Letters the Mad1–Sin3b Interaction Involves a Novel Helical Fold

The Mad1–Sin3B a interaction involves a novel helical fold Christian A. E. M. Spronk1, Marco Tessari1, b Anita M. Kaan2, Jacobus F. A. Jansen1, Michiel Vermeulen2, Hendrik G. Stunnenberg2 and Geerten W. Vuister1

1Department of Biophysical Chemistry, NSR Center, University of Nijmegen, The Netherlands. 2Department of Molecular Biology, Institute of Cellular Signalling, University of Nijmegen, The Netherlands.

Sin3A or Sin3B are components of a corepressor complex that mediates repression by transcription factors such as the helix-loop-helix proteins Mad and Mxi. Members of the Mad/Mxi family of repressors play important roles in the .com transition between proliferation and differentiation by down-regulating the expression of genes that are activated by the proto-oncogene product Myc. Here, we report the solu- Fig. 1 Sin3 architecture and Mad1 binding. a, Schematic diagram of the tion structure of the second paired amphipathic helix (PAH) mSin3 corepressor, showing the four conserved PAH domains (gray domain (PAH2) of Sin3B in complex with a peptide compris- boxes) and their (possible) interacting partners. b, hMad1-SID efficiently ing the N-terminal region of Mad1. This complex exhibits a binds and depletes the mSin3–HDAC complex from HeLa nuclear extracts. ProtG–hMad1-SID(1–35) expressed in E. coli and bound to an novel interaction fold for which we propose the name IgG-Sepharose(IgG-MAD) column or empty beads (IgG) were incubated ‘wedged helical bundle’. Four α-helices of PAH2 form a with HeLa nuclear extract and washed extensively. Polypeptides in the

http://structbio.nature flow through fraction (FT) and bound and eluted polypeptides (Bound) ¥ hydrophobic cleft that accommodates an amphipathic Mad1 α were separated on a SDS acrylamide gel, blotted and analyzed by west- -helix. Our data further show that, upon binding Mad1, sec- ern blotting using polyclonal antibodies against mSin3A, mSin3B, HDAC2 ondary structure elements of PAH2 are stabilized. The and Sap30. PAH2–Mad1 structure provides the basis for determining the principles of protein interaction and selectivity involving PAH domains. The Mad1 repressor belongs to a family of four proteins mSin3–HDAC complex present in HeLa cell extracts, whereas (Mad1, Mxi1, Mad3 and Mad4) that are thought to antagonize a control IgG-Sepharose column did not (Fig. 1b). Western the transcriptional activation, proliferation-promoting and blot analysis of the eluted proteins revealed full depletion of transformation functions of the oncoprotein Myc, and thereby mSin3A, mSin3B and Sap30 from the extract. In contrast, the 2000 Nature America Inc. 1–3

© act as tumor suppressors . Mad proteins act by competing with histone deacetylases HDAC1 and HDAC2, which are known to Myc for a common partner, Max. The Mad–Max and Myc–Max participate in other chromatin-modifying complexes, were heterodimers compete for binding to a common cis element, the only partially depleted (Fig. 1b and data not shown). Similar E-box DNA consensus sequences, and have opposing transcrip- results were obtained when using two mouse nuclear extracts tional activities. Thus, the balance between Mad and Myc regu- (FM3a and MEL cells, data not shown). To delineate the pro- lates the switch between differentiation and proliferation of cells. tein regions involved in the interaction, glutathione-S-trans- Mad proteins recruit the Sin3–histone deacetylase ferase (GST)-pull down and mammalian two-hybrid (Sin3–HDAC) corepressor complex where it is responsible for experiments were performed between PAH2 containing frag- transcriptional silencing by modifying the local chromatin struc- ments of mSin3B and N-terminal fragments of hMad1. These ture4–9. Sin3 is an evolutionarily conserved protein that contains data (A.M.K. & H.G.S., unpublished results) and evidence four repeats of ∼80 amino acids, commonly referred to as paired provided in a separate study12 showed that a minimal hMad1- amphipathic helix (PAH) domains (Fig. 1a). The N-terminus of SID of only 13 amino acids (residues 8–20, NIQML- Mad1 (Mad1-SID, for Sin3 interaction domain) interacts with LEAADYLE) can bind the PAH2 domain of mSin3B or the second PAH domain of both Sin3A and SinB10–12. Several mSin3A. other proteins, such as Sap30 (ref. 13) and N-CoR4,5, have been To elucidate the structural characteristics of the interaction identified as possible binding partners for the other PAH between mSin3B-PAH2 and hMad1-SID, we performed NMR domains, although the exact locations of the contact sites on these studies on the complex between a 105-residue mSin3B-PAH2 proteins have not been clearly defined. construct and a synthetic 13-amino acid hMad1-SID peptide. In our studies we have used the human variant of Mad1-SID The two polypeptides form a specific and stable complex that is in (hMad1-SID) in complex with either mouse Sin3 (mSin3) or slow chemical exchange with the unbound forms. Although the the isolated PAH2 domain of mSin3B (Sin3B-PAH2). It has spectral analysis of the complex was hampered by the presence of been shown that hMad1-SID forms a specific complex with ∼25 disordered residues of PAH2, we were able to perform a com- mSin3 (refs 10,12). A construct containing Mad(1–35) fused plete sequential and nearly complete side chain assignment of to the B1 domain of protein G at the N-terminus bound to an PAH2 and hMad1-SID using isotope labeling techniques. From IgG-Sepharose column efficiently and specifically retained the the analysis of edited and filtered NOESY experiments we were

a b

c d .com

Fig. 2 Solution structure of the PAH2–hMad1-SID complex. a, Stereo view of the superposition of 20 NMR derived structures. Structures were super- imposed on the well-ordered regions. hMad1-SID is shown in red. The four α-helices of PAH2 are indicated as H1–H4. b, Backbone trace of a single structure viewed along the hMad1-SID helix axis showing hydrophobic residues at the interface between PAH2 and hMad1-SID. The side chains of hydrophobic residues of PAH2 (residues 7, 10, 11, 13, 14, 31, 32, 35, 38, 58, 76, 79 and 80) and hMad1-SID (residues 9, 11–13, 15, 16, 18 and 19) are col-

http://structbio.nature ored in gold and magenta, respectively. c, Same view as in (b) showing the polar residues at the interface between PAH2 and hMad1-SID. The side ¥ chains of positively charged, negatively charged and polar residues in PAH2 (residues 6, 9, 36, 39, 40–42, 44–46, 48 and 51) and hMad1-SID (residues 8, 10, 14, 17 and 20) are colored in red, cyan and yellow, respectively. d, Schematic representation of the intermolecular NOEs observed in the PAH2–hMad1-SID complex. Helices 1 and 2 of PAH2 are shown at the bottom of the figure, and helices 3 and 4 of PAH2 are shown at the top of the figure. Residues involved in intermolecular NOEs are indicated with connecting lines. Helical regions and solvent accessibility, as calculated by the program PROCHECK-NMR22, are indicated by the yellow ribbon and the blue bars, respectively (dark blue indicates low solvent accessibility).

able to derive an ensemble of high resolution solution structures helix 4 of PAH2 (Fig. 2b). Helix 3, for which we found no unam- of the PAH2–hMad1-SID complex. biguous NOEs to hMad1-SID, is not directly involved in the interaction with hMad1-SID. The amphipathic hMad1-SID 2000 Nature America Inc.

© Description of the structure helix is accommodated in a hydrophobic cleft between helices 1 A stereo view of the ensemble of structures is shown in Fig. 2a. The and 2 and exposes its polar and charged residues to the solvent PAH2 domain folds into a four-helix bundle topology in which and charged loop between helices 2 and 3 (Fig. 2b,c). Without residues 5–20, 25–42, 55–65 and 70–79 form the α-helices. taking into consideration the conformational transitions in Residues 21–24 and 66–69 fold into turns and the region between PAH2 and hMad1-SID upon binding we calculate that ∼460 Å2 helices 2 and 3 is a partially disordered loop. NMR relaxation of hydrophobic surface is buried when the complex is formed. In experiments show that the disorder in this loop results from local addition to these extensive van der Waals contacts, it is likely that backbone mobility (see below). The hMad1-SID peptide was intermolecular hydrogen bonds and salt bridges at the shown to be unfolded in aqueous solution but has inherent helical hydrophilic interface contribute to the binding affinity and propensity as shown by circular dichroism (CD) experiments con- specificity. However, unambiguous conclusions pertaining to ducted in trifluoroethanol (TFE)12. Our results show that upon these interactions cannot be drawn due to the inherent flexibility complex formation the hMad1-SID is folded into a well-defined of the PAH2 residues in this hydrophilic interface. stable α-helix. A striking feature of the structure of the complex is the position of this amphipathic α-helix. It is positioned at an Structural biology of sequence conservation angle of ∼45° relative to helix 2 of PAH2 and displaces helix 1 out- The protein fold of the PAH2–hMad1-SID complex is expected wards from the four-helix bundle. Whereas helices 2 and 3 are to be present in the complexes of other PAH domains with their essentially antiparallel, helix 1 is tilted at an angle of ∼15° with respective binding partners. The core of the PAH2–hMad1-SID respect to helix 4. Structural comparison of the PAH2–hMad1- complex is formed by hydrophobic residues that share ∼45% SID structure with known protein folds using the DALI server sequence similarity within the four PAH domains of a given (http://www2.ebi.ac.uk/dali) shows that this spatial arrangement mSin3 protein12. An alignment of the PAH1 and PAH2 domains of the five α-helices in the complex is a novel protein structure, for of various mSin3A and mSin3B proteins clearly illustrates the which we propose the name ‘wedged helical bundle’. very high similarity found at positions of crucial aliphatic and The NOE data (Fig. 2d) and the resulting structures show that aromatic amino acids (Fig. 3a). To gain insight into the role of the intermolecular interactions in the complex are mainly conserved residues in the complex we projected the degree of hydrophobic, involving helix 1, helix 2 and, to a lesser extent, conservation of residues in 10 PAH2 domains onto the

Fig. 3 Sequence conservation in PAH1 and PAH2 domains. a, Sequence b alignment of the PAH1 and PAH2 domains of several species. Hydrophobic and polar residues involved in the PAH2–Mad1 interaction are indicated by stars and closed circles, respectively. Helical regions and solvent accessibility are indicated as in Fig. 2d. The sequence used in the current studies is indicated with an asterisk. b, Ribbon diagram of the .com PAH2–hMad1-SID complex, with color intensity reflecting the degree of sequence conservation of residues of the PAH2 domains. hMad1-SID is shown in gray.

further strengthened by the observation that helix disrupting mutants of hMad1-SID are detrimental to binding affinity10,12. Mutation of residues at the hydrophobic face of the hMad1-SID

http://structbio.nature α ¥ -helix to Asp severely impaired binding to PAH2 (ref. 12),

which is in full agreement with the structures presented here. mSin3B-PAH2 structure (Fig. 3b). It is evident that the highest Interestingly, three of the charged residues in close proximity to conservation is found among residues of helices 1 and 2, which hMad1-SID — that is, Glu 6, Asn 9 and Glu 41 — are highly form the hydrophobic cleft in which hMad1-SID binds, as well conserved within the PAH2 sequences (Fig. 3a). Equivalent as of helix 4, which is almost 100% conserved. The third helix of residues in the PAH1 sequences are also highly conserved, albeit PAH2, the flexible loop and the turns, however, show much less with a charge opposite to those in the PAH2 sequences. conservation, indicating that these regions have lower structur- Potentially, these are the specificity defining positions of the al significance. PAH domain. Interestingly, the flexible loop between helices 2 and 3 of 2000 Nature America Inc.

© PAH2 is not present in all PAH2 domains, nor is it present in Structural aspects of complex formation any of the known PAH1 domains (Fig. 3a). This loop contains To gain insight into the molecular mechanisms of hMad1-SID several positively charged residues that may be involved in elec- binding to PAH2, we recorded NMR spectra of PAH2 in the trostatic interactions with the negatively charged surface of unbound state. A comparison of the chemical shifts of the back- hMad1-SID (Fig. 2c). Variations in electrostatic interactions bone Cα atoms of PAH2 in the two states shows relatively large could provide a mechanism for selecting the binding partners of differences for the first two turns (residues 5–11) of helix 1 and the different PAH domains of mSin3A and mSin3B. Recent somewhat smaller differences for helices 2 and 4 (Fig. 4a). Cα studies reveal a central role for electrostatic interactions in pro- chemical shifts are affected both by changes in the chemical tein–protein association, in which they influence both selectivi- environment upon complex formation as well as changes in sec- ty and kinetics14. Eilers et al.12 showed that the Q10R mutation ondary structure15. Our data show gradual changes in Cα chem- in hMad1-SID, which in our structure points towards the flexi- ical shifts for residues 5–11 of PAH2 from random coil values in ble loop, resulted in a four-fold reduced affinity. This reduction the free state to values that are typical of α-helices when PAH2 is can be explained by a decreased complementarity between the bound to hMad1-SID. Additional evidence for the presence of a proteins in the charged part of the protein–protein interface. partly unfolded first helix of PAH2 in the free state was obtained The net positive charge on the flexible loop of PAH2 due to the from 1H-15N heteronuclear NOE experiments (Fig. 4b). These presence of Arg and Lys residues would not favor interacting relaxation experiments provide a measure of the mobility of the with the positively charged residues on the polar side of hMad1- backbone of PAH2 and, in close agreement with the Cα chemi- SID. On the other hand, a two- to three-fold increase in binding cal shift data, show that residues in the first two turns of helix 1 affinity for PAH2 was observed for Q10A, E14A, and D17A are more flexible in the unbound state than in the bound state. mutants, and a mutant bearing all three substitutions12. The Based on these findings, we speculate that a partly unfolded first slight increase in affinity for these mutants may be explained by helix of PAH2 facilitates the accommodation of hMad1-SID the loss of possible intramolecular α-helix destabilizing electro- into the PAH2 structure. Upon binding of hMad1-SID in the static interactions of the negatively charged hMad1-SID upon hydrophobic cleft of PAH2, a folding transition occurs that sta- replacement of the hydrophilic residues by Ala residues. The bilizes the bound state by increasing the number of hydrophobic importance of a stable hMad1-SID helix for binding to PAH2 is interactions between hMad1-SID and PAH2. From our current

a b

Fig. 4 Structural differences between unbound PAH2 and PAH2 in complex with hMad1-SID. a, Ribbon diagram showing the differences in Cα chemical shifts. b, {1H-15N}-NOE values of the unbound (black) and bound (red) PAH2. Increasing negative values indicate increasing flexibility of the backbone. Helical regions are indicated by asterisks at the top. .com data it further appears that the interactions involving residues aliphatic and aromatic residues that form the hydrophobic core, Phe 7, Ala 10 and Ile 11 of PAH2 and the hydrophobic residues the domains are likely to interact with distinct sets of factors. of hMad1-SID are responsible for the stabilization of helix 1 of Other proteins reported to interact with PAH domains include PAH2. The folding transition occurring upon complex forma- the 91 C-terminal residues of Sap30, which interacts with PAH3 tion thus involves both the helix formation in hMad1-SID as (ref. 13), and regions of N-CoR comprising residues 1–312 and well as the stabilization of the secondary structure of PAH2. 1,829–1,840, which interact with PAH3 and PAH1, respective- Given the very high degree of sequence identity between the ly4,5. Interestingly, the Ala-rich region between residues 1,833 http://structbio.nature

¥ PAH2 domains of mSin3A and mSin3B, we predict that their and 1,845 of N-CoR has been shown to be essential for the

structures, modes of interaction and repertoires of interacting N-CoR–PAH1 interaction5. This region can potentially form an partner proteins will be very similar. Thus far, however, the only amphipathic α-helix of similar length to the hMad1-SID helix interaction studied in detail involves the PAH2 domain with and, given the high degree of similarity between PAH1 and Mad family members. Although the similarity between various PAH2, could potentially bind PAH1 in the same manner as PAH domains is very significant, in particular with respect to the hMad1-SID binds to PAH2. In conclusion, the interaction between PAH2 and hMad1-SID involves a novel fold, the wedged helical Table 1 Structural statistics of the PAH2–hMad1-SID complex1 bundle. The PAH2–Mad1 structure provides the basis Total number of experimental distance restraints2 2,176 (1,916 / 260) for determining the principles of protein interaction 2000 Nature America Inc. 2

© Intra-PAH2 1,863 (1,652 / 211) and selectivity involving PAH domains. Intra-MAD2 206 (173 / 33) Intermolecular2 107 (91 / 16) Hydrogen bonding distance restraints 27 Methods Cloning and expression. The fragment containing the ϕ Dihedral -angle restraints 26 PAH2 domain (residues 148–252) of mSin3B was Average r.m.s. deviation from all distance restraints (Å) 0.022 ± 0.002 obtained by PCR and cloned into pGEX2T. Using site Average r.m.s. deviation from dihedral ϕ-angle restraints (º) 0.42 ± 0.08 directed mutagenesis, the C241A mutation was intro- Average r.m.s. deviations from idealized covalent geometry duced in the PAH2 domain to prevent disulfide bridge Bonds (Å) 0.0022 ± 0.0002 mediated dimer formation in solution. Surface plasmon resonance and NMR data indicated that this mutation Angles (º) 0.38 ± 0.01 had no notable effect on the binding of PAH2 to hMad1- Impropers (º) 0.25 ± 0.01 SID. The resulting GST-PAH2(C241A) fusion protein was Pairwise Cartesian positional r.m.s. deviations (Å) expressed in Escherichia coli strain pBL21 using 2xYT Backbone heavy atoms 0.60 ± 0.10 medium supplemented with 0.5% (w/v) glucose. Further All heavy atoms 1.30 ± 0.10 details concerning expression, as well as assignment of Ramachandran quality parameters the PAH2 domain, will be reported elsewhere (C.A.E.M.S., J.F.A.J., M.T., A.M.K., E. Lasonder, J. Aelen, Residues in favored regions 92.3% H.G.S. & G.W.V., unpublished results). Residues in allowed regions 7.2% Synthetic oligonucleotides encoding human Mad1- Residues in accepted regions 0.2% SID(5–35) were cloned into the plasmid Gev2 in frame Residues in disallowed regions 0.3% with the GB1 domain of streptococcal protein G. GB1–hMad1-SID(5–35) was expressed in E. coli strain 1 Structural statistics of the final ensemble of 20 structures. Residues included in the BL21 DE3 and bound to an IgG-Sepharose column as rec- analysis were 5–42, 55–80 of PAH2 and all residues of hMad1-SID. None of the struc- ommended by the manufacturer (Pharmacia). HeLa tures contained distance restraint violations >0.5 Å and dihedral angle restraint violations >5º. nuclear extracts were prepared according to Dignam et 16 2Values in parentheses are the number of ARIA assigned unambiguous / ambiguous al. and passed over the GB1–hMad1-SID(5–35) column. restraints. The column was washed three times with PBS (phos- phate buffered saline, 0.12MKPi, 0.15 M NaCL, pH 7.4)

0.5 % Triton-X 100 (v/v) and twice with PBS containing 0.5 % Triton- 20 lowest energy structures were selected for analysis (Table 1). All X 100 supplemented with 500 mM KCl. Bound proteins were eluted figures were generated with the program Molmol20. Structures with 200 mM glycine and separated by SDS gel electrophoresis. The were analyzed using the programs WHATIF21 and PROCHECK- proteins were transferred and probed by western blot analysis with NMR22. antibodies against mSin3A, mSin3B, HDAC2 (Santa Cruz Biotechnology) and Sap30. Coordinates and chemical shifts. The coordinates of the 20 structures and the chemical shifts have been deposited in the NMR spectroscopy. NMR samples of PAH2 in complex with Protein Data Bank (accession code 1E91) and the BioMagResBank hMad1-SID typically contained 1–2 mM of complex (1:1 stoichiome- (accession code 4841), respectively. try) in 50 mM KPi buffer at pH 6.3. NMR samples of unbound PAH2 contained 1 mM protein, 50 mM KPi buffer at pH 6.5 and 100 mM

KCl. All NMR samples were prepared in a H2O/D2O (95%/5%) mix- Acknowledgments ture and contained trace amounts of NaN3 as a preservative. All We thank J. Aelen for technical assistance and purification of the labeled NMR spectra were acquired at 20 ºC on Varian Inova 500, 750 MHz proteins. We wish to thank the members of our Departments and J. Betz for and Bruker DRX600 spectrometers. Distance restraints for structure suggestions and critical reading of the manuscript. We thank B. Eisenman for his calculations were obtained from 3D 13C NOESY-HSQC, 3D 15N generous gift of the mSin3 cDNAs, D. Reinberg for Sap30 antibody, and A. NOESY-HSQC, 2D 13C/15N-filtered-NOESY, and 3D 15N NOESY-HMQC- Gronenborn for the pGev2 construct. The research of C. Spronk and M. NOESY experiments. Intermolecular NOEs were distinguished from Vermeulen is financially supported by the Netherlands Organization for Scientific intramolecular NOEs using 3D (13C-filtered)-NOESY-(13C-edited)- Research (NWO). HSQC experiments. The NOE mixing time in all NOESY experiments was set to 100 ms. A 3D HNHA experiment was used to obtain dihe- Correspondence should be addressed to G.W.V. email: [email protected] dral ϕ-angle restraints. Heteronuclear 1H-15N NOE experiments were performed at 10.4 T. All data were processed using 17 Received 14 August, 2000; accepted 13 October, 2000. .com NMRPipe . 1. Schreiber-Agus, N. & DePinho, R.A. Bioessays 20, 808–818 (1998). 2. Foley, K.P. & Eisenman, R.N. Biochim. Biophys. Acta 1423, M37–M47 (1999). Structure calculations. Initial structure calculations of the com- 3. Xu L., Glass, C.K. & Rosenfeld, M.G. Curr. Opin. Genet. Dev. 9, 140–147 (1999). plex were performed with a torsion angle dynamics simulated 4. Heinzel, T. et al. Nature 387, 43–48 (1997). annealing protocol using the program CNS18 based on 1,516 manu- 5. Alland, L. et al. Nature 387, 49–55 (1997). ally assigned NOE restraints, 26 dihedral ϕ-angle restraints and 27 6. Hassig, C.A., Fleischer, T.C., Billin, A.N., Schreiber, S.L. & Ayer, D.E. Cell 89, 341–347 (1997). 15 α hydrogen bonding restraints derived from the analysis of C 7. Laherty, C.D. et al. Cell 89, 349–356 (1997). chemical shifts. The original sequence was renumbered to 1–105. 8. Zhang, Y., Iratni, R., Erdjument-Bromage, H., Tempst, P. & Reinberg, D. Cell 89, The flexible C-terminus (residues 86–105) of PAH2 was not included 357–364 (1997). in the calculations. An initial ensemble of structures of the complex 9. Nagy, L. et al. Cell 89, 373–380 (1997). http://structbio.nature 10. Ayer D.E., Lawrence, Q.A. & Eisenman, R.N. Cell 80, 767–776 (1995). ¥ was calculated. This displayed an average positional root mean 11. Schreiber-Agus, N. et al. Cell 80, 777–786 (1995). square (r.m.s.) deviation of ∼1.3 Å for the backbone atoms of the 12. Eilers, A.L., Billin, A.N., Liu, J. & Ayer, D.E. J. Biol. Chem. 274, 32750–32756 (1999). well-ordered part of the structures (residues 5–41, 55–80 of PAH2 13. Laherty, C.D. et al. Mol. Cell 2, 33–42 (1998). 14. Sheinerman, F.B., Norel, R. & Honig, B. Curr. Opin. Struct. Biol. 10, 153–159 (2000). and all residues of hMad1-SID). Subsequently, the structures were 15. Wishart, D.S. & Sykes, B.D. J. Biomol. NMR 4, 171–180 (1994). refined using the automated procedure for NOE assignment and 16. Dignam, J.D., Lebovitz, R.M. & Roeder, R.G. Nucleic Acids Res. 11, 1475–1489 structure calculation in the program ARIA19 (version 1.0). In this pro- (1983) 17. Delaglio, F. et al. J. Biomol. NMR 6, 277–293 (1995). cedure, we used the PROLSQ force field parameters. Energy terms 18. Brunger A.T. et al. Acta Crystallogr. D 54, 905–921 (1998). dealing with dihedral angles were included to improve side chain 19. Nilges, M., Macias, M.J., O’Donoghue, S.I. & Oschkinat, H. J. Mol. Biol. 269, conformations. The iterative assignment of NOEs resulted in a total 408–422 (1997). of 2,176 distance restraints (Table 1), which were used to calculate 20. Koradi, R., Billeter, M. & Wüthrich, K. J. Mol. Graph. 14, 51–55 (1996). 21. Vriend, G. J. Mol. Graph. 52, 29–36 (1990). 100 structures. Thirty structures had no distance restraint violations

2000 Nature America Inc. 22. Laskowski, R.A., Rullmann, J.A., MacArthur, M.W., Kaptein, R. & Thornton, J.M. J. >0.5 Å and no dihedral angle violations >5°. Of these structures, the Biomol. NMR 8, 477–486 (1996). ©

1104 nature structural biology ¥ volume 7 number 12 ¥ december 2000