climate change in the Arctic. Science 297, 14971502 (2002) Wu, Q. & Straus, D. M. AO, COWL, and observed climate trends. J. Clim. 17, 21392156 (2004) Quadrelli, R. & Wallace, J. M. A simplified linear framework for interpreting patterns of Northern Hemisphere wintertime climate variability. J. Clim. 17, 37283744 (2004) Wang, X. & Key, J. R. Arctic surface, cloud, and radiation properties on the AVHRR polar pathfinder dataset. Part II: Recent trends. J. Clim. 18, 25752593 (2005) Manabe, S. & Weatherald, R. T. The effect of doubling the CO2 concentrations on the climate of a general circulation model. J. Atmos. Res. 32, 315 (1975) Hansen, J. et al. Dangerous human-made interference with climate: a GISS modelE study. Atmos. Chem. Phys. 7, 22872312 (2007) Graversen, R. G. Do changes in the midlatitude circulation have any impact on the Arctic surface air temperature trend? J. Clim. 19, 54225438 (2006) Thompson, D. W. J. & Wallace, J. M. The Arctic Oscillation signature in the wintertime geopotential height and temperature fields. Geophys. Res. Lett. 9, 12971300 (1998) Intrieri, J. M. et al. An annual cycle of Arctic surface cloud forcing at SHEBA. J. Geophys. Res. 107 doi: 10.1029/2000JC000439 (2002) Uppala, S. M. et al. The ERA-40 re-analysis. Q. J. R. Meteorol. Soc. 131, 29613012 (2005) Jones, P. D., New, M., Parker, D. E., Martin, S. & Rigor, I. G. Surface air temperature and its variations over the last 150 years. Rev. Geophys. 37, 173199 (1999) Oort, A. H. & Peixóto, J. P. Global angular momentum and energy balance requirements from observations. Adv. Geophys. 25, 355490 (1983) Trenberth, K. E. Climate diagnostics from global analysis: Conservation of mass in ECMWF analysis. J. Clim. 4, 707721 (1991)
doi: 10.1038/nature06502

Effects of acoustic waves on stick–slip in granular media and implications for earthquakes p.57

version="1.0" encoding="utf-8"?> ]>
Nature 451 7174 20080103 57604 0028-0836 1476-4687 2007Nature Publishing Group Supplementary Figures

The file contains Supplementary Figures S1-S2 with Legends.

Effects of acoustic waves on stick–slip in granular media and implications for earthquakes Paul A.JohnsonP A HeatherSavageH MattKnuthM JoanGombergJ ChrisMaroneC Geophysics Group EES-11, Los Alamos National Laboratory of the University of California, MS D443, Los Alamos, New Mexico 87545, USA Department of Geosciences, Pennsylvania State University, University Park, Pennsylvania 16802, USA Department of Earth and Planetary Science, University of California, Santa Cruz, California 95064, USA Department of Geology and Geophysics, University of Wisconsin, Madison, Wisconsin 53706, USA US Geological Survey, University of Washington, Department of Earth and Space Sciences, Box 351310, Seattle, Washington 98195-1310, USA Correspondence and requests for materials should be addressed to P.A.J. (paj@lanl.gov). &e080103-12;

It remains unknown how the small strains induced by seismic waves can trigger earthquakes at large distances, in some cases thousands of kilometres from the triggering earthquake, with failure often occurring long after the waves have passed. Earthquake nucleation is usually observed to take place at depths of 10–20 km, and so static overburden should be large enough to inhibit triggering by seismic-wave stress perturbations. To understand the physics of dynamic triggering better, as well as the influence of dynamic stressing on earthquake recurrence, we have conducted laboratory studies of stick–slip in granular media with and without applied acoustic vibration. Glass beads were used to simulate granular fault zone material, sheared under constant normal stress, and subject to transient or continuous perturbation by acoustic waves. Here we show that small-magnitude failure events, corresponding to triggered aftershocks, occur when applied sound-wave amplitudes exceed several microstrain. These events are frequently delayed or occur as part of a cascade of small events. Vibrations also cause large slip events to be disrupted in time relative to those without wave perturbation. The effects are observed for many large-event cycles after vibrations cease, indicating a strain memory in the granular material. Dynamic stressing of tectonic faults may play a similar role in determining the complexity of earthquake recurrence.

Laboratory studies of granular friction have emerged as a powerful tool for investigating tectonic fault zone processes and earthquake phenomena, including post-seismic slip, interseismic frictional restrengthening and earthquake nucleation. Here we explore experimentally the effects of dynamic loading on stick–slip behaviour and discuss how our results may affect understanding of earthquake processes—in particular dynamic earthquake triggering and stick–slip recurrence. Dynamic earthquake triggering involves seismic waves from one earthquake promoting or inhibiting failure on the faults they disturb. Dynamic triggering has been clearly documented in a few cases far from an earthquake source, at distances much greater than the fault radius of the triggering source (outside the traditional ‘aftershock zone’), and increasing evidence suggests that it commonly occurs near the earthquake source.

Experiments on sheared layers of glass beads (like those shown in Fig. 1, and described in the Methods section) exhibit stick–slip that varies with shear displacement rate, confining stress, relative humidity, granular media thickness, and particle characteristics; however, for fixed experimental conditions stick–slip characteristics are remarkably constant (Fig. 2). Stick–slip events are characterized by sudden, periodic shear stress drops that range from 10–30% of the maximum frictional strength. Leading up to steady-state strength, which takes several tens of seconds and shear strains of ∼0.4–0.5, we observe a material dilation and nonlinear shear-stress increase accompanied by intermittent failure. During steady-state frictional behaviour, major stick–slip events recur very regularly but include rare, small events (for example, at 1,375 s in Fig. 2b). Each major stick–slip event is followed by elastic and then inelastic stress build-up and layer dilation. The dilation is manifested by increasing layer thickness (Fig. 2b). Layers dilate to a point of instability at which catastrophic dynamic failure and compaction occur (Fig. 2b).

The top curve of Fig. 3a shows results from an experiment identical to that of Fig. 2 except that we applied acoustic waves during shearing, commencing a few seconds before expected stick–slip failure, and continuing until the major failure event. The lower curve of Fig. 3a shows the rectified peak strain amplitude measured by the accelerometer attached to the sample (Fig. 1), along with three intervals for which dynamic waves were applied and signals from acoustic emission from both small and large stick–slip failures.

Vibration perturbs the recurrence period of inelastic stress increase before the failure of major events and induces small-amplitude stick–slip events. In many cases one or more small stick–slip events occur during vibration, as well as cascades of delayed, small-amplitude stick–slip events (Fig. 3a, grey shading). In all cases, application of acoustic waves—even for brief intervals—has a lasting effect, such that successive major stick–slip events exhibit a strain memory of applied vibration manifest by delayed failure, disruption of recurrence interval and extended aseismic creep, despite the violent mechanical re-set that occurs during major stick–slip events (Fig. 3). We find that post-vibration, the regular recurrence does not recover.

We also apply acoustic pulses, rather than the longer-duration waves described above. Pulses are more analogous to a single seismic wave in Earth, whereas vibration may be more analogous to the near-source region where quasi-continuous-wave energy may exist for significant periods of time in the form of aftershocks. Our data show that continuous and pulse modes of dynamic triggering yield similar behaviour. See Supplementary Fig. 1, where we show a typical sequence of stick–slip events in the presence of acoustic pulses.

When we apply vibration or pulsed sound at stresses below ∼95% of the failure strength there is little or no effect on stick–slip. This implies that the system must be in a critical state to be susceptible to dynamic triggering, which is consistent with seismic data on earthquake triggering and recent modelling. Qualitatively, wave strain amplitudes must exceed approximately 10-6 for the above effects to be observed (Supplementary Fig. 2), consistent with dynamic triggering observations for real earthquakes. When the system is driven with vibration amplitudes corresponding to strains <10-6 there is no obvious effect on stick–slip; however, we emphasize that this should be further quantified in future experiments.

Analysis of the primary stick–slip recurrence intervals for otherwise identical experiments with and without vibration shows that failure becomes progressively more erratic and, on average, lengthens with time in experiments with wave excitation (Fig. 4a). Repeated experiments with both vibration and pulse-mode verify that this effect is real. We find that the scatter relative to the mean increases with accumulated time in experiments with vibration. Also, although the primary stick–slip recurrence interval increases significantly due to acoustic waves, the stress-drop magnitude and variation increase only slightly (Fig. 4b).

We have described three primary experimental observations: (1) acoustic waves disrupt recurrence intervals and, to a lesser degree, stress drops of large magnitude events; (2) acoustic waves trigger immediate and delayed small-magnitude events, some aseismic; and (3) strain memory of acoustic perturbation is maintained through successive large-magnitude stick–slips. We assess the implications of these results for dynamic earthquake triggering by considering that the primary slick–slip events represent tectonic earthquakes and that the vibration-induced events represent triggered earthquakes.

The overall trend of increasing stress drop (and maximum frictional strength) with recurrence interval is consistent with a large body of previous laboratory and field observations showing that maximum frictional strength increases linearly with the log of recurrence interval between slip events. Vibration diminishes the rate at which stress drop increases with inter-event time, notably creating greater irregularity in stick–slip recurrence interval. The commonly used class of rate-state frictional models explicitly predicts that the rate of strengthening is proportional to the product of the normal stress and the frictional constitutive parameters. Because we hold the normal stress constant, this implies that vibration alters frictional properties, despite the fact that perturbation amplitudes (∼104 Pa) are less than 1% of the normal load. We suggest that the irregularity in recurrence that we observe in our experiments mimics that observed for tectonic faults in the Earth’s crust, and reflects a complex process of disrupting the internal fault zone structure.

We find that vibration has measurable effects only when the system is in a critical state, approaching failure (for example, see Fig. 3a). Application of acoustic waves also has a measurable effect only for experiments conducted at relatively small normal stresses, approximately 4–5 MPa. We have explored higher horizontal loads (up to ∼18 MPa) for which we did not see a vibrational effect; perhaps the vibration amplitude was not sufficiently large to produce an effect. Nevertheless, the laboratory experiments do imply that dynamic earthquake triggering at seismic strain amplitudes is most efficient at low effective stress (normal load minus pore pressure) for faults in a critical state. Some field-based studies also point to a connection between earthquake triggering and low effective stress and/or faults near failure, although this is a point of some debate.

One mystery regarding dynamic earthquake triggering is that it can take place minutes, hours or days after the seismic perturbation. Our experiments show delayed failures following acoustic perturbations, frequently manifesting as cascades of small events. We do not yet understand the physics responsible for this observation; however, we speculate that triggered events, as well as the recurrence and stress-drop disruptions are manifestations of frictional contact mechanics coupled with granular processes. Previous work shows that stick–slip initiates as failure of a contact junction between beads in highly stressed chains of particles. Granular memory effects are presumably the result of similar processes.

We posit that acoustic waves disrupt granular force chains, leading to material softening and simultaneous weakening (granular flow), similar to what is described in a recently proposed phenomenological model. The manifestation of the acoustic disruption may take place immediately or later in time (strain ‘memory’). The vibration-induced memory itself may be maintained as frictional instability at a number of grain contacts that persist through one or more stick–slip cycles, and is reminiscent of dynamically induced strain memory, known as ‘slow dynamics’, observed in nonlinear dynamical experiments on glass bead packs. The memory is also suggestive of statically induced rate-dependent effects observed in sheared granular materials, such as ‘ageing’. We attempted to erase vibration-induced memory by ceasing shear loading to allow the material to heal, as well as by changing normal stress to repack the grains, but neither approach succeeded.

Our previous work shows that permanent damage to the grains themselves is negligible and therefore cannot be the origin of the behaviours observed. Moreover, acoustical studies in three-dimensional glass bead packs under similar wave strain amplitudes, and under (smaller) static stresses of 0.02–0.1 MPa, show no evidence for grain rearrangement; however, the material exhibits very small, irreversible compaction as well as nonlinear-induced modulus softening and slow dynamics. Hertz–Mindlin contact mechanics describe these observations. The compaction we measure in our experiments without vibration is small and does not lead to instability. The addition of vibration shows additional compaction but it is extremely small. Taken together, the observations suggest that minute compaction plays a part in what we observe, but there is no clear evidence suggesting that it is the cause. Our data do not rule out the possibility that instability is abetted, or initiated, by localized compaction (for example, within a shear band in the layer), which would be invisible to our measurements. Local compaction within a granular material would reduce normal stress at contact junctions, which could lead to stick–slip instability. For the moment, the origin of what we observe when stick–slip is combined with vibration remains unknown.

The origin of dynamic earthquake triggering by transient seismic waves is a complex problem. Our results show that granular-friction processes are consistent with two as-yet-unexplained observations in earthquake seismology: (1) small-amplitude waves can trigger both immediate failure and delayed failure relative to the strain transient, and (2) earthquake recurrence patterns are complex. Understanding the role of vibration-induced disruption of earthquake recurrence could have significant implications for seismic hazard assessment and reliable forecasting of earthquakes.

Methods Summary

In our experimental study of acoustic waves interacting with a laboratory-scale fault system, we employ a double-direct shear configuration to shear 4-mm layers of glass beads at constant normal stress (1–18 MPa), using shearing rates of 1–100 µm s-1 (Fig. 1). Class IV bead dimensions are 105–149 µm in diameter. Layers are subject to either continuous vibration or wave pulses of 10–20 cycles at 1–20 kHz, with strain amplitudes ranging from <5 × 10-7 to 8 × 10-6, or alternatively, to no wave excitation. An acoustic source and accelerometer are mounted directly on the central shearing block (Fig. 1b). We measure stresses, displacements and wave-induced strains continuously throughout shearing.

We use a double-direct shear configuration in a biaxial load frame, which applies a horizontal stress to three steel forcing blocks that contain symmetric layers of glass beads at the block interfaces (Fig. 1). A vertical piston drives the central block downward at a constant displacement rate to create shear. The apparatus is servo-controlled so that constant horizontal load and vertical displacement rate are maintained to ±0.1 kN and ±0.1 &mgr;m s-1, respectively. The applied stresses on the shearing layers are measured with strain-gauge load cells in series with each of the loading axes. The apparatus is controlled via computer, and recordings of the load, displacement and stresses are monitored throughout an experiment. The nominal frictional contact dimensions are 10 cm × 10 cm, the vertical displacement of the central block is 5 &mgr;m s-1, corresponding to a strain rate of approximately 1.2 × 10-3 s-1, and the horizontal stress is 4 MPa. The blocks are composed of 17-4 stainless steel with serrated faces of 1 mm wavelength and 0.75 mm depth, adjacent to the glass bead packs. The beads are class IV spheres and range in dimension from 105–149 &mgr;m, meaning the layers are each about 30 beads wide. The sample assembly is sheathed by a latex sleeve. In the experiments, horizontal loads of 1–15 MPa are explored as well. Background noise emanating from the building and instrument is well under 5 × 10-7 strain and at much lower frequency (<∼100 Hz) than the applied acoustic perturbations.

Vibration is applied via an acoustic source (Fig. 1): a Matec M50-2, 50-KHz central-frequency piezoceramic attached mechanically with clamps to the central block using vacuum grease as couplant and driven by a Samson 150, 75-watt amplifier. The signal is detected on the opposite face of the central block using a Brüel and Kjaer model 4393 accelerometer attached with beeswax, amplified by a Brüel and Kjaer 2635 charge amplifier, and recorded on computer. Acoustic frequencies range from 1–20 kHz. Such high frequencies are not part of the seismic spectrum in nature, which extends to 10–100 Hz at maximum, but are used to provide laboratory-scale physical insight that can be applied in nature. No frequency effect is observed in the observations presented. We initiate waves at a shear stress equal to ∼95% of the failure strength by first measuring the stick–slip recurrence interval without wave excitation for approximately 30 events and then timing the initiation of vibration from the end of the previous stick–slip.

In the pulse experiments, a toneburst of 10–20 cycles with frequency ranging from 6.1–8.67 kHz is applied. We use tonebursts of approximately 3.3 ms duration and a centre frequency of 6,100 Hz for the results shown. In general, sound is applied every third stick–slip cycle after steady-state conditions are reached, but experiments were also conducted in which we applied sound at shorter and longer stick–slip intervals.

The strain amplitudes we apply range from about 5 × 10-7 to 8 × 10-6. A strain wave of 10-6 applies a pressure of the order of 104 Pa, which is of the order of 1% of the normal stress. Elastic wave strain is estimated as follows. In a harmonic wave, strain ε = du/dx is related to acceleration ü =  d2u/dx2for the time-average amplitude. We digitize the acceleration data and record the absolute value of the sinusoidal waveform with a sampling rate of 10 kHz at 16 bits.

Experimental apparatus.

a, Apparatus, showing horizontal piston applying constant normal stress, and vertical piston applying a constant (vertical) displacement rate, which drives shear. The dashed circle shows the sample assembly. b, Sample assembly showing three-block arrangement of the double-direct shear configuration (front and side views). We note the location of the acoustic wave source and accelerometer in relation to the glass bead layers and normal stress (horizontal).

Stick–slip behaviour under constant shearing rate, without vibration.

a, Shear stress versus experiment time for a typical run. Note that maximum stick–slip stress drops are ∼30% of the shear strength. Over the total duration of the experiment, there is a small but progressive compaction of about 1% of the glass bead layer thickness (not shown). b, Detail of the stick–slip cycles (top) and change in layer thickness (bottom). The layer thickness has had the overall trend removed. We note consistent failure strength, recurrence interval, and creep before stick–slip. p1108 refers to experiment number.

Stick–slip with and without vibration.

a, Stick–slip behaviour under constant shearing rate, with vibration. Shear stress versus experiment time (upper curve); and measured, rectified strain amplitudes of the detected acoustic waves (lower curve). The letter ‘V’ denotes times and thick black horizontal bars indicate the durations of vibration. Vibration has a marked influence on the stick–slip behaviour. For instance, the applied vibration at ∼2,050 s produces an immediate, small-magnitude stick–slip. The two successive major stick–slips that follow exhibit longer recurrence times as well as multiple small stick–slip events in between—these are triggered events. Regions of triggered events are shaded light grey. Similarly, irregular cycles occur following the applied vibration at 2,155 s. Vibration applied at ∼2,255 s produces an immediate small-magnitude stick–slip event and an increased major-event recurrence interval. b, Comparison of non-vibration versus vibration, emphasizing increased recurrence and irregular behaviour, including triggering, due to acoustic waves. p870 and p1108 refer to experiment numbers.

Stick-slip recurrence time and stress drop comparing vibration and non-vibration experiments.

a, Recurrence versus experiment time for runs with vibration (solid circles) and without. The shaded region and dashed lines show the mean recurrence interval of ±1 standard deviation. Data trend removed. Compared to the non-vibration experiments, both the scatter and average recurrence interval increases progressively in experiments with vibration. b, Stress-drop variation versus recurrence for experiments conducted with and without vibration. We cannot compare stress-drop amplitudes directly owing to minor differences from one experiment to the next; however, when we compare the variation of stress drop to the experimental mean, we see a clear trend of longer recurrence interval for a given change in stress drop.

Funding was provided by Institutional Support (LDRD) at Los Alamos and the DOE Office of Basic Energy Science (P.A.J.), by the National Science Foundation (C.M., H.S., M.K.), and by the United States Geological Survey (J.G.). We thank E. Brodsy, B. Behringer, N. Beeler and X. Jia for comments and reviews.

Author Contributions P.A.J., M.K., H.S. and C.M. designed the study. M.K., P.A.J. and C.M. designed and carried out the data collection procedure. P.A.J. and H.S. did most of the data analyses. P.A.J. and C.M. did most of the writing. P.A.J., H.S., M.K. and C.M. did the laboratory work and J.G. and C.M. did much of the writing interpretation. All authors contributed to the interpretation and writing.

Hill, D. P. et al. Seismicity remotely triggered by the magnitude 7.3 Landers, California, earthquake. Science 260, 16171623 (1993) Gomberg, J., Bodin, P., Larson, K. & Dragert, H. Earthquake nucleation by transient deformations caused by the M = 7.9 Denali, Alaska earthquake. Nature 427, 621624 (2004) Brodsky, E., Karakostas, V. & Kanamori, H. A. New observation of dynamically triggered regional seismicity: Earthquakes in Greece following the August, 1999, Ismit, Turkey earthquake. Geophys. Res. Lett. 27, 27412744 (2000) Hough, S. E. Triggered earthquakes and the 1811–1812 New Madrid, Central United States earthquake sequence. Bull. Seismol. Soc. Am. 91, 15741581 (2001) Gomberg, J., Bodin, P. & Reasenberg, P. A. Observing earthquakes triggered in the near field by dynamic deformations. Bull. Seismol. Soc. Am 93, 118138 (2003) Brodsky, E. & Prejean, S. G. New constraints on mechanisms of remotely triggered seismicity at Long Valley Caldera. J. Geophys. Res. 110, 43024316 (2005) Marone, C. Laboratory-derived friction laws and their application to seismic faulting. Ann. Rev. Earth Planet. Sci. 26, 643696 (1998) Scholz, C. H. The Mechanics of Earthquakes and Faulting 2nd edn (Cambridge Univ. Press, 2002) Felzer, K. & Brodsky, E. Evidence for dynamic aftershock triggering from earthquake densities. Nature 441, 735738 (2006) Nasuno, S., Kudrolli, A., Bak, A. & Gollub, J. P. Time-resolved studies of stick–slip friction in sheared granular layers. Phys. Rev. E 58, 21612171 (1998) Mair, K., Frye, K. M. & Marone, C. Influence of grain characteristics on the friction of granular shear zones. J. Geophys. Res. 107, ECV41-9 (2002) Anthony, J. L. & Marone, C. Influence of particle characteristics on granular friction. J. Geophys. Res. 110 B08409 doi: 10.1029/2004JB003399 (2005) Freed, A. M. Earthquake triggering by static, dynamic, and postseismic stress transfer. Annu. Rev. Earth Planet. Sci. 33, 335367 (2005) Ziv, A. What controls the spatial distribution of remote aftershocks? Bull. Seismol. Soc. Am. 96, 22312241 (2007) Gomberg, J. & Johnson, P. A. Dynamic triggering of earthquakes. Nature 473, 830 (2005) Zoback, M. D. & Beroza, G. Evidence for near-frictionless faulting in the 1989 (M 6.9) Loma Prieta, California, earthquake and its aftershocks. Geology 21, 181185 (1993) Beeler, N. M. & Lockner, D. A. Why earthquakes correlate weakly with the solid Earth tides: effects of periodic stress on the rate and probability of earthquake occurrence. J. Geophys. Res. 108, 23912408 (2003) Scholz, C. H. Evidence for a strong San Andreas fault. Geology 28, 163166 (2000) Johnson, P. A. & Jia, X. Nonlinear dynamics, granular media and dynamic earthquake triggering. Nature 473, 871874 (2005) Hartley, R. R. & Behringer, R. P. Logarithmic rate dependence of force networks in sheared granular materials. Nature 421, 928931 (2003) Brunet, T., Jia, X. & Johnson, P. Transitional, elastic-nonlinear behaviour in dense granular media. Geophys. Res. Lett. (submitted) Morgan, J. K. & Boettcher, M. S. Numerical simulations of granular shear zones using the distinct element method: I. Shear zone kinematics and the micromechanics of localization. J. Geophys. Res. 104, 27032719 (1999)
doi: 10.1038/nature06440

Sparse optical microstimulation in barrel cortex drives learned behaviour in freely moving mice p.61

version="1.0" encoding="utf-8"?> ]>
Nature 451 7174 20080103 61640 0028-0836 1476-4687 2007Nature Publishing Group Supplementary Information 1

The file contains Supplementary Methods, Supplementary Figures 1-7 with Legends and Legends to Supplementary Movies 1-2.

Supplementary Video 1

The file contains Supplementary Movie 1.

Supplementary Video 2

The file contains Supplementary Movie 2.

Sparse optical microstimulation in barrel cortex drives learned behaviour in freely moving mice DanielHuberD LeopoldoPetreanuL NimaGhitaniN SachinRanadeS TomášHromádkaT ZachMainenZ KarelSvobodaK Howard Hughes Medical Institute, Janelia Farm Research Campus, Ashburn, Virginia 20147, USA Cold Spring Harbor Laboratory, Cold Spring Harbor, New York 11724, USA Correspondence and requests for materials should be addressed to K.S. (svobodak@janelia.hhmi.org).

Electrical microstimulation can establish causal links between the activity of groups of neurons and perceptual and cognitive functions. However, the number and identities of neurons microstimulated, as well as the number of action potentials evoked, are difficult to ascertain. To address these issues we introduced the light-gated algal channel channelrhodopsin-2 (ChR2) specifically into a small fraction of layer 2/3 neurons of the mouse primary somatosensory cortex. ChR2 photostimulation in vivo reliably generated stimulus-locked action potentials at frequencies up to 50 Hz. Here we show that naive mice readily learned to detect brief trains of action potentials (five light pulses, 1 ms, 20 Hz). After training, mice could detect a photostimulus firing a single action potential in approximately 300 neurons. Even fewer neurons (approximately 60) were required for longer stimuli (five action potentials, 250 ms). Our results show that perceptual decisions and learning can be driven by extremely brief epochs of cortical activity in a sparse subset of supragranular cortical pyramidal neurons.

We used in utero electroporation to introduce ChR2 fused to a green fluorescent protein (GFP) (ChR2–GFP) together with a red fluorescent cytosolic marker (RFP) into neocortical pyramidal neurons (Fig. 1a, Methods). In the adult brain, ChR2–GFP expression was restricted to pyramidal cells in layers 2/3 (more than 99.4%), mainly in the barrel cortex (Figs 1a and 2a). In vivo two-photon imaging and retrospective immunohistology revealed that ChR2–GFP was localized to the neuronal plasma membrane. ChR2–GFP was expressed in about half (48.9 ± 5.3%, n = 10, five mice; see Methods) of red fluorescent layer 2/3 neurons (Supplementary Movie 1). ChR2–GFP invaded the soma, dendrites and axons (Fig. 1b, c). ChR2–GFP expression was stable for at least 8 months and did not seem to perturb neuronal morphology (Fig. 1a–c, Methods).

We next characterized the responses of ChR2–GFP-expressing neurons to photostimulation in anaesthetized mice. To sample from the entire population of ChR2–GFP-expressing neurons, unbiased by ChR2–GFP expression level, we recorded from red fluorescent neurons using two-photon targeted loose-patch recordings (Fig. 1c, d). Photostimuli consisted of light pulses, produced by a blue miniature light-emitting diode (LED; 470 nm), centred on the recording window (Fig. 1d). At maximum light intensities (Imax = 11.6 mW mm-2 at the surface of the brain, centred on the diode; 1–10 ms duration) about half (51%) of the patched red neurons (n = 39/77, eight mice) responded reliably to single photostimuli with at most one action potential. Increasing the photostimulus duration beyond 10 ms did not reveal additional responsive neurons. The other half of the patched neurons did not fire spikes time-locked to the photostimuli, and presumably corresponded to ChR2–GFP-negative neurons. These measurements indicate that most ChR2–GFP-positive neurons can be driven to spiking using our photostimulation system; furthermore, excitation of layer 2/3 neurons through indirect synaptic pathways was weak.

When stimulated with 1 ms light pulses, ChR2–GFP-expressing neurons were able to follow frequencies up to 20 Hz (Fig. 1e) and in some cases up to 50 Hz (Fig. 1f). These frequencies are comparable to, or higher than, typical spike rates in the barrel cortex. Action potentials followed the photostimuli with short delays (range 3–11 ms) and little jitter (Supplementary Fig. 1).

We next determined the relation between photostimulus intensity and the probability of spiking of ChR2–GFP-expressing neurons. During cell-attached recordings we stimulated with 1 ms light pulses while varying the photostimulus. With decreasing light intensity, neurons switched abruptly from firing action potentials with high probability to firing no action potentials. The photostimulus intensity required to trigger action potentials varied substantially across the population of ChR2–GFP-expressing neurons (Fig. 1g). Control experiments in brain slices revealed that the brightness of ChR2–GFP measured in individual cells was inversely correlated with firing threshold (Supplementary Fig. 2); in contrast, the firing threshold was independent of the depth of the recorded neuron in vivo (Supplementary Fig. 3). The variability in firing threshold in terms of photostimulus intensity therefore primarily reflects heterogeneity in the expression level of ChR2–GFP in individual neurons. These results confirm that ChR2 can transduce photostimuli into precisely timed spike trains in vivo. Furthermore, the fraction of activated neurons can be tuned by modulating the excitation light intensity (Fig. 1h).

Can awake mice learn to report photostimulation of layer 2/3 pyramidal neurons in the barrel cortex? To address this question we delivered light pulses to ChR2–GFP-expressing neurons in freely moving animals (Fig. 2a). We first implanted a window above the barrel cortex, which provided optical access for photostimulation and screening the density of electroporated neurons. We next mounted the miniature LED centred on the imaging window (Fig. 2a; Methods). During the behavioural sessions the mice were temporarily connected to an LED controller (Methods). Mice were trained in a detection task to associate photostimulation of ChR2–GFP-expressing neurons (five light pulses, 20 Hz, 1 ms duration) with water reward on one of two choice ports (Fig. 2b, left port). After four to seven training sessions (200–800 trials per session) all animals expressing ChR2–GFP (n = 9) reliably reported photostimulation; in the presence (absence) of a photostimulus, mice chose the left (right) port (Fig. 3a, range 72–93% correct, defined as hits + correct rejections, divided by total number of trials; Supplementary Movie 2). Control mice without electroporated neurons (n = 6) performed at chance levels (50.1%, P > 0.70, t-test), even after 25 training sessions (Fig. 3a and Supplementary Fig. 4). These experiments demonstrate that photostimulation of layer 2/3 neurons can drive robust behaviour.

How many action potentials triggered by photostimulation are necessary for perception? To address this issue we further trained five mice to respond to one, two and five photostimuli at 20 Hz (example in Fig. 2c). Although performance decreased with fewer pulses, all ChR2–GFP-expressing mice were able to detect single action potentials in the activated cells, even at modest photostimulus intensities (Fig. 3b, red lines).

To determine the relation between performance and the number of neurons directly activated by light, we measured behaviour as a function of light intensity (Fig. 3b). As expected, behavioural performance decreased with decreasing photostimulus intensity, although the psychometric curves varied from animal to animal. For example, at the lowest intensities probed (10% of Imax) some animals continued to discriminate, whereas others performed at chance levels.

We counted the number of ChR2–GFP-positive somata and measured their positions (Fig. 3c; Methods). Between 594 and 1430 ChR2–GFP-positive neurons were found in a 2 mm diameter window (Fig. 3b, c; Methods). The number of ChR2–GFP-positive neurons under the photostimulation window correlated with the performance of individual animals.

For each animal we then estimated the number of active neurons as a function of normalized intensity (Io = intensity/Imax) as:

Here rk is the horizontal position of the kth ChR2–GFP-positive cell and f is the fraction of ChR2-positive cells activated at intensity Ioi (Fig. 1h). i(r) is the spatial distribution of the normalized light intensity in the tissue (horizontal full-width at half maximum = 2.17 mm, 250 &mgr;m below the pia) (Supplementary Methods; Supplementary Fig. 5). For trains of five action potentials, an average of 61 neurons (range 6–197) was sufficient to drive reliable performance (more than 65% of correct choices), whereas 297 (range 135–381) active neurons were required with single action potentials (Fig. 3d). The total number of action potentials required for a given level of performance was independent of the stimulus pattern (Supplementary Fig. 6).

Two factors make us believe that our estimates of the number of active neurons should be interpreted as an upper bound. First, the measured spatial distribution of light in the tissue is likely broader than the actual distribution of light (see Supplementary Methods). Second, we did not consider possible deterioration of the optical path (thickening of the dura, bone growth, and so on) over the long timescales required for the behavioural experiments compared with the more favourable conditions of the calibration measurements. Therefore the actual number of activated neurons in the behavioural experiments might have been lower than the numbers cited above.

Not surprisingly, triggering more action potentials yields better detection accuracy (Fig. 3d). However, performance reached asymptotic levels at remarkably low numbers of directly activated neurons; the range between minimal detection and saturating performance was only a few hundred neurons.

Activated ChR2–GFP-positive neurons were distributed over most of the barrel cortex, with a smattering in adjacent sensory areas. The activated cortical region contains at least 40,000 layer 2/3 neurons (approximately 2,000 per barrel column, unpublished data) implying that synchronous action potentials in less than 1% of layer 2/3 neurons can be robustly perceived. These data imply that mechanisms exist to read out extremely sparse codes from primary sensory areas. Because of convergence in the L2/3 → L5 pathway, it is possible that even fewer activated L5 cells could be detected by behaving mice. We also note that the detection threshold could vary considerably based on the state of the animal.

We have shown that ChR2-based optical microstimulation can be used to dissect the impact of precisely timed action potentials in a few genetically defined neurons on mammalian behaviour. Our data show that the favourable characteristics of ChR2 reported previously in vitro, in vivo and in invertebrate systems—including the ability to generate precisely timed action potentials—are maintained in awake conditions and can be used effectively to drive learning and behaviour.

Photostimulation of genetically defined neurons has key advantages compared with electrical microstimulation. Under typical experimental conditions, electrical microstimulation excites axons non-discriminately, probably including diverse local and long-range axons. Therefore, the cell type and cell location that drive behaviour in classical microstimulation experiments are poorly defined. Photostimulation of genetically defined neural populations naturally overcomes these problems. Our estimates of the number of directly activated cortical neurons necessary to drive perception is lower than previous estimates based on electrical microstimulation. Our stimuli might be functionally more potent because a pure population of excitatory neurons is activated, whereas electrical microstimulation drives a mixture of diverse excitatory and inhibitory neurons. The robust associative learning induced by ChR2-assisted photostimulation opens the door to study the circuit basis of perception and cognition in vivo.

Methods Summary In utero electroporation

DNA solution (ChR2–GFP and either mCherry or DsRedexpress (‘RFP’); 4:1 molar ratio; final concentration 2 µg µl-1) was injected into the right lateral ventricle of embryonic mice (E16). Layer 2/3 progenitor cells were transfected by in utero electroporation.

Photostimulation and behaviour

An imaging window was implanted on the electroporated mice at postnatal day 40–50. A miniature blue high-power LED (470 nm peak wavelength, NFSB036BT, Nichia, Japan) was mounted on the imaging window with black dental acrylic. The timing and intensity of the LED was computer-controlled with a custom-built, low-noise current-source circuit (see Methods). Mice were trained on a detection task to report photostimulation (see Methods). The training protocol consisted of several phases; transitions from one phase to the next were triggered by performance at 65% correct or above. Mice had restricted access to drinking water to maintain 80–85% of their pre-training weight. For calibrations using targeted cell-attached recordings mice were placed under a custom-made two-photon laser-scanning microscope controlled by ScanImage software. For the photostimulation the objective was removed and a miniature blue high-power LED (as described above) was placed on the centre of the recording window (see Methods).


After completion of behavioural experiments, the brain from each animal was cut into coronal or tangential sections (40–60 µm thick) on a cryostat (Leica, CM 3050S). The localization of ChR2–GFP-positive cell bodies was measured using Neurolucida software (MBF Bioscience).

All experimental protocols were conducted according to the National Institutes of Health guidelines for animal research and were approved by the Institutional Animal Care and Use Committee at Cold Spring Harbor Laboratory and HHMI Janelia Farm Research Campus.

In utero electroporation

Venus or GFP was fused to the carboxy (C) terminus of the first 315 amino acids of channelrhodopsin-2 (gift from G. Nagel). The construct (‘ChR2–GFP’) was inserted into pCAGGS vector modified for in utero electroporation. DNA was purified and concentrated using Qiagen plasmid preparation kits and dissolved in 10 mM Tris–HCl (pH 8.0). Layer 2/3 progenitor cells were transfected by in utero electroporation. E16 timed-pregnant C57BL/6J mice (Charles River, Wilmington, Massachusetts) were deeply anaesthetized by using an isoflurane–oxygen mixture (2% isoflurane/O2 by volume). The abdomen was opened and the uterine horns were exposed. Approximately 1 µl of DNA solution coloured with Fast Green (Sigma, St. Louis, Missouri) was pressure injected (Picospritzer, General Valve, Fairfield, New Jersey) through a pulled glass capillary pipette (Warner Instruments, Hamden, Connecticut) into the right lateral ventricle of each embryo. The DNA solution contained a mixture of plasmids encoding ChR2–GFP and either mCherry or DsRedexpress (‘RFP’) in a 4:1 molar ratio, at a final concentration of 2 µg/µl. The DNA was electroporated into the right lateral ventricular wall of the embryo’s brain by applying five pulses of 45 V (duration 50 ms, frequency 1 Hz) through a pair of custom-made tweezer-electrodes, with the positive plate contacting the right side of the head. Approximately 50% of the surviving pups were strongly positive for transgene expression.


A chronic imaging window was implanted on the electroporated mice at postnatal age 40–50 days. The mice were anaesthetized by using an isoflurane-oxygen mixture (2% isoflurane/O2 by volume) delivered by an anaesthesia regulator (SurgiVet, Waukesha, Wisconsin) and mounted on a stereotaxic frame (Stoelting, Wood Dale, Illinois). The animals were screened through the exposed skull for expression of RFP by using a fluorescent dissecting scope (MVX10, Olympus, Tokyo, Japan). Expression centred on the barrel cortex was found in about 30% of the positive mice. A thin cover of cyanoacrylate adhesive (Vetbond, 3M, St Paul, Minnesota) was applied to allow subsequent adhesion of the dental acrylic. A 2 mm diameter skull flap overlying right barrel cortex (centred on 1.5 mm caudal, 3.5 mm lateral of bregma) was removed by using a fine motorized drill. Special care was taken to leave the dura intact. The opening was covered with a thin layer of warm 1% agarose (Sigma, St Louis, Missouri), and a 5 mm diameter round cover glass (Warner Instruments, Hamden, Connecticut) was sealed on top of the agarose with black dental acrylic (Lang Dental, Wheeling, Illinois). A titanium bar was embedded rostral to the window above the midline to allow fixation to the microscope. After the initial surgery, the animals were left to recover for five days. Animals of the same age that were not electroporated were used as controls. A miniature blue high-power LED (470 nm peak wavelength, NFSB036BT, Nichia, Tokyo, Japan) was mounted on the imaging window. The assembly was then covered with black dental acrylic to prevent leakage of light. The timing and intensity of the LED was computer-controlled with a custom-built, low-noise current-source circuit. All LEDs were tested before implantation and after termination of the experiments.

The distribution of light intensity at the surface of the brain was measured using a beam profiler (Supplementary Methods; Supplementary Fig. 5a, b). The distribution of light intensity in the brain, which was used to estimate the number of activated neurons (equation (1)), was estimated in a block of freshly cut brain. The dorsal surface, with mounted skull and LED, was on the side. The cut surface was imaged from the top by using a CCD (charge-coupled device) (Supplementary Fig. 5). Because the light that reaches the camera is smeared by scattering on its way out of the tissue, the width of the distribution (Supplementary Fig. 5c) on the cut surface is likely an overestimate of the actual width of the distribution in the brain.

In vivo two-photon imaging and cell-attached recordings

For targeted cell-attached recordings, similar surgery was performed as described above except that the skull opening was only 1.5 mm and a custom-shaped semilunar coverglass was sealed in place using dental acrylic, leaving the lateral edge of the exposed dura accessible to the recording electrode. To monitor the level of anaesthesia, an electrocorticogram was recorded by inserting a thin Teflon-coated silver wire between the dura and skull in the contralateral hemisphere. A reference wire was inserted above the cerebellum. In vivo imaging was performed by using a custom-made two-photon laser-scanning microscope controlled by ScanImage software. The light source was a pulsed Ti:sapphire laser (wavelength, 920–1010 nm; power, 50–200 mW in the objective back-focal plane; MaiTai, Spectra-Physics, Mountain View, California). Red and green fluorescence photons were separated by using a 565 nm dichroic mirror (Chroma Technology, Brattleboro, Vermont) and barrier filters (green, BG22; red, 607/45; Chroma). Signals were collected using photomultiplier tubes (3896, Hamamatsu, Hamamatsu City, Japan). We used an objective lens (×40, 0.8 NA) from Olympus (Tokyo, Japan). RFP-positive neurons were targeted for loose seal cell-attached recordings by using the two-photon fluorescence image. The recording pipette contained (in mM): 10 KCl, 140 K-gluconate, 10 HEPES, 2 MgCl2, 2 CaCl2, 0.05 sulphorhodamine 101, pH 7.25, 290 milli-osM. The signals were recorded using a patch clamp amplifier (Axoclamp 200B, Axon Instruments, Foster City, California). For the photostimulation, the objective was removed and a miniature blue high power LED (470 nm) was placed on the centre of the recording window. To determine the spike threshold, single light pulses of 1 ms or 10 ms duration were initially presented at maximal intensity. In addition, a series of longer pulses (20–100 ms) were tested for each recording without revealing any additional responsive neurons. All neurons in the analysis displayed evoked or spontaneous activity. If a neuron fired with a 1 ms pulse, the threshold was probed by presenting varying light intensity. The light pulses were presented every 15 s with increasing amplitude. Six to ten intensity values were repeated five times for each condition. Data collection and LED stimulation were controlled by custom-written physiology software in MATLAB (Mathworks, Natick, Massachusetts). All data of neurons responding to light durations above 1 ms were pooled for the analysis.

Training procedures

Mice with implanted LEDs had free access to food but had restricted access to drinking water to maintain 80–85% of their pre-training weight. Water was only available during and immediately after the behavioural sessions, with a minimum of 1.5 ml per day. Body weight was monitored daily before the training. The mice were kept at a reversed 12 h light/dark cycle and sessions were performed during the dark cycle. The behavioural box consisted of a white Plexiglas chamber (200 mm × 140 mm × 200 mm) with three ports mounted on one wall. The ports were conical and equipped with an infrared phototransistor–photodiode pair that signalled the interruption of the beam when the mouse entered his snout. The floor was a washable plastic kitchen cutting board. Both box and floor were cleaned with 70% ethanol after each animal. The box was placed in a sound- and light-proof cabinet that was constantly illuminated with bright white light. The box was covered with a transparent Plexiglas plate in which an infrared camera (for monitoring) and a bright masking light (high-power blue LED 470 nm, Luxeon V, Lumileds, San Jose, California) were mounted along the midline. The white light illumination of the behavioural box and the bright blue light flash were designed to mask any scattered light potentially reaching the retina through the skull or brain. The masking flash consisted of a series of bright 2 ms flashes (30 Hz, for 300 ms) illuminating the entire box. This mask flash was presented during every stimulation period (independently of whether a stimulus was presented).

Detection task

Training consisted of several phases. Transitions from one phase to the next were triggered by performance at 65% correct or above.

1. Mice were habituated to the behavioural box and trained for one week to get water from either left or right water port by breaking the light beam inside the port. The availability of water in a port was signalled with a white noise click from a loudspeaker (synchronized with the 5 ms valve opening). Water was delivered with a gravitational system and the drop size was controlled with solenoid valves (Neptune Research, West Caldwell, New Jersey). Single drop volume was approximately 4 µl. All components of the behaviour box were controlled with custom MATLAB software that was interfaced with a real-time processor system (RP2 or RM1, Tucker-Davis Technologies, Gainesville, Florida).

2. The snout of the mouse had to enter the centre port (trial initiation) to make water reward available in either the left or the right port for 10 s.

3. The LED was connected to the behavioural control system before the mouse was placed in the box. The cable to the LED controller ran through a hole in the Plexiglas cover and was mounted on a rotating hook 60 cm above the mouse. Reward was only available in the left port after photostimulation, whereas water was available on the right port in the absence of photostimulation (Fig. 2b). Stimuli were presented in an interleaved series of 20 trials, with each type occurring with a probability of 0.5 (pseudo-randomly, excluding runs of more than four consecutive identical trials). Occasional response bias for one port over the other was corrected by repeating the stimulus associated with the ignored port until the correct response was achieved.

4. Animals were trained to respond to fewer stimuli and decreased light intensity.


After completion of behavioural experiments, the mice were deeply anaesthetized and transcardially perfused with 4% paraformaldehyde in 0.1 M phosphate buffer, pH 7.4. The brain was carefully removed from each animal and cut into coronal or tangential sections ranging from 40 to 60 µm on a cryostat (CM 3050S, Leica Microsystems, Wetzlar, Germany). ChR2–GFP-positive neurons were detected with an anti-GFP polyclonal antibody (Rb-anti-GFP, AB3080P, dilution 1:700, Chemicon, Temecula, California), subsequently amplified with a biotin-conjugated antibody (Gt-anti-Rb, 111-066-003, Jackson ImmunoResearch Laboratories, West Grove, Pennsylvania) and revealed by using a standard ABC-kit (Vector Laboratories, Burlingame, California) with diaminobenzidine (DAB) precipitation. The sections were then mounted, dehydrated with increasing alcohol series, and coverslipped in DPX mounting medium. The localization of the cell bodies and serial reconstruction of the brain volume was performed in Neurolucida software (MBF Bioscience, Williston, Vermont). Estimates of the number of neurons under the window area were performed in MATLAB by calculating the number of neurons inside a sphere of diameter 2 mm positioned above the window centre. To define the degree of co-localization of ChR2–GFP and RFP, the ChR2–GFP was revealed with a fluorescent anti-GFP antibody (Rb-anti-GFP-488, A21311, Invitrogen, Carlsbad, California) whereas the cytosolic red signal was detected directly. Thick sections (40 &mgr;m) were imaged under a confocal microscope (LSM 510, Carl Zeiss, Jena, Germany). The individual red or green neurons were then classified and counted by hand on the image stacks using Object-Image software (http://simon.bio.uva.nl/object-image.html). Based on in vivo two-photon imaging or confocal and bright field microscopy of fixed brain sections, we could not detect any morphological changes due to the expression of ChR2 in layer 2/3 neurons.

ChR2-assisted photostimulation of layer 2/3 barrel cortex neurons in vivo.

a, Coronal section through the electroporated mouse somatosensory cortex after immunohistochemical staining for ChR2–GFP. b, Individual layer 2/3 neuron, side view. c, Maximum value projection (top view) of an image stack in vivo (see Supplementary Movie 1) showing layer 2/3 neurons expressing ChR2–GFP and cytosolic RFP. d, Schematic of the recording geometry. e, Action potentials recorded from one ChR2–GFP-positive neuron. Blue bars indicate photostimuli (1 ms duration, 11.6 mW mm-2, 20 Hz). f, Same as e, 50 Hz. g, Probability of spiking as a function of light intensity (1 ms duration, five repetitions per condition, 15 s between stimuli) (Imax = 11.6 mW mm-2). Each line corresponds to a different neuron, each colour to a different animal. Neurons that could only be driven with photostimuli longer than 1 ms were pooled at the far right (above). h, Cumulative fraction of recorded neurons firing at various threshold intensity levels (computed from the data in g).

Photostimulation in freely moving mice performing a detection task.

a, Schematic of the photostimulation setup (see Methods). b, Schematic of the behavioural apparatus and reward contingencies. The mouse initiates a trial by sticking its snout into the central port. Photostimuli are applied during a stimulation period (300 ms) accompanied by a series of bright blue light flashes delivered to the behavioural arena (30 Hz, 300 ms) to mask possible scattered light from the portable light source. The mouse then decides to enter either the left or the right port for a water reward. If a photostimulus was present, the choice of the left port was rewarded with a drop of water (hit, green star) whereas the choice of the right port lead to a short timeout (4 s, miss, red star). If the stimulus was absent, only the choice of the right port was rewarded with reward (correct reject, green circle) whereas the left port lead to a timeout (4 s, false alarm, red circle). c, Data from one session (200 trials) with a single stimulus (1 ms) with decreasing light intensities. Each horizontal line delineates 20 trials at fixed light intensity. Blue dots indicate the presence or absence of a photostimulus. Stimulated and non-stimulated trials were presented pseudo-randomly with a probability of 0.5.

Behavioural detection of photostimulation.

a, Comparison of the performance ((hits + correct rejections)/total trials) in mice expressing ChR2–GFP (n = 9) and control mice (n = 6) after training with five photostimuli (P < 0.001, t-test). b, Performance as a function of light intensity (as percentage of Imax = 11.6 mW mm-2) for five light pulses (1 ms, 20 Hz, blue lines), two light pulses (1 ms, 20 Hz, green lines) and a single light pulse (1 ms, red lines). Dotted lines: mean across five sessions (200–1000 trials per session). Error bars: binomial standard error. The number of ChR2–GFP-positive neurons located under the window area is indicated for each mouse. c, Location of ChR2-expressing neurons in serial reconstruction of the sectioned brain (coronal sections). The blue cone illustrates the light source over the window. Arrows indicate rostral (r) and dorsal (d) orientation. d, Performance as a function of the number of activated neurons. Thick lines, mean performance across all five animals for one (red), two (green) and five (blue) light pulses. Dotted lines indicate mean values of individual animals from Fig. 3b.

We thank B. Burbach, D. Flickinger, H. Kessels, D. O’Connor, T. Sato, R. Weimer and A. Zador for help with experiments, and D. O’Connor for comments on the manuscript. This work was supported by the Swiss National Science Foundation (to D.H.), the National Institutes of Health and the Howard Hughes Medical Institute.

Author Contributions D.H. and K.S. designed the experiments. D.H. performed the behavioral and in vivo physiological experiments. L.P., D.H. and K.S. performed the brain slice measurements. N.G. performed histology. S.R., T.H., Z.M. and K.S. provided advice and equipment. D.H. and K.S. wrote the paper. All authors discussed the results and commented on the manuscript.

Penfield, W. & Boldery, P. Somatic motor and sensory representation in the cerebral cortex of man as studied by electrical stimulation. Brain 60, 389443 (1937) Salzman, C. D., Britten, K. H. & Newsome, W. T. Cortical microstimulation influences perceptual judgements of motion direction. Nature 346, 174177 (1990) Romo, R., Hernandez, A., Zainos, A. & Salinas, E. Somatosensory discrimination based on cortical microstimulation. Nature 392, 387390 (1998) Libet, B. in Handbook of Sensory Physiology (ed. Iggo, A.) 743790 (Springer, Berlin, 1973) Leal-Campanario, R., Delgado-Garcia, J. M. & Gruart, A. Microstimulation of the somatosensory cortex can substitute for vibrissa stimulation during Pavlovian conditioning. Proc. Natl Acad. Sci. USA 103, 1005210057 (2006) Butovas, S. & Schwarz, C. Detection psychophysics of intracortical microstimulation in rat primary somatosensory cortex. Eur. J. Neurosci. 25, 21612169 (2007) Tehovnik, E. J. Electrical stimulation of neural tissue to evoke behavioral responses. J. Neurosci. Methods 65, 117 (1996) Ranck, J. B. Which elements are excited in electrical stimulation of mammalian central nervous system: a review. Brain Res. 98, 417440 (1975) Nagel, G. et al. Channelrhodopsin-2, a directly light-gated cation-selective membrane channel. Proc. Natl Acad. Sci. USA 100, 1394013945 (2003) Boyden, E. S., Zhang, F., Bamberg, E., Nagel, G. & Deisseroth, K. Millisecond-timescale, genetically targeted optical control of neural activity. Nature Neurosci. 8, 12631268 (2005) Li, X. et al. Fast noninvasive activation and inhibition of neural and network activity by vertebrate rhodopsin and green algae channelrhodopsin. Proc. Natl Acad. Sci. USA 102, 1781617821 (2005) Ishizuka, T., Kakuda, M., Araki, R. & Yawo, H. Kinetic evaluation of photosensitivity in genetically engineered neurons expressing green algae light-gated channels. Neurosci. Res. 54, 8594 (2006) Bi, A. et al. Ectopic expression of a microbial-type rhodopsin restores visual responses in mice with photoreceptor degeneration. Neuron 50, 2333 (2006) Hatanaka, Y., Hisanaga, S., Heizmann, C. W. & Murakami, F. Distinct migratory behavior of early- and late-born neurons derived from the cortical ventricular zone. J. Comp. Neurol. 479, 114 (2004) Petreanu, L., Huber, D., Sobczyk, A. & Svoboda, K. Channelrhodopsin-2-assisted circuit mapping of long-range callosal projections. Nature Neurosci. 10, 663668 (2007) Margrie, T. W. et al. Targeted whole-cell recordings in the mammalian brain in vivo. Neuron 39, 911918 (2003) Fee, M. S., Mitra, P. P. & Kleinfeld, D. Central versus peripheral determinants of patterned spike activity in rat vibrissa cortex during whisking. J. Neurophysiol. 78, 11441149 (1997) Arenkiel, B. R. et al. In vivo light-induced activation of neural circuitry in transgenic mice expressing channelrhodopsin-2. Neuron 54, 205218 (2007) Gray, N. W., Weimer, R. M., Bureau, I. & Svoboda, K. Rapid redistribution of synaptic PSD-95 in the neocortex in vivo. PLoS Biol. 4, e370 (2006) DeWeese, M. R., Wehr, M. & Zador, A. M. Binary spiking in auditory cortex. J. Neurosci. 23, 79407949 (2003) Petersen, R. S., Panzeri, S. & Diamond, M. E. Population coding in somatosensory cortex. Curr. Opin. Neurobiol. 12, 441447 (2002) Ferezou, I., Bolea, S. & Petersen, C. C. Visualizing the cortical representation of whisker touch: voltage-sensitive dye imaging in freely moving mice. Neuron 50, 617629 (2006) Zhang, Y. P. & Oertner, T. G. Optical induction of synaptic plasticity using a light-sensitive channel. Nature Methods 4, 139141 (2006) Nagel, G. et al. Light activation of channelrhodopsin-2 in excitable cells of Caenorhabditis elegans triggers rapid behavioral responses. Curr. Biol. 15, 22792284 (2005) Wang, H. et al. High-speed mapping of synaptic connectivity using photostimulation in channelrhodopsin-2 transgenic mice. Proc. Natl Acad. Sci. USA 104, 81438148 (2007) Schroll, C. et al. Light-induced activation of distinct modulatory neurons triggers appetitive or aversive learning in Drosophila larvae. Curr. Biol. 16, 17411747 (2006) Lima, S. Q. & Miesenbock, G. Remote control of behavior through genetically targeted photostimulation of neurons. Cell 121, 141152 (2005) Salzman, C. D., Murasugi, C. M., Britten, K. H. & Newsome, W. T. Microstimulation in visual area MT: effects on direction discrimination performance. J. Neurosci. 12, 23312355 (1992) Tehovnik, E. J., Tolias, A. S., Sultan, F., Slocum, W. M. & Logothetis, N. K. Direct and indirect activation of cortical neurons by electrical microstimulation. J. Neurophysiol. 96, 512521 (2006) Pologruto, T. A., Sabatini, B. L. & Svoboda, K. ScanImage: flexible software for operating laser-scanning microscopes. Biomed. Eng. Online 2, 13 (2003)
doi: 10.1038/nature06445

Behavioural report of single neuron stimulation in somatosensory cortex p.65

version="1.0" encoding="utf-8"?> ]>
Nature 451 7174 20080103 65680 0028-0836 1476-4687 2007Nature Publishing Group Supplementary Information

This file contains Supplementary Figures S1-S8 with Legends illustrating additional analysis of single cell stimulation experiments in somatosensory cortex.

Behavioural report of single neuron stimulation in somatosensory cortex Arthur R.HouwelingA R MichaelBrechtM Bernstein Center for Computational Neuroscience and Humboldt University Berlin, Philippstrasse 13, House 6, 10115 Berlin, Germany Department of Neuroscience, Erasmus Medical Center, PO Box 2040, 3000 CA, Rotterdam, The Netherlands Correspondence and requests for materials should be addressed to A.R.H. (arthur.houweling@bccn-berlin.de) or M.B. (michael.brecht@bccn-berlin.de). &nature06447-s1;

Understanding how neural activity in sensory cortices relates to perception is a central theme of neuroscience. Action potentials of sensory cortical neurons can be strongly correlated to properties of sensory stimuli and reflect the subjective judgements of an individual about stimuli. Microstimulation experiments have established a direct link from sensory activity to behaviour, suggesting that small neuronal populations can influence sensory decisions. However, microstimulation does not allow identification and quantification of the stimulated cellular elements. The sensory impact of individual cortical neurons therefore remains unknown. Here we show that stimulation of single neurons in somatosensory cortex affects behavioural responses in a detection task. We trained rats to respond to microstimulation of barrel cortex at low current intensities. We then initiated short trains of action potentials in single neurons by juxtacellular stimulation. Animals responded significantly more often in single-cell stimulation trials than in catch trials without stimulation. Stimulation effects varied greatly between cells, and on average in 5% of trials a response was induced. Whereas stimulation of putative excitatory neurons led to weak biases towards responding, stimulation of putative inhibitory neurons led to more variable and stronger sensory effects. Reaction times for single-cell stimulation were long and variable. Our results demonstrate that single neuron activity can cause a change in the animal’s detection behaviour, suggesting a much sparser cortical code for sensations than previously anticipated.

Based on its volume and density of neurons, rat somatosensory cortex contains an estimated two million neurons. The detection of single-cell stimulation might therefore be a difficult task, and we adopted a behavioural paradigm designed for observing single-cell effects (Fig. 1a). We first trained animals to report short (200 ms) trains of microstimulation pulses. Stimulation of somatosensory cortex evokes tactile sensations in humans, and animal studies have demonstrated an interchangeability of tactile stimuli and cortical microstimulation. We mainly targeted deep cortical layers, where microstimulation detection thresholds are lowest in rat barrel cortex. Tongue lick responses were rewarded with a drop of water and counted as a hit if a lick occurred within 100–1200 ms from stimulus onset (Fig. 1b). Animals typically learned this microstimulation report task in a single session and detection thresholds decreased to 2–5 µA within days (Supplementary Fig. 1). These values are comparable to the lowest cortical microstimulation detection thresholds reported in humans and animals. To be able to detect potentially weak effects of single-cell stimulation, we encouraged guessing (a non-conservative response criterion) by introducing only mild negative reinforcement (a 1.5 s time-out) for false-positive responses (licks without preceding stimulation).

Once animals responded consistently to low microstimulation currents, we approached a cortical neuron closely with a glass pipette and evoked short (200 ms) trains of action potentials by juxtacellular stimulation, a technique developed to label individual neurons. Juxtacellular stimulation currents (3–43 nA, mean 12.6 nA) strongly modulated action potential firing in barrel cortex neurons (Fig. 1c, d). On average, we evoked 14.2 ± 6.5 (s.d.) action potentials during current injection, a 25-fold increase over the average spontaneous firing rate. The close apposition of neuron and pipette in the juxtacellular configuration in behaving animals typically resulted in short experimental sessions per cell. Microstimulation in the vicinity of the neuron (at an average distance of about 75 µm) was adjusted such that current intensities were close to the animal’s detection threshold, resulting in an average hit (detection) rate of 75%. The action potential firing of most cells was affected during and after microstimulation (Supplementary Fig. 2).

Microstimulation and single-cell stimulation trials were randomly interleaved with ‘catch’ trials (with no or subthreshold current injection) (Fig. 1b). In paradigms with random stimulus presentation times, catch trials can be used to estimate chance performance and to guard against inadvertent cues. To assess single-cell effects when the animal was attentive, we confined our analysis to those single-cell stimulation and catch trials flanked by correct microstimulation responses.

Figure 2 shows an experiment on a regular spiking layer 5b pyramidal neuron with a slender apical dendrite (Fig. 2a). Juxtacellular stimulation evoked on average 9.1 action potentials during the current injection (Fig. 2b top). Lick responses (red squares) occurred mainly after single-cell stimulation (Fig. 2b top) and microstimulation (Fig. 2b bottom), but only once after no stimulation catch trials (Fig. 2b middle). Quantification of responses (Fig. 2c) suggests that the animal reported single pyramidal cell activity. Even though this neuron was one of the cells with the strongest behavioural effects, this effect was not significant on the single neuron level (P = 0.099, Fisher’s exact test). This is not unexpected given the limited number of trials (see Methods).

A population analysis revealed, however, that single-cell stimulation biased animals towards responding. Figure 3a shows, for 51 neurons, that animals responded significantly more often in single-cell stimulation trials (mean hit rate 22.0%) than in no-current-injection catch trials (mean false-positive rate 17.9%; P = 0.022). To test if single-cell detection was dependent on the firing of the stimulated neuron rather than on inadvertent cues associated with the current injection, we stimulated a further set of 19 neurons. We applied single-cell stimulation as usual, but instead of the no-current catch trials we presented subthreshold (10% of the single-cell stimulation current) catch trials. Subthreshold current injections activated neurons only weakly or not at all. Animals also responded significantly more often in single-cell stimulation trials than in subthreshold catch trials (Fig. 3b; mean hit rate 27.4%; mean false-positive rate 20.6%; P = 0.019). Stimulation effects were distributed evenly across animals (Supplementary Fig. 3). Having verified that single-cell stimulation led to significant biases in two independent sets of neurons (Fig. 3a, b), we wanted to confirm that this effect did not result from the inadvertent stimulation of neighbouring neurons or other nonspecific effects. Thus, we injected current (25 nA, twice the average current applied with juxtacellular stimulation) through the pipette into extracellular space (instead of applying it to a neuron). These control experiments showed that animals did not report current injection into extracellular space (Fig. 3c; n = 90; mean hit rate 18.7%; mean false-positive rate 19.0%; P = 0.598). To test if single-cell stimulation effects (Fig. 3a, b) were different from those of extracellular current injection (Fig. 3c), we compared effect size (hit rate - catch trial response rate) across those two data sets and observed a significant difference (P = 0.008). Finally, we tested if single-cell stimulation effects were specific to the attended (and trained) cortical area. As before, microstimulation was applied to barrel cortex, but we now stimulated single neurons in visual cortex. Animals did not report single-cell stimulation in visual cortex (Fig. 3d; n = 21; mean hit rate 21.2%; mean false-positive rate 20.1%; P = 0.319), suggesting that stimulation effects are specific to the attended cortical area.

Further observations show that the animals’ responses were caused by the stimulation of single and not multiple neurons. (1) Juxtacellular stimulation currents were approximately three orders of magnitude lower (3–43 nA) than those required for evoking motor or sensory responses with microstimulation (2–200 µA). (2) Although we occasionally observed the inadvertent stimulation of a second neuron by the appearance of a second large action potential waveform in our recordings, such inadvertent stimulation was rare (accounting for only about 1% of evoked action potentials across experiments; Supplementary Fig. 4). All results presented here were also significant when single-cell stimulation trials with secondary action potentials were excluded. (3) Firing rates of more distant cells (with action potentials less than 0.5 mV) were not modulated (Supplementary Fig. 5). (4) Juxtacellular labelling typically fills single neurons.

Because microstimulation in barrel cortex can evoke whisker movements, we combined stimulation experiments with whisker tracking to assess if rats sense single-cell stimulation indirectly by detecting movements. Our data argue against such an indirect mechanism: near detection threshold microstimulation did not evoke movements even though it was reported (Supplementary Fig. 6a) and single-cell stimulation did not evoke whisker movements (Supplementary Fig. 6b).

The bias towards responding evoked by single-cell stimulation was weak on average (approximate 5% effect size: single-cell stimulation hit rate – catch trial response rate). As illustrated in Supplementary Fig. 7, the strength of the effect depended greatly on the animal’s overall response rate. When animals were conservative (low response rate) the effect was small (and not significant); at high response rates, however, the effect size was about 9% (and significant). Most interestingly, the variance of stimulation-evoked sensory effects was greater for putative interneurons than putative excitatory cells (P = 0.0023, see Methods), which we distinguished based on spike width and firing pattern (Supplementary Fig. 8). Whereas stimulation of putative excitatory neurons led to weak but significant biases towards responding, stimulation of putative inhibitory neurons led to stronger and more variable sensory effects (Fig. 3e). In particular, in 3 out of 11 putative interneurons the effect was stronger than in any of the 59 putative excitatory cells or 90 control experiments. In two of these three putative interneurons most hits were observed for trials in which the evoked firing rates were not higher than the population average, suggesting that interneuron action potentials are in some cases more readily detected than action potentials of excitatory cells.

Reaction times for single-cell stimulation were long and variable (Figs 2 and 4a, b) compared with microstimulation responses (Fig. 4c). Although effect size could be considerable in individual cells (Fig. 2), we did not observe single-cell responses with close to 100% hit rates and short reaction times as observed often in microstimulation trials. Thus, single cells never led to a strong, perceptually saturating signal. Microstimulation at 2–8 µA presumably activates multiple neurons, which may account for the difference between single-cell stimulation and microstimulation. Even smaller currents (less than 2 µA) are known to activate cortical neurons. It remains to be seen if such currents lead to a microstimulation performance comparable to that of single-cell stimulation.

The combination of single-cell stimulation and control experiments shows that the activity of single sensory cortical neurons can lead to a behaviourally reportable effect. It has been estimated that a single barrel cortical column contains approximately 8,500 excitatory cells that generate about 1,550 spontaneous action potentials in a 200 ms period and about 4,000 action potentials in response to a small (3.3°) whisker deflection that is close (about 60% hit rate) to the animal’s detection threshold. Given these numbers it is surprising that adding approximately 14 action potentials over 200 ms in a neuron is detectable. Other measurements suggest lower rates of ongoing and sensory-evoked cortical action potential activity. The detectability of single-cell stimulation might therefore be related to the sparseness of cortical activity. A single cortical pyramidal cell connects to several thousand postsynaptic neurons, but most of these connections are weak. Likewise, a single inhibitory neuron connects to thousands of local neurons. Depending on ongoing membrane potential fluctuations, variable sets of postsynaptic cells may become activated or suppressed, which might contribute to the variable reaction times. Some models of sensory decision making contain a temporal integration step during which sensory evidence is accumulated. It is conceivable that the weak single-cell signals require longer temporal integration, thereby contributing to the long reaction times. The mechanisms that hold the sensory information between single-cell stimulation and reaction remain to be determined.

Our finding that stimulation of putative interneurons (which project locally) can lead to strong sensory effects suggests that (1) local circuits are involved in the read-out of single-cell activity and that (2) read-out mechanisms are sensitive to suppression of action potentials. Cortical microstimulation evokes pronounced and long-lasting inhibition (Fig. 2 and Supplementary Fig. 2). Thus, our animals might have been trained to detect the suppression of activity. The present results, with the classic single afferent stimulation experiments by Vallbo and colleagues and single-cell stimulation experiments in rat motor cortex, demonstrate the behavioural relevance of single neuron activity. These studies establish a reverse physiology approach in which one analyses responses to cellular activity rather than cellular responses as in classical physiology. In further studies it should be possible to establish how the frequency and number of action potentials are related to the evoked sensations.

Methods Summary

Animals were trained to report microstimulation (40 cathodal pulses at 200 Hz, 0.3 ms pulse duration) applied to the barrel cortex through a tungsten microelectrode at a depth of 1,500 µm. In the first training session, current intensities no greater than 200 µA were applied; subsequently current intensity was decreased according to detection performance. During training animals were put on a water restriction schedule with daily access to water ad libitum for one hour after the experiment. Once the animal performed at detection thresholds no greater than 5 µA in at least one block of trials on two consecutive days, we switched to the single-cell stimulation report task; here microstimulation currents were on average 5.0 ± 1.6 µA. To stimulate single neurons close to the microstimulation site, we glued a tungsten microelectrode close to the tip of a glass pipette (average tip separation approximately 75 µm). The construct was inserted through the intact dura to a mean depth of 1,400 ± 271 µm, whereby the actual depth from the cortical surface was less because of dimpling and oblique penetrations. From the histologically identified neurons it appears that most single-cell stimulation experiments were performed in cortical layers 4, 5A and 5B. Cells were classified as fast spiking neurons (putative interneurons) if the action potential width was no greater than 0.4 ms (peak to trough) and/or if they responded with at least 50 action potentials (that is, at least 250 Hz) during at least one 200 ms current injection (see Supplementary Fig. 8). The remainder of cells were classified as non-fast spiking (putative excitatory) cells.

Experimental procedures

We used standard surgical and electrophysiological techniques to prepare animals (n = 15 Wistar rats, about P35 at the day of surgery) for chronic, head-fixed recording of the barrel cortex (P3, L5 relative to bregma). Glass pipettes for single-cell stimulation were filled either with intracellular solution containing (in mM): K-gluconate 135, HEPES 10, Na2-phosphocreatine 10, KCl 4, MgATP 4, and Na3GTP 0.3 (pH 7.2), or Ringer’s solution containing NaCl 135, KCl 5.4, HEPES 5, CaCl2 1.8, and MgCl2 1 (pH 7.2). Penetrations were targeted to the E, D and C whisker row representation of barrel cortex. All experimental procedures were performed according to Dutch and German guidelines on animal welfare under the supervision of local ethics committees.


Cells were included if at least five catch trials and five single-cell stimulation trials had been presented that satisfied our inclusion criterion. All single-cell stimulation and catch trials were included for which the animal responded to both the preceding and the succeeding microstimulation. We also included all trials where the animal responded only to either the immediately preceding or immediately succeeding microstimulation. All numbers on single-cell stimulation experiments refer to these included trials. For our barrel cortex experiments an average of 30.2 single-cell stimulation trials and 17.7 catch trials were included per cell. Applying the inclusion criterion led to the exclusion of many trials with low response rate (on average 14.0 single-cell stimulation trials and 7.7 catch trials). When all trials were considered, animals still responded significantly more often in single-cell stimulation trials than in catch trials (data not shown). From these trial numbers it becomes clear that it is difficult to obtain a significant result on the single neuron level. Monte Carlo simulations show that approximately 480 single-cell stimulation trials (about 16 times more than our average) are required to obtain a significant outcome with 90% probability (one-sided binomial test by normal approximation, &agr; = 0.05) assuming a 10% effect size (30% hits, 20% false positives).

As we trained animals to report stimulation of the barrel cortex, we tested the prediction that single-cell stimulation led to responses (hits). Thus, differences between hit rates and false-positive rates were evaluated by using a one-sided, paired t-test, and differences between single-cell stimulation experiments and control experiments by using a one-sided, unpaired t-test. However, all results presented here were also significant when applying two-sided t-tests or when using Monte Carlo simulations of the statistical distributions. To test non-parametrically if the variance of sensory effects was greater for putative interneurons than excitatory neurons (see Fig. 3e), we ranked all neurons on effect size and computed the variance of interneuron ranks and compared this with a Monte Carlo distribution based on random permutations of the ranks.

It is our impression that the interneuron inclusion criteria (action potential width no greater than 0.4 ms and/or a response of at least 50 action potentials during at least one 200 ms current injection) were conservative and may have led to the false classification of potential interneurons as excitatory cells. The difference in sensory effects between putative interneurons and excitatory cells persisted when more inclusive interneuron inclusion criteria were chosen (data not shown). Further observations support our classification scheme. First, morphologically identified excitatory neurons were correctly classified as putative excitatory cells (Supplementary Fig. 8). Second, maximum discharge patterns of putative excitatory cells showed significantly more irregular spike intervals than those of putative interneurons (P = 0.037, two-sided t-test on the coefficient of variation), which showed very little or no accommodation (Supplementary Fig. 8) in agreement with in vitro studies. Spontaneous firing rates (measured in a 30 s period at the onset of each experiment before any current injections) were similar between putative interneurons (mean 2.1 Hz) and putative excitatory neurons (1.6 Hz).


On the last day of experiments with an animal, we included biocytin (1.5%) in the stimulation pipette and processed brains as described previously. Even though combining cell recovery and behavioural measurements was difficult, we recovered ten neurons or fragments of neurons. Three of these entered our data set (satisfied the trial number inclusion criterion) and were identified as excitatory neurons (see Supplementary Fig. 8). Combining cell recovery and behavioural measurements is difficult because one is limited to a single session per animal, and even aborted single-cell stimulation attempts can stain neurons, which complicates neuron identification. We evaluated the tissue damage by electrode constructs in 8 out of 15 animal brains. In these brains the damage to the barrel cortex target area was rated as either minimal or weak.

Behavioural setup and single-cell stimulation.

a, Stimulation experiments were performed in the barrel cortex of awake rats. Animals responded to stimulation by interrupting a light beam (dashed line) with multiple tongue licks. The time of the first lick was taken as the reaction time and reward was delivered for correct responses (right). Top, single-cell stimulation pipette with stimulation current wave form (upper) and tungsten microelectrode with stimulation pulse train (lower). b, Three types of stimulus were presented at random intervals (Poisson process, mean 3 s): microstimulation (2–8 &mgr;A) (40% probability), juxtacellular single-cell stimulation (40%) and no (or subthreshold) current injection ‘catch’ trials (20%). Licks within the interstimulus interval led to an additional 1.5 s delay to presentation of the next stimulus (left box) and were rewarded after a stimulus (right box) for all three trial types. c, Single-cell stimulation trial by juxtacellular current injection. Triangles indicate stimulation onset and offset artefacts. d, Evoked action potentials (open circles) in a series of stimulation trials. Spontaneous action potentials (solid circles) were quantified for 1 s before each stimulation. The left y axis label applies to both spontaneous and evoked action potentials; the right y axis label applies to evoked action potentials.

Behavioural responses to stimulation of a single layer 5b pyramidal neuron.

a, Reconstruction of the stimulated neuron with dendritic tree (red) and axon (blue, incompletely filled). Superimposed is a micrograph of a stimulation pipette and a tungsten microstimulation electrode aligned along the electrode track. Barrel rows (brown) are labelled with letters. L, layer; WM, white matter. b, Action potential (ticks) raster plots and first lick responses (red squares) during juxtacellular single-cell stimulation trials (top), no-current-injection catch trials (middle) and 19 randomly selected microstimulation trials (bottom). The neuron was inhibited during and after microstimulation (stimulation current, 4 µA). c, Quantification of responses to single-cell stimulation, catch trials and microstimulation.

Initiation of action potentials in single barrel cortex neurons causes biases towards responding.

a, Response rates for single-cell stimulation trials (hits) versus no-current-injection catch trials (false positives) (n = 51 neurons; note several points coincide). Fast spiking, putative interneurons (filled circles); non-fast spiking, putative excitatory neurons (empty circles). b, Response rates for single-cell stimulation trials (hits) versus subthreshold current injection catch trials (n = 19 neurons). Conventions as in a. c, Response rates for trials in which we applied 25 nA (twice the average juxtacellular stimulation current) into extracellular space (hits) versus no-current-injection catch trials (false positives) (n = 90 stimulation sites). d, Response rates for single-cell stimulation trials (hits) versus no-current-injection catch trials (false positives) for visual cortex neurons (n = 21), while microstimulation was applied in barrel cortex. Animals had been trained to report microstimulation in barrel cortex. e, Distribution of sensory effects (single-cell stimulation hit rate - catch trial response rate) across putative interneurons and putative excitatory neurons in barrel cortex single-cell stimulation experiments (conventions as in a).

Reaction times for single-cell stimulation are long and variable compared with microstimulation responses.

a, Cumulative distribution of reaction times for microstimulation (dashed), single-cell stimulation (solid) and catch trials (dotted). b, Difference of the cumulative distributions of reaction times for single-cell stimulation and catch trials. This isolates the contribution of single-cell stimulation from false-positive responses. The vertical line marks the time where 50% of the peak difference is reached; the grey area marks the time from 25% to 75% of the peak difference. c, Difference of the cumulative distributions of reaction times for microstimulation and catch trials. Conventions as in b.

We thank B. Sakmann for suggesting the juxtacellular stimulation approach, J. van der Burg, E. Haasdijk and G. Maas for technical contributions, P. den Iseger and A. Lee for discussions, and G. Borst, M. Frens, C. Hansel, L. Herfst, A. Lee, B. Voigt and J. Wolfe for comments on the manuscript. This work was supported by the Bernstein Center for Computational Neuroscience and Humboldt University Berlin, Erasmus MC, and VIDI (NWO) and HFSP grants to M.B.

Rieke, F., Warland, D., de Ruyter Van Steveninck, R. R. & Bialek, W. Spikes: Exploring the Neural Code (MIT Press, Cambridge, Massachusetts, 1997) Britten, K. H., Newsome, W. T., Shadlen, M. N., Celebrini, S. & Movshon, J. A. A relationship between behavioral choice and the visual responses of neurons in macaque MT. Vis. Neurosci. 13, 87100 (1996) Afraz, S. R., Kiani, R. & Esteky, H. Microstimulation of inferotemporal cortex influences face categorization. Nature 442, 692695 (2006) Salzman, C. D., Britten, K. H. & Newsome, W. T. Cortical microstimulation influences perceptual judgements of motion direction. Nature 346, 174177 (1990) Murasugi, C. M., Salzman, C. D. & Newsome, W. T. Microstimulation in visual area MT: effects of varying pulse amplitude and frequency. J. Neurosci. 13, 17191729 (1993) Tehovnik, E. J. Electrical stimulation of neural tissue to evoke behavioral responses. J. Neurosci. Methods 65, 117 (1996) Zilles, K. & Wree, A. in The Rat Nervous System (ed. Paxinos, G.) 649685 (Academic Press, London, 1995) Beaulieu, C. Numerical data on neocortical neurons in adult rat, with special reference to the GABA population. Brain Res. 609, 284292 (1993) Penfield, W. & Rasmussen, T. The Cerebral Cortex of Man: A Clinical Study of Localization of Function (Macmillan, New York, 1950) Romo, R., Hernandez, A., Zainos, A. & Salinas, E. Somatosensory discrimination based on cortical microstimulation. Nature 392, 387390 (1998) Krauss, A., Manns, I. D. & Brecht, M. in Annu. Meeting Soc. Neurosci. 58.19. (2003) Schmidt, E. M. et al. Feasibility of a visual prosthesis for the blind based on intracortical microstimulation of the visual cortex. Brain 119, 507522 (1996) Bartlett, J. R. & Doty, R. W. An exploration of the ability of macaques to detect microstimulation of striate cortex. Acta Neurobiol. Exp. (Wars.) 40, 713727 (1980) Pinault, D. A novel single-cell staining procedure performed in vivo under electrophysiological control: morpho-functional features of juxtacellularly labeled thalamic cells and other central neurons with biocytin or Neurobiotin. J. Neurosci. Methods 65, 113136 (1996) Talwar, S. K. & Gerstein, G. L. Auditory frequency discrimination in the white rat. Hear. Res. 126, 135150 (1998) Swadlow, H. A. Monitoring the excitability of neocortical efferent neurons to direct activation by extracellular current pulses. J. Neurophysiol. 68, 605619 (1992) de Kock, C. P., Bruno, R. M., Spors, H. & Sakmann, B. Layer- and cell-type-specific suprathreshold stimulus representation in rat primary somatosensory cortex. J. Physiol. (Lond.) 581, 139154 (2007) Stuttgen, M. C., Ruter, J. & Schwarz, C. Two psychophysical channels of whisker deflection in rats align with two neuronal classes of primary afferents. J. Neurosci. 26, 79337941 (2006) Kerr, J. N., Greenberg, D. & Helmchen, F. Imaging input and output of neocortical networks in vivo. Proc. Natl Acad. Sci. USA 102, 1406314068 (2005) Margrie, T. W., Brecht, M. & Sakmann, B. In vivo, low-resistance, whole-cell recordings from neurons in the anaesthetized and awake mammalian brain. Pflugers Arch. 444, 491498 (2002) Waters, J. & Helmchen, F. Background synaptic activity is sparse in neocortex. J. Neurosci. 26, 82678277 (2006) Brecht, M. & Sakmann, B. Dynamic representation of whisker deflection by synaptic potentials in spiny stellate and pyramidal cells in the barrels and septa of layer 4 rat somatosensory cortex. J. Physiol. (Lond.) 543, 4970 (2002) Moore, C. I. & Nelson, S. B. Spatio-temporal subthreshold receptive fields in the vibrissa representation of rat primary somatosensory cortex. J. Neurophysiol. 80, 28822892 (1998) Olshausen, B. A. & Field, D. J. Sparse coding of sensory inputs. Curr. Opin. Neurobiol. 14, 481487 (2004) Warren, R. A. & Dykes, R. W. Population analysis of single neurons in cat somatosensory cortex. Somatosens. Mot. Res. 9, 297312 (1992) DeFelipe, J. & Farinas, I. The pyramidal neuron of the cerebral cortex: morphological and chemical characteristics of the synaptic inputs. Prog. Neurobiol. 39, 563607 (1992) Song, S., Sjostrom, P. J., Reigl, M., Nelson, S. & Chklovskii, D. B. Highly nonrandom features of synaptic connectivity in local cortical circuits. PLoS Biol. 3, e68 (2005) Gold, J. I. & Shadlen, M. N. Representation of a perceptual decision in developing oculomotor commands. Nature 404, 390394 (2000) Vallbo, A. B., Olsson, K. A., Westberg, K. G. & Clark, F. J. Microstimulation of single tactile afferents from the human hand. Sensory attributes related to unit type and properties of receptive fields. Brain 107, 727749 (1984) Brecht, M., Schneider, M., Sakmann, B. & Margrie, T. W. Whisker movements evoked by stimulation of single pyramidal cells in rat motor cortex. Nature 427, 704710 (2004) Connors, B. W. & Gutnick, M. J. Intrinsic firing patterns of diverse neocortical neurons. Trends Neurosci. 13, 99104 (1990)
doi: 10.1038/nature06447

TRPC channel activation by extracellular thioredoxin p.69

version="1.0" encoding="utf-8"?> ]>
Nature 451 7174 20080103 69724 0028-0836 1476-4687 2007Nature Publishing Group Supplementary Information

The file contains Supplementary Methods, Supplementary Figures 1-14 with Legends, Supplementary Results and Discussion and additional references.

TRPC channel activation by extracellular thioredoxin Shang-ZhongXuS PiruthiviSukumarP FanningZengF JingLiJ AmitJairamanA AnneEnglishA JacquelineNaylorJ CozianaCiurtinC YasserMajeedY Carol J.MilliganC J Yahya M.BahnasiY M EmanAl-ShawafE Karen E.PorterK E Lin-HuaJiangL PaulEmeryP AsipuSivaprasadaraoA David J.BeechD J Institute of Membrane and Systems Biology, Garstang Building, Faculty of Biological Sciences, and, School of Medicine, University of Leeds, Leeds LS2 9JT, UK Academic Unit of Musculoskeletal Disease, Chapel Allerton Hospital, Leeds LS7 4SA, UK These authors contributed equally to this work. Present address: Postgraduate Medical Institute & Hull York Medical School, University of Hull, Cottingham Road, Hull HU6 7RX, UK. Correspondence and requests for materials should be addressed to D.J.B. (d.j.beech@leeds.ac.uk). &nature06414-s1;

Mammalian homologues of Drosophila melanogaster transient receptor potential (TRP) are a large family of multimeric cation channels that act, or putatively act, as sensors of one or more chemical factor. Major research objectives are the identification of endogenous activators and the determination of cellular and tissue functions of these channels. Here we show the activation of TRPC5 (canonical TRP 5) homomultimeric and TRPC5–TRPC1 heteromultimeric channels by extracellular reduced thioredoxin, which acts by breaking a disulphide bridge in the predicted extracellular loop adjacent to the ion-selectivity filter of TRPC5. Thioredoxin is an endogenous redox protein with established intracellular functions, but it is also secreted and its extracellular targets are largely unknown. Particularly high extracellular concentrations of thioredoxin are apparent in rheumatoid arthritis, an inflammatory joint disease that disables millions of people worldwide. We show that TRPC5 and TRPC1 are expressed in secretory fibroblast-like synoviocytes from patients with rheumatoid arthritis, that endogenous TRPC5–TRPC1 channels of the cells are activated by reduced thioredoxin, and that blockade of the channels enhances secretory activity and prevents the suppression of secretion by thioredoxin. The data indicate the presence of a previously unrecognized ion-channel activation mechanism that couples extracellular thioredoxin to cell function.

TRPC5 is markedly activated by extracellular lanthanide ions. The effects of these ions depend on a glutamic acid residue at position 543 (ref. 14) in the predicted extracellular loop adjacent to the ion pore (Supplementary Figs 1 and 2). This structural feature may therefore have functional importance in enabling extracellular factors to activate the channels. Because lanthanides are unlikely to be physiological activators, we were interested in alternatives and developed a hypothesis based on amino acid sequence alignment, which showed two cysteine residues near glutamic acid 543 that are conserved in TRPC5, TRPC4 and TRPC1 (Supplementary Fig. 2), a subset of the seven TRPC channels. TRPC5 and TRPC4 have similar functional properties and both form heteromultimers with TRPC1 (refs 3–5), a subunit that has weak targeting to the plasma membrane when expressed in isolation.

Pairs of cysteine residues may be covalently linked by a disulphide bridge that can be cleaved by reduction. We therefore applied the chemical reducing agent dithiothreitol (DTT) to HEK-293 cells expressing TRPC5 (refs 15, 16). There was channel activation with the characteristic current–voltage (IV) relationship of TRPC5 and blocking by 2-aminoethoxydiphenyl borate (2-APB), an inhibitor of TRPC5 (ref. 5) (Fig. 1a, b, d). Current recovered on wash-out of DTT (data not shown). Similarly, the membrane-impermeable disulphide reducing agent Tris (2-carboxyethyl) phosphine hydrochloride (TCEP; Fig. 1c, d) activated TRPC5, whereas the thiol reagent [2-(trimethylammonium) ethyl]methanethiosulphonate bromide (MTSET) had no effect (Fig. 1d). TRPC5 was inhibited by cadmium ions only after pretreatment with DTT (Fig. 1e, f), which is consistent with the metal ions acting by re-engaging cysteine residues. Other TRP channels lacking the cysteine pair in a similar position were unresponsive to DTT (Supplementary Figs 2 and 3). The data support the hypothesis that the cysteine pair in TRPC5 normally engages in a disulphide bridge that constrains the channel in a state of limited opening probability, enabling enhanced channel activity when the bridge is broken.

To test the hypothesis further, we expressed TRPC5 mutants containing alanine in place of cysteine. Such mutants were constitutively active and were not stimulated by reducing agent or lanthanide (Fig. 1g and Supplementary Figs 4 and 5). Ionic currents for the single mutants (C553A, C553S or C558A) and double mutant (C553A + C558A) were not significantly different, suggesting that the two cysteine residues have a joint role (Fig. 1g). Expression of wild-type TRPC1 together with the TRPC5 double mutant led to smaller constitutive currents that were not affected by DTT or lanthanide, which is consistent with TRPC1 suppressing the current amplitude but not conferring a functional effect of reducing agents (Supplementary Fig. 6). Dimers of TRPC5 were not detected under non-reducing conditions, suggesting an intra-subunit rather than inter-subunit disulphide bridge (Supplementary Fig. 7).

Thioredoxin is an important redox protein with established biological roles including those in cancer, ischaemic reperfusion injury, inflammation and ageing. It is both an intracellular and secreted protein. It is reduced by the NADPH-dependent flavoprotein thioredoxin reductase and in this form has the capability of breaking disulphide bridges. Extracellular reduced thioredoxin (rTRX) acts similarly to DTT, causing TRPC5 activation (Fig. 2a, b). We therefore proposed that rTRX is a previously unrecognized endogenous extracellular regulator of TRPC5. In taking this idea forward we also considered TRPC1 because many cells endogenously expressed TRPC5 and TRPC1 together, leading to TRPC5–TRPC1 heteromultimers. The TRPC5–TRPC1 channel is also activated by rTRX or DTT (Fig. 2c). Consistent with previous reports was our observation that the TRPC5 and TRPC5–TRPC1 channels had distinct ‘fingerprint’ IV relationships (for example Fig. 3b and Supplementary Fig. 14).

Thioredoxin concentrations up to a mean of 0.41 &mgr;g ml-1 (maximum 1.2 &mgr;g ml-1) have been detected in serum and synovial fluid from patients with rheumatoid arthritis. Furthermore, reducing capability of thioredoxin exists in serum; thioredoxin reductase occurs in human joints and its activity is correlated with disease severity (see also Supplementary Results and Supplementary Discussion). We therefore considered whether the activation of TRPC5 by rTRX is relevant to the cells that secrete synovial fluid, the CD55-positive fibroblast-like synoviocytes (FLS cells). CD55-positive FLS cells (Supplementary Fig. 8) showed a non-selective cationic current in response to DTT or rTRX (Fig. 2d–g and Supplementary Figs 9 and 10). The mean current evoked by rTRX at -80 mV in FLS cells from the knee joint of patients with rheumatoid arthritis was -0.85 ± 0.42 nA (mean ± s.e.m.; n = 14). Oxidized TRX (oTRX) had no effect (Fig. 2f). The effective concentrations of rTRX indicate a possible relevance to rheumatoid arthritis (Fig. 2g). Nitric oxide is an alternative endogenous regulator of cysteine residues; however, it failed to evoke current in FLS cells, even at a concentration 100-fold that required to evoke vasorelaxation (Supplementary Fig. 10, Supplementary Results and Supplementary Discussion).

There have been no previous reports on the expression of TRPC channels in synovial joints, so we explored synovial tissue biopsies from patients with rheumatoid arthritis. TRPC5 and TRPC1 proteins were detected and localized together with CD55 (Fig. 3a). Similarly, the FLS cells used in our electrophysiological experiments expressed messenger RNAs encoding TRPC5 and TRPC1, western blotting indicated the presence of TRPC5 and TRPC1 proteins, and immunolabelling revealed TRPC5 and TRPC1 at the cell surface (Supplementary Figs 8, 11 and 13).

The IV relationship of the rTRX-evoked current in FLS cells was similar to that of the TRPC5–TRPC1 heteromultimeric channel (Fig. 3b). Furthermore, experiments with lanthanum ions showed unusual and striking similarity between the endogenous current and the current of overexpressed TRPC5–TRPC1: in the absence of a reducing agent, lanthanum ions stimulated current in both HEK-293 cells (exogenously expressing TRPC5–TRPC1) and FLS cells, whereas after the induction of current by rTRX, lanthanum ions were inhibitory in both cases (Fig. 3c and Supplementary Fig. 9). Also consistent with the involvement of TRPC channels were the observations that the rTRX-evoked current of FLS cells was blocked by 2-APB and that the inward current was suppressed when most of the extracellular Na+ was replaced by the bulky and impermeant cation N-methyl-d-glucamine (Fig. 2e and Supplementary Fig. 9). As a further test of the involvement of TRPC5 and TRPC1, FLS cells were transfected with a dominant-negative ion-pore mutant of TRPC5 that inhibited native channels capable of interacting with TRPC5 (refs 16, 22). The mutant suppressed current evoked by rTRX (Fig. 3d).

Further evidence that TRPC5 and TRPC1 contribute to the endogenous rTRX-responsive channel of FLS cells came from studies with anti-TRPC5 (T5E3) and anti-TRPC1 (T1E3) antibodies, which target the predicted extracellular loop region and specifically block the functions of TRPC5 and TRPC1, respectively. T5E3 and T1E3 antibodies labelled unpermeabilized FLS cells, unlike antibody targeted to the intracellular carboxy terminus of TRPC5, which labelled only permeabilized cells (Supplementary Figs 8 and 11), indicating that TRPC5 and TRPC1 are transmembrane proteins with extracellular epitopes. Like dominant-negative mutant TRPC5, T5E3 or T1E3 suppressed rTRX-evoked current (Fig. 3e). Antibody targeted to CD55, which is a membrane protein unrelated to TRP, had no significant effect (n = 7; data not shown). Gene expression, electrophysiology, pharmacology, recombinant DNA and antibody studies therefore yielded data consistent with the carrying of rTRX-evoked current in FLS cells by a channel containing TRPC5 and TRPC1.

One of the functions of FLS cells is to secrete matrix metalloproteinases (MMPs), which are associated with tissue remodelling and the progression of arthritis. The use of zymography to detect gelatinase activities of MMP-2 and MMP-9 secreted from rabbit FLS cells (Supplementary Fig. 12) revealed that T5E3 and T1E3 antibodies have large stimulatory effects (Fig. 4a, b). Human FLS cells showed greater MMP-2 secretion than that of MMP-9 (compare Supplementary Fig. 12 with Fig. 4a). Enzyme-linked immunosorbent assays (ELISAs) for human MMP-2 enabled the quantification of the absolute concentration of total MMP-2 secreted; again, either T5E3 or T1E3 antibody had a profound stimulatory effect (Fig. 4c). Similarly, knockdown of expression of the genes encoding TRPC1 and TRPC5 by RNA-mediated interference enhanced the secretion of MMP-2 (Supplementary Fig. 13). Inhibition of MMP secretion by the addition of exogenous reducing TRX was lost in the presence of T5E3 (Fig. 4d). Similar data were obtained for pro-MMP-1 secretion from human FLS cells (Fig. 4e, f) and MMP-9 measured by zymography in rabbit FLS cells (n = 6; data not shown). The data therefore reveal constitutive and rTRX-evoked activity of TRPC5 and TRPC1 channels that inhibits the secretion of MMP from FLS cells.

The data of this study indicate that secreted TRX is a type of ion channel agonist that acts through its reduced form to break a restraining intra-subunit disulphide bridge between cysteine residues in TRPC5, thereby stimulating the channel either as a homomeric assembly or as a heteromultimer with TRPC1. A transduction mechanism is therefore revealed that can directly couple cell activity to extracellular reduced thioredoxin. This mechanism may have particular relevance in conditions such as rheumatoid arthritis, in which TRX concentrations are strongly elevated, but the broad distributions of TRX and the channels indicate that the mechanism could be widely used.

Methods Summary Cells

Synovial tissue biopsies were obtained with informed consent from patients diagnosed with rheumatoid arthritis at the Academic Unit of Musculoskeletal Disease, Chapel Allerton Hospital, Leeds. Ethical approval was given by the local ethics committee. Human synovial tissue biopsies were washed with PBS and digested in 0.25% type 1A collagenase for 4 h at 37 °C, after which FLS cells were cultured in DMEM/F-12 + Glutamax (Gibco). HEK-293 cells were grown in DMEM-F12 (Gibco) and rabbit FLS cells (HIG82; ATCC) were grown in Ham’s F12 (Gibco). Culture media contained 10% fetal calf serum, 100 IU ml-1 penicillin and 100 &mgr;g ml-1 streptomycin. Cells were maintained at 37 °C in a humidified atmosphere of 5% CO2 in air and replated on coverslips or 24-well plates before experiments.


Whole-cell patch-clamp recordings were performed at 21 ± 2 °C using patch pipette solution containing (in mM): 115 CsCl, 10 EGTA, 2 MgCl2, 5 Na2ATP, 0.1 NaGTP, 10 HEPES, 5.7 CaCl2; the pH was adjusted to 7.2 with CsOH. The standard bath solution contained (in mM): 130 NaCl, 5 KCl, 8 d-glucose, 10 HEPES, 1.2 MgCl2 and 1.5 CaCl2; the pH was adjusted to 7.4 with NaOH.

Data analysis

Ionic currents are shown as positive values when they increased in response to a treatment, and as negative values when they decreased. Data are expressed as means and s.e.m., where n is the number of individual experiments. Data sets were compared by using paired or unpaired Student’s t-tests, with a significant difference indicated by P < 0.05 (asterisk) and no difference by n.s. All human tissue or cell data are derived from, or are representative of, at least three independent experiments on samples from three patients.

Complementary DNA clones, mutagenesis and cell transfection

HEK-293 cells stably expressing tetracycline-regulated human TRPC5 have been described. Expression was induced by 1 &mgr;g ml-1 tetracycline (Tet+; Sigma) for 24–72 h before recording. Non-induced cells without addition of tetracycline (Tet-) were controls. Human TRPC1 cDNA was expressed transiently from the bicistronic vector pIRES EYFP. Point mutations in human TRPC5 were introduced by using QuikChange site-directed mutagenesis (Stratagene) and appropriate primer sets. Dominant-negative TRPC5 is a triple alanine mutation of the conserved LFW sequence in the ion pore (Supplementary Fig. 2). The mutations were confirmed by direct sequencing of the entire reading frame. cDNAs were transiently transfected into HEK-293 cells or synoviocytes with FuGENE 6 transfection reagent (Roche) or Lipofectamine 2000 (Invitrogen) 48 h before recording. cDNA encoding GFP or YFP was cotransfected to identify transfected cells.


A salt-agar bridge was used to connect the ground Ag–AgCl wire to the bath solution. Signals were amplified with an Axopatch 200B patch clamp amplifier and controlled with pClamp software v. 6.0 (Axon) or Signal software v. 3.05 (CED). A 1-s ramp voltage protocol from -100 mV to +100 mV was applied at a frequency of 0.1 Hz from a holding potential of -60 mV. Current signals were filtered at 1 kHz and sampled at 3 kHz. Patch pipettes were made from borosilicate tubing which, after fire-polishing and filling with pipette solution, had a resistance of 3–5 M&OHgr;. The osmolarity of the pipette solution was adjusted to ∼290 mosM with mannitol and the calculated free Ca2+ was 200 nM. ATP and GTP were omitted when recording from cells expressing TRPC5 alone. When we were studying TRPC5 in HEK-293 cells, gadolinium chloride (Gd3+, 1–5 &mgr;M) was included in the bath solution to block background currents, which evoked submaximal TRPC5 current in some recordings before other agents were applied. The effect of reducing agents was not dependent on the presence of Gd3+. The recording chamber had a volume of 150 &mgr;l and was perfused at a rate of about 2 ml min-1. Recordings from human FLS cells used the Patchliner (Nanion) planar patch-clamp system with rapid bath solution exchange. For antibody treatment experiments, cells were treated with one of T1E3 at 1:500 dilution (ref. 23), T5E3 at 1:100 dilution (refs 24, 25), boiled (10 min) T1E3 at 1:500 dilution, T5E3 antibody at 1:100 dilution preabsorbed on its antigenic peptide (10 &mgr;M) or anti-CD55 antibody (see below), which were diluted in F12 Ham’s medium and incubated with cells for 2–3 h at 37 °C before patch-clamp recording.


Sections 4 &mgr;m thick were obtained from snap-frozen synovial tissue biopsy samples of patients suffering from rheumatoid arthritis, fixed with acetone and stored at -80 °C until use. Staining was in accordance with standard protocols. In brief, sections were incubated with primary antibody overnight at 4 °C and with secondary antibody (goat anti-rabbit IgG conjugated with fluorescein isothiocyanate (FITC) (Sigma) and donkey anti-mouse IgG conjugated with Cy3 (Jackson)) for 1 h at 21 ± 2 °C. For cell labelling, FLS cells adhered to coverslips were fixed for 13 min in 4% paraformaldehyde and, unless indicated, permeabilized for 2 h with 0.1% Triton X-100 in 1% BSA. Incubation in primary antibody was overnight at 4 °C and with secondary antibody (goat anti-rabbit IgG-FITC) for 2 h at room temperature. For control experiments, antibodies were preabsorbed on their antigenic peptide (10 &mgr;M) or omitted, as specified. Slides were mounted with 4,6-diamidino-2-phenylindole hardest mounting medium (Vector Labs) and analysed with a Zeiss confocal microscope. T1E3, T5E3, T5C3, anti-CD55 (Serotec) and CD68 (Dako) antibodies were used at 1:500, 1:100, 1:500 and 1:200 dilutions, respectively.

Secretion assays

FLS cells were cultured for 24 h in 24-well plates and serum-starved for 24 h; fresh serum-free medium was then added that contained a TRX cocktail, which included TRX (0.4 &mgr;g ml-1), NADPH-dependent flavoprotein TRX reductase (0.5 &mgr;g ml-1) and NADPH (2 &mgr;g ml-1) for 12 h. Omission of TRX was the control. Incubations with antibodies occurred for 2 h before addition of the TRX cocktail and were maintained in the presence of the TRX cocktail. Supernatants were collected, frozen and analysed by zymography or ELISA. For zymography the supernatant was mixed with 2× non-reducing SDS–PAGE sample buffer and resolved through a 7.4% polyacrylamide gel impregnated with 1.5 mg ml-1 gelatin. After electrophoresis, gels were washed, incubated and stained as described previously. The relative density of gelatinolytic bands was determined from scanned images of gels by using ImageQuant software (Amersham). MMP-2 or MMP-1 concentrations in supernatants from human cells were quantified with Quantikine Human total MMP-2 and pro-MMP-1 ELISA kits in accordance with the manufacturer’s instructions (R&D Systems).


All salts and reagents were from Sigma or BDH. Gadolinium (Gd3+) chloride, lanthanum (La3+) chloride, cadmium (Cd2+) chloride, DTT, 2-APB, NADPH and NADPH-dependent flavoprotein thioredoxin reductase (Escherichia coli) were from Sigma. Recombinant thioredoxin (TRX; Sigma) was from E. coli (unless specified) or human (no differences in effect were observed compared with E. coli TRX) and purchased from Sigma. MTSET was from Toronto Research Chemicals and TCEP was from Pierce Biotech. MTSET, TCEP and NADPH were prepared fresh for each experiment. Collagenase was from StemCell Technologies Inc. 2-APB (75 mM) stock solution was in 100% dimethylsulphoxide. To prepare reduced thioredoxin (rTRX), TRX (1 mg) was dissolved in 1 ml of the binding buffer (10 mM HEPES, 1 mM EDTA, 50 mM NaCl, pH 7.0) and 0.25 ml of this was mixed with 2.5 &mgr;l of 1 M DTT, incubated at room temperature for 30 min and then added to 0.25 ml of pre-equilibrated resin (DEAE-Sephadex; Sigma). The mixture was centrifuged for 30 s and then washed three times with binding buffer to remove DTT completely. Elution buffer (0.25 ml; 10 mM HEPES, 1 mM EDTA, 1 M NaCl, pH 7.0) was added and centrifuged to harvest the supernatant. The final TRX concentration was determined by Bradford assay. rTRX was diluted from cold stocks (on ice) immediately before use.

Functional disulphide bridge in TRPC5.

Whole-cell recordings from HEK-293 cells. a, In a cell expressing TRPC5, response to bath-applied 10 mM DTT and 75 &mgr;M 2-APB. b, IV relationship from a. c, As for b but with 1 mM TCEP. d, Currents at -80 mV evoked by 10 mM DTT (n = 8), 1 mM TCEP (n = 5) or 5 mM MTSET (n = 6) in cells expressing TRPC5. DTT had no effect without TRPC5 (n = 5). e, Inhibition of current at -80 mV by 0.1 mM Cd2+ in TRPC5-expressing cells with and without DTT treatment. f, As for e but typical IV relationships. g, Currents at -80 mV after transfection with green fluorescent protein (GFP) alone (no TRPC5, n = 8) or GFP plus wild-type TRPC5 (n = 7) or the TRPC5 mutants C553A (n = 11), C558A (n = 6), C553A + C558A (double, n = 3) or C553S (n = 6). Gd3+ (100 &mgr;M) activated wild-type TRPC5 but had no effect on mutants. All currents were blocked by 2-APB (see, for example, Supplementary Fig. 5). Where error bars are shown, results are expressed as means and s.e.m. Asterisk, P < 0.05; n.s., no significant difference.

Ionic current induced by rTRX.

ac, Whole-cell current data from HEK-293 cells expressing TRPC5 alone (a, b), TRPC1 alone or TRPC5 plus TRPC1 (c). a, Effect of 4 &mgr;g ml-1 rTRX. b, Current at -80 mV in response to elution buffer diluted 1:100 (n = 4) or rTRX with (n = 8) and without (n = 3) TRPC5 expression. c, Responses to rTRX or 10 mM DTT (n = 5 for each). d, Effect of rTRX on a human FLS cell. eg, Data for rabbit FLS cells. e, rTRX induced IV relationships in standard bath (Na+) or N-methyl-d-glucamine (NMDG+) solution (see Supplementary Fig. 9). f, Currents evoked at -80 mV. g, Data for human rTRX with a fitted Hill equation (concentration giving half-maximal response 0.20 &mgr;g ml-1, slope 2.64). Open symbols are control data and shaded areas are the concentrations of TRX in patients without arthritis (1) or with osteoarthritis (1) or rheumatoid arthritis (2). In f and g, n = 5 per data point. Where error bars are shown, results are expressed as means and s.e.m. Asterisk, P < 0.05.

Endogenous TRPC expression and function.

a, Tissue sections from joints of patients with rheumatoid arthritis stained with T1E3 or T5E3 (green) or anti-CD55 (red) antibodies. Controls were omission of anti-CD55 antibody, and T1E3 or T5E3 preadsorbed on its antigenic peptide. b, Normalized rTRX-evoked IV relationships for rabbit FLS cells (n = 3) and HEK-293 cells expressing TRPC5 and TRPC1 (n = 5). c, Changes in currents at -80 mV in response to 10 &mgr;M La3+ before or after treatment with 4 &mgr;g ml-1 rTRX (FLS cells, n = 6; HEK cells, n = 4). d, Current at -80 mV in FLS cells transfected with dominant-negative (DN) TRPC5 plus yellow fluorescent protein (YFP) or YFP alone (n = 5 for each). e, As in d but showing the effects of anti-TRPC antibodies (n = 5 for each). Pep., antigenic peptide. Where error bars are shown, results are expressed as means and s.e.m. Asterisk, P < 0.05.

Relevance to secretion from FLS cells.

a, Zymogram showing MMP-9 (pro and active) and MMP-2 from rabbit. Ab, antibody. b, As for a but mean data after normalization of rabbit MMP-9 band intensity to the control group without antibody (n = 3 for each). c, ELISA data for human MMP-2 (n = 4). d, Effect of T5E3 (n = 4) on inhibition of human MMP-2 secretion by exogenous TRX cocktail. For each group, secretion in TRX was normalized to that in its absence (control). e, f, As for c, d, but for secretion of human pro-MMP-1. Results are expressed as means and s.e.m. Asterisk, P < 0.05; n.s., no significant difference.

This work was supported by Wellcome Trust grants to D.J.B. and A.S., and a Physiological Society Junior Fellowship to C.C. P.S. has an Overseas Research Scholarship and University Studentship, J.N. has a Biotechnology and Biological Sciences Research Council PhD studentship, Y.M. a university studentship and Y.B. a scholarship from the Egyptian Ministry of Higher Education.

Flockerzi, V. An introduction on TRP channels. Handb. Exp. Pharmacol. 179, 119 (2007) Nilius, B., Owsianik, G., Voets, T. & Peters, J. A. Transient receptor potential cation channels in disease. Physiol. Rev. 87, 165217 (2007) Strubing, C., Krapivinsky, G., Krapivinsky, L. & Clapham, D. E. TRPC1 and TRPC5 form a novel cation channel in mammalian brain. Neuron 29, 645655 (2001) Plant, T. D. & Schaefer, M. TRPC4 and TRPC5: receptor-operated Ca2+-permeable nonselective cation channels. Cell Calcium 33, 441450 (2003) Beech, D. J. Canonical transient receptor potential 5. Handb. Exp. Pharmacol. 179, 109123 (2007) Rubartelli, A., Bajetto, A., Allavena, G., Wollman, E. & Sitia, R. Secretion of thioredoxin by normal and neoplastic cells through a leaderless secretory pathway. J. Biol. Chem. 267, 2416124164 (1992) Arner, E. S. & Holmgren, A. Physiological functions of thioredoxin and thioredoxin reductase. Eur. J. Biochem. 267, 61026109 (2000) Burke-Gaffney, A., Callister, M. E. & Nakamura, H. Thioredoxin: friend or foe in human disease? Trends Pharmacol. Sci. 26, 398404 (2005) Schwertassek, U. et al. Selective redox regulation of cytokine receptor signaling by extracellular thioredoxin-1. EMBO J. 26, 30863097 (2007) Maurice, M. M. et al. Expression of the thioredoxin–thioredoxin reductase system in the inflamed joints of patients with rheumatoid arthritis. Arthritis Rheum. 42, 24302439 (1999) Yoshida, S. et al. Involvement of thioredoxin in rheumatoid arthritis: its costimulatory roles in the TNF-&agr;-induced production of IL-6 and IL-8 from cultured synovial fibroblasts. J. Immunol. 163, 351358 (1999) Jikimoto, T. et al. Thioredoxin as a biomarker for oxidative stress in patients with rheumatoid arthritis. Mol. Immunol. 38, 765772 (2002) Smolen, J. S., Aletaha, D., Koeller, M., Weisman, M. H. & Emery, P. New therapies for treatment of rheumatoid arthritis. Lancet advance online publication. doi: 10.1016/S0140-6736(07)60784-3 (13 June 2007) Jung, S. et al. Lanthanides potentiate TRPC5 currents by an action at extracellular sites close to the pore mouth. J. Biol. Chem. 278, 35623571 (2003) Zeng, F. et al. Human TRPC5 channel activated by a multiplicity of signals in a single cell. J. Physiol. (Lond.) 559, 739750 (2004) Xu, S. Z. et al. A sphingosine-1-phosphate-activated calcium channel controlling vascular smooth muscle cell motility. Circ. Res. 98, 13811389 (2006) Elliott, D. J. et al. Molecular mechanism of voltage sensor movements in a potassium channel. EMBO J. 23, 47174726 (2004) Riccio, A. et al. mRNA distribution analysis of human TRPC family in CNS and peripheral tissues. Brain Res. Mol. Brain Res. 109, 95104 (2002) Lemarechal, H. et al. High redox thioredoxin but low thioredoxin reductase activities in the serum of patients with rheumatoid arthritis. Clin. Chim. Acta 367, 156161 (2006) Lemarechal, H. et al. Impairment of thioredoxin reductase activity by oxidative stress in human rheumatoid synoviocytes. Free Radic. Res. 41, 688698 (2007) Yoshida, T. et al. Nitric oxide activates TRP channels by cysteine S-nitrosylation. Nature Chem. Biol. 2, 596607 (2006) Strubing, C., Krapivinsky, G., Krapivinsky, L. & Clapham, D. E. Formation of novel TRPC channels by complex subunit interactions in embryonic brain. J. Biol. Chem. 278, 3901439019 (2003) Xu, S. Z. & Beech, D. J. TrpC1 is a membrane-spanning subunit of store-operated Ca2+ channels in native vascular smooth muscle cells. Circ. Res. 88, 8487 (2001) Xu, S. Z. et al. Generation of functional ion-channel tools by E3 targeting. Nature Biotechnol. 23, 12891293 (2005) Xu, S. Z., Boulay, G., Flemming, R. & Beech, D. J. E3-targeted anti-TRPC5 antibody inhibits store-operated calcium entry in freshly isolated pial arterioles. Am. J. Physiol. Heart Circ. Physiol. 291, H2653H2659 (2006) Burrage, P. S., Mix, K. S. & Brinckerhoff, C. E. Matrix metalloproteinases: role in arthritis. Front. Biosci. 11, 529543 (2006) Porter, K. E. et al. Simvastatin inhibits human saphenous vein neointima formation via inhibition of smooth muscle cell proliferation and migration. J. Vasc. Surg. 36, 150157 (2002)
doi: 10.1038/nature06414

Trisomy represses -mediated tumours in mouse models of Down’s syndrome p.73

version="1.0" encoding="utf-8"?> ]>
Nature 451 7174 20080103 73753 0028-0836 1476-4687 2007Nature Publishing Group Supplementary Information

The file contains Supplementary Tables 1-2 and Supplementary Figures 1-4 with Legends.

Trisomy represses ApcMin-mediated tumours in mouse models of Down’s syndrome Thomas E.SussanT E AnnanYangA FuLiF Michael C.OstrowskiM C Roger H.ReevesR H Department of Physiology and The Institute for Genetic Medicine, The Johns Hopkins University School of Medicine, Baltimore, Maryland 21205, USA Department of Molecular and Cellular Biochemistry, Ohio State University, Columbus, Ohio 43210, USA Present address: Department of Environmental Health Sciences, Bloomberg School of Public Health, Baltimore, Maryland 21205, USA. Correspondence and requests for materials should be addressed to R.H.R. (rreeves@jhmi.edu). &e080103-7; &nature06446-s1;

Epidemiological studies spanning more than 50 yr reach conflicting conclusions as to whether there is a lower incidence of solid tumours in people with trisomy 21 (Down’s syndrome). We used mouse models of Down’s syndrome and of cancer in a biological approach to investigate the relationship between trisomy and the incidence of intestinal tumours. ApcMin-mediated tumour number was determined in aneuploid mouse models Ts65Dn, Ts1Rhr and Ms1Rhr. Trisomy for orthologues of about half of the genes on chromosome 21 (Hsa21) in Ts65Dn mice or just 33 of these genes in Ts1Rhr mice resulted in a significant reduction in the number of intestinal tumours. In Ms1Rhr, segmental monosomy for the same 33 genes that are triplicated in Ts1Rhr resulted in an increased number of tumours. Further studies demonstrated that the Ets2 gene contributed most of the dosage-sensitive effect on intestinal tumour number. The action of Ets2 as a repressor when it is overexpressed differs from tumour suppression, which requires normal gene function to prevent cellular transformation. Upregulation of Ets2 and, potentially, other genes involved in this kind of protective effect may provide a prophylactic effect in all individuals, regardless of ploidy.

The most widely used model of Down’s syndrome is the Ts65Dn mouse, which is trisomic for orthologues of about 100 Hsa21 genes and recapitulates in detail several phenotypes of Down’s syndrome (Supplementary Fig. 1). Mice that are heterozygous for the ApcMin mutation accumulate tumours analogous to those in familial adenomatous polyposis along the wall of the small intestine and colon. APC is also mutated in a high proportion of spontaneous intestinal cancers in human beings. Although the mouse mutation is completely penetrant, the number of tumours that develop is dependent on both genetic modifier genes and environmental factors.

Female Ts65Dn mice were crossed to ApcMin males and the number of tumours in the small intestine was determined in their trisomic and euploid ApcMin progeny at 120 days of age (Supplementary Fig. 2). Trisomic mice showed a significant 44% reduction in the number of tumours compared to their euploid, ApcMin littermates, from 45.4 to 23.8 tumours (Table 1). This establishes a biological basis for the effects of trisomy on tumour formation and shows that trisomy for orthologues of about half of the genes on Hsa21 is sufficient to reduce tumour incidence in this model.

We reanalysed these data considering the inheritance of susceptible or resistant alleles of the modifier of Min 1 (Mom1) locus that result in higher or lower tumour number (Mom1s and Mom1r, respectively; genetic background of all crosses is shown in Supplementary Fig. 3). The inheritance of a single Mom1r allele reduced the average tumour number from 62.6 to 21.3 in euploid mice (66%) as expected, and a similar 59% reduction occurred in Ts65Dn (Table 1). Ts65Dn, Mom1s/s mice had a highly significant 50% reduction in small intestine tumour number compared to euploid Mom1s/s mice (P = 0.0028). Trisomic mice that inherited a Mom1r allele (Mom1s/r) also had substantially reduced tumour numbers relative to euploid mice, although this reduction did not reach a statistically significant level in the small sample of Ts65Dn, Mom1s/r mice available for this post-hoc analysis. Thus the Mom1r effect seems to be additive with the protective effect of trisomy, suggesting that independent mechanisms are involved.

We analysed Ts1Rhr mice to narrow the candidate region for the gene or genes responsible for reduced tumour number. These mice have segmental trisomy for 33 of the genes that are triplicated in Ts65Dn (Supplementary Fig. 1). These genes represent a ‘critical region’ of Hsa21, previously thought to be sufficient to cause several phenotypes of Down’s syndrome. Ts1Rhr, ApcMin mice had a significant 26% reduction in the average number of tumours in the small intestine when compared to euploid, ApcMin mice (Table 1).

When ApcMin mice were crossed to Ms1Rhr, which have segmental monosomy for the 33 genes that are triplicated in Ts1Rhr, we observed a significant 101% increase in tumour number in the monosomic mice compared to euploid (Table 1). These results demonstrate that a gene (or combination of genes) in this region is dosage sensitive in both directions with respect to the effect on tumour number.

The 33 genes at dosage imbalance in Ts1Rhr and Ms1Rhr mice include several possible candidates for the tumour number effect (Supplementary Table 1), including the Ets2 ‘proto-oncogene’. Although generally considered a ‘pro-cancer’ gene, Ets2 has several activities consistent with a role in repressing the early stages of transformation. We performed a three-way cross to produce mice carrying ApcMin that were either euploid or had the Ts1Rhr segmental trisomy, and which segregated an allele of Ets2 that deletes exons 3–5 and fails to produce functional Ets2 protein (F.L. and M.C.O., manuscript in preparation). Tumours were counted at 120 days (Fig. 1). This independent cohort of mice replicated the observation (Table 1) that trisomy for three copies of Ets2 and 32 flanking genes in Ts1Rhr results in a significantly reduced tumour incidence, from a mean of 100.8 to 53.9 (P = 0.001). However, when Ets2 was returned to the normal two copy level in mice that were still trisomic for the 32 flanking genes (ApcMin, Ets2+/-, Ts1Rhr), average tumour number increased significantly to 81.2 (P = 0.012). Thus, a substantial portion though not all of the tumour repression in Ts1Rhr is accounted for by the extra copy of Ets2.

Mice that carried a single copy of Ets2 in a euploid background showed a substantial, 20% increase in tumour frequency (P = 0.075), reminiscent of the increase in tumours in Ms1Rhr mice, which carry a single copy of this gene. These mice developed severe disease much earlier than mice of other genotypes and several did not survive long enough for tumours to be counted. Thus this difference in tumour number is probably under-represented. Ets2 messenger RNA and protein levels corresponded directly to gene copy number in all of the genotypes (Supplementary Fig. 4).

The size of tumours in a given genetic background provides one indicator of tumour initiation and growth rates. We compared the size of tumours between trisomic and euploid ApcMin mice (Fig. 2a). Ts65Dn, Mom1s/s mice showed a significant 34% reduction in average and median tumour size at 120 days compared to euploid (P < 0.005). Note that Ts65Dn mice in this experiment had 48% fewer tumours than did euploid animals, a significantly lower level that replicates in this independent cross the reduction in tumour number reported for the independent cohort of mice represented in Table 1 (P < 0.04, N = 4 euploid, 5 Ts65Dn).

To determine whether this difference was evident earlier in the course of tumour formation, intestines of trisomic and euploid mice carrying the ApcMin allele were immunostained for &bgr;-catenin at 60 days of age (Supplementary Fig. 2). As at 120 days of age, the number and average size of tumours in Ts65Dn mice was significantly less than in their euploid counterparts (Fig. 2b). No tumours were seen at 30 days of age in two euploid or one trisomic ApcMin mouse after &bgr;-catenin staining. Thus the repression of tumour number and size in Ts65Dn mice was evident early in tumour formation.

In contrast to Ts65Dn mice, tumour size was not different from euploid in either Ts1Rhr or Ms1Rhr mice (data not shown). The absence of a tumour size phenotype even though tumour number is reduced in Ts1Rhr mice indicates that multiple genes on Mmu16 (and Hsa21) may contribute to different aspects of tumour repression caused by trisomy.

For 50 yr, epidemiological studies examining rates of solid tumours in individuals with Down’s syndrome have reached discrepant conclusions about whether trisomy is protective against cancer (Supplementary Table 2). Although our demonstration of tumour repression owing to gene dosage applies specifically to the role of trisomy and especially Ets2 dosage in Apc-induced tumours, it provides biological evidence supporting the protective effect of trisomy. It will be important to determine the range of cancer types and the range of dosage-sensitive genes that contribute to this protective effect in different tissues.

Notable among the Hsa21 genes that have been implicated in pro- or anti-tumorigenesis is endostatin, an inhibitor of angiogenesis that has been shown to be a potent inhibitor of tumour growth in model systems. Elevated expression of another Hsa21 gene, RCAN1, can reduce endothelial cell proliferation and angiogenesis, affecting size and vascularity of xenografted tumours. However, Rcan1 is not trisomic in Ts1Rhr, and the Col18a1 gene (which encodes endostatin) is not triplicated in either Ts65Dn or Ts1Rhr. Therefore, these genes do not account for the reduction in tumour number seen here.

Two general implications that stem from the observation that trisomy and specifically Ets2 dosage can repress or promote tumour growth are worth special note. First, repression of tumorigenesis when Ets2 expression is elevated may in fact be a characteristic of many genes identified previously as oncogenes or tumour suppressor genes. Natural variation in average expression levels of ETS family (or other) repressor genes may exist in tumour-prone families without a known molecular basis for a high cancer frequency (reduced expression of Ets2) or in cancer-resistant families (elevated expression). This phenomenon might be exploited to identify a pharmacological-based approach to tumour protection.

Second, previous observations about the role of the ETS2 proto-oncogene in cancer could not have predicted that elevation of expression beyond euploid levels would provide a natural repression of tumour formation and growth. If trisomy for Hsa21 was not viable, the correlation of increased gene expression with lower solid tumour frequency would not occur in a systematic manner and may not have been observed for some time. The implication for promoting tumour resistance in all people on the basis of gene dosage of ‘oncogenes’ is thus a product of the genetic heritage of those with Down’s syndrome.

Methods Summary

C57BL/6J-ApcMin mice (herein ApcMin) and B6EiC3Sn a/A-Ts(1716)65Dn (herein Ts65Dn) mice were purchased from the Jackson Laboratory and genotyped as described. B6.Dup(Cbr1-ORF9)1Rhr mice (herein Ts1Rhr) were backcrossed eight or more generations onto C57Bl/6J (B6). B6C.3Del(16Cbr1-ORF9)1Rhr (herein Ms1Rhr) and Ts65Dn mice were maintained as an advanced intercross between B6 and C3H. For tumour analysis, mice were euthanized at 120 ± 2 days, intestines were placed in fresh PBS, and tumours counted under 20× magnification. Tumour size was determined for the longest axis, using an eyepiece reticule. Statistical significance was determined using a Student’s t-test. Detailed methods are in Supplementary Information and Methods.


C57BL/6J-ApcMin mice (herein ApcMin) were purchased from the Jackson Laboratory (Bar Harbor, ME) and maintained by repeated backcrossing to C57Bl/6J (B6) mice. B6EiC3Sn a/A-Ts(17)65Dn (herein Ts65Dn) mice were purchased from the Jackson Laboratory. B6C3Del(16Cbr1-ORF9)1Rhr (herein Ms1Rhr) were maintained in our colony where both Ms1Rhr and Ts65Dn mice were maintained as an advanced intercross by crossing to (B6 × C3H/HeJ)F1 mice. B6.Dup(Cbr1-ORF9)1Rhr mice (herein Ts1Rhr) were backcrossed eight or more generations onto C57Bl/6J. Mice carrying a null allele of Ets2 (herein Ets2+/- mice, F.L. and M.C.O., in preparation) were backcrossed for more than nine generations onto B6 before being used in these experiments. The genetic backgrounds of all mice produced for this study are shown in Supplementary Fig. 3. In general, groups of euploid and trisomic littermates from related mothers were used in crosses that generated aneuploid mice to minimize genetic variation.


ApcMin, Ts1Rhr and Ms1Rhr mice were genotyped by PCR as described. Ts65Dn mice were identified by fluorescent in situ hybridization (FISH) as described.

For Mom1, PCR primers were designed to amplify the wild-type (Mom1r) (Mom Common-TGGGGAAATGATTTGGCTTA, MomWT-TGGCATCCTTGGGGGAT) and mutant (Mom1s) (Mom Common, Mom MUT-TGGCATCCTTGGGGGAA) alleles. These primers were used with the LightCycler FastStart DNA Master SYBR Green I kit (Roche Diagnostics Corporation), with conditions: 95 °C 10 min, (95 °C 10 s, 58 °C 5 s, 72 °C 20 s) × 55. Presence of an allele resulted in a five-cycle shift in the amplification curve. This result was confirmed by melting curve analysis which yielded distinct profiles for Mom1r and Mom1s .

PCR was used to type Ets2+/- mice. The wild-type allele was detected using primers Ets2I2P10 (CGCTTGCTAGGCAAGTGCTCTACC) and Ets2I2P9 (GCTGACACAGGGTTTTGGTGTCATGC), and the Ets2 deleted allele was detected using primers Ets2I2P10 and Est2I25P3 (CTAAGCCAGCCTGGCTACAGAACC), under the following cycling conditions: 95 °C 2 min, (94 °C 45 s, 55 °C 45 s, 72 °C 1 min) for 35 cycles, 72 °C 10 min. The wild-type band was 300 bp and the deleted band was 600 bp.

Tumour analysis

All animals were assessed blind to genotype in all assays. Groups of littermate mice from closely related mothers (and inbred fathers) were euthanized at 120 ± 2 days of age. Intestines were removed and rinsed then cut longitudinally and placed in fresh PBS. Tumours were counted under 20× magnification across the entire length of the small intestine. For Table 1, tumours were scored if they were ≥0.4 mm in diameter; small tumours that did not involve multiple crypts were excluded. Because the process of identifying tumours is disruptive and tumour tissue rapidly degrades under dissection conditions, multiple observers are not used for the same mice in the ApcMin tumour assay. Rather, independent crosses were assessed by independent observers to confirm the effects of aneuploidy on tumorigenesis. The Ts65Dn ApcMin tumour analysis was done three times by two observers (T.E.S. and A.Y.)(data in Table 1, Fig. 2a and Fig. 2b) and the Ts1Rhr × ApcMin analysis was performed twice by two observers (Table 1 and Fig. 1). A summary of the crosses and data collection process is in Supplementary Fig. 3.

For visible tumours (at 20× magnification), tumour size was determined for the longest axis of the tumour using an eyepiece reticule. Statistical significance was determined using a Student’s t-test. For microscopic tumours, intestines were recovered from ApcMin and Ts65Dn, ApcMin littermates 60 days of age. Intestines were removed, washed in 4 °C PBS several times and then cut into three sections (proximal to distal). Each section was cut open longitudinally and fixed overnight in 10% formalin. The next day the intestine was rolled up and embedded in paraffin as a ‘Swiss roll’, and ten slides each containing 3 sections of 6 microns thick were recovered at an interval of 50 microns. Slides were deparaffinized, stained with &bgr;-catenin antibody (BD Biosciences Clone 14, Vector M.O.M immunodetection Kit) and co-stained with haematoxylin, and tumours from ten slides per mouse were measured under a light microscope with an eye piece reticule. Tumour size and number were counted and results compiled (Supplementary Fig. 2).

RNA and protein analysis

Mouse embryo fibroblasts (MEFs) were established from fetuses at E13.5. Fetuses were removed and the visceral tissue separated. Remaining tissue was minced in Trypsin/EDTA and incubated at 37 °C for an hour. Trypsin was neutralized by addition of medium (DMEM plus 10% serum and antibiotics) and cells collected and plated, taking care to avoid transfer of larger pieces of tissue. The next day, cells were re-fed, then passaged as they reached confluence. For these experiments, cells were used between 6–8 passages.

Total RNA was isolated from mouse small intestine or MEFs with TRIzol reagent (invitrogen) and RNeasy Mini Kit (Qiagen), including a DNase I treatment step. RNA concentration was determined by UV spectrophotometry and 1 &mgr;g was reverse transcribed with GeneAmp RNA PCR kit (Applied Biosystems). After dilution, 10 ng of complementary DNA was amplified by real-time PCR with SYBR Green PCR master mix (Applied Biosystems) using specific primers for Ets2 (Forward, AGAGAAGGGAGCACAGCAAA; Reverse, AAGAACATGGACCAAGTGGC) (http://mouseprimerdepot.nci.nih.gov/) and &bgr;-actin (Forward, AGTGTGACGTTGACATCCGTA; Reverse, GCCAGAGCAGTAATCTCCTTCT). Real-time PCR was carried out under the following conditions: 10 min at 95 °C, followed by 40 cycles of 15 s at 95 °C and 1 min at 60 °C (Applied Biosystems 7500 System). Ct values were determined by subtracting the average &bgr;-actin Ct value from the average Ets2 Ct value. The s.d. of the difference was calculated from the s.d. s of Ets2 and &bgr;-actin values. After each real-time RT–PCR, a melting profile was done to rule out non-specific contributions from PCR products and primer dimers.

For western blots, whole cell lysates were prepared by lysing MEFs with RIPA buffer, and 100 &mgr;g of protein from each sample was separated by 8% SDS PAGE. The membrane was blotted overnight with anti-Ets2 (ref. 21) 1:1,000, 5% milk in 0.05% TBST (TBST is 0.05% Tween-20, 20mM Tris-HCl pH7.6 and 150mM NaCl), probed with anti-Rabbit HRP and developed for ECL. Blots were stripped and reprobed with anti-tubulin (1:1,000 in 5% milk in 0.05% TBST) antibody. Scanned images of each blot were inverted by NIH Image J and the density calculated for Ets2 and tubulin in each sample. The background was measured and subtracted and the ratio of Ets2/tubulin density was used to compare protein expression level of Ets2 in MEFs of different genotypes. The average level of Ets2:tubulin in euploid mice was arbitrarily set at 1.0 and Ets2 levels in other genotypes were calculated in proportion.

Ets2 dosage is substantially responsible for tumour number repression or increase.

Average tumour number at 120 days is measured for the four genotypes, error bars indicate s.d. Number of mice analysed, P value and the gene copy number of Ets2 in each strain are indicated. *, statistical significance by Student’s t-test of the designated pair. Although the increased tumour number in euploid Ets2+/- mice at 120 days did not reach a formal level of statistical significance, this result underestimates the impact of reduced Ets2 dosage, because four ApcMin, Ets2+/- mice became sick and were euthanized before tumours could be counted at 120 days. None of the 47 mice representing the other 3 genotypes died before 120 days.

Tumour growth and number are reduced in Ts65Dn mice.

Distribution of tumour sizes for trisomic (open bars) and euploid (closed bars) mice. a, At 120 days of age, tumour number is reduced and tumours are significantly smaller in Ts65Dn, Mom1s/s than in euploid mice. Mean tumour size is reduced by 34%; Ts65Dn = 0.91 mm, euploid = 1.38 mm (P = 0.005, N = 347 and 577 tumours for Ts65Dn and euploid, respectively). b, At 60 days, the number of tumours identified after staining with &bgr;-catenin is significantly reduced in Ts65Dn (P = 0.037) and mean tumour size is reduced 36%; mean = 18 &mgr;m in trisomic and 28 &mgr;m in euploid mice (P = 0.029, n = 346 and 636 tumours for Ts65Dn and euploid, respectively). Arrows indicate mean tumour size in each genotype.

Average numbers of intestinal tumours in aneuploid and euploid mice at 120 days of age Average no. of tumours s.d. No. of mice t-test significance (P value)

*Genetic background is shown in parentheses. Ts65Dn and euploid controls are B6/C3H (Supplementary Fig. 3).

Either Mom1 allele Euploid 45.4 29.9 24 0.008 Ts65Dn 23.8 14.2 10 Mom1s/s Euploid 62.6 26.4 14 0.0028 Ts65Dn 31.2 13.7 6 Mom1 s/r Euploid 21.3 13.3 10 0.105 Ts65Dn 12.8 5.0 4 Segmental aneuploidies* Euploid (B6) 107.3 45.0 16 0.043 Ts1Rhr (B6) 79.6 29.9 21 Euploid (B6/C3H) 37.0 16.0 9 0.048 Ms1Rhr (B6/C3H) 74.4 39.7 7

The authors thank L. Siracusa for advice regarding the ApcMin system, C. Dang for important discussions regarding cancer models, and R. Roper for statistical advice. This work was supported by National Institute of Child Health and Development and National Cancer Institute awards (M.C.O. and R.H.R.).

Author Contributions T.E.S. and R.H.R. designed the experiments. T.E.S. and A.Y. managed husbandry and collected tumour data, which were analysed by T.E.S., A.Y. and R.H.R.; F.L. and M.C.O. designed the Ets2 conditional knockout mice; and A.Y., F.L. and M.C.O. analysed Ets2 expression. R.H.R. wrote the paper with substantial input from all authors.

Satgé, D. et al. A tumor profile in Down syndrome. Am. J. Med. Genet. 78, 207216 (1998) Yang, Q., Rasmussen, S. A. & Friedman, J. M. Mortality associated with Down's syndrome in the USA from 1983 to 1997: a population-based study. Lancet 359, 10191025 (2002) Reeves, R. H. et al. A mouse model for Down Syndrome exhibits learning and behaviour deficits. Nature Genet. 11, 177183 (1995) Gardiner, K., Fortna, A., Bechtel, L. & Davisson, M. T. Mouse models of Down syndrome: how useful can they be? Comparison of the gene content of human chromosome 21 with orthologous mouse genomic regions. Gene 318, 137147 (2003) Su, L. K. et al. Multiple intestinal neoplasia caused by a mutation in the murine homolog of the APC gene. Science 256, 668670 (1992) Yamada, Y. & Mori, H. Multistep carcinogenesis of the colon in ApcMin/+) mouse. Cancer Sci. 98, 610 (2007) Dietrich, W. F. et al. Genetic identification of Mom-1, a major modifier locus affecting Min-induced intestinal neoplasia in the mouse. Cell 75, 631639 (1993) MacPhee, M. et al. The secretory phospholipase A2 gene is a candidate for the Mom1 locus, a major modifier of ApcMin-induced intestinal neoplasia. Cell 81, 957966 (1995) Olson, L. E., Richtsmeier, J. T., Leszl, J. & Reeves, R. H. A chromosome 21 critical region does not cause specific Down syndrome phenotypes. Science 306, 687690 (2004) Hsu, T., Trojanowska, M. & Watson, D. K. Ets proteins in biological control and cancer. J. Cell. Biochem. 91, 896903 (2004) Wolvetang, E. J. et al. Overexpression of the chromosome 21 transcription factor Ets2 induces neuronal apoptosis. Neurobiol. Dis. 14, 349356 (2003) Roberts, R. B. et al. Importance of epidermal growth factor receptor signaling in establishment of adenomas and maintenance of carcinomas during intestinal tumorigenesis. Proc. Natl Acad. Sci. USA 99, 15211526 (2002) Day, S. M., Strauss, D. J., Shavelle, R. M. & Reynolds, R. J. Mortality and causes of death in persons with Down syndrome in California. Dev. Med. Child Neurol. 47, 171176 (2005) Hill, D. A. et al. Mortality and cancer incidence among individuals with Down syndrome. Arch. Intern. Med. 163, 705711 (2003) Patja, K., Pukkala, E., Sund, R., Iivanainen, M. & Kaski, M. Cancer incidence of persons with Down syndrome in Finland: a population-based study. Int. J. Cancer 118, 17691772 (2006) Goldacre, M. J., Wotton, C. J., Seagroatt, V. & Yeates, D. Cancers and immune related diseases associated with Down's syndrome: a record linkage study. Arch. Dis. Child. 89, 10141017 (2004) O'Reilly, M. S. et al. Endostatin: an endogenous inhibitor of angiogenesis and tumor growth. Cell 88, 277285 (1997) Minami, T. et al. Vascular endothelial growth factor- and thrombin-induced termination factor, Down syndrome critical region-1, attenuates endothelial cell proliferation and angiogenesis. J. Biol. Chem. 279, 5053750554 (2004) Moore, C. S. et al. Integration of cytogenetic with recombinational and physical maps of mouse chromosome 16. Genomics 59, 15 (1999) The Jackson Laboratory. Genotyping protocol for Apc. 〈http://jaxmice.jax.org/pub-cgi/protocols/protocols.sh?objtype=protocol&protocol_id=529#133 Smith, J. L. et al. ets-2 is a target for an akt (Protein kinase B)/jun N-terminal kinase signaling pathway in macrophages of motheaten-viable mutant mice. Mol. Cell. Biol. 20, 80268034 (2000)
doi: 10.1038/nature06446

NUMB controls p53 tumour suppressor activity p.76

version="1.0" encoding="utf-8"?> ]>
Nature 451 7174 20080103 76805 0028-0836 1476-4687 2007Nature Publishing Group Supplementary Information

This file contains Supplementary Discussion and additional references; Supplementary Table S1 and Supplementary Figures S1-S6 with Legends.

NUMB controls p53 tumour suppressor activity Ivan N.ColalucaI N DanielaTosoniD PaoloNuciforoP FrancescaSenic-MatugliaF VivianaGalimbertiV GiuseppeVialeG SalvatorePeceS Pier PaoloDi FioreP P IFOM, the FIRC Institute for Molecular Oncology Foundation, Via Adamello 16, 20139, Milan, Italy European Institute of Oncology, Via Ripamonti 435, 20141 Milan, Italy Dipartimento di Medicina, Chirurgia ed Odontoiatria, Università degli Studi di Milano, 20122 Milan, Italy Correspondence and requests for materials should be addressed to S.P. (salvatore.pece@ifom-ieo-campus.it) or P.P.D.F (pierpaolo.difiore@ifom-ieo-campus.it). &e080103-13; &nature06412-s1;

NUMB is a cell fate determinant, which, by asymmetrically partitioning at mitosis, controls cell fate choices by antagonising the activity of the plasma membrane receptor of the NOTCH family. NUMB is also an endocytic protein, and the NOTCH–NUMB counteraction has been linked to this function. There might be, however, additional functions of NUMB, as witnessed by its proposed role as a tumour suppressor in breast cancer. Here we describe a previously unknown function for human NUMB as a regulator of tumour protein p53 (also known as TP53). NUMB enters in a tricomplex with p53 and the E3 ubiquitin ligase HDM2 (also known as MDM2), thereby preventing ubiquitination and degradation of p53. This results in increased p53 protein levels and activity, and in regulation of p53-dependent phenotypes. In breast cancers there is frequent loss of NUMB expression. We show that, in primary breast tumour cells, this event causes decreased p53 levels and increased chemoresistance. In breast cancers, loss of NUMB expression causes increased activity of the receptor NOTCH. Thus, in these cancers, a single event—loss of NUMB expression—determines activation of an oncogene (NOTCH) and attenuation of the p53 tumour suppressor pathway. Biologically, this results in an aggressive tumour phenotype, as witnessed by findings that NUMB-defective breast tumours display poor prognosis. Our results uncover a previously unknown tumour suppressor circuitry.

p53 is one of the major tumour suppressor proteins. Mutations in the p53 gene are detected in ∼50% of human cancers. Indirect mechanisms also lead to p53 inactivation. HDM2 binds to p53 and hinders its transcriptional activity. In addition, HDM2 regulates p53 half-life through its E3 ubiquitin-ligase activity. The tumour suppressor ARF (alternative reading frame, and encoded by the INK4a/ARF locus; INK4a is also called CDKN2A) binds to HDM2 and interferes with its activity, thereby stabilizing p53 (ref. 11). Thus, in human tumours, amplification of the HDM2 gene or loss of the ARF protein result in defective p53 activity. In breast tumours, p53 mutations and amplification of HDM2 are not as frequent as in other tumours, being detected in ∼20% and <10% of cases, respectively. Similarly, homozygous deletions or mutations of the INK4a/ARF locus are rare. Although epigenetic mechanisms might contribute to altered expression of HDM2 or ARF, other unknown circuitries of regulation of p53 might be subverted in breast tumours. NUMB is a candidate for this role, in that it binds to HDM2 (ref. 20) and is frequently underexpressed in breast cancers.

Previous work has shown interaction in vivo between overexpressed NUMB and HDM2 (ref. 20). Also the two endogenous proteins can be co-immunoprecipitated from cellular lysates of normal mammary MCF10A cells (Fig. 1a). In addition, the four described isoforms of NUMB all interact with HDM2 (Supplementary Fig. 1a). NUMB also co-immunoprecipitates with p53 (Fig. 1a). This is compatible with the existence in vivo of multiple binary complexes (NUMB–HDM2, NUMB–p53 and HDM2–p53), or of a tricomplex NUMB–HDM2–p53. In in vitro binding assays with purified proteins, all three binary complexes could form, indicating direct interactions between the three proteins (Supplementary Fig. 1b). The presence of nutlin, which inhibits the HDM2–p53 interaction, prevented the formation of the HDM2–p53 complex, but not of the HDM2–NUMB or p53–NUMB complexes (Fig. 1b). Thus, the NUMB–p53 and NUMB–HDM2 surfaces of interaction are distinct, at least in part, from that of the HDM2–p53 interaction. This pointed to the possibility of formation of a tricomplex in vivo (which might coexist with the various binary complexes)—a possibility confirmed by sequential immunoprecipitation experiments with endogenous (Fig. 1c) or overexpressed (Supplementary Fig. 1c) proteins. Nutlin and cisplatin strongly decreased the NUMB–HDM2, but not the NUMB–p53, interaction (Supplementary Fig. 1d, e). These results argue that the stability of the tricomplex is affected by agents disrupting the HDM2–p53 interaction and that it is regulated in response to stress signals.

We analysed the effects of NUMB knockdown (NUMB-KD) on p53 by targeting two different sequences in NUMB using short interfering RNA (siRNA) or short hairpin RNA (shRNA), respectively. Both methods yielded 80–90% decrease in NUMB levels and an approximately twofold decrease in the p53 steady-state levels (Fig. 1d and Supplementary Fig. 2a). This effect was NUMB-specific because ablation of the related protein NUMBL (NUMB-like) did not affect p53 levels (Fig. 1d). Importantly, approximately threefold-higher doses of genotoxic drugs (Fig. 1e, f and Supplementary Fig. 2b) were needed to induce, in knockdown cells, levels of p53 comparable to those induced in wild-type cells. Similar results were obtained in primary normal human mammary cells (Supplementary Fig. 2c). The reduced levels of p53 on cisplatin treatment were paralleled by a reduction in the levels of Ser 15-phosphorylated p53 (Fig. 1f), a marker of the activation status of p53. Furthermore, we observed marked reductions in the expression of several p53 transcriptional targets (Fig. 1f, g). Thus, in NUMB-KD cells, the overall activation status of p53 after DNA damage is reduced.

In NUMB-KD cells, p53 messenger RNA levels were not altered (Fig. 2a), arguing for NUMB-mediated regulation of p53 at the post-transcriptional level. Indeed, the half-life of p53 was reduced in NUMB-KD versus control cells, from ∼60 to ∼20 min (Fig. 2b, c). The half-life of HDM2 was not affected, arguing that NUMB-KD does not primarily affect HDM2 stability (Fig. 2b, c).

The simultaneous silencing of NUMB and HDM2 restored p53 to levels indistinguishable from that of control cells or HDM2-KD cells (Fig. 2d). Nutlin stabilizes and activates p53 (ref. 21). Nutlin treatment of MCF10A cells resulted in increased steady-state levels of p53 and of HDM2 (Fig. 2e), which confirms that the accumulated p53 protein is transcriptionally active. More importantly, nutlin reversed the effects of NUMB-silencing on p53 levels (Fig. 2e). Thus, there is a requirement for HDM2 in the regulation of p53 by NUMB.

HDM2 regulates p53 turnover through its E3 activity. In NUMB-KD cells, we detected enhanced ubiquitination of p53, which was inhibited by nutlin (Fig. 2f), arguing for a direct counteraction of NUMB over the p53-ubiquitinating activity of HDM2. This was confirmed in in vitro assays in which glutathione S-transferase (GST)–p53 was used as a substrate for the E3 activity of HDM2. Under these conditions, ubiquitinated p53 was readily detectable (Fig. 2g). However, this effect was abolished in the presence of purified NUMB or nutlin (Fig. 2g; see Supplementary Fig. 3 for possible models for action of NUMB).

NUMB inhibits NOTCH activity. Thus, it was important to prove that the observed effects were not a consequence of deregulated NOTCH activity. We treated NUMB-KD cells with inhibitors of presenilin/&ggr;-secretase (known as &ggr;-secretase inhibitors, GSI) to abolish NOTCH activity. GSI had no significant effect on p53 levels in NUMB-KD or control cells (Fig. 2h and Supplementary Fig. 4a), ruling out participation of NOTCH to the observed NUMB regulation of p53. This was confirmed using gain-of-function mutants of NOTCH (Supplementary Fig. 4b).

We tested whether the NUMB–HDM2 counteraction was relevant to p53-dependent transcriptional activity. HDM2 significantly inhibited the trans-activating ability of p53 on a luciferase reporter gene. However, NUMB restored, albeit not completely, this ability in a dose-dependent fashion (Fig. 2i).

Perturbation of NUMB levels should result in alterations in p53-mediated responses to DNA damage. To monitor DNA damage, we analysed the levels of phosphorylation at serine 139 of histone H2AFX (&ggr;-H2AX) in cells treated with cisplatin, because prolonged persistence of &ggr;-H2AX is considered to be a marker of persistent DNA damage. MCF10A cells were treated and then washed free of the drug and monitored for up to 48 h. In both p53-KD (Fig. 3a) and NUMB-KD (Fig. 3b, c) cells, we detected higher levels of &ggr;-H2AX than in control cells, and demonstrated persistence of &ggr;-H2AX during the cisplatin chase.

We also tested the effects of NUMB or p53 ablation on perturbations of the cell cycle induced by genotoxic drugs. In MCF10A cells, cisplatin induced an S-phase block; this was, however, independent of p53 or NUMB, and thus not informative for our purposes (Fig. 3d and Supplementary Fig. 5). However, doxorubicin and SN38 induced a G2/M block and an S-phase block, respectively, which were partially rescued by NUMB-KD or p53-KD (Fig. 3d). This effect is better illustrated by correcting for the initial percentage of cells in the various phases of the cycle (Fig. 3d, right-most panel), because the ablation of NUMB or p53 caused some alterations in the cell cycle already at steady state.

After treatment and washout with doxorubicin or SN38, MCF10A cells did not efficiently re-enter the cell cycle for at least 20 h (Fig. 3e), probably owing to checkpoint activation. Conversely, NUMB-KD cells and p53-KD cells displayed accelerated exit from the blocked phase (Fig. 3e). Moreover, doxorubicin and SN38 caused a marked block of cell proliferation, which was partly alleviated by the silencing of NUMB or p53. In addition, even after washout of the drugs, the proliferation rate was more sustained in NUMB-KD and p53-KD cells compared to control MCF10A cells (Fig. 3f).

We then performed experiments under conditions of NUMB overexpression. Expression of NUMB–GFP (green fluorescent protein) in MCF10A cells increased the levels of p53 in both unstressed cells and cisplatin- or doxorubicin-treated cells (Supplementary Fig. 6a, b). The moderate, albeit reproducible, increase in p53 possibly reflects the fact that proliferating cells can tolerate only limited amounts of p53. NUMB overexpression also enhanced p53-dependent transcriptional activity, and prolonged p53 half-life (from ∼60 to ∼120 min) (Supplementary Fig. 6a–d).

The above results demonstrate that NUMB overexpression increases p53 stability and activity, predicting enhancement of p53-mediated responses to genotoxicity, such as apoptosis. Thus, we monitored activation of caspases. NUMB–GFP-transfected MCF10A cells displayed an approximately threefold-higher level of activated caspase-3 compared to control cells in response to cisplatin-induced DNA damage—an effect that was abolished by silencing of p53 (Supplementary Fig. 6e, f). These results show that NUMB levels are relevant to the p53-mediated cellular responses.

In breast tumours, loss of NUMB expression is frequently detected. These tumours should harbour reduced p53 levels and impaired p53-mediated responses. We addressed these issues in primary human breast tumour cells. These cells can be cultivated from tumours displaying low or absent levels of NUMB (class 1 tumour cells) or normal levels of NUMB (class 3). We selected eight primary cultures (four of each for class 1 and class 3). The p53 coding sequence was normal in all selected primary cells (not shown).

In class 1 compared with class 3 cells, the steady-state levels of p53 were reduced (Fig. 4a) owing to increased proteasomal degradation, as shown by comparable p53 mRNA levels in class 1 and 3 (Fig. 4b), and to restoration of p53 in class 1 cells by the proteasome inhibitor MG132 (Fig. 4a). The reduction in p53 levels was caused by loss of NUMB, by means of HDM2. Indeed, forced re-expression of NUMB–GFP (Fig. 4c) or silencing of HDM2 (Fig. 4d) restored normal p53 levels in class 1, whereas it had a limited effect, as expected, in class 3 cells.

Deficient p53 activity is associated with resistance to the cytotoxic effects of chemotherapy. Thus, NUMB-defective breast tumours should show resistance to genotoxic anticancer drugs. Indeed, class 1 cells exhibited higher resistance to cisplatin than class 3 cells (Fig. 4e). Re-expression of NUMB in class 1 cells restored responsiveness to cisplatin to levels comparable to those of control class 3 cells (Fig. 4e). NUMB-silencing in class 3 cells increased resistance to the drug to levels comparable to those of control class 1 cells (Fig. 4e). Nutlin restored the susceptibility of class 1 cells to cisplatin (Fig. 4e), and reverted the effects of NUMB ablation in class 3 cells (Fig. 4e), again implicating the HDM2–p53 circuitry.

Finally, we analysed a cohort of 443 breast cancer patients who received adjuvant chemotherapy. We found that NUMB status was inversely correlated with the major clinical and pathological parameters indicative of biologically aggressive neoplastic disease (Supplementary Table 1), and that it behaved as an independent predictor of poor prognosis (Fig. 4f). In conclusion, in breast tumours there is frequent loss of NUMB expression, and this event causes decreased p53 activity. Moreover, loss of NUMB expression is associated with poor prognosis, further arguing its clinical relevance. We note that lack of NUMB also leads to increased NOTCH activity. Thus, the alteration of NUMB concomitantly leads to the activation of an oncogene, NOTCH, and to the attenuation of a tumour suppressor, p53.

Many questions await answers. It remains to be established where NUMB, HDM2 and p53 interact. This is not trivial because HDM2 and p53 are by-and-large nuclear proteins whereas NUMB is in the cytoplasm, mostly associated to biomembranes. Similarly, our findings ask whether endocytosis participates to the regulation of p53, because NUMB is an endocytic protein. Of note, it has been shown that other endocytic proteins control various aspects of p53-mediated functions: dynamin 2 induces p53-dependent apoptosis, and the clathrin heavy chain promotes p53-mediated transcription (see also Supplementary Discussion).

Finally, because NUMB is involved in binary fate decisions, a relevant question is whether NUMB, HDM2 or p53 are involved in the homeostasis of mammary stem cells, and in its subversion in tumours. The role of p53 in stem cells has been widely investigated, mostly in light of the induction of cellular senescence by p53, which in turn can be linked to the depletion of stem cells and to organism ageing. Our data raise the possibility that p53, because of the NUMB liaison, is involved in the initial process that sits at the heart of stem cell fate (that is, asymmetric cell division). Indeed, a role for p53 as a cell-autonomous asymmetric kinetics control gene has been proposed, which might be due to its involvement in regulating the linked phenomenon of immortal DNA strand cosegregation. Thus, our data put forward the possibility that an additional mechanism of tumorigenesis, caused by the lack of the p53/NUMB axis, is the skewing of stem cell division towards a symmetric pattern.

Methods Summary

Cultivation of primary epithelial cells was performed as described previously. Procedures for immunofluorescence, immunoblotting and immunoprecipitation were also performed as described previously. A list of the antibodies and reagents used is in Supplementary Information.

pG13–Luc and expression vectors for p53 and HDM2 were a gift of K. Helin. p53 shRNA pSUPER was extracted from a shRNA library (a gift from R. Bernard). The retroviral PINCO–NUMB–GFP vector was as described. The lentiviral pLL3.7 vector was a gift from L. Van Parijs. All constructs used in this study were sequence-verified. Procedures for retroviral infection and luciferase assays were also described. Procedures for lentiviral infection are shown at http://web.mit.edu/ccr/labs/jacks/protocols/pll37.htm.

Specific siRNA for NUMB and controls were as described previously. NUMBL- and HDM2-specific siRNA were from Dharmacon: HDM2, 5′-GCCACAAAUCUGAUAGUAU-3′; NUMBL, 5′-ACGCCUUCUGCUCAGCCGC-3′.

Real-time PCR for p53, HDM2, p21 (also known as CDKN1A), p53R2 (also known as RRM2B), PUMA (also known as BBC3) and FAS was performed using the TaqMan Gene Expression Assays Indentification: Hs00153349-m1, Hs00242813-m1, Hs00355782-m1, Hs00153085_m1, Hs00248075_m1 and Hs00163653_m1, respectively (Applied Biosystems).

Cells and reagents

Cultivation of primary normal and tumour human mammary epithelial cells was as described previously. These cells can be cultivated from tumours displaying low or absent levels of NUMB (class 1) or from tumours showing normal levels of NUMB (class 3). Primary cultures are used within 7–10 days from explant to prevent adaptation to the cell culture conditions, and constitute, therefore, a rather faithful representation of the parenchymal component of breast tumours.

For survival assays (Fig. 4e), primary tumour cells were plated in six-well plates. Subconfluent cells were treated with cisplatin (12 &mgr;g ml-1) for 15 h and then grown in cisplatin-free medium (plus or minus nutlin) for 96 h. Cells were then stained with 0.05% crystal violet for 10 min and then extensively rinsed with water. The crystal violet retained by live cells was leached in acetic acid (10%) and absorbance was read at 595 nm.

Antibodies were: anti-Flag M2-agarose Affinity Gel from Sigma; anti-p53 (DO-1 and FL-393), anti-NOTCH1 (c-20), anti-HDM2 (SMP14) and anti-PUMA (N-19) from Santacruz Biotechnology; anti-phospho-p53–Ser 15 and anti-cleaved caspase-3 (Asp 175) from Cell Signaling; anti-ubiquitin (FK2, BioMol); anti-phospho-histone H2AX (Ser 139) from Upstate; and anti-NUMB monoclonal antibody (Ab21, generated against amino acids 537–551 of human NUMB). Fluorochrome-conjugated secondary antibodies were from Jackson ImmunoResearch Laboratories. MG132 (Affinity) was used at 10 &mgr;M. The GSIs DAPT (N-[N-(3,5-difluorophenacetyl)-L-alanyl]-S-phenylglycine t-butyl ester), and DFP-AA (N-[N-(3,5-difluorophenacetyl)]-L-alanyl-3-amino-1-methyl-5phenyl-1,3-dihydro-benzo[e](1,4)diazepin-2-one) (Calbiochem) were used at 1 &mgr;M. Nutlin-3 (Cayman) was used at 8 &mgr;M. In the experiments shown in Fig. 1b, Nutlin was used at 20 &mgr;M.

Cell cycle analysis

Cells were washed with PBS and were subsequently fixed in 70% ethanol and either stored at 4 °C or directly used for staining with propidium iodide. For propidium iodide staining, cells were washed once in PBS supplemented with 1% BSA and were counterstained overnight with 5 &mgr;g ml-1 propidium iodide and 40 &mgr;g ml-1 RNase. The samples were analysed with a Becton Dickinson FACScan.

Engineering of vectors, siRNA experiments, and quantitative PCR

pG13–Luc, a luciferase reporter controlled by multimeric p53 binding sites, and expression vectors for p53 and HDM2 were a gift from K. Helin. A retroviral vector to silence p53 (p53-shRNA pSUPER) was extracted from a shRNA library (a gift from R. Bernard). The retroviral PINCO–NUMB–GFP vector was as described. The lentiviral pLL3.7 vector (a gift from L. Van Parijs) was used to clone sh-RNAs under the control of the mouse U6 promoter, upstream of a CMV–GFP expression cassette. Target sequences were: 5′-GGTTAAGTACCTTGGCCATGT-3′ and 5′-AGACGAACAAGTCACCGACTT-3′ for NUMB and control, respectively. Constructs harbouring GST–NUMB, hN&Dgr;E and hNICD (which represent two different gain-of-function mutants of the NOTCH1 receptor, see Supplementary Fig. 4) were engineered by PCR starting from a pcDNAIII FL-NUMB or pcDNAIII FL-NOTCH1 plasmid. All constructs were sequence-verified. Details are available on request. Procedures for retroviral infection and luciferase assays were also described. Procedures for lentiviral infection are at http://web.mit.edu/ccr/labs/jacks/protocols/pll37.htm.

Specific siRNAs for NUMB and the corresponding control were described previously. NUMBL- and HDM2-specific siRNA were obtained from Dharmacon: HDM2, 5′-GCCACAAAUCUGAUAGUAU-3′; NUMBL, 5′-ACGCCUUCUGCUCAGCCGC-3′. Cells were transfected using Oligofectamine (Invitrogen) for 72 h (siRNA concentration, 100 nM). For double NUMB and HDM2 ablation, MCF10A cells were infected with NUMB shRNA (or control shRNA) for 120 h and were then transfected with HDM2 siRNA (or control siRNA) oligonucleotides for 48 h.

The relative quantity of mRNA transcripts for p53, HDM2, p21, p53R2, PUMA and FAS was determined by real-time PCR with TaqMan Gene Expression Assays Identification: Hs00153349-m1, Hs00242813-m1, Hs00355782-m1, Hs00153085_m1, Hs00248075_m1 and Hs00163653_m1, respectively (Applied Biosystems).

Biochemical studies

Procedures for immunofluorescence, immunoblotting and immunoprecipitation were also performed as described.

For the binding assays in Fig. 1b (and in Supplementary Fig. 1b), NUMB, p53 and HDM2 were bacterially expressed as GST-fusion proteins. GST fusions harbouring NUMB and HDM2 were cleaved and purified with the Thrombin Cleavage Capture Kit (Novagen). p53–GST was not cleaved, and was used as such. Purified proteins were incubated for 6 h at 4 °C in 300 &mgr;l of binding buffer (25 mM Tris-Cl, pH 7.2, 50 mM NaCl and 0.2% Nonidet P-40) with continuous shaking. Recovery of proteins was achieved (4 h at 4 °C) with either glutathione agarose beads (Pharmacia; for p53–GST pulldown) or protein G beads (Zymed) preconjugated to NUMB (AB21) or HDM2 (SMP14, Santa Cruz) antibody (for NUMB or HDM2 immunoprecipitation, respectively). Beads were washed three times with a large excess of washing buffer (100 mM Tris-Cl, pH 8.0, 100 mM NaCl and 1% Nonidet P-40), and were boiled in protein sample buffer. Eluted proteins were resolved by SDS–PAGE.

For the in vitro ubiquitination assay in Fig. 2g, NUMB, p53 and HDM2 were bacterially expressed as GST-fusion proteins. GST fusions harbouring NUMB and HDM2 were cleaved and purified with the Thrombin Cleavage Capture Kit (Novagen). Assays were performed as described. In brief, reaction mixtures contained purified enzymes (150 ng E1 (a ubiquitin-activating enzyme), 150 ng purified His-tagged UbcH5B and 0.4 &mgr;g HDM2), 2.0 &mgr;g of GST–p53 fusion and 5 &mgr;g of ubiquitin in ubiquitination buffer (25 mM Tris-HCl, pH 7.6, 5 mM MgCl2, 100 mM NaCl, 1 mM DTT, 2 mM ATP). When appropriate, 2.0 &mgr;g NUMB were added to the reaction mixture. Reactions were incubated at 30 °C for 1 h, followed by four washes in 1% Triton/0.1% SDS buffer, and were subjected to detection in immunoblot.

For sequential immunoprecipitations, total cellular lysates from HEK293 cells overexpressing HDM2, Flag–NUMB and p53 (Supplementary Fig. 1c), or total cellular lysates from U2OS (Fig. 1c, endogenous proteins), were used. After the first immunoprecipitation step using an anti-Flag M2 affinity resin (for the overexpressed Flag–NUMB, Supplementary Fig. 1c) or anti-NUMB (AB21) antibody (for the endogenous protein, Fig. 1c), immunoprecipitates were eluted with the Flag peptide (Sigma) or with the immunogenic NUMB peptide (amino acids 537–551) (Eurogentec), and were then further immunoprecipitated and immunoblotted as described in the legends to the Figures. That panels shown in Fig. 1c were assembled from different lanes of the same blots by splicing out lanes loaded with additional controls.

Experiments with inhibitors of presenilin/&ggr;-secretase

The GSI DFP-AA (also known as Compound E) and DAPT (Calbiochem) were used at 1 &mgr;M in the experiments shown in Fig. 2h and Supplementary Fig. 4a. These compounds are benzodiazepine-type compounds that act as highly potent, selective and non-competitive inhibitors of &ggr;-secretase/presenilin by binding the active site of presenilin-1 and presenilin-2. This enzymatic activity is critical for the physiological activation of NOTCH, because it executes the cleavage of the NOTCH receptor at the so-called S3 site (located at the end of the transmembrane region) to release the NOTCH intracytoplasmic tail; this then translocates to the nucleus and executes NOTCH function.

Selection of breast cancer patients and statistical analysis

To establish a possible correlation between NUMB status of breast tumours and patient outcome, we used data from 443 breast cancer patients enrolled in a surgical trial conducted at the European Institute of Oncology between March 1998 and December 1999. Immunohistochemical analysis of NUMB was performed by tissue microarray, as described. In brief, breast tumours displaying NUMB expression in less than 10% of tumour cells were scored as NUMB-negative (114 patients), whereas tumours showing more than 10% NUMB-expressing cells were scored as NUMB-positive (329 patients).

The 443 patients were followed for a median of 54.8 months (range 4.3–60). During the follow-up period, a total of 40 new events were registered. Ten patients developed a loco-regional event, 9 developed a controlateral carcinoma, and 21 developed a distant metastasis. Association between the clinical/pathological features of the tumours and NUMB expression was evaluated by the Pearson chi-squared test (see Supplementary Table 1). Plots of the cumulative incidence of events according to NUMB expression were drawn using the Kaplan–Meier method and compared by the Log-rank test (see Fig. 4f). Univariate and multivariate analysis was carried out by means of the Cox proportional hazards method to assess the prognostic value of NUMB status before and after correction for well-recognized prognostic factors, including age at diagnosis of the tumour, pathological stage, tumour grade of differentiation, hormone-receptor status, nodal status, Ki-67 and HER-2/neu expression. SAS statistical software was used for all the analysis (SAS Institute, Inc.). A P value of less than 0.05 was considered as significant.

NUMB interacts with and regulates p53.

a, MCF10A lysates (3 mg) were immunoprecipitated (IP) and immunoblotted (IB). The control was an irrelevant antibody. b, Pure HDM2, GST–p53 and NUMB were mixed (3.2 nM each) and the solution was immunoprecipitated and immunoblotted as shown. c, Left, lysates (40 mg) from U2OS cells were immunoprecipitated with anti-NUMB (control, irrelevant antibody) and an aliquot (one-twenty-fifth) of the immunoprecipitate and immunoblot as shown. Right, the immunoprecipitate was eluted with the immunogenic NUMB peptide (amino acids 537–551); the immunoprecipitate and immunoblot were as indicated. Inp, one-fortieth of the eluate; Lys., lysate. d, MCF10A lysates, transfected as shown, were immunoblotted as indicated. e, f, MCF10A cells, transfected as shown, were exposed to cisplatin (24 h). The immunoblot was as indicated. Right, quantification (mean of three experiments) of p53 induction in control (open circles) and NUMB-KD cells (filled circles). g, Quantitative PCR with reverse transcription (RT–PCR) in control siRNA (open bars) and NUMB-KD (filled bars) MCF10A cells. Values represent mean (control siRNA/no cisplatin = 1) ± s.d. from two experiments. Cisplatin, 8 h.

NUMB regulates HDM2-mediated degradation of p53.

a, p53 mRNA levels in control-siRNA and NUMB-siRNA MCF10A cells. Values represent the mean ± s.d. (control siRNA = 1) from two experiments. b, c, MCF10A cells, transfected as shown, were treated with cycloheximide (CHX). Immunoblot was as indicated. In c, quantification of p53 and HDM2 levels in control-siRNA (open circles) and NUMB-siRNA (filled circles) cells are shown; values are expressed relative to time 0 (normalized to vinculin), and represent, in the case of p53, the mean ± s.d. of three experiments. d, e, f, Lysates from MCF10A cells, transfected and treated as indicated, were immunoprecipitated and immunoblotted as shown. In f, p53 levels were normalized by loading proportionally different amounts of cell extracts. g, GST–p53 was subjected to in vitro ubiquitination assay as indicated. Detection was in the immunoblot (Ub, anti-ubiquitin antibody). h, Lysates from MCF10A cells, transfected and treated as shown, were immunoblotted as indicated. i, Luciferase assay in U2OS cells transfected as indicated. Results represent mean ± s.d. from three experiments.

NUMB silencing alters the p53-mediated response to DNA damage.

a, b, MCF10A cells, in which p53 (p53 shRNA, a) or NUMB (NUMB shRNA, b) had been silenced, were treated with cisplatin (6 &mgr;g ml-1 for 15 h) and released in cisplatin-free medium. Immunoblot was as indicated. NT, not treated. c, &ggr;-H2AX (red) in MCF10A cells, infected and treated as in b. Green (GFP), transduced cells. Asterisks, GFP-positive cells with persistent &ggr;-H2AX staining. Arrowheads, non-infected cells with loss of &ggr;-H2AX staining. Blue, 4,6-diamidino-2-phenylindole (DAPI). Original magnification, ×40. d, MCF10A, silenced with p53 shRNA (p53-KD), NUMB shRNA (NUMB-KD) or control shRNA (Control) were mock-treated (-) or treated (+) for 24 h (cisplatin, 9 &mgr;g ml-1; doxorubicin, 0.05 &mgr;g ml-1; SN38, 10 ng ml-1). Bars: solid, G1; grey, S; empty, G2/M. In the far-right ‘Block’ graph, values of the blocked phases are reported (cisplatin and SN38, S; doxorubicin, G2/M) in NUMB-KD (N), p53-KD (53) or control (C) cells. Values are expressed after subtracting cells in the same phase under non-treated conditions. e, MCF10A cells, silenced and treated as in d, were released in drug-free medium (washout) for 6 or 20 h, followed by fluorescence-activated cell sorting. For the ‘Exit from the blocked phase’ graph, values of the blocked phases (doxorubicin, G2/M; SN38, S) are reported versus time of drug washout (0, 6 or 20 h, triangles). f, MCF10A cells, silenced as in d, were treated for 20 h and then released in drug-free medium for 18 h. Cells were counted at the indicated times. df are representative of at least three experiments in triplicates.

Loss of NUMB in human breast tumours determines decreased p53, enhanced chemoresistance and predicts poor prognosis.

a, Class 1 and class 3 cells were treated with MG132 (+). Immunoblot was as indicated. b, p53 transcripts. Results represent mean (normalized to class 3) ± s.d. from four tumours. c, Class 1 and class 3 cells were transduced as shown. Immunoblot was as indicated. d, Class 1 and class 3 cells were transfected with HDM2 siRNA (+) or control siRNA (-). Immunoblot was as indicated. e, Class 1 and class 3 cells were transduced as indicated, treated with cisplatin and nutlin, and analysed for cell viability. Results represent mean ± s.d. from triplicate points. In a, c, d and e, results are representative of four class 1 and four class 3 cultures, from different patients. f, NUMB status (as evaluated by immunohistochemistry) and prognosis (as evaluated by cumulative probability of any secondary event) in patients with breast cancer. Kaplan–Meier curves were compared by the Log-rank test. Hazard ratio before adjustment for clinical and pathological features (Cox proportional hazard method), 0.610 (P, 0.0068); hazard ratio after adjustment, 0.651 (P, 0.0291); Log-rank, 0.00387.

We thank K. Helin for the p53 and HDM2 reagents; L. Van Parijis for the pLL3.7 lentiviral vector; R. Bernard for the p53 shRNA pSUPER vector; G. Matera for technical assistance; P. Maisonneuve and G. Goisis for statistical analysis; the Imaging Service at IEO; and the Real Time PCR Service at IFOM. This work was supported by grants from the Associazione Italiana per la Ricerca sul Cancro and MIUR to S.P. and P.P.D.F., from the European Community (VI Framework), The Ferrari Foundation, the Monzino Foundation and the CARIPLO Foundation to P.P.D.F., and from the G. Vollaro Foundation to S.P.

Author Contributions I.N.C., D.T., F.S.-N. and S.P. performed experimental work. P.N., V.G. and G.V. performed the clinical part of the work (patient selection, histology and data analysis of the patient’s case collection). S.P. and P.P.D.F. planned and supervised the project, performed data analysis and wrote the manuscript.

Roegiers, F. & Jan, Y. N. Asymmetric cell division. Curr. Opin. Cell Biol. 16, 195205 (2004) Santolini, E. et al. Numb is an endocytic protein. J. Cell Biol. 151, 13451352 (2000) Berdnik, D., Torok, T., Gonzalez-Gaitan, M. & Knoblich, J. A. The endocytic protein &agr;-adaptin is required for numb-mediated asymmetric cell division in Drosophila. Dev. Cell 3, 221231 (2002) Hutterer, A. & Knoblich, J. A. Numb and &agr;-adaptin regulate Sanpodo endocytosis to specify cell fate in Drosophila external sensory organs. EMBO Rep. 6, 836842 (2005) Pece, S. et al. Loss of negative regulation by Numb over Notch is relevant to human breast carcinogenesis. J. Cell Biol. 167, 215221 (2004) Vousden, K. H. & Prives, C. p53 and prognosis: new insights and further complexity. Cell 120, 710 (2005) Hollstein, M., Sidransky, D., Vogelstein, B. & Harris, C. C. p53 mutations in human cancers. Science 253, 4953 (1991) Momand, J., Zambetti, G. P., Olson, D. C., George, D. & Levine, A. J. The mdm-2 oncogene product forms a complex with the p53 protein and inhibits p53-mediated transactivation. Cell 69, 12371245 (1992) Haupt, Y., Maya, R., Kazaz, A. & Oren, M. Mdm2 promotes the rapid degradation of p53. Nature 387, 296299 (1997) Kubbutat, M. H., Jones, S. N. & Vousden, K. H. Regulation of p53 stability by Mdm2. Nature 387, 299303 (1997) Zhang, Y., Xiong, Y. & Yarbrough, W. G. ARF promotes MDM2 degradation and stabilizes p53: ARF-INK4a locus deletion impairs both the Rb and p53 tumor suppression pathways. Cell 92, 725734 (1998) Oliner, J. D., Kinzler, K. W., Meltzer, P. S., George, D. L. & Vogelstein, B. Amplification of a gene encoding a p53-associated protein in human sarcomas. Nature 358, 8083 (1992) Sherr, C. J. Tumor surveillance via the ARF–p53 pathway. Genes Dev. 12, 29842991 (1998) McCann, A. H. et al. Amplification of the MDM2 gene in human breast cancer and its association with MDM2 and p53 protein status. Br. J. Cancer 71, 981985 (1995) Pharoah, P. D., Day, N. E. & Caldas, C. Somatic mutations in the p53 gene and prognosis in breast cancer: a meta-analysis. Br. J. Cancer 80, 19681973 (1999) Vestey, S. B. et al. p14ARF expression in invasive breast cancers and ductal carcinoma in situ—relationships to p53 and Hdm2. Breast Cancer Res. 6, R571R585 (2004) Sharpless, N. E. & DePinho, R. A. The INK4A/ARF locus and its two gene products. Curr. Opin. Genet. Dev. 9, 2230 (1999) Silva, J. et al. Analysis of genetic and epigenetic processes that influence p14ARF expression in breast cancer. Oncogene 20, 45864590 (2001) Silva, J. et al. Concomitant expression of p16INK4a and p14ARF in primary breast cancer and analysis of inactivation mechanisms. J. Pathol. 199, 289297 (2003) Juven-Gershon, T. et al. The Mdm2 oncoprotein interacts with the cell fate regulator Numb. Mol. Cell. Biol. 18, 39743982 (1998) Vassilev, L. T. et al. In vivo activation of the p53 pathway by small-molecule antagonists of MDM2. Science 303, 844848 (2004) Rogakou, E. P., Pilch, D. R., Orr, A. H., Ivanova, V. S. & Bonner, W. M. DNA double-stranded breaks induce histone H2AX phosphorylation on serine 139. J. Biol. Chem. 273, 58585868 (1998) Paull, T. T. et al. A critical role for histone H2AX in recruitment of repair factors to nuclear foci after DNA damage. Curr. Biol. 10, 886895 (2000) Reis, T. & Edgar, B. A. Negative regulation of dE2F1 by cyclin-dependent kinases controls cell cycle timing. Cell 117, 253264 (2004) Lowe, S. W., Ruley, H. E., Jacks, T. & Housman, D. E. p53-dependent apoptosis modulates the cytotoxicity of anticancer agents. Cell 74, 957967 (1993) Fish, K. N., Schmid, S. L. & Damke, H. Evidence that dynamin-2 functions as a signal-transducing GTPase. J. Cell Biol. 150, 145154 (2000) Enari, M., Ohmori, K., Kitabayashi, I. & Taya, Y. Requirement of clathrin heavy chain for p53-mediated transcription. Genes Dev. 20, 10871099 (2006) Sharpless, N. E. & DePinho, R. A. Telomeres, stem cells, senescence, and cancer. J. Clin. Invest. 113, 160168 (2004) Sherley, J. L., Stadler, P. B. & Johnson, D. R. Expression of the wild-type p53 antioncogene induces guanine nucleotide-dependent stem cell division kinetics. Proc. Natl Acad. Sci. USA 92, 136140 (1995) Rambhatla, L., Ram-Mohan, S., Cheng, J. J. & Sherley, J. L. Immortal DNA strand cosegregation requires p53/IMPDH-dependent asymmetric self-renewal associated with adult stem cells. Cancer Res. 65, 31553161 (2005) Rubinson, D. A. et al. A lentivirus-based system to functionally silence genes in primary mammalian cells, stem cells and transgenic mice by RNA interference. Nature Genet. 33, 401406 (2003) Woelk, T. et al. Molecular mechanisms of coupled monoubiquitination. Nature Cell Biol. 8, 12461254 (2006) Veronesi, U. et al. A randomized comparison of sentinel-node biopsy with routine axillary dissection in breast cancer. N. Engl. J. Med. 349, 546553 (2003)
doi: 10.1038/nature06412

Poly(ADP-ribose)-binding zinc finger motifs in DNA repair/checkpoint proteins p.81

version="1.0" encoding="utf-8"?> ]>
Nature 451 7174 20080103 81855 0028-0836 1476-4687 2007Nature Publishing Group Supplementary Information

This file contains Supplementary Methods, Supplementary Figures 1-2 and Supplementary Table 1.

Supplementary Video 1

This file contains Supplementary Video 1.

Supplementary Video 2

This file contains Supplementary Video 2.

Supplementary Video 3

This file contains Supplementary Video 3.

Poly(ADP-ribose)-binding zinc finger motifs in DNA repair/checkpoint proteins IvanAhelI DraganaAhelD TakahiroMatsusakaT Allison J.ClarkA J JonathonPinesJ Simon J.BoultonS J Stephen C.WestS C Genetic Recombination and, DNA Damage Response Laboratories, Cancer Research UK London Research Institute, Clare Hall Laboratories, South Mimms, Herts EN6 3LD, UK Wellcome Trust/Cancer Research UK Gurdon Institute, Tennis Court Road, Cambridge CB2 1QR, UK These authors contributed equally to this work. Correspondence and requests for materials should be addressed to S.J.B. (simon.boulton@cancer.org.uk) or S.C.W. (stephen.west@cancer.org.uk). &nature06420-s1;

Post-translational modification (PTM) of proteins plays an important part in mediating protein interactions and/or the recruitment of specific protein targets. PTM can be mediated by the addition of functional groups (for example, acetylation or phosphorylation), peptides (for example, ubiquitylation or sumoylation), or nucleotides (for example, poly(ADP-ribosyl)ation). Poly(ADP-ribosyl)ation often involves the addition of long chains of ADP-ribose units, linked by glycosidic ribose–ribose bonds, and is critical for a wide range of processes, including DNA repair, regulation of chromosome structure, transcriptional regulation, mitosis and apoptosis. Here we identify a novel poly(ADP-ribose)-binding zinc finger (PBZ) motif in a number of eukaryotic proteins involved in the DNA damage response and checkpoint regulation. The PBZ motif is also required for post-translational poly(ADP-ribosyl)ation. We demonstrate interaction of poly(ADP-ribose) with this motif in two representative human proteins, APLF (aprataxin PNK-like factor) and CHFR (checkpoint protein with FHA and RING domains), and show that the actions of CHFR in the antephase checkpoint are abrogated by mutations in PBZ or by inhibition of poly(ADP-ribose) synthesis.

Damaged DNA and the mitotic apparatus (mitotic spindle, centromeres and centrosome) represent major sites of poly(ADP-ribose) accumulation. The specific targeting of proteins to these sites is dependent on the recognition of poly(ADP-ribose) (PAR) by defined PAR-binding motifs or modules. Until now, only two such motifs have been described. One is found in proteins such as p53, histones and XRCC1, and is characterized by a 20 amino acid motif containing a basic residue-rich cluster and a pattern of hydrophobic amino acids interspersed with basic residues; the second is a conserved ∼190-residue domain known as the macro domain, and is found in the poly(ADP-ribose) polymerases PARP9, PARP14 and PARP15 (ref. 8).

CHFR is a ubiquitin ligase that functions in the antephase checkpoint by actively delaying passage into mitosis in response to microtubule poisons. It is frequently mutated in human epithelial cancers, and CHFR-deficient mice develop spontaneous tumours, but a detailed understanding of its function is still emerging. Analysis of the primary sequence of CHFR revealed a conserved putative C2H2 zinc-finger motif at its carboxy terminus. Using homology and pattern searches, similar motifs were found in other eukaryotic DNA repair and checkpoint control proteins (Supplementary Fig. 1a, b). Similarities between a subset of these proteins have been noted, but the function of the motif was not elucidated.

The putative C2H2 zinc-finger is separated by a 6–8 amino acid spacer and has the consensus [K/R]xxCx[F/Y]GxxCxbbxxxxHxxx[F/Y]xH (Supplementary Fig. 1b). On the basis of the data that follows, the motif will be referred to as PBZ. The phylogenetic distribution of the PBZ motif is limited to eukaryotes, excluding yeast, and, as such, its occurrence coincides with the presence of poly(ADP-ribose) polymerases (PARPs). Because the majority of proteins containing PBZ motifs are either directly or indirectly associated with PAR metabolism, we analysed the PAR-binding ability of two human representatives, CHFR and APLF. APLF (C2 or f13) is an FHA-domain protein involved in the DNA damage response. The modular structures of CHFR and APLF revealed one and two PBZ motifs, respectively (Fig. 1a). Additionally, Caenorhabditis elegans DNA ligase III, containing a single PBZ motif at its C terminus, was also analysed. Purified recombinant proteins were dot-blotted onto a nitrocellulose membrane and tested for their ability to bind 32P-labelled poly(ADP-ribose). PAR binding was observed with all three proteins, and was resistant to extensive washing with 1 M salt (Fig. 1b, lanes 1–3). The interaction with PAR was equal to, or better than, that observed with XRCC1, which binds PAR with high affinity (lane 4).

Mutation of the conserved cysteine residues in the single putative PBZ motif within CHFR (Fig. 1a, indicated in red) resulted in the inability of the CHFR*PBZ mutant to bind PAR (Fig. 1c, compare lanes 5 and 6). With APLF, PAR-binding was abolished by mutation of both putative zinc-finger motifs (APLF*PBZ), but not by mutation of a single motif (APLF*PBZ1 and APLF*PBZ2) (Fig. 1c, lanes 1–4). The tandem motifs of APLF, when purified as a recombinant glutathione S-transferase (GST)-tagged protein, exhibited PAR binding (Fig. 1c, lane 7).

Depletion of zinc, by incubation of wild-type CHFR and APLF with the metal-chelating agent EDTA, resulted in a severe reduction in the ability of each protein to bind PAR (Fig. 1d, lanes 2 and 5). Subsequent incubation with excess zinc, however, restored PAR-binding ability to CHFR (lane 6) and C. elegans DNA ligase III (data not shown), but not to APLF which was irreversibly inactivated (lane 3). We conclude that PAR binding is dependent on the presence of zinc, and define this newly identified motif as a PAR-binding zinc-finger or PBZ module. To our knowledge, this is the first description of a zinc finger that is involved in PAR binding.

Interactions between PAR and recombinant CHFR and APLF were further analysed by surface plasmon resonance (Fig. 1e). Wild-type CHFR and APLF bound PAR efficiently, and the kinetics of binding and dissociation are shown in Fig. 1e and Supplementary Table 1. The interactions of PAR with CHFR and APLF were significantly more stable than that observed with XRCC1. The binding of PAR by CHFR*PBZ or APLF*PBZ was not detectable (Fig. 1e).

To investigate whether APLF and CHFR were themselves substrates for poly(ADP-ribosyl)ation, they were incubated with PARP1 in the presence of 32P-labelled NAD (Fig. 1f). PARP1 poly(ADP-ribosyl)ated wild-type APLF and CHFR, as well as APLF*PBZ2 (lanes 2, 3 and 5), whereas the CHFR*PBZ and APLF*PBZ mutants remained unmodified (lanes 4 and 6). Thus, an intact PBZ motif is required for poly(ADP-ribosyl)ation.

Mutational analysis of the PBZ motif revealed that the conserved arginine preceding the zinc finger was required for PAR-binding in both APLF (APLF*R1) and CHFR (CHFR*R1) (Fig. 1g, lanes 6 and 12). Furthermore, mutations at residues following the second cysteine of PBZ compromised PAR binding to CHFR*R2, CHFR*Q, APLF*Y, APLF*R2 and APLF*R2K (lanes 2, 3, 8, and 10, and Supplementary Fig. 2). All APLF mutants deficient for PAR-binding were not poly(ADP-ribosyl)ated by PARP1 in vitro (Supplementary Fig. 2).

We next determined whether APLF and CHFR associate with PAR in vivo. When Flag-tagged APLF and CHFR proteins were transiently expressed in HEK293T cells, we found that the Flag pull downs of each protein contained PAR, as detected by western blotting (Fig. 2a, lanes 3 and 7). The PBZ-inactivating mutations severely reduced or abolished the associations of both APLF and CHFR with PAR (Fig. 2a, lanes 4 and 8, and Fig. 2b, lanes 3, 4, 6 and 7). Interestingly, PARP1 was present in both the APLF and the APLF*PBZ immunoprecipitates (Fig. 2a, lanes 3 and 4), but not in the CHFR pull downs (lanes 7 and 8). We also found that expression of green fluorescent protein (GFP)–APLF and GFP–CHFR fusion proteins in HEK293T cells resulted in the formation of distinct nuclear foci that co-localized with PAR (Fig. 2c), even in the absence of DNA damage. The co-localization of PAR with GFP–APLF and GFP–CHFR, but not with their PBZ mutant derivatives, indicates that the overexpressed wild-type proteins are poly(ADP-ribosyl)ated in vivo. We do not, however, rule out the possibility that the overexpressed proteins might recruit other poly(ADP-ribosyl)ated proteins leading to the observed signals. In control experiments, PAR failed to localize with GFP when expressed without CHFR or APLF (data not shown).

To examine the functional importance of the PBZ motif, we analysed its requirement for the CHFR-dependent antephase checkpoint in Ptk1 cells. Following treatment with microtubule poisons (such as colcemid), late G2 and prophase cells with an intact checkpoint delay entry to mitosis and decondense their chromosomes, while the nuclear envelope remains intact. However, a CHFR deletion mutant lacking the amino-terminal FHA domain (CHFR&Dgr;FHA) acts as a trans-dominant inhibitor of endogenous CHFR function, and its expression abrogates the mitotic delay. We found that mutations of the cysteine residues in the PBZ motif of CHFR&Dgr;FHA (CHFR*PBZ&Dgr;FHA) abolished its ability to act as a trans-dominant inhibitor (Fig. 3a, b, and Supplementary videos 1 and 2). All prophase cells expressing CHFR&Dgr;FHA failed to exhibit an antephase checkpoint when exposed to colcemid (9 out of 9 cells), whereas those expressing CHFR*PBZ&Dgr;FHA returned to interphase (6 out of 6 cells). Thus, the ability of CHFR&Dgr;FHA to act as a dominant negative relies on PBZ, which in turn suggests an involvement of the PBZ motif in regulating CHFR actions.

HeLa cells do not express CHFR and therefore lack an intact antephase checkpoint, leading us to use them as the parental cell line in a direct test of the physiological significance of PBZ in the antephase checkpoint. When HeLa cells were transfected with either wild-type or PBZ-mutated cyan fluorescent protein (CFP)-tagged CHFR and treated with colcemid, we found that expression of wild-type but not mutant CHFR restored the checkpoint defect, as indicated by a decrease in the mitotic index (Fig. 3c). Importantly, the auto-ubiquitylation activity of CHFR, both in vivo and in vitro, was unaffected by mutations in the PBZ motif (Fig. 2a, lanes 7 and 8, and Fig. 3d, lanes 7 and 11). Furthermore, auto-ubiquitylation of wild-type CHFR did not impair its PAR-binding potential (Fig. 3d, lanes 1 and 2). Hence, the disparity between CHFR and its PBZ-mutated variant in the antephase checkpoint most probably reflects differences in their PAR-related functions.

Finally, a direct link between PAR metabolism and the antephase checkpoint was established by treating Ptk1 cells with the specific PARP inhibitor KU-0058948. We found that the PARP inhibitor compromised the ability of Ptk1 cells to delay nuclear envelope breakdown in response to microtubule poisons (Fig. 3e and Supplementary video 3). Instead, the majority of the treated cells continued into mitosis. These results show that inhibition of PAR synthesis compromises the antephase checkpoint, and demonstrate a connection between the PAR-related functions of PBZ and its requirement for CHFR checkpoint regulation.

In this work, we have defined a novel PAR-interaction motif present in a number of proteins associated with the DNA damage response and checkpoint regulation. Although two functionally equivalent domains have previously been reported, this is the first example of a zinc-dependent motif implicated in PAR binding and poly(ADP-ribosyl)ation. Zinc fingers were originally identified as nucleic acid recognition elements, but can also mediate protein–protein interactions. Owing to its chemical composition, PAR may be considered as the third type of nucleic acid, a notion supported by the base stacking and hydrogen-bonding potential of its constituting ADP-ribose residues. Moreover, long PAR chains were postulated to adopt helical conformations, reminiscent of those found with DNA and RNA. In light of this, the evolution of zinc-fingers into PAR-binding elements may seem a suitable consequence of diversification.

The use of the PBZ motif is widespread amongst eukaryotes, and is particularly prominent in Dictyostelium discoideum (Supplementary Fig. 1). The absence of the motif in organisms lacking PARP metabolism (such as prokaryotes and yeasts) may suggest the co-evolution of this motif with PARPs. Importantly, all the PBZ motifs identified in this study were found in proteins potentially regulated by poly(ADP-ribosyl)ation. The majority are DNA damage response proteins, including several PARPs, PARP-related proteins, Ku, Chk2, RAD17, APLF, and proteins involved in single-strand break and base-excision repair (for example, tyrosyl-DNA phosphodiesterase, DNA ligase III and uracil DNA glycosylase). The modulation of DNA ligase III activity by PAR and interactions with poly(ADP-ribosyl)ated PARP1 have been described previously. Similarly, Ku and PARP1 form a complex, the properties of which are changed on its poly(ADP-ribosyl)ation.

Using CHFR, we established the functional importance of the PBZ motif, demonstrating that specific PBZ-targeted mutations abrogate CHFR function in the antephase checkpoint and that treatment with a PARP inhibitor abolished this checkpoint in CHFR-proficient cells. Thus, PAR assumes a major role in modulating CHFR activity, and consequently in regulation of the antephase checkpoint in response to microtubule poisons. The physiological importance of the PBZ motif is further supported by observations that APLF localizes at sites of DNA damage, by a mechanism dependent on the region of APLF containing the PBZ motif and on PAR synthesis. Given that APLF interacts directly with poly(ADP-ribosyl)ated PARP1, we propose that this association defines a role for APLF in DNA break repair.

In general, PAR modifications regulate a dynamic network of intermolecular associations. It has been estimated that PARP1-associated PAR constitutes the major fraction of PAR within the cell. Consequently, automodified PARP1 is likely to attract proteins with PAR-binding motifs, the subsequent poly(ADP-ribosyl)ation of which may be a secondary effect that provides an additional level of regulation. This would be consistent with our results demonstrating efficient binding of PAR by APLF and CHFR, and the ability of these proteins to be poly(ADP-ribosyl)ated by PARP1. Collectively, these data define a novel poly(ADP-ribose)-binding zinc finger and indicate a mechanism by which cells use modification-dependent interactions to orchestrate the assembly of regulatory pathways.

Methods Summary

All proteins were purified after expression in Escherichia coli. Modification by PARP1 was carried out using a PARP activity assay kit (Trevigen). PAR binding was assessed in dot-blot assays and quantitated by Surface Plasmon Resonance using a BIACORE 3000. Off-rates were determined in the presence of PAR. Transient expression of Flag-tagged proteins in human embryonic kidney 293T cells allowed the co-immunoprecipitation of protein–PAR and protein–protein complexes. Transiently expressed GFP-/CFP-tagged proteins and PAR-specific antibodies were used in immunofluoresence studies. CHFR checkpoints in Ptk1 and Hela cells treated with colcemid were analysed by time-lapse differential interference contrast (DIC) and fluorescence microscopy. In vitro CHFR auto-ubiquitylation was performed as described. Detailed experimental procedures are found in Supplementary Information and Methods.


APLF and CHFR clones were obtained from the RZPD German Resource Centre for Genome Research, whereas a C. elegans DNA ligase III construct was purchased from Geneservice. Tandem zinc finger motifs were subcloned from APLF complementary DNA (bases 1,102–1,332) to construct GST–PBZ. Mutations in APLF and CHFR were introduced using the QuickChange II site-directed mutagenesis kit (Stratagene). APLF and DNA ligase III were overexpressed as N-terminally His-tagged proteins using the Gateway pDEST17 vector (Invitrogen), whereas GST- and His-tagged CHFR and GST-PBZ proteins were produced using pET41a (Novagen). All proteins were expressed in E. coli BL21-CodonPlus cells. The cultures were induced at 30 °C with 0.2 mM IPTG for 2 h for APLF expression, or overnight at 18 °C with 0.01 mM IPTG for CHFR, GST–PBZ and DNA ligase III. Recombinant proteins were purified over nickel-NTA-agarose (Qiagen) (APLF and DNA ligase III), or on glutathione Sepharose (GE Healthcare) (CHFR, GST–PBZ). Mutant proteins were overexpressed and purified according to the procedures defined for the wild-type variants. XRCC1 protein was a gift from T. Lindahl.


Rabbit anti-APLF polyclonal antibodies were raised against purified recombinant APLF protein prepared from E. coli. Mouse monoclonal and rabbit polyclonal anti-PAR antibodies (Trevigen), rabbit anti-CHFR antibody and anti-PARP1 (Abcam), and monoclonal anti-ubiquitin antibody (Abcam) were purchased.

Modification by PARP1 in vitro

Recombinant proteins were modified using a PARP activity assay kit (Trevigen). Typically, reactions (10 µl) contained 30 ng of purified PARP1, 1 µM substrate protein, and 100 µM of NAD+ spiked with [32P]-labelled NAD+ (Amersham Biosciences). Reactions were incubated for 10 min at room temperature. Modified proteins were analysed by SDS–PAGE and visualized by autoradiography.

PAR-binding assay

Proteins (2 pmol) were spotted onto a nitrocellulose membrane, which was subsequently blocked with TBS-T buffer (Tris-HCl, pH 7.5, 150 mM NaCl, 0.05% Tween) supplemented with 5% milk. Radioactively labelled PAR was prepared from 200 ng of automodified PARP1 as described. PAR polymer was detached from PARP1 using DNase I and proteinase K and extracted using phenol–chloroform. The water-soluble polymer was then diluted in 10 ml with TBS-T and incubated with the nitrocellulose membrane. The membrane was extensively washed with TBS-T, and TBS-T containing 1 M NaCl, then air-dried and subjected to autoradiography.

For the zinc-depletion experiments, proteins were incubated with 20 mM EDTA overnight at 4 °C. For zinc rebinding, the proteins were desalted and incubated with 1 mM ZnSO4.

Surface plasmon resonance

Biotinylated PAR was coupled to streptavidin-coated BIACORE sensor chips, and assays were carried in 20 mM HEPES, pH 7.2, 150 mM NaCl and 0.005% P20 surfactant. Off rates were measured in the presence of 5 µM PAR (defined in monomeric units) using the coinject function of the Biocore. Biotinylated PAR was produced using 10 µM biotinylated NAD (Trevigen) and 100 µM NAD, as described for the modification by PARP1 in vitro. Proteins to be analysed (1 nM–1 µM) were injected at a flow rate of 20 µl min-1. Binding events were measured in response units.


Human embryonic kidney 293T cells were transiently transfected using Lipofectamine 2000 (Invitrogen), according to the manufacturer’s specifications, with Flag-tagged wild-type and mutant APLF and CHFR constructs. Following transfection (24 h), cells were solubilized in lysis buffer (50 mM Tris-HCl, pH 8.0, 200 mM NaCl, 1% Triton X-100, 1 mM dithiothreitol, 1 mM EDTA) supplemented with 50 U µl-1 benzonase (Novagen) and protein inhibitors (Sigma). Whole cell extracts were clarified by centrifugation and incubated with anti-Flag M2 agarose (Sigma) for 2 h at 4 °C. Following repeated washes with lysis buffer, the immunoprecipitates were boiled in SDS–PAGE loading buffer and analysed by immunoblotting.


HEK293T cells grown on glass coverslips were transfected with GFP-fusion constructs of the wild-type and mutant APLF and CHFR proteins. Lipofectamine 2000 was used as a transfecting agent. Post transfection (24 h), the cells were fixed in 4% paraformaldehyde for 10 min, permeabilized with 0.1% Triton X-100 in PBS and blocked with 2% BSA in PBS. Following incubation with the monoclonal anti-PAR antibody, and Alexa Fluor 546 goat anti-mouse IgG secondary antibody (Invitrogen), the cells were analysed using a Deltavision system.

Ubiquitylation assays

Auto-ubiquitylation was performed essentially as described. Typically, a 100 µl reaction contained 10 ng E1, 200 ng E2 ubiquitin-conjugating enzyme (UbcH5B), 1 µg ubiquitin, 5 pmol GST-tagged CHFR, 4 mM ATP, 20 mM Tris-HCl, pH 7.5, 50 mM NaCl, 0.2 mM DTT and 10 mM MgCl2. Reactions were incubated for 30 min at 30 °C and half the sample was analysed for PAR binding by dot blotting. The other half was mixed with glutathione Sepharose beads for retention of GST–CHFR, and after extensive washing the beads were boiled in SDS–PAGE loading buffer and the proteins resolved by gel electrophoresis.

CHFR checkpoint assays

Ptk1 cells were cultured on 0.15 mm &Dgr;T dishes (Bioptechs) at 37 °C in Ham’s F-12 medium (Gibco), 10% FBS, 100 U ml-1 penicillin, 0.1 mg ml-1 streptomycin, 1 mM Na Pyruvate, and 0.1% (v/v) Fungizone (Gibco). Approximately 5% of the cell volume of CFP-tagged wild-type or mutant CHFR&Dgr;FHA constructs (30 ng µl-1) was injected into Ptk1 cells using a semiautomatic microinjector (Eppendorf) attached to a microscope (model DMIRBE; Leica). Once cells entered into prophase, they were treated with colcemid (15 µM) and their progress followed by time-lapse DIC and fluorescence microscopy at 3-min intervals. Early prophase Ptk1 cells were identified by the beginnings of chromosome condensation, as visualized by DIC microscopy.

HeLa cells were transfected with CFP-tagged wild-type or mutant CHFR by electoroporation (250 V, 1,500 mF, Easyject, Equibio). Following electroporation (18 to 24 h), the medium was replaced with Leibovitz’s L-15 medium (Invitrogen). Colcemid (Sigma) was added to the medium at time 0 at a concentration of 15 µM and the DIC and fluorescence images were taken every 15 min for 18 h on a Leica DMIRB microscope equipped with a 40× 1.2 numerical aperture lens and a CoolSNAP HQ camera (Photometrics) controlled by Slidebook software (Intelligent Imaging Innovations). In these experiments (Fig. 3c), the data are presented as mean ± standard deviation. The statistical significance was determined using a pairwise comparison at 2 and 4 h using a t-test (the antephase delay is about 4 h). The data indicated a significant difference between the untransfected cells and those transfected with wild-type CHFR (P = 0.067 and 0.15).

In the PARP inhibition experiments, Ptk1 cells were pre-treated with 1 µM KU-0058948 or PBS for 1 h and challenged with 15 µM colcemid. Prophase cells that decondensed their chromosomes were scored as returning to interphase, and cells that broke down their nuclear envelopes were scored as continuing to mitosis.

PAR binding/modification mediated by the PBZ zinc finger motif.

a, Schematic diagram of the PBZ motifs in CHFR and APLF. Mutations are indicated in red. b, PAR binding by APLF, CHFR and C. elegans DNA ligase III (lig3) as determined by dot-blot analysis. XRCC1 and BSA were used as controls. c, PAR-binding is abolished by mutations in the PBZ motif. WT, wild type. d, PAR-binding is dependent on zinc. e, Analysis of PAR binding by surface plasmon resonance. f, In vitro poly(ADP-ribosyl)ation of APLF and CHFR by PARP1. g, PAR binding by the APLF and CHFR PBZ mutants.

Interactions of APLF and CHFR with PAR are mediated by the PBZ motif.

a, Immunoprecipitation of Flag-tagged APLF and CHFR from HEK293T extracts. Inputs (10%) and Flag-precipitates were blotted using PAR, APLF and CHFR antibodies. Ubi-CHFR indicates ubiquitylated CHFR. b, Association of PAR with wild-type and PBZ-mutated Flag-tagged APLF. c, Co-localization of wild-type APLF and CHFR proteins with PAR in HEK293T cells transfected with GFP-tagged APLF and CHFR wild-type and mutant constructs. Scale bar, 10 µm.

Checkpoint functions, but not ubiquitylation, of CHFR are affected by PBZ mutations.

a, Ptk1 cells expressing CFP-tagged CHFR&Dgr;FHA (top panels) or CHFR*PBZ&Dgr;FHA (lower panels) were treated with 15 µM colcemid (0 min) and their behaviour was monitored by time-lapse DIC and fluorescence microscopy at 3-min intervals. Scale bar, 10 µm. b, Quantification of the data from a. c, HeLa cells expressing wild-type or mutant CHFR and untransfected cells were treated with colcemid and the mitotic indices (±s.d., n = 3) were determined at two-hourly intervals. d, Auto-ubiquitylation of CHFR does not impair PAR-binding. Wild-type and mutant CHFR were ubiquitylated in vitro and analysed for PAR binding (upper panels) or by western blotting (lower panels). e, Analysis of Ptk1 cells pre-treated with the PARP inhibitor KU-0058948 (1 µM), or PBS for 1 h, and challenged with 15 µM colcemid. Their behaviour was monitored by time-lapse DIC.

We thank T. Lindahl (LRI, CRUK) for XRCC1, G. Smith for the PARP inhibitor KU-0058948, and J. Gannon for assistance with the Biacore. This work was supported by Cancer Research UK, the EU DNA Repair Consortium and the Louis-Jeantet Foundation. I.A. and D.A. are supported by EMBO fellowships.

Author Contributions I.A. and D.A. discovered the PBZ motif and performed most of the experiments. A.J.C. carried out supporting analyses. T.M. and J.P. defined the role of PBZ in the antephase checkpoint. S.J.B. and S.C.W. are joint senior authors who managed the project and helped write the manuscript.

Walsh, C. T. Postranslational Modification of Proteins: Expanding Nature's Inventory (Roberts and Co., Greenwood Village, Colorado, 2006) Seet, B. T. et al. Reading protein modifications with interaction domains. Nature Rev. Mol. Cell Biol. 7, 473483 (2006) D'Amours, D. et al. Poly(ADP-ribosyl)ation reaction in the regulation of nuclear functions. J. Biochem. 342, 249268 (1999) Kim, M. Y., Zhang, T. & Kraus, W. L. Poly(ADP-ribosyl)ation by PARP-1: 'PAR-laying' NAD+ into a nuclear signal. Genes Dev. 19, 19511967 (2005) Hassa, P. O. et al. Nuclear ADP-ribosylation reactions in mammalian cells: where are we today and where are we going? Microbiol. Mol. Biol. Rev. 70, 789829 (2006) Pleschke, J. M. et al. Poly(ADP-ribose) binds to specific domains in DNA damage checkpoint proteins. J. Biol. Chem. 275, 4097440980 (2000) Gagne, J. P. et al. A proteomic approach to the identification of heterogeneous nuclear ribonucleoproteins as a new family of poly(ADP-ribose)-binding proteins. Biochem. J. 371, 331340 (2003) Karras, G. I. et al. The macro domain is an ADP-ribose binding module. EMBO J. 24, 19111920 (2005) Matsusaka, T. & Pines, J. CHFR acts with the p38 stress kinases to block entry to mitosis in mammalian cells. J. Cell Biol. 166, 507516 (2004) Scolnick, D. M. & Halazonetis, T. D. CHFR defines a mitotic stress checkpoint that delays entry into metaphase. Nature 406, 430435 (2000) Kang, D. et al. The checkpoint protein CHFR is a ligase that ubiquitinates Plk1 and inhibits Cdc2 at the G2 to M transition. J. Cell Biol. 156, 249259 (2002) Yu, X. et al. CHFR is required for tumour suppression and Aurora A regulation. Nature Genet. 37, 401406 (2005) Iles, N. et al. APLF (C2orf13) is a novel protein involved in the cellular response to chromosomal DNA strand breaks. Mol. Cell. Biol. 27, 37933803 (2007) Kanno, S.-I. et al. A novel human AP endonuclease with conserved zinc-finger-like motifs involved in DNA strand break responses. EMBO J. 26, 20942103 (2007) Bekker-Jensen, S. et al. Human XIP1 (C2orf13) is a novel regulator of cellular responses to DNA strand breaks. J. Biol. Chem. 282, 1963819643 (2007) Masson, M. et al. XRCC1 is specifically associated with poly(ADP-ribose) polymerase and negatively regulates its activity following DNA-damage. Mol. Cell. Biol. 18, 35633571 (1998) Rieder, C. L. & Cole, R. Entry into mitosis in vertebrate somatic cells is guarded by a chromosome damage checkpoint that reverses the cell cycle when triggered during early but not late prophase. J. Cell Biol. 142, 10131022 (1998) Rieder, C. L. & Cole, R. Microscopy-induced radiation damage, microtubules, and progression through the terminal stage of G2 (prophase) in vertebrate somatic cells. Cold Spring Harb. Symp. Quant. Biol. 65, 369376 (2000) Farmer, H. et al. Targeting the DNA repair defect in BRCA mutant cells as a therapeutic strategy. Nature 434, 917921 (2005) Miller, J., McLachlan, A. D. & Klug, A. Repetitive zinc-binding domains in the protein transcription factor IIIA from Xenopus oocytes. EMBO J. 4, 16091614 (1985) Gamsjaeger, R. et al. Sticky fingers: zinc fingers as protein recognition motifs. Trends Biochem. Sci. 32, 6370 (2007) Minaga, T. & Kun, E. Probable helical conformation of poly(ADP-ribose). The effect of cations on spectral properties. J. Biol. Chem. 258, 57265730 (1983) Leppard, J. B. et al. Physical and functional interaction between DNA ligase III&agr; and poly(ADP-ribose) polymerase 1 in DNA single-strand break repair. Mol. Cell. Biol. 23, 59195927 (2003) Galande, S. & Kohwi-Shigematsu, T. Poly(ADP-ribose) polymerase and Ku autoantigen form a complex and synergistically bind to matrix attachment sequences. J. Biol. Chem. 274, 2052120528 (1999) Burkle, A. Poly(ADP-ribose). The most elaborate metabolite of NAD+. FEBS J. 272, 45764589 (2005) Bothos, J. et al. The CHFR checkpoint protein functions with Ubc13-Mms2 to form Lys63-linked polyubiquitin chains. Oncogene 22, 71017107 (2003)
doi: 10.1038/nature06420

Non-fermentative pathways for synthesis of branched-chain higher alcohols as biofuels p.86

version="1.0" encoding="utf-8"?> ]>
Nature 451 7174 20080103 86894 0028-0836 1476-4687 2007Nature Publishing Group Supplementary Information

The file contains Supplementary Figures 1-4 with Legends and Supplementary Tables 1-4.

Non-fermentative pathways for synthesis of branched-chain higher alcohols as biofuels ShotaAtsumiS TaizoHanaiT James C.LiaoJ C Department of Chemical and Biomolecular Engineering, UCLA-DOE Insitute for Genomics and Proteomics, University of California, Los Angeles, 5531 Boelter Hall, 420 Westwood Plaza, Los Angeles, California 90095, USA Correspondence and requests for materials should be addressed to J.C.L. (liaoj@seas.ucla.edu). &e080103-14; &nature06450-s1;

Global energy and environmental problems have stimulated increased efforts towards synthesizing biofuels from renewable resources. Compared to the traditional biofuel, ethanol, higher alcohols offer advantages as gasoline substitutes because of their higher energy density and lower hygroscopicity. In addition, branched-chain alcohols have higher octane numbers compared with their straight-chain counterparts. However, these alcohols cannot be synthesized economically using native organisms. Here we present a metabolic engineering approach using Escherichia coli to produce higher alcohols including isobutanol, 1-butanol, 2-methyl-1-butanol, 3-methyl-1-butanol and 2-phenylethanol from glucose, a renewable carbon source. This strategy uses the host’s highly active amino acid biosynthetic pathway and diverts its 2-keto acid intermediates for alcohol synthesis. In particular, we have achieved high-yield, high-specificity production of isobutanol from glucose. The strategy enables the exploration of biofuels beyond those naturally accumulated to high quantities in microbial fermentation.

Ethanol is not an ideal fuel because it has a lower energy density than gasoline, and its hygroscopicity poses a problem for storage and distribution. Higher alcohols (C4 and C5), on the other hand, have energy densities closer to gasoline, are not hygroscopic, and are less volatile compared with ethanol. Except for 1-butanol, none of the C4 and C5 alcohols has been produced from a renewable source in a yield high enough to be considered as a gasoline substitute. No microorganisms have been identified to produce, from glucose, higher alcohols such as isobutanol, 2-methyl-1-butanol or 3-methyl-1-butanol to industrially relevant quantities, although small amounts have been identified as microbial by-products.

Here, we devised a synthetic approach to produce the above-mentioned longer chain alcohols as next-generation biofuels. This strategy was implemented in E. coli, although other user friendly hosts such as Saccharomyces cerevisiae are readily applicable. These host organisms have fast growth rates and are facultative anaerobes, allowing for a flexible and economical process design for large-scale production. However, importing and the expression of non-native pathways may lead to metabolic imbalance, whereas the accumulation of the heterologous metabolites may cause cytotoxicity. To achieve high productivity of the target foreign products, it is desirable to seek pathways that are compatible to the host. Therefore, we took advantage of the existing metabolic capability in E. coli and the broad substrate range of the last two steps in the Ehrlich pathway for 2-keto acid degradation from other organisms.

2-Keto acids are intermediates in amino acid biosynthesis pathways. These metabolites can be converted to aldehydes by broad-substrate-range 2-keto-acid decarboxylases (KDCs) and then to alcohols by alcohol dehydrogenases (ADHs). Using this strategy, only two non-native steps were needed to produce biofuels by shunting intermediates from amino acid biosynthesis pathways to alcohol production (Fig. 1a). Amino acid biosynthesis pathways produce various 2-keto acids (Fig. 1b). In this work, six different 2-keto acids for alcohol production were used. The isoleucine biosynthesis pathway generates 2-ketobutyrate and 2-keto-3-methyl-valerate, which can be converted to 1-propanol and 2-methyl-1-butanol, respectively. The valine biosynthesis pathway produces 2-keto-isovalerate, which is the precursor for isobutanol. The leucine biosynthesis pathway generates 2-keto-4-methyl-pentanoate, which is the substrate for 3-methyl-1-butanol. The phenylalanine biosynthesis pathway produces phenylpyruvate, which can lead to 2-phenylethanol. The norvaline biosynthesis pathway, which is a side-reaction of the leucine biosynthesis, produces a substrate for 1-butanol, 2-ketovalerate.

A critical enzyme in this alcohol production strategy is KDC, which is common in plants, yeasts and fungi but less so in bacteria. The aldehydes produced can then be converted to alcohols by an ADH, which is commonly found in many organisms. Some of the KDCs have broad substrate ranges, whereas others are more specific. To test the capability of the endogenous 2-keto acids as a substrate for KDC in E. coli, five KDCs (Pdc6 (ref. 16), Aro10 (ref. 17), Thi3 (ref. 5) from S. cerevisiae, Kivd from Lactococcus lactis, and Pdc from Clostridium acetobutylicum) with alcohol dehydrogenase 2 (Adh2) of S. cerevisiae were overexpressed. E. coli cultures expressing these foreign genes were grown in a minimal media with 0.2 M glucose. Gas chromatography–mass spectrometry (GC–MS) analysis (Table 1) revealed that the strains expressing either Kivd or Aro10 produced all of the expected alcohols. S. cerevisiae Pdc6 and C. acetobutylicum Pdc were not as versatile, whereas S. cerevisiae Thi3 did not have any expected activity. In all of these cases, aldehydes were detected only in trace amounts, indicating sufficient activity of Adh2. These results demonstrate that Kivd is the most active and versatile decarboxylase tested and, therefore, suited for our objectives. Furthermore, addition of various 2-keto acids (Table 2) to the E. coli culture expressing Kivd confirmed the specific production of the corresponding alcohols by 2- to 23-fold. The supply of 2-keto acids also decreased the production of the other alcohols markedly. These results indicate that increasing the flux to the 2-keto acids could improve both the productivity and specificity of production of the alcohols.

The existing E. coli metabolic pathways were then genetically modified to increase the production of the specific 2-keto acid so that the desired alcohol could be produced. To produce isobutanol, the ilvIHCD genes under the control of the PLlacO1 (ref. 20) promoter on a plasmid were overexpressed to enhance 2-ketoisovalerate biosynthesis (Fig. 1b). The amplified ilv pathway was then combined with the alcohol producing pathway (Kivd and Adh2) to achieve isobutanol production. As a result of the ilvIHCD overexpression, the strain produced 23 mM isobutanol, which is a ∼5-fold increase over the strain without ilvIHCD overexpression (Fig. 2a and Supplementary Table 3). These results demonstrate that the synthetic pathway was functional and capable of supplying the 2-ketoisovalerate required for the efficient production of isobutanol. To increase further the isobutanol production, genes that contribute to by-product formation, including adhE, ldhA, frdAB, fnr and pta, were deleted. These deletions could increase the level of pyruvate available for the ilvIHCD pathway. Indeed, this strain produced 30 mM isobutanol, indicating that these deletions were beneficial for isobutanol production. In addition, this strain converted glucose to isobutanol with a yield of 0.21 g of isobutanol per gram of glucose between 16 h and 24 h (Fig. 2a, right panel).

To improve isobutanol production further, the alsS gene from Bacillus subtilis was used instead of ilvIH of E. coli. AlsS of B. subtilis has high affinity for pyruvate, whereas E. coli IlvIH has higher preference for 2-ketobutyrate. The strain with the alsS pathway produced ∼50 mM of isobutanol, which is a ∼1.7-fold increase over the strain using IlvIH (Supplementary Fig. 1). In addition, pflB was deleted to decrease further the competition for pyruvate. The combined effects of these manipulations led to ∼300 mM (22 g l-1) of isobutanol under micro-aerobic conditions (Fig. 2b, left panel and Supplementary Fig. 2). In this experiment, 0.5% yeast extract was supplied in the medium to obtain higher cell density. As a control, this strain produced a negligible amount of isobutanol with 0.5% yeast extract without glucose (Supplementary Fig. 3). The yield reached 0.35 (g isobutanol per g glucose) between 40 h and 112 h (Fig. 2b, right panel), which is 86% of the theoretical maximum. This result demonstrates the potential of this strategy, as high-yield production was achieved even without detailed optimization of the pathways and production conditions.

To demonstrate the generality of this approach, the same strategy was also applied to 1-butanol production. Some clostridial species produce 1-butanol during fermentative growth and many of the enzymes in this pathway are oxygen-sensitive and CoA-dependent. We found that by overexpressing Kivd or Aro10 in E. coli, which does not have the 1-butanol fermentative pathway, the cell produced a small amount of 1-butanol (Table 1) from glucose in a non-fermentative growth, indicating the existence of a corresponding 2-keto acid precursor, 2-ketovalerate. Unfortunately, 2-ketovalerate is not a common metabolite in E. coli. To increase the amount of synthesized 2-ketovalerate, we took advantage of the broad substrate specificity of the leuABCD pathway, the natural substrate of which is 2-ketoisovalerate (Fig. 1b). By using a smaller substrate, 2-ketobutyrate, which has one less methyl group than 2-ketoisovalerate (Fig. 1a), we attempted to synthesize 2-ketovalerate in a manner similar to the steps used in leucine biosynthesis. 2-Ketobutyrate can be generated from l-threonine by the threonine dehydratase, encoded by the ilvA gene, or from an alternative pathway identified in Leptospira interrogans and Methanocaldococcus jannaschii. In the latter pathway, 2-ketobutyrate is generated from citramalate by the enzymes isopropylmalate isomerase (LeuCD) and &bgr;-isopropylmalate dehydrogenase (LeuB).

Therefore, to produce 1-butanol, the operon encoding the ilvA–leuABCD pathway under the control of PLlacO1 (ref. 20) was constructed. It was found that the strain with the ilvA–leuABCD pathway produced 0.6 mM 1-butanol, which is a ∼3-fold increase compared with the strain without overexpression of this pathway (Fig. 2c and Supplementary Table 4). When the media was supplemented with 8 g l-1 l-threonine, a marked increase of 1-butanol production to 3.2 mM was observed, suggesting that 2-ketovalerate could be produced from l-threonine by means of an IlvA-mediated reaction (Fig. 2d).

To improve 1-butanol production further, the ilvD gene was deleted. This gene encodes dihydroxy-acid dehydratase, an enzyme that produces both 2-ketoisovalerate (a precursor for leucine and valine) and 2-keto-3-methyl-valerate (a precursor for isoleucine). This deletion could be beneficial for two reasons. First, the deletion of ilvD eliminates the native substrate, 2-ketoisovalerate, for the leuABCD pathway, thus reducing inhibition by the competitive substrate. Second, the deletion of ilvD eliminates competing substrates for Kivd: 2-keto-3-methyl-valerate and 2-keto-4-methyl-pentanoate. As expected, deletion of ilvD improved 1-butanol production (Fig. 2d).

Because strains of E. coli that hyperproduce l-threonine have been developed for commercial production, it would be straightforward to modify a threonine producing strain with the above strategy for 1-butanol production. For further improvement, it would be necessary to increase the activity of the leuABCD pathway towards the non-native substrate, 2-ketobutyrate, and to raise the specificity of Kivd for 2-ketovalerate. Because 2-ketobutyrate is also the substrate for 1-propanol (Fig. 1b), increasing 2-ketobutyrate availability also enhances the production of 1-propanol (Fig. 2c, d, right). Therefore, increasing the LeuABCD activity and the specificity of KDC would be crucial for high-efficiency 1-butanol production.

Non-native hosts such as E. coli lack tolerance to high alcohols. Isobutanol is slightly less toxic to microorganisms than 1-butanol. The native 1-butanol producers can tolerate concentrations of 1-butanol up to about 2% (w/v) (ref. 1). To show the potential for improving tolerance, we conducted serial transfer of cultures to enrich for isobutanol-tolerant strains. We found that a wild-type E. coli strain (JCL16) was inhibited by 1.5% (w/v) isobutanol. However, after only five rounds of culture transfer with increasing isobutanol concentrations, mutants were found to grow in the presence of 2% (w/v) isobutanol (Supplementary Fig. 4). This level of solvent tolerance is comparable or better than native producers of 1-butanol, suggesting that E. coli can adapt to high concentrations of long-chain alcohols. Other strategies such as global transcription machinery engineering can be used for further improvement of tolerance.

The strategy described above opens up an unexplored frontier for biofuels production, both in E. coli and in other microorganisms. This strategy takes advantage of the well-developed amino acid production technology, and channels the amino acid intermediates to the 2-keto acid degradation pathway for alcohol production. The strategy avoids CoA-mediated chemistry, which is commonly used in alcohol production in native organisms, and enables the synthesis of other higher and complex alcohols on large scales. Specific strategies for producing other alcohols can be readily devised based on the synthetic pathways and metabolic physiology. These strategies can also be implemented in yeast or other industrial microorganisms. In the case of isobutanol production, the complete pathway is CoA-independent and requires only pyruvate as a precursor. This feature avoids the mitochondria compartmentalization issue of acetyl-CoA when implementing the strategy in yeast.

Methods Summary Strains and plasmids

The JCL16 strain is BW25113 (rrnBT14 &Dgr;lacZWJ16 hsdR514 &Dgr;araBADAH33 &Dgr;rhaBADLD78) with F′ transduced from XL-1 blue to supply lacIq. JCL88 is JCL16 with &Dgr;adh, &Dgr;ldh, &Dgr;frd, &Dgr;fnr and &Dgr;pta. JCL260 is the same as JCL88 but with &Dgr;pfl&Bgr;. A list of the strains used is given in Supplementary Table 1. Construction of plasmids is described in Methods, and the primers used are listed in Supplementary Table 2.

Medium and cultivation

Unless stated otherwise, M9 medium containing 0.2 M glucose and 1,000th dilution of Trace Metal Mix A5 (2.86 g H3BO3, 1.81 g MnCl2·4H2O, 0.222 g ZnSO4·7H2O, 0.39 g Na2MoO4·2H2O, 0.079 g CuSO4·5H2O, 49.4 mg Co(NO3)2·6H2O per litre water) was used for cell growth. Ampicillin (100 &mgr;g ml-1) and kanamycin (30 &mgr;g ml-1) were added as appropriate. l-Valine (35 &mgr;g ml-1), l-isoleucine (39.5 &mgr;g ml-1) and l-leucine (39.5 &mgr;g ml-1) were used to culture strains with &Dgr;ilvD. Pre-culture in test tubes containing 3 ml of medium was performed at 37 °C overnight on a rotary shaker (250 r.p.m.). Overnight culture was diluted 1:100 into 20 ml of fresh medium in a 250-ml conical flask. For Fig. 2b, 250-ml screw-cap conical flasks were used. Cells were grown to an optical density at 600 nm of 0.8 at 37 °C, followed by adding 0.1 mM isopropyl-&bgr;-d-thiogalactoside (IPTG). For 1-butanol production (Fig. 2c, d), 8 g l-1 l-threonine was added together with IPTG. Cultivation was performed at 30 °C on a rotary shaker (250 r.p.m.). Gas chromatography–mass spectrometry (GC–MS) and gas chromatography–flame ionization detector (GC–FID) analyses are described in Methods.


Restriction enzymes, Klenow fragment and Antarctic phosphatase were from New England Biolabs. Rapid DNA ligation kit was from Roche. KOD DNA polymerase was from EMD Chemicals. 2-Ketobutyrate, 2-ketoisovalerate, 2-ketovalerate, 2-keto-3-methyl-valerate, 2-keto-4-methyl-pentanoate, phenylpyruvate and glucose assay reagent were from Sigma. Oligonucleotides were from Invitrogen.

KDC and ADH plasmid construction

A list of the oligonucleotides used is given in Supplementary Table 2. To clone PDC6, we used genomic DNA of Saccharomyces cerevisiae (ATCC) as a PCR template with a pair of primers A65 and A66. PCR products were digested with Acc65I and SphI and cloned into pZE12-luc (ref. 20) cut with the same enzyme, creating pSA46.

To clone ADH2, genomic DNA of S. cerevisiae (ATCC) was used as a PCR template with a pair of primers A67 and A68. PCR products were digested with SphI and XbaI and cloned into pSA46 cut with the same enzyme, creating pSA49.

To clone kivd, genomic DNA of Lactococcus lactis (ATCC) was used as a PCR template with a pair of primers A96 and A97. PCR products were digested with Acc65I and SphI and cloned into pSA49 cut with the same enzyme, creating pSA55.

To clone ARO10, we used genomic DNA of S. cerevisiae (ATCC) as a PCR template with a pair of primers A98 and A99. PCR products were digested with Acc65I and SphI and cloned into pSA49 cut with the same enzyme, creating pSA56.

To clone THI3, we used genomic DNA of S. cerevisiae (ATCC) as a PCR template with a pair of primers A100 and A101. PCR products were digested with Acc65I. pSA49 was digested with SphI and blunted with Klenow fragment, followed by digestion with Acc65I. This backbone was ligated with PCR products, creating pSA57.

To clone the pdc gene of Clostridium acetobutylicum, we used genomic DNA of C. acetobutylicum (ATCC) as a PCR template with a pair of primers A102 and A103. PCR products were digested with Acc65I and SphI and cloned into pSA49 cut with the same enzyme, creating pSA58.

ilvIHCD plasmid construction

To replace PLtetO1 of pZE21-MCS1 (ref. 20) with PLlacO1, pZE12-luc was digested with AatII and Acc65I. The shorter fragment was purified and cloned into plasmid pZE21-MCS1 cut with the same enzymes, creating pSA40.

To clone ilvC, genomic DNA of E. coli MG1655 was used as a PCR template with a pair of primers A71 and A72. PCR products were digested with SalI and XmaI and cloned into pSA40 cut with the same enzyme, creating pSA45.

To clone ilvD, genomic DNA of E. coli MG1655 was used as a PCR template with a pair of primers A74 and A84. PCR products were digested with BspEI and MluI and cloned into pSA45 cut with SalI and MluI, creating pSA47.

To clone ilvI and ilvH, genomic DNA of E. coli MG1655 was used as a PCR template with a pair of primers A70 and A83. PCR products were digested with BsaI and SalI and cloned into pSA40 cut with Acc65I and SalI, creating pSA51.

To clone ilvC and ilvD downstream of ilvH, pSA47 was digested with SalI and MluI. The shorter fragment was purified and cloned into plasmid pSA51 cut with the same enzymes, creating pSA52.

To replace replication origin with p15A, pZA31-luc (ref. 20) was digested with SacI and AvrII. The shorter fragment was purified and cloned into plasmid pSA52 cut with the same enzymes, creating pSA54.

alsSilvCD plasmid construction

pSA66 includes the 3′ fragment of an alsS sequence. The alsS sequence was obtained using the genomic DNA of Bacillus subtilis as a PCR template with a pair of primers A123 and A124. PCR products were digested with Acc65I and SalI and cloned into pSA40 cut with the same enzyme.

pSA67 includes alsS sequence. The 5′ fragment of the alsS sequence was obtained using the genomic DNA of B. subtilis as a PCR template with a pair of primers A125 and A126. PCR products were digested with BsrGI and XbaI and cloned into pSA66 cut with Acc65I and XbaI.

pSA68 includes ilvC and ilvD sequence downstream of alsS. pSA47 was digested with SalI and MluI. The shorter fragment was purified and cloned into plasmid pSA67 cut with the same enzymes.

pSA69 was created by transferring the p15A replication origin from pZA31-luc, digested with SacI and AvrII, to plasmid pSA68.

ilvAleuABCD plasmid construction

To clone leuABCD, genomic DNA of E. coli MG1655 was used as a PCR template with a pair of primers A106 and A109. PCR products were digested with SalI and BglII and cloned into pSA40 cut with SalI and BamHI, creating pSA59.

To clone ilvA, genomic DNA of E. coli MG1655 was used as a PCR template with a pair of primers A104 and A105. PCR products were digested with Acc65I and XhoI and cloned into pSA59 cut with Acc65I and SalI, creating pSA60.

To replace replication origin with p15A, pZA31-luc (ref. 20) was digested with SacI and AvrII. The shorter fragment was purified and cloned into plasmid pSA60 cut with the same enzymes, creating pSA62.

GC–MS analysis

Alcohol compounds produced by our strains were identified by GC–MS. The system consisted of model 6890N network GC system (Agilent Technologies), a model 7883B injector and autosampler (Agilent Technologies) and a model 5973 network mass selective detector (Agilent Technologies). A DB-5ms capillary column (30 m, 0.25-mm internal diameter, 0.25-&mgr;m film thickness; Agilent Technologies) was used, with helium (1 ml min-1) as the carrier gas. An oven temperature was programmed from 75 °C (2.6 min) to 200 °C at 30 °C min-1. The injector and detector were maintained at 250 °C. Alcohol compounds were isolated by solvent extraction. Three-hundred microlitres of supernatant of culture broth after centrifugation was extracted with 150 &mgr;l GC standard grade toluene (Fluka). A 1 &mgr;l sample was injected in split injection mode with a 30:1 split ratio.

GC–FID analysis

The produced alcohol compounds were quantified by a gas chromatograph equipped with flame ionization detector. The system consisted of a model 5890A gas chromatograph (Hewlett Packard) and a model 7673A automatic injector, sampler and controller (Hewlett Packard). The separation of alcohol compounds was carried out by A DB-FFAP capillary column (30 m, 0.32-mm internal diameter, 0.25-&mgr;m film thickness; Agilent Technologies). GC oven temperature was initially held at 40 °C for 2 min and raised with a gradient of 5 °C min-1 until 45 °C and held for 4 min. And then it was raised with a gradient 15 °C min-1 until 230 °C and held for 4 min. Helium was used as the carrier gas with 14 p.s.i. inlet pressure. The injector and detector were maintained at 225 °C. A 0.5-&mgr;l sample was injected in splitless injection mode. Methanol was used as the internal standard.

Production of higher alcohols through the synthetic non-fermentative pathways.

a, Various 2-keto acid precursors lead to corresponding alcohols through 2-ketoacid decarboxylase and alcohol dehydrogenase. b, The synthetic networks for the non-fermentative alcohol production in engineered E. coli. Red arrows represent the 2-keto acid decarboxylation and reduction pathway. Blue enzyme names represent amino acid biosynthesis pathways. The double lines represent a side pathway leading to norvaline and 1-butanol biosynthesis.

Summary of results for isobutanol and 1-butanol production in E. coli.

The cells were grown in M9 medium containing 36 g l-1 glucose in shake flasks at 30 °C with or without other nutrients indicated, and induced with 0.1 mM IPTG. Overexpressed genes and nutrient supplementation are indicated below the axis. Error bars indicate s.d. a, Left panel, isobutanol production; right panel, isobutanol yield per g of glucose. The theoretical maximum yield of isobutanol is 0.41 g g-1. Knockout, &Dgr;adh, &Dgr;ldh, &Dgr;frd, &Dgr;fnr and &Dgr;pta. b, Isobutanol production with B. subtilis alsS and yeast extract (5 g l-1) supplementation to increase cell density. The host is JCL260. Detailed results are shown in Supplementary Fig. 2. c, d, Left panel, 1-butanol production; right panel, 1-propanol production in the same strain. The host strain is JCL16, with or without &Dgr;ilvD. l-Threonine, l-threonine (8 g l-1) supplementation.

Alcohol production with KDC and ADH in <i>E. coli</i> Product (&mgr;M) KDC/plasmid Kivd/pSA55 Aro10/pSA56 Pdc6/pSA49 Thi3/pSA57 Pdc (C. acetobutylicum)/pSA58

The strain was JCL16 with various kdc genes and S. cerevisiae ADH2 expressed from plasmids. Culture was grown in M9 medium with 0.2 M glucose plus 0.1 mM IPTG at 30 °C for 40 h. These products were identified by GC–MS and quantified by GC–FID (see Methods). ND, not detectable.

1-Propanol 520 290 125 ND ND Isobutanol 5,242 2,094 260 ND 75 1-Butanol 220 95 ND ND ND 2-Methyl-1-butanol 766 652 56 ND ND 3-Methyl-1-butanol 1,495 1,099 92 ND ND 2-Phenylethanol 324 469 ND ND 175
Alcohol production with the supply of 2-keto acids Product (&mgr;M) 2-Ketobutyrate 2-Keto-isovalerate 2-Ketovalerate 2-Keto-3-methyl-valerate 2-Keto-4-methyl-pentanoate Phenylpyruvate

Strains and culture conditions are the same as described in Table 1. A total of 8 g l-1 of 2-keto acids was added, except for 2-ketovalerate, where 1 g l-1 was added because of its toxicity. ND, not detectable.

1-Propanol 2,138 ND ND ND ND 8 Isobutanol 98 10,016 ND ND ND 64 1-Butanol 492 ND 3,926 ND ND 23 2-Methyl-1-butanol 1,315 ND ND 5,284 ND ND 3-Methyl-1-butanol ND ND 52 ND 3,756 105 2-Phenylethanol 26 109 66 ND ND 7,269

This work was partially supported by UCLA-DOE Institute for Genomics and Proteomics. We are grateful to H. Bujard for plasmids, and members of the Liao laboratory for discussion and comments on the manuscript.

Author Contributions S.A. and J.C.L. designed experiments; S.A. and T.H. performed the experiments; S.A. and J.C.L. analysed the data; and S.A. and J.C.L. wrote the paper.

The University of California has filed a provisional patent application based on this work, and has licensed the technology to Gevo, a biofuel company. J.C.L. serves on the Scientific Advisory Board of Gevo. No financial support was received from Gevo to support this work. Lin, Y. L. & Blaschek, H. P. Butanol production by a butanol-tolerant strain of Clostridium acetobutylicum in extruded corn broth. Appl. Environ. Microbiol. 45, 966973 (1983) Nair, R. V., Bennett, G. N. & Papoutsakis, E. T. Molecular characterization of an aldehyde/alcohol dehydrogenase gene from Clostridium acetobutylicum ATCC 824. J Bacteriol. 176, 871885 (1994) Ingram, L. O. et al. Enteric bacterial catalysts for fuel ethanol production. Biotechnol. Prog. 15, 855866 (1999) Sentheshanuganathan, S. The mechanism of the formation of higher alcohols from amino acids by Saccharomyces cerevisiae. Biochem. J. 74, 568576 (1960) Dickinson, J. R. et al. A 13C nuclear magnetic resonance investigation of the metabolism of leucine to isoamyl alcohol in Saccharomyces cerevisiae. J. Biol. Chem. 272, 2687126878 (1997) Dickinson, J. R., Harrison, S. J. & Hewlins, M. J. An investigation of the metabolism of valine to isobutyl alcohol in Saccharomyces cerevisiae. J. Biol. Chem. 273, 2575125756 (1998) Dickinson, J. R., Harrison, S. J., Dickinson, J. A. & Hewlins, M. J. An investigation of the metabolism of isoleucine to active Amyl alcohol in Saccharomyces cerevisiae. J. Biol. Chem. 275, 1093710942 (2000) Dickinson, J. R. et al. The catabolism of amino acids to long chain and complex alcohols in Saccharomyces cerevisiae. J. Biol. Chem. 278, 80288034 (2003) Farmer, W. R. & Liao, J. C. Improving lycopene production in Escherichia coli by engineering metabolic control. Nature Biotechnol. 18, 533537 (2000) Khosla, C. & Keasling, J. D. Metabolic engineering for drug discovery and development. Nature Rev. Drug Discov. 2, 10191025 (2003) Barbirato, F., Grivet, J. P., Soucaille, P. & Bories, A. 3-Hydroxypropionaldehyde, an inhibitory metabolite of glycerol fermentation to 1,3-propanediol by enterobacterial species. Appl. Environ. Microbiol. 62, 14481451 (1996) Zhu, M. M., Lawman, P. D. & Cameron, D. C. Improving 1,3-propanediol production from glycerol in a metabolically engineered Escherichia coli by reducing accumulation of sn-glycerol-3-phosphate. Biotechnol. Prog. 18, 694699 (2002) Pitera, D. J., Paddon, C. J., Newman, J. D. & Keasling, J. D. Balancing a heterologous mevalonate pathway for improved isoprenoid production in Escherichia coli. Metab. Eng. 9, 193207 (2007) Sentheshanmuganathan, S. & Elsden, S. R. The mechanism of the formation of tyrosol by Saccharomyces cerevisiae. Biochem. J. 69, 210218 (1958) Konig, S. Subunit structure, function and organisation of pyruvate decarboxylases from various organisms. Biochim. Biophys. Acta 1385, 271286 (1998) Hohmann, S. Characterization of PDC6, a third structural gene for pyruvate decarboxylase in Saccharomyces cerevisiae. J. Bacteriol. 173, 79637969 (1991) Vuralhan, Z., Morais, M. A., Tai, S. L., Piper, M. D. & Pronk, J. T. Identification and characterization of phenylpyruvate decarboxylase genes in Saccharomyces cerevisiae. Appl. Environ. Microbiol. 69, 45344541 (2003) de la Plaza, M., Fernandez de Palencia, P., Pelaez, C. & Requena, T. Biochemical and molecular characterization of &agr;-ketoisovalerate decarboxylase, an enzyme involved in the formation of aldehydes from amino acids by Lactococcus lactis. FEMS Microbiol. Lett. 238, 367374 (2004) Russell, D. W., Smith, M., Williamson, V. M. & Young, E. T. Nucleotide sequence of the yeast alcohol dehydrogenase II gene. J. Biol. Chem. 258, 26742682 (1983) Lutz, R. & Bujard, H. Independent and tight regulation of transcriptional units in Escherichia coli via the LacR/O, the TetR/O and AraC/I1–I2 regulatory elements. Nucleic Acids Res. 25, 12031210 (1997) Gollop, N., Damri, B., Chipman, D. M. & Barak, Z. Physiological implications of the substrate specificities of acetohydroxy acid synthases from varied organisms. J. Bacteriol. 172, 34443449 (1990) Woods, D. R. The genetic engineering of microbial solvent production. Trends Biotechnol. 13, 259264 (1995) Bogosian, G. et al. Biosynthesis and incorporation into protein of norleucine by Escherichia coli. J. Biol. Chem. 264, 531539 (1989) Calhoun, D. H., Rimerman, R. A. & Hatfield, G. W. Threonine deaminase from Escherichia coli. I. Purification and properties. J. Biol. Chem. 248, 35113516 (1973) Charon, N. W., Johnson, R. C. & Peterson, D. Amino acid biosynthesis in the spirochete Leptospira: evidence for a novel pathway of isoleucine biosynthesis. J. Bacteriol. 117, 203211 (1974) Howell, D. M., Xu, H. & White, R. H. (R)-citramalate synthase in methanogenic archaea.. J. Bacteriol. 181, 331333 (1999) Xu, H. et al. Isoleucine biosynthesis in Leptospira interrogans serotype lai strain 56601 proceeds via a threonine-independent pathway. J. Bacteriol. 186, 54005409 (2004) Flint, D. H., Emptage, M. H., Finnegan, M. G., Fu, W. & Johnson, M. K. The role and properties of the iron-sulfur cluster in Escherichia coli dihydroxy-acid dehydratase. J. Biol. Chem. 268, 1473214742 (1993) Miwa, K. et al. Construction of L-threonine overproducing strains of Escherichia coli K-12 using recombinant DNA techniques. Agric. Biol. Chem. 47, 23292334 (1983) Alper, H., Moxley, J., Nevoigt, E., Fink, G. R. & Stephanopoulos, G. Engineering yeast transcription machinery for improved ethanol tolerance and production. Science 314, 15651568 (2006)
doi: 10.1038/nature06450

Distinct domains of tRNA synthetase recognize the same base pair p.90

version="1.0" encoding="utf-8"?> ]>
Nature 451 7174 20080103 90934 0028-0836 1476-4687 2007Nature Publishing Group Supplementary Figures

The file contains Supplementary Figures 1-6 with Legends.

Distinct domains of tRNA synthetase recognize the same base pair KirkBeebeK MarissaMockM EveMerrimanE PaulSchimmelP Department of Molecular Biology and Chemistry and The Skaggs Institute for Chemical Biology, The Scripps Research Institute, 10550 North Torrey Pines Road, La Jolla, California 92037, USA These authors contributed equally to this work. Correspondence and requests for materials should be addressed to P.S. (schimmel@scripps.edu). &nature06454-s1;

Synthesis of proteins containing errors (mistranslation) is prevented by aminoacyl transfer RNA synthetases through their accurate aminoacylation of cognate tRNAs and their ability to correct occasional errors of aminoacylation by editing reactions. A principal source of mistranslation comes from mistaking glycine or serine for alanine, which can lead to serious cell and animal pathologies, including neurodegeneration. A single specific G·U base pair (G3·U70) marks a tRNA for aminoacylation by alanyl-tRNA synthetase. Mistranslation occurs when glycine or serine is joined to the G3·U70-containing tRNAs, and is prevented by the editing activity that clears the mischarged amino acid. Previously it was assumed that the specificity for recognition of tRNAAla for editing was provided by the same structural determinants as used for aminoacylation. Here we show that the editing site of alanyl-tRNA synthetase, as an artificial recombinant fragment, targets mischarged tRNAAla using a structural motif unrelated to that for aminoacylation so that, remarkably, two motifs (one for aminoacylation and one for editing) in the same enzyme independently can provide determinants for tRNAAla recognition. The structural motif for editing is also found naturally in genome-encoded protein fragments that are widely distributed in evolution. These also recognize mischarged tRNAAla. Thus, through evolution, three different complexes with the same tRNA can guard against mistaking glycine or serine for alanine.

Mistranslation results from insertion of amino acids at wrong codons. Aminoacyl-tRNA synthetases (aaRSs) provide the main mechanism for production of mischarged tRNAs. These enzymes catalyse attachment of amino acids in a two-step reaction: where the amino acid (aa) is condensed with ATP to form a tightly bound aminoacyl adenylate and PPi is released (equation (1)). The activated aminoacyl group is then transferred from the adenylate to the 3′-end of the tRNA to form aa–tRNA with liberation of AMP and regeneration of enzyme (equation (2)). However, some aaRSs make errors during the amino acid activation step owing to the inherent physiochemical limitations on closely discriminating similar amino acid side chains. An example is alanyl-tRNA synthetase (AlaRS), which misactivates Gly and Ser to yield Gly–tRNAAla and Ser–tRNAAla, respectively. These errors are cleared by a second active site, which is specifically designed for hydrolytic editing: Thus, this special activity prevents mistranslation.

Figure 1a schematically illustrates the design of Escherichia coli AlaRS (encoded by alaS), a polypeptide consisting of 875 amino acids. The amino-terminal 461 amino acids encode the catalytic domain for aminoacylation, which is strictly dependent on a G3˙U70 base pair in the acceptor stem of tRNAAla for aminoacylation. The G3˙U70-dependent recognition of tRNAAla is conserved from bacteria to humans and is so robust that transfer of G3˙U70 into a non-alanine tRNA can be sufficient to confer aminoacylation with alanine. The three-dimensional structure of the aminoacylation domain of Aquifex aeolicus AlaRS corresponds essentially to the N-terminal 461-amino-acid fragment of E. coli AlaRS (AlaRS(1–461)). This fragment is sufficient for aminoacylation in vitro and in vivo, and the determinants (marked in Fig. 1a) for specific (G3˙U70-dependent) recognition of tRNAAla include Asp 235, Asp 285, Arg 314 and Ala 409 and are embedded in the structural format of a class II tRNA synthetase with its seven-stranded &bgr;-structure and flanking &agr;-helices.

For class II tRNA synthetases such as ThrRS (encoded by thrS) and AlaRS, a special insertion or fusion provides the centre for editing. For E. coli AlaRS, this editing centre is encoded by the region from 553–705 (Fig. 1a). Mild editing defects, arising from mutations in the germ line, can be vertically transmitted in the population. Indeed, a minor defect (twofold) in the hydrolytic editing activity of AlaRS leads to heritable ataxia in the mouse. This ataxia is caused by neuronal degeneration of Purkinje cells in the cerebellum, associated with mistranslation-induced triggering of the unfolded protein response. The extreme sensitivity of cells (from bacteria to mammals) to editing defects provides a straightforward rationale for why the editing domain of AlaRS is conserved through evolution. This domain is believed to have been present in the last common ancestor of the tree of life that split into the three great domains: Archaea, Bacteria and Eukarya.

Genome-encoded, active free-standing fragments homologous to editing domains of tRNA synthetases, including AlaRS, are widely distributed in nature. The domains that most closely resemble the editing domain of AlaRS are referred to as AlaXps, and we classify them into two different types on the basis of sequence (Fig. 1a). Type I AlaXps have the core editing domain and a modest amount of flanking sequence. Type II AlaXps have all of the sequence of type I AlaXps, but in addition have an extended carboxy domain much like that found in AlaRSs. Because in our preliminary work several type II AlaXps were active for editing in catalytic amounts, we considered the possibility that the active site for aminoacylation was not needed for capture of mischarged tRNA. Instead, we reasoned that simple fusion of the type II AlaXp-like piece was a straightforward way to provide an editing function for AlaRS. Not clear was how either the type I or the type II AlaXps recognized mischarged tRNAAla and, furthermore, whether that recognition mechanism was also used by the editing site of AlaRS.

To separate out the functional components of AlaRS required for editing and tRNA recognition, deletion mutants lacking the catalytic site for aminoacylation were created. The design of these mutants was roughly based on the sequences of AlaXps, either type I (E. coli AlaRS(438–730)) or type II (E. coli AlaRS(438–875)) (see Fig. 1b). Because the natural AlaXp fragments deacylate misacylated tRNAAla, deletions based on their sequences provided a logical framework for construction of deletions in AlaRS.

Both E. coli AlaRS(438–730) (homologous to type I AlaXp) and E. coli AlaRS(438–875) (homologous to type II AlaXp) were tested for their ability to deacylate Ser–tRNAAla. Whereas E. coli AlaRS(438–875) was fully active for clearance of Ser–tRNAAla, E. coli AlaRS(438–730) was inactive (Fig. 1c). Using the nitrocellulose filter RNA-binding assay, we demonstrated that the inactivity of E. coli AlaRS(438–730) correlated with a lack of binding of tRNAAla (Supplementary Fig. 1). However, at much higher concentrations, E. coli AlaRS(438–730) was capable of specifically deacylating misacylated tRNAAla (Fig. 1c, inset). Thus, the catalytic site for editing was not disrupted in E. coli AlaRS(438–730); instead, the reduction in editing activity resulted from a loss of affinity for tRNA.

Still unclear was whether fragment E. coli AlaRS(438–875), which lacks the ĠU-specific determinants of the aminoacylation domain, would be specific for tRNAAla. Although E. coli AlaRS(438–875) cleared mischarged tRNAAla with robust activity, it failed to deacylate Ser–tRNAThr (Fig. 2a), mirroring the specificity observed with isolated type I AlaXp and full-length AlaRS. Because E. coli AlaRS(438–875) deacylated diverse chimaeric tRNAs that had the acceptor stem of tRNAAla but in which the rest of the sequence was swapped for a different tRNA sequence (Supplementary Fig. 2a, b), the specificity of E. coli AlaRS(438–875) must be governed by the acceptor stem of tRNAAla. As shown in Fig. 1c (inset), high concentrations of E. coli AlaRS(438–730) deacylated Ser–tRNAAla; however, the same or higher concentrations of this fragment failed to deacylate Ser–tRNAThr (Fig. 2a).

Thus, specific determinants for the recognition of tRNAAla are embedded in E. coli AlaRS(438–730). These determinants are inherent to the core editing domain and are, presumably, shared with both type I and II AlaXps, which deacylate misacylated tRNAs with high efficiency (K.B. and P.S., unpublished). Separate from these specificity determinants are extra C-terminal non-specific RNA-binding elements between amino acids 730 and 875 of AlaRS(438–875) (Fig. 1c and Supplementary Fig. 1). To compensate for lacking the extra RNA-binding determinants of the C domain of AlaRSs and type II AlaXps, type I AlaXps have adapted the core editing domain with basic residues to interact more strongly with tRNAAla (refs 25, 26 and Supplementary Fig. 3a, b). In the case of the C domain of AlaRSs and type II AlaXps, we showed in separate experiments that the needed tRNA interaction energy is further localized to non-specific RNA-binding determinants located in the region between amino acids 808 and 875 (Supplementary Fig. 4).

To investigate further the specificity of the newly found determinants for tRNA recognition in the region outside the domain for aminoacylation, we installed the G3˙U70 base pair into Ser–tRNAThr to produce Ser–tRNAThrG3˙U70. Even though E. coli AlaRS(438–875) and the Methanosarcina mazeii type I AlaXp lack the aminoacylation domain of AlaRS, both deacylated Ser–tRNAThrG3˙U70, suggesting that they recognize tRNA for deacylation based, at least in part, on a G3˙U70 base pair (Fig. 2b). At the same concentration of enzyme used to deacylate Ser–tRNAThrG3U70, these enzymes failed to significantly deacylate Ser–tRNAThr (Fig. 2b, inset). A region important for tRNA-specificity was further localized to a predicted strand–loop–strand motif within E. coli AlaRS(438–875). In particular, Arg 693 in the strand–loop–strand motif is highly conserved between AlaRSs and AlaXps and, on the basis of existing structural information, can be modelled to be close to the 3˙70 base pair of tRNAAla (Supplementary Fig. 5). Notably, mutant R693K E. coli AlaRS(438–875) had relaxed specificity for tRNAThr, and deacylated Ser–tRNAThr (Fig. 2c). Thus, the E. coli AlaRS editing domain and M. mazei type I AlaXp share a second, independent way to recognize tRNAAla.

This work shows that AlaRS contains two protein motifs for specific recognition of tRNAAla: one well-studied set of amino acids in the aminoacylation domain and a second, unrelated, structural motif within the editing domain between residues 680 and 699 (Fig. 3). Notably, the data in Figs 1c and 2 show that each of these motifs can recognize tRNAAla in the absence of the other. The need for two distinct motifs for recognition of tRNAAla in the same tRNA synthetase can be rationalized from the severe neurodegeneration in the mouse resulting from even a mild level of mistranslation, in which both glycine and serine were confused for alanine. In addition to the two tRNAAla recognition elements imbedded within AlaRS (Fig. 3), in many organisms (including the mouse) a third mechanism for capture of tRNAAla is provided by genome-encoded AlaXp fragments (Fig. 3). Even though type I AlaXps lack the N-terminal-specific tRNA-binding elements of AlaRSs, we showed here an example from M. mazei that specifically recognized tRNAAla. Mouse AlaXp (also known as Aarsd1) also deacylated misacylated tRNAAla (Supplementary Fig. 6; K.B. and P.S, unpublished). However, we could detect no complex (by pull-down assays) between mouse AlaXp and mouse AlaRS (Aars) (as reported for E. coli ProRS, and the Haemophilus influenzae free-standing editing fragment YbaK), consistent with AlaXp working in isolation in the mouse. It is of interest to determine whether multiple checkpoints guarding against mistranslation are operative through other tRNAs, such as the occasional confusion of serine for threonine that comes from mischarging of tRNAThr by ThrRS, which has an AlaRS-like editing domain for clearing serine from tRNAThr.

Methods Summary Preparation of materials

Constructs described were prepared by PCR of the targeted sequence and cloning of the PCR product. Recombinant protein was produced by E. coli overexpression and Ni-NTA purification. The concentration of purified proteins was determined by Bradford assay. Transfer RNA was produced by either in vivo overexpression or in vitro transcription. Correctly acylated tRNAs were produced by extraction and size exclusion purification of a mixture containing tritiated cognate amino acid, aaRS and the purified tRNA. An aaRS bearing a mutation in the editing site was used for producing incorrectly acylated tRNA. The quantity of acylated tRNA was determined by A260.

Deacylation assays

Assays were performed at 25 °C (pH 7.5) with assay buffer (50 mM HEPES (pH 7.5), 20 mM KCl, 2 mM DTT and 10 mM MgCl2) in 96-well plates as described. Enzyme dilution buffer was added instead of enzyme to determine background hydrolysis. To determine 100% product (and therefore percentage of aa–tRNA remaining), NaOH (20–50 mM) was added to the reaction to liberate all aa from the tRNA during the time course.

The enzyme concentration in Fig. 1c was 10 nM (inset, 20 &mgr;M). For Fig. 2a, 10 nM was used for all enzymes except E. coli AlaRS(438–730) (25 &mgr;M). The assay in Fig. 2b used 200 nM (inset, same concentration). Finally, for the assay in Fig. 2c, 625 nM was used (inset, same concentration). Each enzyme activity towards each substrate was independently verified a minimum of three times (the exception being the ThrRS positive control in Fig. 2a).

Plasmid construction

Plasmids for expression of E. coli alaS deletion mutants were constructed through PCR amplification of the targeted region of the alaS gene with oligonucleotides containing either an Nde1 or an Xho1 site and ligated into pET21b as described above to generate pET21b-EcAlaRS(438–875), pET21b-EcAlaRS(438–730) and pET21b-EcAlaRS(438–808). The plasmid for expression of M. mazei AlaXp was constructed through PCR amplification of genomic DNA (from ATCC) with oligonucleotides containing either an Nde1 or an Xho1 site as well as M. mazei AlaXp primer-specific regions. The PCR product was digested with Nde1 and Xho1, and ligated into pET21b (Novagen) to create plasmid pET21b-mmAlaXp-H6. The plasmid for production of mouse AlaXp (also known as Aarsd1) was constructed by PCR of an Invitrogen complementary DNA clone with Gateway compatible primers. A BP Gateway reaction with the PCR product and pDONR207 resulted in a subclone suitable for combining with pH8GW in a Gateway LR reaction to generate mouse pH8GW-AlaX. Plasmids for in vitro transcription of E. coli tRNAThr (GCTGATATAGCTCAGTTGGTAGAGCGCACCCTTGGTAAGGGTGAGGTCGGCAGTTCGAATCTGCCTATCAGCACCA) and E. coli tRNAThrG3˙U70 (GCGGATATAGCTCAGTTGGTAGAGCGCACCCTTGGTAAGGGTGAGGTCGGCAGTTCGAATCTGCCTATCTGCACCA) were constructed with four overlapping nucleotides in a manner described previously. Mutant variants of the above genes were generated by QuikChange mutagenesis (Stratagene). The coding sequences within the plasmids were confirmed by DNA sequencing.

RNA preparation

Transfer RNA (E. coli tRNAAla(GGC) or E. coli tRNAThr) was produced by in vivo transcription with plasmid pWW-ectRNAAla for the overproduction of E. coli tRNAAla(GGC) and with pWFW1015 for E. coli tRNAThr (ref. 27). Chimaeric tRNAs were produced by in vitro transcription precisely as described. E. coli tRNAThr(G3˙U70) and E. coli tRNAThr were produced by in vitro transcription similar to the method described in ref. 28 but using a MegashortScript kit (Ambion). In vivo transcribed tRNA was used in all the assays except those depicted in Fig. 2b and Supplementary Fig. 2, for which in vitro transcribed tRNA was used.

Production of acylated and misacylated tRNA

The tRNAAla (∼20 &mgr;M) was aminoacylated in the presence of 3 &mgr;M E. coli AlaRS (wild type or C666A/Q584H), 8–30 &mgr;M [3H]-aa, 2 mM ATP and aminoacylation buffer (50 mM HEPES, pH 7.5, 20 mM KCl, 2 mM DTT and 10 mM MgCl2). Similarly, tRNAThr was aminoacylated by wild-type or H64A ThrRS. Reactions were incubated for 20 min at room temperature (23–25 °C), quenched with 0.3 M sodium acetate (pH 5), phenol-extracted once, passed through a PD-10 column (GE Healthcare), ethanol-precipitated, resuspended in 10 mM sodium acetate (pH 5), and stored at –80 °C before use. The concentration of acylated tRNA was determined by absorbance at 260 nm.

Protein expression and purification

E. coli AlaRS, all deletion mutants, and M. mazei AlaXp were prepared by gene expression from plasmid pET21b in BL21-CodonPlus(DE3)-RIL cells (Stratagene). Expression and purification were as described previously except cells were induced with 500 &mgr;M IPTG for 3–5 h. Wild-type and mutant ThrRS were expressed and purified as described. Mouse AlaXp was produced similarly from plasmid pH8GW-mAlaX. Protein concentrations were determined by the Bradford assay.

Deacylation and tRNA-binding assays

Deacylation assays are described in the Methods Summary. Binding of tRNAAla was determined using nitrocellulose filter-binding assays as detailed.

Sequence alignments and three-dimensional structure analysis

An open BLAST search was initiated using the sequence of M. mazei AlaXp. From all of the hits obtained in this search, known AlaRS and ThrRS sequences were discarded. Next, type II AlaXps, which contain the tRNA-binding C-domain, were manually removed. Finally, sequences lacking the two known catalytic motifs (HxxxH and CGGxxH) were discarded. This left approximately three groupings of sequences (group one representing ∼70%; group two, ∼20%; and group three, ∼10%). To select approximately ten sequences for a representative alignment, six sequences of group one were randomly selected, two sequences from group two were selected, and M. mazei (group one) and P. horikoshi (group three) sequences were added. The AlaRS sequences of the corresponding organisms were then selected. The sequence alignments in Supplementary Fig. 5 were generated by BLAST-searching with the E. coli AlaRS sequence and selecting representative sequences from diverse species. The first arginine of the ‘RR’ motif is >90% conserved, and the second one is absolutely conserved (in current genomic databases).

For generation of an AlaRS–tRNA docking model, coordinates of a model of E. coli AlaRS were obtained through the Phyre server (Imperial College, London, http://www.sbg.bio.ic.ac.uk/phyre/). The model generated was based on the coordinates of P. horikoshi AlaXp (Protein DataBank (PDB) accession number, 1WNU). The E. coli AlaRS model and P. horikoshi AlaXp were superimposed on the coordinates of the E. coli ThrRS–tRNAThr complex structure (PDB accession number, 1QF6). Renderings were produced with Chimera (University of California San Francisco, http://www.rbvi.ucsf.edu/chimera/).

Domains of AlaRS and AlaXp and deacylation activity of selected fragments.

a, Domains of E. coli AlaRS and type I and II AlaXps. The domain for amino acid activation and G3˙U70 tRNAAla recognition is red (vertical bars show determinants for acceptor stem and G3˙U70 recognition). The editing domain is violet and the C domain is blue (with the highly conserved portion in dark blue). b, Constructs used in this study. c, Deacylation of Ser–tRNAAla by wild-type AlaRS (circles), E. coli AlaRS(438–875) (upright triangles), E. coli AlaRS(438–730) (squares) or a no-enzyme control (inverted triangles). Inset, specific deacylation of Ser–tRNAAla (squares) and Ala–tRNAAla (triangles) by 2,000× the concentration of AlaRS(438–730), or a no-enzyme control (circles). Data shown represent a typical experiment.

AlaRS and its fragments are specific for tRNAAla and influenced by the G3˙U70 base pair.

a, Deacylation of Ser–tRNAThr with ThrRS (diamonds), E. coli AlaRS(438–875) (circles), M. mazei AlaXp (a type I AlaXp) (inverted triangles), 2,500× concentration of E. coli AlaRS(438–730) (upright triangles) or a no-enzyme control (squares). b, Installing the G3˙U70 base pair into tRNAThr (tRNAThrG3U70) triggers recognition by E. coli AlaRS(438–875) (squares) and type I M. mazei AlaXp (triangles). A no-enzyme control is shown above (inverted triangles). Inset, control showing lack of deacylation of Ser–tRNAThr by E. coli AlaRS(438–875) (triangles) and type I M. mazei AlaXp (circles). c, Perturbation of a conserved ‘RR’ motif leads to loss in G3˙U70 selectivity. Deacylation activity of E. coli AlaRS(438–875) (triangles), E. coli R693K AlaRS(438–875) (diamonds), or a no-enzyme control (squares) towards Ser–tRNAThr. Inset, deacylation of Ser–tRNAAla by E. coli AlaRS(438–875) (triangles), E. coli R693K AlaRS(438–875) (inverted triangles) or a no-enzyme control (squares). Data shown represent a typical experiment.

Multiple checkpoints of tRNAAla recognition for prevention of mistranslation.

The first checkpoint occurs in the N-terminal domain in recognition of tRNAAla and discrimination of alanine versus serine (or glycine). The result is the occasional production of Ser–tRNAAla (refs 1, 3, 14). This minor product is then recognized in a tRNAAla-dependent manner and cleared by a second checkpoint (the editing domain). Finally, any residual Ser–tRNAAla that remains can be cleared by a third checkpoint (AlaXp) that also recognizes tRNAAla. The net effect is that the quantity of misacylated tRNAAla is reduced, and mistranslation from confusing serine or glycine for alanine is prevented.

We thank P. O’Maille for his gift of plasmid pH8GW and M. Sokabe for discussions about the model of the tRNA complex with Pyrococcus horikoshi AlaXp. This work was supported by grants from the National Institutes of Health, the Skaggs Foundation, and the National Foundation for Cancer Research.

Author Contributions K.B., M.M. and E.M. performed experiments and produced all materials. K.B., M.M. and P.S. conceived ideas, designed experiments, and wrote and edited the manuscript. All authors reviewed and approved the final manuscript.

Beebe, K., Ribas De Pouplana, L. & Schimmel, P. Elucidation of tRNA-dependent editing by a class II tRNA synthetase and significance for cell viability. EMBO J. 22, 668675 (2003) Doring, V. et al. Enlarging the amino acid set of Escherichia coli by infiltration of the valine coding pathway. Science 292, 501504 (2001) Lee, J. W. et al. Editing-defective tRNA synthetase causes protein misfolding and neurodegeneration. Nature 443, 5055 (2006) Nangle, L. A., De Crecy Lagard, V., Doring, V. & Schimmel, P. Genetic code ambiguity. Cell viability related to the severity of editing defects in mutant tRNA synthetases. J. Biol. Chem. 277, 4572945733 (2002) Nangle, L. A., Motta, C. M. & Schimmel, P. Global effects of mistranslation from an editing defect in mammalian cells. Chem. Biol. 13, 10911100 (2006) Hou, Y. M. & Schimmel, P. A simple structural feature is a major determinant of the identity of a transfer RNA. Nature 333, 140145 (1988) Hou, Y. M. & Schimmel, P. Evidence that a major determinant for the identity of a transfer RNA is conserved in evolution. Biochemistry 28, 68006804 (1989) McClain, W. H. & Foss, K. Changing the acceptor identity of a transfer RNA by altering nucleotides in a “variable pocket”. Science 241, 18041807 (1988) Swairjo, M. A. et al. Alanyl-tRNA synthetase crystal structure and design for acceptor-stem recognition. Mol. Cell 13, 829841 (2004) Ahel, I., Korencic, D., Ibba, M. & Soll, D. Trans-editing of mischarged tRNAs. Proc. Natl Acad. Sci. USA 100, 1542215427 (2003) An, S. & Musier-Forsyth, K. Trans-editing of Cys-tRNAPro by Haemophilus influenzae YbaK protein. J. Biol. Chem. 279, 4235942362 (2004) An, S. & Musier-Forsyth, K. Cys-tRNAPro editing by Haemophilus influenzae YbaK via a novel synthetasėYbaK tRNA ternary complex. J. Biol. Chem. 280, 3446534472 (2005) Pauling, L. in Festschrift fur Prof. Dr. Arthur Stoll 597–602 (Birkhauser, Basel, 1957) Tsui, W. C. & Fersht, A. R. Probing the principles of amino acid selection using the alanyl-tRNA synthetase from Escherichia coli. Nucleic Acids Res. 9, 46274637 (1981) Putney, S. D. et al. Primary structure of a large aminoacyl-tRNA synthetase. Science 213, 14971501 (1981) Jasin, M., Regan, L. & Schimmel, P. Modular arrangement of functional domains along the sequence of an aminoacyl tRNA synthetase. Nature 306, 441447 (1983) Cusack, S. et al. A second class of synthetase structure revealed by X-ray analysis of Escherichia coli seryl-tRNA synthetase at 2.5 Å. Nature 347, 249255 (1990) Ruff, M. et al. Class II aminoacyl transfer RNA synthetases: crystal structure of yeast aspartyl-tRNA synthetase complexed with tRNA(Asp). Science 252, 16821689 (1991) Dock-Bregeon, A. et al. Transfer RNA-mediated editing in threonyl-tRNA synthetase. The class II solution to the double discrimination problem. Cell 103, 877884 (2000) Sankaranarayanan, R. et al. Zinc ion mediated amino acid discrimination by threonyl-tRNA synthetase. Nature Struct. Biol. 7, 461465 (2000) Karkhanis, V. A., Boniecki, M. T., Poruri, K. & Martinis, S. A. A viable amino acid editing activity in the leucyl-tRNA synthetase CP1-splicing domain is not required in the yeast mitochondria. J. Biol. Chem. 281, 3321733225 (2006) Roy, H., Ling, J., Irnov, M. & Ibba, M. Post-transfer editing in vitro and in vivo by the &bgr; subunit of phenylalanyl-tRNA synthetase. EMBO J. 23, 46394648 (2004) Schimmel, P. & Ribas de Pouplana, L. Formation of two classes of tRNA synthetases in relation to editing functions and genetic code. Cold Spring Harb. Symp. Quant. Biol. 66, 161166 (2001) Regan, L., Bowie, J. & Schimmel, P. Polypeptide sequences essential for RNA recognition by an enzyme. Science 235, 16511653 (1987) Fukunaga, R. & Yokoyama, S. Structure of the AlaX-M trans-editing enzyme from Pyrococcus horikoshii. Acta Crystallogr. D 63, 390400 (2007) Sokabe, M. et al. Molecular basis of alanine discrimination in editing site. Proc. Natl Acad. Sci. USA 102, 1166911674 (2005) Waas, W. F. & Schimmel, P. Evidence that tRNA synthetase-directed proton transfer stops mistranslation. Biochemistry 46, 1206212070 (2007) Beebe, K., Merriman, E. & Schimmel, P. Structure-specific tRNA determinants for editing a mischarged amino acid. J. Biol. Chem. 278, 4505645061 (2003) Beebe, K. et al. A universal plate format for increased throughput of assays that monitor multiple aminoacyl transfer RNA synthetase activities. Anal. Biochem. 368, 111121 (2007)
doi: 10.1038/nature06454

Structure of a tyrosyl-tRNA synthetase splicing factor bound to a group I intron RNA p.94

version="1.0" encoding="utf-8"?> ]>
Nature 451 7174 20080103 94974 0028-0836 1476-4687 2007Nature Publishing Group Supplementary Information

This file contains Supplementary Notes on structure validation, Supplementary Tables 1-3, Supplementary Figures 1-11 with Legends, and additional references.

Structure of a tyrosyl-tRNA synthetase splicing factor bound to a group I intron RNA Paul J.PaukstelisP J Jui-HuiChenJ ElaineChaseE Alan M.LambowitzA M Barbara L.GoldenB L Institute for Cellular and Molecular Biology, Department of Chemistry and Biochemistry, and Section of Molecular Genetics and Microbiology, School of Biological Sciences, University of Texas at Austin, Austin, Texas 78712, USA Department of Biochemistry, Purdue University, West Lafayette, Indiana 47907, USA These authors contributed equally to this work. Correspondence and requests for materials should be addressed to A.M.L. (lambowitz@mail.utexas.edu) or B.L.G. (barbgolden@purdue.edu). &e080103-15; &nature06413-s1;

The ‘RNA world’ hypothesis holds that during evolution the structural and enzymatic functions initially served by RNA were assumed by proteins, leading to the latter’s domination of biological catalysis. This progression can still be seen in modern biology, where ribozymes, such as the ribosome and RNase P, have evolved into protein-dependent RNA catalysts (‘RNPzymes’). Similarly, group I introns use RNA-catalysed splicing reactions, but many function as RNPzymes bound to proteins that stabilize their catalytically active RNA structure. One such protein, the Neurospora crassa mitochondrial tyrosyl-tRNA synthetase (TyrRS; CYT-18), is bifunctional and both aminoacylates mitochondrial tRNATyr and promotes the splicing of mitochondrial group I introns. Here we determine a 4.5-Å co-crystal structure of the Twort orf142-I2 group I intron ribozyme bound to splicing-active, carboxy-terminally truncated CYT-18. The structure shows that the group I intron binds across the two subunits of the homodimeric protein with a newly evolved RNA-binding surface distinct from that which binds tRNATyr. This RNA binding surface provides an extended scaffold for the phosphodiester backbone of the conserved catalytic core of the intron RNA, allowing the protein to promote the splicing of a wide variety of group I introns. The group I intron-binding surface includes three small insertions and additional structural adaptations relative to non-splicing bacterial TyrRSs, indicating a multistep adaptation for splicing function. The co-crystal structure provides insight into how CYT-18 promotes group I intron splicing, how it evolved to have this function, and how proteins could have incrementally replaced RNA structures during the transition from an RNA world to an RNP world.

The group I intron catalytic core has a conserved three-dimensional structure consisting of two extended RNA domains, P4–P6 and P3–P9, which interact to form the intron’s active site (Fig. 1a, b). This active site aligns the splice sites and guanosine substrate and uses specifically bound Mg2+ ions to catalyse splicing by means of guanosine-initiated transesterification reactions. CYT-18 recognizes highly conserved secondary and tertiary structural features of the catalytic core of the intron RNA and is unique among group I intron splicing factors in being able to promote the splicing of various group I introns provided that the catalytic core is accessible. CYT-18 and homologous bacterial TyrRSs consist of a nucleotide-binding-fold domain followed by &agr;-helical and C-terminal domains, and they function as homodimers, with the nucleotide-binding fold of one subunit binding the acceptor stem of the tRNA, and the &agr;-helical and C-terminal domains of the other subunit binding the anticodon and variable arms of the tRNA. However, only the N. crassa mitochondrial TyrRS and that of the closely related fungus Podospora anserina have been found to function in group I intron splicing, suggesting that adaptation of the conserved structure is required for splicing activity.

We recently determined a crystal structure of CYT-18/&Dgr;424–669, which lacks the flexibly attached C-terminal domain but still promotes the splicing of most group I introns. Models based on this structure combined with biochemical data suggested that the protein does not recognize tRNA-like features of the intron as such, but uses instead a distinct RNA-binding surface that includes an &agr;-helical amino-terminal extension (H0) and two other small insertions (Ins1 and Ins2), which are absent from non-splicing bacterial TyrRSs.

Here we determine by molecular replacement a co-crystal structure of CYT-18/&Dgr;424–669 bound to the bacteriophage Twort orf142-I2 group I ribozyme, using data extending to 4.5 Å resolution. At this resolution, interacting regions of the protein and RNA are clearly discernible, but local conformational changes could go undetected (Fig. 1a and Supplementary Fig. 1). The asymmetric unit is composed of four nearly identical complexes related by non-crystallographic symmetry, which significantly improves the parameter-to-observation ratio (Supplementary Table 1). Except for RNA regions involved in crystal contacts or not previously visible, the protein and RNA structures in the complex deviate little from those for the unbound molecules (root mean squared deviations 0.909 and 0.996 Å for C&agr; atoms of the two protein subunits, and 1.56 Å for the RNA; Supplementary Fig. 2). Because the individual RNA and protein structures were determined at significantly higher resolution and seem largely unchanged on binding, the data yield a pseudo-atomic model of the complex. Functional binding of CYT-18/&Dgr;424–669 to the Twort ribozyme is indicated by a 17-fold increased kcat for RNA substrate cleavage under single-turnover conditions in reaction medium containing a low (1 mM) Mg2+ concentration (from 0.015 to 0.26 min-1; not shown). Further, the crystal structure is supported by large amounts of biochemical data identifying RNA–protein interaction sites (Supplementary Table 3), including distance restraints determined by site-directed hydroxyl radical cleavage of the RNA from 11 different amino-acid positions (Supplementary Fig. 3).

The structure shows that the Twort RNA binds across the two CYT-18 subunits (denoted A and B), with the protein contacting both the P4–P6 and P3–P9 domains of the catalytic core of the intron but not peripheral RNA structures. The intron RNA-binding surface is electropositive but does not overlap that which binds tRNATyr (Supplementary Fig. 4). Overall, the RNA–protein interface excludes 1,721 Å2 of solvent-accessible surface area, with most being due to subunit B interactions (1,568 Å2).

The P4–P6 domain of group I introns is a rod-like structure formed by the stacking of the P4 and P6 helices (Fig. 2). Previous biochemical studies indicated that CYT-18 interacts extensively with the P4–P6 domain and promotes its assembly in part by helping to establish the correct geometry around the P4–P6 helical junction. The structure shows that CYT-18 binds along one face of the coaxially stacked P4–P6 helices, with the insertions unique to splicing-competent mitochondrial TyrRSs helping to create RNA-binding pockets (Fig. 2 and Supplementary Fig. 5). The protein binds to the P4–P6 domain as follows: first, in the minor groove at the P4–P6 junction; second, below the P4–P6 junction with Ins1-B jutting into the major groove of P6 and P6a; and third, above the P4–P6 junction in the minor groove of P5a. Notably, the P4–P6 junction binds in a pocket formed from &bgr;-strand D and helix 5 of subunit B, and P5a binds in the same pocket of subunit A (Fig. 2 and Supplementary Fig. 5a). Collectively, these interactions form a clamp that may stabilize the correct conformation of the P4–P6 stacked helices.

In addition to contacting the P4–P6 domain, biochemical studies indicated that CYT-18 also contacts the P3–P9 domain to stabilize the two domains in the correct relative orientation to form the intron’s active site. The relative orientation of the P4–P6 and P3–P9 domains in group I introns is determined principally by four tertiary interactions: the J3/4 and J6/7 nucleotide triples in the minor groove of P6 and major groove of P4, respectively; docking of P3 in the minor groove of J6/6a; and the L9–P5 tetraloop-receptor interaction. Strikingly, the structure shows that CYT-18 binds in a position to promote and/or stabilize all of these interactions, again with key roles for the CYT-18-specific insertions (Fig. 3).

The J3/4 junction region is seen in the structure to interact with the N-terminal extension H0-B and Ins1-B of CYT-18 (Fig. 3). Tyr 41-B is found between U48 and U49, potentially inducing a kink that forces J3/4 into the P6 minor groove for triple formation (Supplementary Fig. 6a), in agreement with previous findings that CYT-18 promotes the J3/4 nucleotide triple interaction.

In J6/7, the first two nucleotides (G117 and C118) form major-groove triples with the first two base pairs of P4, and the last nucleotide (A119) interacts with P7 to help in forming the guanosine-binding site. The CYT-18 core interacts with the same P4 base pairs in the minor groove (see above) and H0-B interacts directly with P7. Thus, the protein potentially stabilizes both the J6/7 interactions and the guanosine-binding site (Fig. 3 and Supplementary Fig. 6a).

The P3–J6/6a interaction may be promoted by CYT-18 in two ways. First, the previously described docking of Ins1-B in the major groove of P6 and P6a (Fig. 2a) may help to establish the correct conformation of J6/6a for the docking of P3 in the minor groove. Second, biochemical evidence suggests that the C-terminal domain of CYT-18, which is absent from CYT-18/&Dgr;424–669, binds P3 (Figs 1b and 3).

Finally, the interaction between the L9 tetraloop and the minor groove of P5 seems to be stabilized by direct contacts with the protein. The L9 tetraloop is positioned by the binding of Ins2-B within the minor groove of P9 and by contacts to H5-A, whereas P5 is oriented by the previously described interaction between subunit A and P5a (Fig. 3 and Supplementary Fig. 6b).

The N. crassa mitochondrial group I introns that are natural substrates for CYT-18 form by themselves most of the conserved secondary structure, but they form little tertiary structure even at high Mg2+ concentrations. The crystal structure is consistent with previous biochemical and genetic analyses, which suggested that CYT-18 binds first to the P4–P6 domain to promote its assembly and then makes additional contacts with the P3–P9 domain to stabilize the correct relative orientation of the two domains to form the active site of the intron RNA. The latter step could occur by CYT-18 serving as a scaffold for tertiary structure nucleation, tertiary structure capture, or some combination of these mechanisms.

The structure suggests that CYT-18 interacts almost exclusively with the intron’s phosphodiester backbone, with the only potential base contacts between residues in Ins2-B and P9, and between H0-B and J3/4. These findings agree with previous chemical footprinting experiments, which showed similarly positioned phosphate-backbone protections in the N. crassa ND1, mitochondrial large subunit rRNA (LSU) and yeast bI5 introns, and few if any base contacts (Fig. 1b shows the correspondence of phosphate-backbone contacts for the ND1 intron). Recognition of the three-dimensional structure of the phosphodiester backbone enables CYT-18 to bind various group I introns, which have little primary sequence similarity but highly conserved secondary and tertiary structures.

The co-crystal structure provides a striking snapshot of how structural functions of RNA can be assumed by proteins. Previous work showed that CYT-18 could replace the peripheral RNA structure P5abc to promote the splicing at low Mg2+ concentrations of a Tetrahymena thermophila LSU intron derivative lacking this structure. Both CYT-18 and P5abc bind the length of the P4–P6 domain with contacts at P5, the P4–P6 junction and a distal region of P6 (P6a for CYT-18 and J6a/6b for P5abc), enabling them to stabilize the backbone conformation on both sides of the P4–P6 junction (Fig. 4). However, P5abc is sequence specific, stabilizing the P4–P6 junction by means of A–minor contacts in the minor groove, whereas CYT-18 seems to contact only the phosphodiester backbone in this region. CYT-18 also differs from P5abc in contacting the P3–P9 domain in addition to the P4–P6 domain. These further interactions explain the ability of CYT-18 to compensate not only for mutations in the P4–P6 domain but also for those that weaken tertiary interactions between the P4–P6 and P3–P9 domains.

The structure also shows how the unique structural adaptations of the N. crassa mitochondrial TyrRS are related to splicing activity. Thus, the CYT-18-specific insertions H0, Ins1 and Ins2, which had been implicated previously in splicing activity, are all seen to interact directly with the intron RNA and potentially to stabilize key tertiary interactions. Further, the protein core has additional structural adaptations, including basic amino-acid substitutions relative to non-splicing bacterial TyrRSs that contribute to group I intron binding (Supplementary Table 3).

Finally, analysis of genome sequences showed that H0, Ins1 and Ins2 are uniquely characteristic of mitochondrial TyrRSs of fungi belonging to the same subphylum as N. crassa and P. anserina (Pezizomycotina), and we confirmed that several of these other mitochondrial TyrRSs have group I intron splicing activity (P.J.P. and A.M.L., unpublished observations). These findings suggest that these mitochondrial TyrRSs adapted to function in splicing after the divergence of the Pezizomycotina and the Saccharomycotina about 360 million years ago. A plausible evolutionary model is that the initial interaction was between the nucleotide-binding fold and the P4–P6 domain, and that H0, Ins1 and Ins2 were acquired subsequently to stabilize interactions with the P3–P9 domain, permitting further degeneration of the RNA structure. The conclusion is that an extended scaffold for the group I intron catalytic core developed in multiple steps on a previously unused protein surface in a relatively short period of evolutionary time. The adaptation of an essential mitochondrial TyrRS for group I intron splicing may have been dictated both by tractable features of the protein and by a distinctive genome surveillance mechanism in these fungi, namely repeat-induced point mutations, that effectively prevents functional gene duplications and thereby limits evolutionary options. The fungal mitochondrial TyrRSs now provide a unique model system for studying how essential proteins in general, and aminoacyl-tRNA synthetases in particular, can progressively acquire new functions and evolve to bind multiple structurally related RNAs.

Methods Summary Crystallization

The Twort ribozyme and CYT-18/&Dgr;424–669 protein were synthesized and purified as described, with the addition of a final Superdex 200 gel-filtration step to the protein purification. To renature the Twort ribozyme, 0.06 mM Twort RNA, 0.07 mM RNA product analogue (5′-GCUU; Dharmacon), 10 mM potassium cacodylate pH 6.5 and 15 mM MgCl2 were heated at 50 °C for 5 min and equilibrated at 18–20 °C for 10 min. CYT-18 protein was added to the RNA such that the molar ratio of dimeric CYT-18/&Dgr;424–669 to RNA was 1:1. Before crystallization, the buffer was exchanged to 10 mM potassium cacodylate pH 6.5, 15 mM MgCl2, 50 mM KCl. Crystals were grown by the hanging-drop method with the use of a well solution of 50 mM potassium cacodylate pH 6.5 and 1.8 M ammonium sulphate.

Data collection and refinement

Data were collected at beamline 23-ID-D at the Advanced Photon Source, and indexed and integrated with HKL2000 (ref. 28) (Supplementary Table 1). Molecular replacement was performed with Phaser, with the dimeric CYT-18/&Dgr;424–669 structure and the Twort ribozyme structure as search models (Supplementary Table 2). The model was refined with CNS (Supplementary Table 1), care being taken to minimize potential phase bias (see Supplementary Notes and Supplementary Figs 1 and 7–11).

Materials for crystallization

A gene encoding a ribozyme derivative of phage Twort orf142-I2 was synthesized by PCR amplification of pTHS17 (ref. 31). The resulting fragment encodes a T7 promoter, nucleotides 9–250 of the intron, and an EarI site. This fragment was cloned between the XbaI and HindIII sites of pUC19 and propagated in Escherichia coli strain XL-1 Blue. To facilitate transcription and crystallization, the second nucleotide in the ribozyme, U10, was mutated to an A, and nucleotides 100–107 were replaced with 5′-GCGGAAACGC-3′. The Twort ribozyme was generated by runoff transcription and purified as described. The product analogue 5′-GCUU-3′ was purchased from Dharmacon, and deprotected and desalted as recommended by the manufacturer.

Crystallization and data collection

CYT-18 protein was added to the RNA such that the molar ratio of dimeric CYT-18/&Dgr;424–669 to RNA was 1:1. Before crystallization, the buffer was exchanged to 10 mM potassium cacodylate pH 6.5, 15 mM MgCl2, 50 mM KCl and concentrated to 100 &mgr;l with a Centricon YM-30 centrifugal filter (Millipore). This resulted in a solution with an A260 of 44.5 and an A280 of 23.8. Crystals were grown by using the hanging-drop method. To make the drops, 2 &mgr;l of the Twort–CYT-18 solution were mixed with 1 &mgr;l of well solution composed of 50 mM potassium cacodylate pH 6.5 and 1.8 M ammonium sulphate. Crystals appeared after 2 days at 4 °C and continued to grow to maximum dimensions of 0.35 mm × 0.5 mm. The crystals were gradually transferred into a cryostabilizing solution containing 50 mM potassium cacodylate pH 6.5, 30 mM MgCl2, 2 mM spermine, 2.4 M ammonium sulphate and 15% xylitol, and then flash frozen in liquid nitrogen before data collection. Data were collected at 100 K on beamline 23-ID-D at the Advanced Photon Source, using a detector distance of 575 mm and an oscillation angle of 0.25°. Reflections were indexed and integrated with HKL2000 (ref. 28) (Supplementary Table 1). The mosaic spread was refined in batches of 20 frames and ranged from 0.821 to 2.111 depending on the orientation of the crystal. Low redundancy and completeness is due primarily to modestly anisotropic diffraction combined with highly mosaic zones of the crystal that result in poor spot profiles and spot rejections. This loss is most pronounced in the highest-resolution shells as a result of the greater propensity for spot overlap and weak diffraction.

Phasing and refinement

Molecular replacement was performed with Phaser with the dimeric CYT-18/&Dgr;424–669 structure and Twort ribozyme structure as search models. Four copies of the protein dimer were located within the asymmetric unit, followed by the four RNA molecules. Supplementary Table 2 shows log-likelihood gain and Z-score statistics after rotation and translation of each search model. The progressive increases in log-likelihood gain and Z-scores with the addition of each search model provide evidence for the correctness of the molecular replacement solution. We also explored P21 as a possible space group, but Phaser showed a decrease in log-likelihood gain and Z-scores after placement of the second protein dimer.

The model was refined with CNS (Supplementary Table 1). Distance restraints were derived from the molecular replacement search models and used to minimize the parameter-to-observation ratio (Supplementary Table 1). These restraints were based on optimal hydrogen bonds as determined by WHAT IF. Only main-chain hydrogen bonds (NH–O) were used for the protein, whereas all optimal hydrogen bonds were used for the RNA. Non-crystallographic symmetry restraints were used between corresponding protein subunits and RNA molecules. Parameters for the bulk solvent model were determined through automated procedures in CNS. Refinement consisted of rigid-body refinement after molecular replacement, one round of refinement (simulated annealing, conjugate gradient minimization and B-factor refinement) before any manual building, and one round of refinement after building of residues in P5a and correction of the P6a and P9.1 helices. A final round of minimization was performed after minor adjustments to the model. Manual building was done with Xfit.

Co-crystal structure of the Twort orf142-I2 group I intron ribozyme bound to CYT-18/&Dgr;424–669.

a, Ribbon diagram. CYT-18 subunits A and B are coloured magenta and violet, respectively, and CYT-18-specific insertions H0, Ins1 and Ins2 are coloured cyan. b, Secondary structure of the Twort orf142-I2 intron ribozyme showing nucleotide residues within 4 Å of the protein in the co-crystal structure (circled). Boxed nucleotide residues correspond to phosphodiester-backbone positions protected by full-length CYT-18 in the N. crassa ND1 intron. Protections in P3, P5 and P8, which are not seen in the co-crystal structure, are attributable to the C-terminal domain of CYT-18, which is absent from CYT-18/&Dgr;424–669 (see Fig. 3).

CYT-18 binding to the P4–P6 domain.

a, CYT-18 binds the length of the P4–P6 domain, with the CYT-18-specific insertions (cyan) contributing to the formation of binding pockets for different regions. CYT-18 is drawn as a surface model, with subunits A and B coloured magenta and violet, respectively. b, c, Putative contacts between CYT-18 and the P4–P6 domain. The surfaces of protein atoms within 4 Å of the RNA are coloured red, atoms within 4–5 Å yellow, atoms within 5–10 Å peach, and atoms farther away than 10 Å orange. The views in b and c are rotated by 90° around the vertical axis.

CYT-18-specific insertions bridge the P4–P6 and P3–P9 domains.

RNA and protein are coloured as in Fig. 1. The nucleophilic guanosine residue (&ohgr;G) bound at P7 is shown in orange. Spheres show backbone phosphates in P3 that are putatively protected by the C-terminal domain of CYT-18, which is absent from CYT-18/&Dgr;424–669 (see Fig. 1).

CYT-18 and the P5abc peripheral RNA structure found in some group I introns interact similarly with the P4–P6 domain.

a, Orthogonal ribbon diagrams of the CYT-18/&Dgr;424–669–Twort ribozyme co-crystal structure. b, The corresponding views of the T. thermophila LSU intron crystal structure, with the P5abc domain in magenta.

We thank C. Correll, M. Hermodson, R. Russell and J. Tesmer for comments on the manuscript; T. Cech for discussions; the staff of SER-CAT and GM/CA-CAT, N. Sanishvili, D. Khare and M. Oldham for assistance with crystallographic data collection; and H. Kim for performing kinetic assays. Data were collected at GM/CA-CAT and SER-CAT beamlines at the Advanced Photon Source, Argonne National Laboratory. This work was supported by a grant from the National Institutes of Health to A.M.L.

Author Contributions P.J.P. and E.C. prepared materials for crystallization. E.C. crystallized the CYT-18–Twort RNA complex. J-H.C. collected and processed diffraction data. P.J.P. solved the structure. P.J.P., B.L.G. and A.M.L. interpreted data and wrote the paper.

Lambowitz, A. M., Caprara, M. G., Zimmerly, S. & Perlman, P. S. in The RNA World, 2nd edn (eds Gesteland, R.F., Cech, T.R. & Atkins, J.F.) 451485 (Cold Spring Harbor Press, New York, 1999) Lambowitz, A. M. & Perlman, P. S. Involvement of aminoacyl-tRNA synthetases and other proteins in group I and group II intron splicing. Trends Biochem. Sci. 15, 440444 (1990) Akins, R. A. & Lambowitz, A. M. A protein required for splicing group I introns in Neurospora mitochondria is mitochondrial tyrosyl-tRNA synthetase or a derivative thereof. Cell 50, 331345 (1987) Michel, F. & Westhof, E. Modelling of the three-dimensional architecture of group I catalytic introns based on comparative sequence analysis. J. Mol. Biol. 216, 585610 (1990) Adams, P. L., Stahley, M. R., Kosek, A. B., Wang, J. & Strobel, S. A. Crystal structure of a self-splicing group I intron with both exons. Nature 430, 4550 (2004) Guo, F., Gooding, A. R. & Cech, T. R. Structure of the Tetrahymena ribozyme: base triple sandwich and metal ion at the active site. Mol. Cell 16, 351362 (2004) Golden, B. L., Kim, H. & Chase, E. Crystal structure of a phage Twort group I ribozyme–product complex. Nat. Struct. Mol. Biol. 12, 8289 (2005) Mohr, G., Zhang, A., Gianelos, J. A., Belfort, M. & Lambowitz, A. M. The Neurospora CYT-18 protein suppresses defects in the phage T4 td intron by stabilizing the catalytically active structure of the intron core. Cell 69, 483494 (1992) Guo, Q. & Lambowitz, A. M. A tyrosyl-tRNA synthetase binds specifically to the group I intron catalytic core. Genes Dev. 6, 13571372 (1992) Mohr, G., Caprara, M. G., Guo, Q. & Lambowitz, A. M. A tyrosyl-tRNA synthetase can function similarly to an RNA structure in the Tetrahymena ribozyme. Nature 370, 147150 (1994) Caprara, M. G., Lehnert, V., Lambowitz, A. M. & Westhof, E. A tyrosyl-tRNA synthetase recognizes a conserved tRNA-like structural motif in the group I intron catalytic core. Cell 87, 11351145 (1996) Paukstelis, P. J. et al. A tyrosyl-tRNA synthetase adapted to function in group I intron splicing by acquiring a new RNA binding surface. Mol. Cell 17, 417428 (2005) Yaremchuk, A., Kriklivyi, I., Tukalo, M. & Cusack, S. Class I tyrosyl-tRNA synthetase has a class II mode of cognate tRNA recognition. EMBO J. 21, 38293840 (2002) Kämper, U., Kück, U., Cherniack, A. D. & Lambowitz, A. M. The mitochondrial tyrosyl-tRNA synthetase of Podospora anserina is a bifunctional enzyme active in protein synthesis and RNA splicing. Mol. Cell. Biol. 12, 499511 (1992) Kleinjung, J. & Fraternali, F. POPSCOMP: an automated interaction analysis of biomolecular complexes. Nucleic Acids Res. 33, W3426 (2005) Chen, X., Gutell, R. R. & Lambowitz, A. M. Function of tyrosyl-tRNA synthetase in splicing group I introns: an induced-fit model for binding to the P4–P6 domain based on analysis of mutations at the junction of the P4–P6 stacked helices. J. Mol. Biol. 301, 265283 (2000) Caprara, M. G., Myers, C. A. & Lambowitz, A. M. Interaction of the Neurospora crassa mitochondrial tyrosyl-tRNA synthetase (CYT-18 protein) with the group I intron P4–P6 domain. Thermodynamic analysis and the role of metal ions. J. Mol. Biol. 308, 165190 (2001) Caprara, M. G., Mohr, G. & Lambowitz, A. M. A tyrosyl-tRNA synthetase protein induces tertiary folding of the group I intron catalytic core. J. Mol. Biol. 257, 512531 (1996) Webb, A. E., Rose, M. A., Westhof, E. & Weeks, K. M. Protein-dependent transition states for ribonucleoprotein assembly. J. Mol. Biol. 309, 10871100 (2001) Myers, C. A. et al. A tyrosyl-tRNA synthetase suppresses structural defects in the two major helical domains of the group I intron catalytic core. J. Mol. Biol. 262, 87104 (1996) Waldsich, C., Grossberger, R. & Schroeder, R. RNA chaperone StpA loosens interactions of the tertiary structure in the td group I intron in vivo. Genes Dev. 16, 23002312 (2002) Weeks, K. M. & Cech, T. R. Assembly of a ribonucleoprotein catalyst by tertiary structure capture. Science 271, 345348 (1996) Cate, J. H. et al. Crystal structure of a group I ribozyme domain: principles of RNA packing. Science 273, 16781685 (1996) Chen, X., Mohr, G. & Lambowitz, A. M. The Neurospora crassa CYT-18 protein C-terminal RNA-binding domain helps stabilize interdomain tertiary interactions in group I introns. RNA 10, 634644 (2004) Mohr, G., Rennard, R., Cherniack, A. D., Stryker, J. & Lambowitz, A. M. Function of the Neurospora crassa mitochondrial tyrosyl-tRNA synthetase in RNA splicing. Role of the idiosyncratic N-terminal extension and different modes of interaction with different group I introns. J. Mol. Biol. 307, 7592 (2001) Galagan, J. E., Henn, M. R., Ma, L. J., Cuomo, C. A. & Birren, B. Genomics of the fungal kingdom: insights into eukaryotic biology. Genome Res. 15, 16201631 (2005) Selker, E. U. Premeiotic instability of repeated sequences in Neurospora crassa. Annu. Rev. Genet. 24, 579613 (1990) Otwinowski, Z. & Minor, W. Processing of X-ray diffraction data collected in oscillation mode. Methods Enzymol. 276, 307326 (1997) McCoy, A. J., Grosse-Kunstleve, R. W., Storoni, L. C. & Read, R. J. Likelihood-enhanced fast translation functions. Acta Crystallogr. D Biol. Crystallogr. 61, 458464 (2005) Brünger, A. T. et al. Crystallography & NMR system: A new software suite for macromolecular structure determination. Acta Crystallogr. D Biol. Crystallogr. 54, 905921 (1998) Landthaler, M. & Shub, D. A. Unexpected abundance of self-splicing introns in the genome of bacteriophage Twort: introns in multiple genes, a single gene with three introns, and exon skipping by group I ribozymes. Proc. Natl Acad. Sci. USA 96, 70057010 (1999) DeLaBarre, B. & Brünger, A. T. Considerations for the refinement of low-resolution crystal structures. Acta Crystallogr. D Biol. Crystallogr. 62, 923932 (2006) Vriend, G. WHAT IF: A molecular modeling and drug design program. J. Mol. Graph. 8, 5256 (1990) McRee, D. E. XtalView/Xfit—a versatile program for manipulating atomic coordinates and electron density. J. Struct. Biol. 125, 156165 (1999)
doi: 10.1038/nature06413

Structure of the sulphiredoxin–peroxiredoxin complex reveals an essential repair embrace p.98

version="1.0" encoding="utf-8"?> ]>
Nature 451 7174 20080103 981014 0028-0836 1476-4687 2007Nature Publishing Group Supplementary Information

The file contains Supplementary Table S1, Supplementary Figures S1-S9 with Legends and additional references.

Structure of the sulphiredoxin–peroxiredoxin complex reveals an essential repair embrace Thomas J.JönssonT J Lynnette C.JohnsonL C W. ToddLowtherW T Center for Structural Biology and Department of Biochemistry, Wake Forest University School of Medicine, Medical Center Boulevard, Winston-Salem, North Carolina 27157, USA Correspondence and requests for materials should be addressed to W.T.L. (tlowther@wfubmc.edu). &nature06415-s1;

Typical 2-Cys peroxiredoxins (Prxs) have an important role in regulating hydrogen peroxide-mediated cell signalling. In this process, Prxs can become inactivated through the hyperoxidation of an active site Cys residue to Cys sulphinic acid. The unique repair of this moiety by sulphiredoxin (Srx) restores peroxidase activity and terminates the signal. The hyperoxidized form of Prx exists as a stable decameric structure with each active site buried. Therefore, it is unclear how Srx can access the sulphinic acid moiety. Here we present the 2.6 Å crystal structure of the human Srx–PrxI complex. This complex reveals the complete unfolding of the carboxy terminus of Prx, and its unexpected packing onto the backside of Srx away from the Srx active site. Binding studies and activity analyses of site-directed mutants at this interface show that the interaction is required for repair to occur. Moreover, rearrangements in the Prx active site lead to a juxtaposition of the Prx Gly-Gly-Leu-Gly and Srx ATP-binding motifs, providing a structural basis for the first step of the catalytic mechanism. The results also suggest that the observed interactions may represent a common mode for other proteins to bind to Prxs.

Reactive oxygen species, such as hydrogen peroxide (H2O2) and peroxynitrite, have been recognized as compounds capable of modifying protein, DNA and lipids, especially when present at elevated levels. In contrast, low levels of H2O2 can function as a second messenger signal in cell proliferation, differentiation and migration. The dysregulation of these signalling processes are hallmarks of oxidative stress and disease states, including diabetes, cancer and ageing. In this context, the ubiquitous thiol peroxidases, 2-Cys Prxs, function as critical peroxide sensors that can be inactivated through hyperoxidation. The hyperoxidation phenomenon is a fundamental element of the flood-gate hypothesis. Once the Prx molecules are inactivated, through the formation of a Cys sulphinic acid (Cys-SPO, Fig. 1a) during the catalytic cycle, H2O2 can ‘breach the gate’ to initiate signalling events. Two additional scenarios for Prx-mediated signalling include the sulphinic acid form of 2-Cys Prxs acting as a signal itself and the fostering of disulphide bond formation in other proteins. Thus, the unprecedented repair or retroreduction of 2-Cys Prxs by Srx is essential to restore peroxidase activity and the regulation of signalling events.

Structural studies on 2-Cys Prxs have revealed that the active site region can exist in fully folded and locally unfolded states. The hyperoxidized form of human PrxII exists in the fully folded state. In this form, the peroxidatic Cys residue, Cys51-SPO, is located at the amino terminus of an &agr;-helix stabilized by a salt bridge to Arg 127 (Supplementary Fig. 1, residue numbering is one less than in human PrxI) and the resolving Cys-SRH residue is ∼14 Å away. Access to the sulphinic acid moiety is further restricted by the YF and GGLG motifs. The active site helix, however, must locally unfold to allow the formation of a disulphide bond between the Cys-SPH and Cys-SRH residues during the Prx catalytic cycle (Fig. 1a). Given these observations, it is clear that large structural rearrangements must occur in order for the Srx molecule to access the Prx sulphinic acid moiety. This notion is supported by the inability to model the catalytic Cys residues of each enzyme in close proximity to each other. Therefore, the complex between the two enzymes is not readily predictable.

Using X-ray crystallography, we determined the structure of human Srx in complex with PrxI to 2.6 Å resolution after screening many engineered constructs. This complex contained one PrxI dimer (Fig. 1b) and two Srx monomers. The electron density across the disulphide bond that bridges between the active sites was unambiguous (Supplementary Fig. 2). The Srx molecules were sandwiched between the active site surface of one PrxI monomer and the C-terminal tail from the adjacent PrxI monomer. Complex formation resulted in the burial of ∼690 Å2 at each Srx-Prx active site interface and ∼960 Å2 between the C-terminal tail and the ‘backside’ of Srx (Fig. 1c). Phe 50 of PrxI packs within an Srx pocket (Fig. 2a) constituted by Leu 52, Leu 82, Phe 96, Val 118, Val 127 and Tyr 128 (Supplementary Fig. 3). A comparison to the structure of hyperoxidized human PrxII (Fig. 2b, c) shows that the Cys-SO moiety (Csd 51) is distant from Srx. We propose that the hydrophobic surface of Srx triggers the local unfolding of the Prx active site helix to place Phe 50 in the Srx pocket. As a result, Cys52-SO of human PrxI moves ∼10 Å away from Arg 128 to approach Cys 99 of Srx.

A superposition of the model of human Srx with ATP bound to it, based on the ADP complex determined experimentally, onto the Srx–PrxI complex also suggests that the unfolding of the PrxI active site helix would place Cys52-SO near the &ggr;-phosphate of ATP (Fig. 2a and Supplementary Fig. 4). In the ternary complex, the &ggr;-phosphate atom is located 3.0 Å from the S&ggr; atom of PrxI-Cys52 and 3.5 Å from the S&ggr; atom of Srx-Cys99. The oxygen atom of Cys52-SO is positioned correctly to perform an inline attack on the &ggr;-phosphate, as originally proposed. In contrast, Cys 99 of Srx points away from the &ggr;-phosphate. This observation implies that this residue does not transfer the phosphate moiety from ATP to the Prx sulphinic acid (Fig. 1a); this is consistent with the weak phosphorylation of the inactive, C99S human Srx variant. The ternary complex also suggests that the GGLG motif and the preceding residues, Lys 92, Lys 93 and Gln 94, are in a position to generate the second half of the ATP binding site.

A comparison of the Srx-PrxI structure to Prx molecules present in two different oxidation states further supports the necessary flexibility of the GGLG motif, the active site helix containing the Cys-SPH residue, the YF motif, and Cys-SRH movements. The Srx and the YF motif of the adjacent Prx monomer cannot occupy the same space at the same time (Fig. 2b, c). The active site helix containing the sulphinic acid moiety must break its interaction with a conserved Arg residue (Fig. 2c) in order to attack the ATP molecule in the Srx active site, as described above. This locally unfolded state is consistent with the formation of the disulphide bond between Cys-SPH and Cys-SRH (Fig. 2d) during the peroxidase catalytic cycle. The current Srx–PrxI complex also suggests that the ADP molecule would need to be released and that additional structural changes in either of the protein molecules or both would be necessary for Trx or GSH to break down the proposed thiosulphinate intermediate.

The binding of the PrxI C terminus onto the backside of Srx was surprising (Fig. 1). Residues 172–186 of PrxI (Fig. 3a) pack onto a conserved (Supplementary Fig. 5), predominantly hydrophobic groove of Srx. This region of 2-Cys Prxs is also conserved and contains Cys173-SRH (mutated to Ser in this variant), three or more Pro residues, and Trp 177 (Supplementary Fig. 6). This interface suggests that the active site interactions are insufficient for binding, and that the C terminus functions to hold Srx in the correct orientation for catalysis. In order to test this hypothesis and the relevance of the dimeric complex structure to the Prx hyperoxidized decamer, Ile 50, Tyr 92, Phe 93 and Leu 117 (Fig. 3b) of Srx were mutated to Arg to repel the binding of the Prx C terminus. A corresponding alteration or truncation of the PrxI C terminus was not performed, because Prx variants of these types from a variety of organisms become resistant to hyperoxidation and Cys sulphinic acid cannot be formed.

The Srx variants were analysed by circular dichroic spectroscopy (Supplementary Fig. 7) to verify that their global structure had not been significantly compromised. Despite the careful selection of the sites of mutation, the F93R mutant exhibited a loss in structure. The ability of the Srx variants, including the partially unfolded F93R variant, to bind wild-type (WT), decameric PrxI-SO was tested using fluorescence anisotropy (Fig. 3c). WT and I50R Srx bound PrxI-SO with similar affinities, 5.1 ± 0.9 &mgr;M and 7.2 ± 1.3 &mgr;M (mean ± s.d.), respectively. This finding agrees with Ile 50 being the residue farthest from the interface. In contrast, the Y92R and L117R variants of Srx had significantly reduced or no binding, a result similar to that of F93R (data not shown). The catalytic activity of the Srx mutants was also monitored using reverse-phase high-performance liquid chromatography (HPLC; Fig. 3d). WT Srx was able to repair decameric PrxI-SO at a rate of 0.23 min–1 (Supplementary Fig. 8), a value similar to the rate previously reported. The I50R and Y92R variants exhibited 60% and 15% of WT activity, respectively. The L117R mutant and the structurally comprised F93R mutant both exhibited no activity. These observations indicate that decreased binding of Srx to Prx is sufficient to reduce or abolish Srx activity.

The necessity for the C terminus of 2-Cys Prxs to bind and embrace Srx highlights its expanding cellular roles. For example, the interaction of the human PrxI C terminus with the PDZ domain of Omi/HtrA2 is necessary to promote protease activity. The interactions with c-Abl, c-Myc, MIF, phospholipase D1 and the PDGF receptor also raise the possibility that the binding of the Pro-rich C terminus of Prx to Srx represents a general mechanism for 2-Cys Prxs to associate with key regulatory or signalling proteins. Moreover, these latter interactions may modulate the repair process or vice versa.

The importance of Srx is likely to extend beyond the repair of the decameric form of 2-Cys Prxs. The association of Prx decamers into stacks of toroids has been observed via electron microscopy and within the crystal structure of human PrxII-SO (refs 11, 23). Confocal microscopy studies also suggest that human PrxII-SO can form filamentous structures in cell culture, thereby alerting cells to a perturbation in peroxide homeostasis. Sphere-like Prx aggregates have also been shown to switch from a peroxidase activity to a protein chaperone function. In an effort to understand how Srx may interact with the higher-order forms of Prxs, a model of the decameric Srx-PrxI complex was generated. The PrxI dimer of the Srx-Prx complex was superimposed onto each of the five Prx dimers of the PrxII-SO structure (Supplementary Fig. 9). No significant steric clashes were observed, suggesting that the Srx-Prx interaction is not influenced by the oligomeric state. The addition of ten Srx molecules, however, did expand the toroid diameter (∼110 to 125 Å) and thickness (∼45 to 55 Å). These substantial changes suggest that the binding of one or two Srx molecules would be sufficient to destabilize Prx-Prx interactions in higher-order oligomers.

In summary, the embrace observed in the Srx-Prx complex represents an unexpected structural rearrangement fundamentally important for the repair of Prxs in higher organisms. A structural basis is now available for designing future biochemical and cellular studies to dissect additional aspects of the Srx reaction mechanism and the roles of Srx and Prxs in modulating cell signalling.

Methods Summary

One key to stabilizing the Srx–Prx crystals was to mimic the proposed thiosulphinate intermediate (Fig. 1a) with a disulphide bond. An intermolecular disulphide bond was formed between the active site residues, Cys 99 of human Srx and Cys52-SPH of the C71S, C83E, C173S variant of PrxI. Both proteins were separately overexpressed in Escherichia coli, and disulphide bond formation was facilitated by pre-treatment of the PrxI variant with 5,5′-dithiobis-(2-nitrobenzoic acid). The comparable disulphide-bonded species has also been observed in vivo and in vitro. A dimeric form of PrxI was also necessary, and was generated by introducing charged residues, Cys83Glu on each monomer, juxtaposed at the dimer–dimer interface of the PrxI decamer. An N-terminal truncation of human Srx, residues 1–37, was required to remove a non-conserved, glycine-rich region. Crystals were grown by vapour diffusion, and diffraction data collected on beamline X8C at the National Synchrotron Light Source (NSLS). The structure was solved by molecular replacement using the rat PrxI dimer and human Srx as search models. The final model has Rwork and Rfree values of 23.9% and 30.8%, respectively. The binding of Srx variants to the hyperoxidized, decameric form of human PrxI was determined by fluorescence anisotropy by labelling Srx with Oregon Green 514. Hyperoxidized PrxI was generated by forcing the enzyme to go through the catalytic cycle many times by the addition of H2O2 and dithiothreitol. The activity of Srx variants was determined by quantifying the conversion of Prx-SO to Prx-SH by reverse-phase HPLC. Detailed procedures are presented in Supplementary Information.

Protein purification and generation of PrxI-Srx complex crystals

Several approaches were used to obtain crystals of the human Srx–Prx complex. Both His-tagged and untagged recombinant PrxI and human PrxII constructs were simultaneously manipulated to obtain crystals. Attempts to crystallize hyperoxidized PrxI and PrxII decamers in association with Srx and the presence or absence of ATP/ADP were not successful. In order to facilitate crystallization, a covalent disulphide bond between the active site Cys 99 of Srx and the peroxidatic Cys in Prx (Cys 52 in PrxI and Cys 51 in PrxII) was generated as described below. The dimer–dimer interface of the Prx decamer was manipulated by mutating Cys 83 to Glu/Ser/Val in PrxI or the corresponding Thr 82 in PrxII to Glu or Val. Some of these mutant constructs were also truncated at the C terminus (position 170 or 185 in PrxI and the corresponding positions in PrxII). Three different forms of Srx were used in the crystallization attempts as well: full length (residues 1–137), ET-Srx (32–137) or TT-Srx (38–137). More than 15 different disulphide-linked complexes were generated and screened for crystallization. The typical yields for each complex were 2.5–20 mg. The best crystals obtained were from the complex that contained the PrxI variant C71S, C83E, C173S disulphide linked to ET-Srx.

Human Srx was expressed using pET19 (Novagen) or pMALc2 (New England Biolabs) vector derivatives containing a PreScission protease (GE Healthcare) cleavage site between Srx and the N-terminal His-tag or maltose binding protein. The proteins were expressed in C41(DE3) E. coli cells and purified using nickel affinity or amylose resins followed by size-exclusion chromatography. All site-directed mutants were generated using the QuikChange Site-directed Mutagenesis Kit from Stratagene. All Srx variants were analysed by circular dichroic spectroscopy in 20 mM HEPES pH 7.5, 100 mM NaCl at a concentration from 0.3 to 0.45 mg ml-1. Three 20 nm min-1, room temperature scans from a JASCO-720 spectropolarimeter were averaged. The PrxI variant was expressed from the pET17b vector in C41(DE3) cells and purified as previously described with alterations. Briefly, PrxI for crystallographic studies was purified in the presence of 1 mM DTT and sequential chromatography on phenyl sepharose, S-sepharose and hydroxyapatite columns. The protein was then passed through a 250 ml Superdex 200 size exclusion column to facilitate buffer exchange and removal of DTT. Ellman’s reagent (500 &mgr;M), 5,5′-dithiobis-(2-nitrobenzoic acid) (DTNB), was added immediately to all fractions corresponding to the dimeric form of PrxI. Excess DTNB and generated TNB were removed using a 75 ml G25 size exclusion column. ET-Srx was titrated into the PrxI solution until no further release of TNB was observed at 412 nm (<2-fold excess Srx over PrxI). The disulphide-bonded Srx-PrxI complex was passed over the Superdex 200 column to remove excess ET-Srx and TNB. The complex was concentrated, aliquoted, flash frozen with liquid nitrogen, and stored at -80 °C. Crystals of the Srx-PrxI complex were obtained by hanging-drop vapour diffusion. Equal volumes of protein (10–14.5 mg ml-1 in 20 mM HEPES pH 7.5 and 100 mM NaCl) and well solution (14% PEG 8000, 16% ethylene glycol and 100 mM HEPES pH 7.2) were mixed. Crystals were mounted in nylon loops and cryo-cooled at -170 °C for data collection.

Data collection and structure determination

A single wavelength (1 Å) data set was collected on beamline X8C at NSLS. Diffraction intensities were integrated using d*Trek (Rigaku/MSC) and scaled to 2.6 Å resolution (Supplementary Table 1). 5.1% of the reflections were set aside for cross-validation. The space group of the crystal was P212121 with unit cell dimensions a = 54.9 Å, b = 85.0 Å, c = 130.8 Å. The structure of the Srx-PrxI complex was solved by molecular replacement using PHASER. The search models used were the following: the ET-Srx model (residues 38–137, PDB code 1XW3) and only the regions of rat PrxI (PDB entry 1QQ2) that were structurally conserved with human PrxII (PDB entry 1QMV). Importantly, the entire active site helix of PrxI (residues 46–69), the GGLG motif (residues 89–102) and the C terminus (residues 169–199) were removed from the search model. Two monomers of Srx and one dimeric PrxI were found in the asymmetric unit with rotational Z-scores from 5.4 to 10.6 and translational Z-scores from 21.1 to 22.8. Electron densities for the regions not present in the search model were clearly visible except for the residues 188–199. A combination of simulated-annealing composite omit and sigmaA-weighted 3Fobs - 2Fcalc, 2Fobs - Fcalc and Fobs - Fcalc maps were used for model building in O and COOT. The model was refined with CNS using alternating cycles of simulated annealing, positional and B-factor refinement. At the final stage of refinement REFMAC5 was used and ten water molecules were added to 3&sgr; positive features within a Fobs - Fcalc map. Structural figures were generated using PYMOL (DeLano Scientific). Model quality was assessed using MOLPROBITY. Intermolecular surface accessible areas were calculated using AREAIMOL within the CCP4 package.

Srx activity assay

In order to generate the hyperoxidized form of PrxI for Srx to repair, WT PrxI was added to a solution containing 50 mM Tris pH 7.5 and 100 mM KCl. Hyperoxidation was achieved by four step-wise additions of 5 mM H2O2 and 10 mM DTT with incubation for 30 min at 37 °C. Excess DTT and H2O2 were removed by extensive buffer exchange (50 mM Tris-HCl pH 7.5 and 100 mM KCl) via ultrafiltration (Vivaspin by Sartorious). Repair of PrxI-SO was performed by incubating 50 &mgr;M PrxI-SO with 10 &mgr;M WT Srx or variant in a 30 &mgr;l reaction mixture containing 50 mM Tris pH 7.5, 100 mM KCl, 1 mM ATP, 1 mM MgCl2 and 2 mM DTT. The reaction was stopped at various times by the addition of 15 &mgr;l 1 M H3PO4 to prevent disulphide shuffling and potential reoxidation. Five microlitres of the sample was injected using an autosampler onto a Waters analytical HPLC system. The PrxI-SO and PrxI-SH species were separated from each other on a C4 column (Vydac) using a 60–63% acetonitrile/0.1% TFA gradient over 19 min. A similar reaction containing fivefold excess (250 &mgr;M) of Srx over PrxI-SOwas performed to determine the amount of non-repairable Prx-SO present. The peaks from duplicate samples were integrated and the amount of Prx-SO subtracted. The fraction of available PrxSO that was repaired is reported as the mean ± s.d.

Fluorescence anisotropy binding assay

The N-terminal amine of ET-Srx (present in 75 mM NaHCO3 adjusted to pH 8.3) was derivatized by incubation with a twofold molar excess of the fluorophor Oregon Green 514 succinimidyl ethyl ester (Molecular Probes) for 30 min at room temperature in the dark. The reaction was quenched with 200 mM glycine in 75 mM NaHCO3 which had been adjusted to pH 8.3. The labelled protein was separated from excess dye by a G-25 sephadex desalting column (Bio-Rad). The correct labelling of the N terminus was confirmed by limited trypsin digest, which cleaves ET-Srx at residue Arg 37. This truncated version of Srx, TT-Srx, lacked fluorescence as judged by SDS–PAGE separation and imaging. The protein concentrations and the degree of labelling, typically 0.5–1 mol Oregon Green 514 per mol protein, were determined by UV-VIS spectroscopy. Fluorescence intensity and anisotropy measurements were carried out on a Safire II plate reader (Tecan Instruments) equipped with dual monochromators and the ability to rapidly collect data from quadruplicate 20 µl samples in the wells of a microtitre plate. Samples of Oregon Green-labelled ET-Srx mutants were excited at 470 nm and emission measured at 530 nm using 5 nm slits. The instrument computed anisotropy data from samples illuminated with vertically polarized light from the vertically and horizontally polarized components (Iv and Ih respectively) of the emitted light: A = Iv - (GIh/Iv) + GIh. The G-factor, which corrects for differential detector responses to the vertically and horizontally polarized emitted light, was calculated optimally over the plate. Oregon Green 514 (1 nM) was used as the reference. Oregon green labelled ET-Srx (50 nM) was mixed with PrxI-SO for 10 min at 30 °C at the indicated concentrations in a 20 µl reaction mixture containing 50 mM Tris pH 7.5, 100 mM KCl and 1 mM ATP before measurements were performed at the same temperature. In these experiments Mg2+ was omitted from the reaction to prevent Srx-mediated catalysis. Representative, duplicate samples with their s.d. are shown in Fig. 3c.

Peroxiredoxin hyperoxidation and repair by sulphiredoxin.

a, In the typical 2-Cys Prx catalytic cycle (violet), the peroxidatic Cys is depicted as a thiol (SPH) or sulphenic acid (SPOH). The disulphide bond between Cys-SP and the resolving Cys, Cys-SRH, from the adjacent monomer is reduced by Trx. Reaction with a second molecule of H2O2 results in hyperoxidation and sulphinic acid (SPO) formation (yellow). The reaction mediated by Srx (blue) is specific for 2-Cys Prxs and dependent upon ATP, Mg2+ and a Cys thiol. The sulphinic acid moiety is thought to be phosphorylated to form a sulphinic phosphoryl ester (Prx-SO2PO) through either a direct attack on the &ggr;-phosphate of ATP or transfer from Cys 99 of human Srx. The phosphoryl ester is subsequently converted to a thiosulphinate bond (for example, Prx-S(O)-S-Srx, boxed in black) which includes the sulphur atom of Cys 99 of human Srx or possibly glutathione (GSH), and inorganic phosphate (Pi) is released. Thioredoxin (Trx) or GSH reduce this complex to release free Srx and Prx-Cys-SPOH. b, Overall structure of the Srx-PrxI complex. Cartoon representation is shown, with secondary structural elements along the two-fold axis. The two monomers of PrxI and Srx are shown in violet/purple and cyan/blue, respectively. c, Surface representation of the Srx-PrxI complex, illustrating active site and backside interfaces.

Srx-PrxI active site interactions and structural plasticity.

a, ATP modelled (translucent) into the active site of the Srx–PrxI complex containing the disulphide bond between Cys 99 of Srx (cyan) and Cys 52 of PrxI (violet). b, Ribbon diagram of the locally unfolded human PrxI active site in complex with Srx. c, Human PrxII-SO active site in the fully folded state (dark purple) with the YF motif from the adjacent monomer (light purple) overlaying the active site. The active site Cys-SPH residue is in the sulphinic acid form (Csd 51). PDB code 1QMV. Human PrxI and PrxII exhibit 77.8% sequence identity. d, Rat PrxI active site (dark green) present in the oxidized, disulphide form involving the resolving Cys residue, Cys 173 (light green). PDB code 1QQ2. Panels b–d are presented in the same orientation using a superposition of the core &bgr;-sheet structure of each Prx dimer.

PrxI backside interaction with Srx.

a, The C-terminal tail of PrxI binds to Srx. Conserved residues of PrxI are coloured orange, and the conserved surface of Srx is coloured grey. b, Surface residues of Srx that interact with the C terminus of PrxI. c, Srx–PrxI-SO interactions measured in solution by changes in fluorescence anisotropy with Oregon Green 514-labelled Srx variants. Data obtained from representative, duplicate titrations are expressed as the fractional change in anisotropy, (A–A0)/A0, versus the concentration of decameric PrxI-SO added, with the error bars indicating s.d. The data for WT Srx and the I50R mutant were fitted to a single-site, saturable binding model. d, Representative HPLC traces from the activity analysis of Srx variants.

We thank M. Murray for contributions to the structure determination, L. B. Poole, P. A. Karplus and N. H. Heintz for discussions, the staff of the NSLS and beamline X8C for their assistance during data collection and the RapiData course, and R. R. Hantgan for help with the circular dichroism and fluorescence anisotropy experiments. This work was supported by an NIH grant (W.T.L.) and an American Heart Association Postdoctoral Fellowship (T.J.J.). NSLS is supported by the US Department of Energy and NIH.

Author Contributions T.J.J. and L.C.J. performed all biochemical and crystallization experiments. T.J.J. and W.T.L solved the structure. T.J.J. and W.T.L. wrote the paper. All authors discussed the results and commented on the manuscript.

Coordinates and structure factors have been deposited with the Protein Data Bank under the accession number 2RII.

Rhee, S. G. Cell signaling. H2O2, a necessary evil for cell signaling. Science 312, 18821883 (2006) Biteau, B., Labarre, J. & Toledano, M. B. ATP-dependent reduction of cysteine-sulphinic acid by S. cerevisiae sulphiredoxin. Nature 425, 980984 (2003) Finkel, T. & Holbrook, N. J. Oxidants, oxidative stress and the biology of ageing. Nature 408, 239247 (2000) Sundaresan, M., Yu, Z. X., Ferrans, V. J., Irani, K. & Finkel, T. Requirement for generation of H2O2 for platelet-derived growth factor signal transduction. Science 270, 296299 (1995) Kang, S. W., Rhee, S. G., Chang, T. S., Jeong, W. & Choi, M. H. 2-Cys peroxiredoxin function in intracellular signal transduction: Therapeutic implications. Trends Mol. Med. 11, 571578 (2005) Klaunig, J. E. & Kamendulis, L. M. The role of oxidative stress in carcinogenesis. Annu. Rev. Pharmacol. Toxicol. 44, 239267 (2004) Wood, Z. A., Poole, L. B. & Karplus, P. A. Peroxiredoxin evolution and the regulation of hydrogen peroxide signaling. Science 300, 650653 (2003) Bozonet, S. M. et al. Oxidation of a eukaryotic 2-Cys peroxiredoxin is a molecular switch controlling the transcriptional response to increasing levels of hydrogen peroxide. J. Biol. Chem. 280, 2331923327 (2005) Vivancos, A. P. et al. A cysteine-sulfinic acid in peroxiredoxin regulates H2O2-sensing by the antioxidant Pap1 pathway. Proc. Natl Acad. Sci. USA 102, 88758880 (2005) Wood, Z. A., Schröder, E., Harris, R. J. & Poole, L. B. Structure, mechanism and regulation of peroxiredoxins. Trends Biochem. Sci. 28, 3240 (2003) Schröder, E. et al. Crystal structure of decameric 2-Cys peroxiredoxin from human erythrocytes at 1.7 Å resolution. Structure 8, 605615 (2000) Lee, D. Y. et al. Mutagenesis and modeling of the peroxiredoxin (Prx) complex with the NMR structure of ATP-bound human sulfiredoxin implicate aspartate 187 of Prx I as the catalytic residue in ATP hydrolysis. Biochemistry 45, 1530115309 (2006) Jönsson, T. J., Murray, M. S., Johnson, L. C., Poole, L. B. & Lowther, W. T. Structural basis for the retroreduction of inactivated peroxiredoxins by human sulfiredoxin. Biochemistry 44, 86348642 (2005) Jeong, W., Park, S. J., Chang, T. S., Lee, D. Y. & Rhee, S. G. Molecular mechanism of the reduction of cysteine sulfinic acid of peroxiredoxin to cysteine by mammalian sulfiredoxin. J. Biol. Chem. 281, 1440014407 (2006) Koo, K. H. et al. Regulation of thioredoxin peroxidase activity by C-terminal truncation. Arch. Biochem. Biophys. 397, 312318 (2002) Moon, J. C. et al. Oxidative stress-dependent structural and functional switching of a human 2-Cys peroxiredoxin isotype II that enhances HeLa cell resistance to H2O2-induced cell death. J. Biol. Chem. 280, 2877528784 (2005) Sayed, A. A. & Williams, D. L. Biochemical characterization of 2-Cys peroxiredoxins from Schistosoma mansoni. J. Biol. Chem. 279, 2615926166 (2004) Chang, T. S. et al. Characterization of mammalian sulfiredoxin and its reactivation of hyperoxidized peroxiredoxin through reduction of cysteine sulfinic acid in the active site to cysteine. J. Biol. Chem. 279, 5099451001 (2004) Hong, S. K., Cha, M. K. & Kim, I. H. Specific protein interaction of human Pag with Omi/HtrA2 and the activation of the protease activity of Omi/HtrA2. Free Radic. Biol. Med. 40, 275284 (2006) Choi, M. H. et al. Regulation of PDGF signalling and vascular remodelling by peroxiredoxin II. Nature 435, 347353 (2005) Jung, H., Kim, T., Chae, H. Z., Kim, K. T. & Ha, H. Regulation of macrophage migration inhibitory factor and thiol-specific antioxidant protein PAG by direct interaction. J. Biol. Chem. 276, 1550415510 (2001) Xiao, N., Du, G. & Frohman, M. A. Peroxiredoxin II functions as a signal terminator for H2O2-activated phospholipase D1. FEBS J. 272, 39293937 (2005) Harris, J. R. et al. Comparison of the decameric structure of peroxiredoxin-II by transmission electron microscopy and X-ray crystallography. Biochim. Biophys. Acta 1547, 221234 (2001) Phalen, T. J. et al. Oxidation state governs structural transitions in peroxiredoxin II that correlate with cell cycle arrest and recovery. J. Cell Biol. 175, 779789 (2006) Jang, H. H. et al. Two enzymes in one: Two yeast peroxiredoxins display oxidative stress-dependent switching from a peroxidase to a molecular chaperone function. Cell 117, 625635 (2004) Parsonage, D. et al. Analysis of the link between enzymatic activity and oligomeric state in AhpC, a bacterial peroxiredoxin. Biochemistry 44, 1058310592 (2005) Hirotsu, S. et al. Crystal structure of a multifunctional 2-Cys peroxiredoxin heme-binding protein 23 kDa/proliferation-associated gene product. Proc. Natl Acad. Sci. USA 96, 1233312338 (1999) Woo, H. A. et al. Reduction of cysteine sulfinic acid by sulfiredoxin is specific to 2-Cys peroxiredoxins. J. Biol. Chem. 280, 31253128 (2005) Mazur, D. J. & Perrino, F. W. Excision of 3′ termini by the TREX1 and TREX2 3′→5' exonucleases. Characterization of the recombinant proteins. J. Biol. Chem. 276, 1702217029 (2001) McCoy, A. J. Solving structures of protein complexes by molecular replacement with Phaser. Acta Crystallogr. D 63, 3241 (2007) Jones, T. A. & Kjeldgaard, M. Electron-density map interpretation. Methods Enzymol. 277, 173208 (1997) Emsley, P. & Cowtan, K. Coot: Model-building tools for molecular graphics. Acta Crystallogr. D 60, 21262132 (2004) Brunger, A. T. et al. Crystallography & NMR system: A new software suite for macromolecular structure determination. Acta Crystallogr D. 54, 905921 (1998) Murshudov, G. N., Vagin, A. A. & Dodson, E. J. Refinement of macromolecular structures by the maximum-likelihood method. Acta Crystallogr. D 53, 240255 (1997) Davis, I. W., Murray, L. W., Richardson, J. S. & Richardson, D. C. MOLPROBITY: Structure validation and all-atom contact analysis for nucleic acids and their complexes. Nucleic Acids Res. 32, W615W619 (2004) Collaborative Computational Project, Number 4. The CCP4 suite: Programs for protein crystallography. Acta Crystallogr. D 50, 760763 (1994)
doi: 10.1038/nature06415