- Email: [email protected]
A comparative three-dimensional model of the carboxy-terminal domain of the lambda repressor and its use to build intact repressor tetramer models bound to adjacent operator sites
Journal of
Structural Biology Journal of Structural Biology 141 (2003) 103–114 www.elsevier.com/locate/yjsbi
A comparative three-dimensional model of the carboxy-terminal domain of the lambda repressor and its use to build intact repressor tetramer models bound to adjacent operator sitesq Rajagopal Chattopadhyaya* and Kaushik Ghosh Department of Biochemistry, Bose Institute, P-1/12, C.I.T. Scheme VII M, Calcutta 700054, India Received 1 July 2002, and in revised form 21 November 2002
Abstract A model for residues 93–236 of the k repressor (1gfx) was predicted, based on the UmuD0 crystal structure, as part of four intact repressor molecules bound to two adjacent operator sites. The structure of region 136–230 in 1gfx was found to be nearly identical to the independently determined crystal structure of the 132–236 fragment, 1f39, released later by the PDB. Later, two more tetrameric models of the k repressor tetramer bound to two adjacent operator sites were constructed by us; in one of these, 1j5g, the N-domain and C-domain coordinates and hence monomer-monomer and dimer-dimer interactions are almost the same as in 1gfx, but the structure of the linker region is partly based on the linker region of the LexA dimer in 1jhe; in the other, 1lwq, the crystalline tetramer for region 140–236 has been coopted from the crystal structure deposited in 1kca, the operator DNA and N-domain coordinates of which are same as those in 1gfx and 1j5g, but the linker region is partly based on the LexA dimer structures 1jhe and 1jhh. Monomer-monomer interactions at the same operator site are stabilized by exposed hydrophobic side chains in b-strands while cooperative interactions are mostly confined to b6 and some adjacent residues in both 1gfx and 1j5g. Mutational data, existence of a twofold axis relating two C-domains within a dimer, and minimization of DNA distortion between adjacent operator sites allow us to roughly position the C-domain with respect to the N-domain for both 1gfx and 1j5g. The study correlates these models with functional, biochemical, biophysical, and immunological data on the repressor in the literature. The oligomerization mode observed in the crystal structure of 132–236 may not exist in the intact repressor bound to the operator since it is shown to contradict several published biochemical data on the intact repressor. Ó 2003 Elsevier Science (USA). All rights reserved. Keywords: Lambda repressor cI; Gene regulation; Protein structure prediction; Structure–function relationship
1. Introduction A common important feature of gene expression in each of the bacteriophages k, P22, and 434 is the use of a negative regulatory protein called repressor (Ptashne,
q PDB accession numbers for the k repressor first comparative model in RCSB001477 (1gfx) was processed on June 22, 2000, the second model in RCSB001635 (1j5g) processed on May 3, 2002, and third comparative model in RCSB016355 processed May 30, 2002 (1lwq); our related model of LexA repressor in RCSB000408 (1qaa) was processed on February 2, 1999. * Corresponding author. Fax: +91-33-234-3886. E-mail address: [email protected] (R. Chattopadhyaya).
1992; Johnson et al., 1981), important for maintaining lysogeny. The repressors of k, P22, and 434 consist of 236, 216, and 209 residues, respectively, folded into two structural domains, separable by protease digestion (Pabo et al., 1979; De Anda et al., 1983; Anderson et al., 1984). For controlling expression, each repressor molecule forms a dimer that can bind to two sets of three homologous, partially symmetric binding sites (OR,OL) known as operator sites (Johnson et al., 1981) through the amino terminal domains (N-domains), while the carboxy-terminal domains (C-domains) are involved in dimer formation and dimer-dimer interactions crucial for its function. The C-domains also contain both the catalytic and the cleavage sites for autocleavage assisted
1047-8477/03/$ - see front matter Ó 2003 Elsevier Science (USA). All rights reserved. doi:10.1016/S1047-8477(02)00627-5
104
R. Chattopadhyaya, K. Ghosh / Journal of Structural Biology 141 (2003) 103–114
by activated RecA of the host leading to the lytic pathway, triggered by host DNA damage. Crystal structures of all the intact repressor proteins remain hitherto undetermined. Those of the N-domains of k (residues 1–92) (Pabo and Lewis, 1982; Beamer and Pabo, 1992) and 434 (residues 1–69) (Aggarwal et al., 1988; Shimon and Harrison, 1993) have been elucidated. Recently, the crystal structure of fragment 132–236 has been reported (Bell et al., 2000). Independently, we have predicted the three-dimensional structure of fragment 93–236 and combined it with the crystal structure of the N-domain bound to operator DNA (Beamer and Pabo, 1992). The structural classification of protein database available on the web at http://scop.mrc-lmb.cam.ac.uk/ scop/ (Murzin et al., 1995) contains a description of the evolutionary and structural relation of proteins with experimentally determined structures. The protein structure prediction field has received intense interest and high level of participation lately as evidenced by three meetings on the critical assessment of techniques for protein structure prediction in 1994, 1996, and 1998 (Moult et al., 1999). The methodology of protein structure predictions can be of three broad kinds: comparative or homology modeling, fold recognition, and ab initio rediction (Moult et al., 1999). We have used all three methods for predicting the three-dimensional structure of the repressor C-domain. Though the ab initio model of the C-domain was initially used to build a tetramer model bound to two adjacent operator sites, the C-domain in the comparative model 1gfx was found to be almost identical with the subsequently released crystal structure in 1f39. Another model, 1j5g, with a different structure of the cleavage site based on a LexA dimer crystal structure, 1jhe, was constructed using the comparative model as previously deposited in 1gfx. Finally, a crystal structure of the tetramer for region 140–236 taken from 1kca was used to build a third model 1lwq of the lambda repressor tetramer bound to two adjacent operator sites similar to the one partly proposed in Fig. 6 of Bell and Lewis (2001). The present paper analyzes to what extent each of these alternative models for the tetramer satisfies the biochemical and biophysical data accumulated in the literature. The methods here are similar to, but not identical with, those adopted recently for building a model of the LexA dimer bound to the recA operator (Chattopadhyaya et al., 2000).
2. Methods Carboxy-terminal sequence similarities among the repressors. The bacteriophage repressors of k, P22, and 434 share extensive sequence identity in their C-domains (Sauer et al., 1982). In pairwise comparisons identities
are significant (434 and P22, 82/139 or 59%, P22 and k, 62/138 or 45%; 434 and k, 48/139 or 35%). For the three phage repressors, 44 positions are identical and including LexA, an SOS protein (Little and Mount, 1982), 20 positions are still identical (Sauer et al., 1982). In addition, both phage and SOS repressors like LexA, UmuD, and mucA are cleaved by RecA protein at identical AlaGly sequences in a reaction involving RecA binding to the respective C-domains (Slilaty et al., 1986; Sauer et al., 1982; Little and Mount, 1982). These similarities prompted us to hypothesize that the three phage repressors possess C-domains whose three-dimensional structures are more or less identical (Chothia and Lesk, 1986). Phased secondary structure prediction. Secondary structure was predicted for each of the phage primary sequences using the Chou and Fasman (1978) and Garnier et al. (1978) methods independently. To find the common secondary structure of the phage repressors, we emphasized the phased nature of the individual aligned amino acid probabilities (Pa ; Pb ; Pt ) with those of other members of the family assuming a multiple alignment (Sauer et al., 1982) and their magnitudes; the correct secondary structure in any region showed this phasing. In Fig. 1 our secondary structure prediction is seen to match quite well with that found in the UmuD0 crystal structure which was used as a template in our comparative modeling. Template search for comparative modeling. Using the primary sequence of residues 93–236 of the k cI repressor as a query sequence, the structural classification database (Murzin et al., 1995) was searched for its predicted similarities with experimentally solved structures through the website http://stash.mrc-lmb.cam.ac.uk/ PDB_ISL/. Template search for fold recognition. Using the primary sequence of residues 93–236 of the k repressor, a fold recognition program using comparison of predicted secondary structure strings with those of known folds (Fischer and Eisenberg, 1996) was run at the UCLA fold recognition server http://www.mbi.ucla.edu/people/ fischer/BENCH/benchmark1.html. Comparative model building for region 138–230 (1gfx, 1j5g). Both template searches for comparative modeling and fold recognition described above yielded the UmuD0 crystal structure (Peat et al., 1996) as the correct answer. For building the k repressor model the amino acid alignment in Peat et al. (1996) was used as a starting point for UmuD0 residues 49–136 or k residues 138–230. The loop regions 153–158, 185–188, and 196–200 of k repressor had insertions and a deletion at 216–218 with respect to UmuD0 (Fig. 1b), and hence were built using a procedure for generating loops de novo within the Homology module of InsightII (Shenkin et al., 1987; Biosym/MSI, 1995), choosing the best loop by the criterion of lack of steric clash with the rest of the protein
R. Chattopadhyaya, K. Ghosh / Journal of Structural Biology 141 (2003) 103–114
105
Fig. 1. (a) Pairwise amino acid sequence alignment of k repressor residues 102–135 with residues 75–105 of LexA. The LexA crystal structure (Luo et al., 2001) was used only to build region 104–118 in our revised comparative model in 1j5g. (b) Pairwise amino acid sequence alignment of k repressor residues 100–236 with UmuD0 residues 25–139 used in building the comparative model with our phased secondary structure prediction versus that found in the crystal structure (Peat et al., 1996). Identical residues (6 of 31 LexA positions, 31 of 115 UmuD0 positions) are in bold, and stretches of the parent structures, a total of 78 residues, used to build comparative models 1gfx and 1j5g are underlined. Secondary structure nomenclature in Peat et al. (1996) also given for comparison in the lowest line.
structure and subsequent energy refinement within the Discover module of InsightII. The alignment given in Peat et al. (1996) for the region 214–220 between b7 and b8 cannot be correct since that would expose both Cys 215 and Cys 219 to the surface. In our alignment (Fig. 1b) for this region, Cys 215 aligns with position 119 of UmuD0 , rather than 120. This loop is shorter in k compared to UmuD0 by two residues and Cys 219 is also partially hidden as a result, aligned with UmuD0 residue 125. The structure of region 150–160 from 1f39 has been coopted in 1j5g, otherwise the C-domain in 1j5g is the same as in 1gfx. Cleavage site in 1gfx. The conformation of the region 32–48 in UmuD0 was ignored and not used for model building, since from our autocleavage experiments it was concluded that the substrate Ala-Gly must be close to the catalytic side chains of Ser 149 and Lys 192 due to the large value of the preexponential factor (see Results and discussion) and the binding of fragment AB with BC for P22 c2 repressor (De Anda et al., 1983). from the Ser 149 Keeping the Ala-Gly at about 7–8 A the region 109–137 was modeled based on the secondary structure prediction including loops and two brief ahelices whose apolar sides pointed toward the rest of the C-domain. Relative juxtapositioning of the two domains (1gfx). For building the intact repressor, the N-domain–C-domain steric relationship was not decided directly. A dimer consisting of two C-domains (109–236) related by a
2-fold axis was constructed first such that hydrophobicity considerations, thermodynamics of dimerization and biochemical data were satisfied (see points 16,17, and 20 under Results and discussion). Later, the C–C dimer 2-fold was coincided with that of the N–N dimer; while keeping the N-domains and the operator DNA fixed, the C–C dimer can be rotated as a unit (Chattopadhyaya et al., 2000) about the 2-fold, or translated parallel to the 2-fold (i.e., two degrees of freedom). By trial and error, the C-domain dimer was placed on the N-domain dimer such that correct dimer–dimer contacts could be maintained and distortion in the DNA structure minimized. We assumed that the 2-fold relating the C-domains within a dimer coincides with the 2-fold relating the N-domains in the same dimer. The dimensions of the repressor dimer obtained from electron microscopy (Brack and Pirrotta, 1975) places an upper limit to the separation of the C–C dimer from the N–N dimer, while steric clashes placed at a lower limit. After the Cdomains were properly placed with respect to the Ndomains, residues 93–108 were used to join the two domains. Two operator duplexes with two bound intact repressor molecules were placed by trial and error on a graphics terminal such that the regions in the C-domain implicated in cooperative binding (see points 18 and 19 under Results and discussion) were in contact and the DNA could be made continuous using the sequence separating the OR1 and OR2 sites with minimal distortion from ideal B-DNA.
106
R. Chattopadhyaya, K. Ghosh / Journal of Structural Biology 141 (2003) 103–114
2.1. Cleavage site in 1j5g
3. Results and discussion
Since in our autocleavage experiments for the k repressor bound to operator DNA (A. Pal and R. Chattopadhyaya, unpublished, 2001), we find cleavage, we used structure 1jhe of the LexA dimer (Luo et al., 2001) containing the cleavable form of the cleavage site for region 105–118. This region has a different structure compared to the corresponding region in 1gfx. There is a helix in region 125–132 in this model in agreement with our secondary structure prediction, CD and FTIR data. Only 6 out of 34 amino acids in this region are identical between LexA and k cI repressors (Fig. 1a); in addition, cI has 3 insertions compared to LexA—hence it is quite possible that these extra residues will be accommodated as a helix whereas a helix does not exist in LexA at the corresponding location (Luo et al., 2001).
3.1. Comparative and crystal structure-based models
2.2. Relative juxtapositioning of the two domains (1j5g) The two domains and operator DNA are almost unchanged from 1gfx though the C-domains have been renamed. The N–C connection is different in two ways, one being the structure of the cleavage site as described above and the other being a swap type N–C connection unlike in 1gfx. As a result of this connection, if ABCD is the order of four molecules on the operator DNA, molecules A and D are involved in the cooperative contacts, whereas in 1gfx, molecules B and C were involved in the cooperative contacts. 2.3. Cleavage site in 1lwq Since region 140–236 of each molecule forming a tetramer is taken from the crystal structure in 1kca, it was found that the ‘‘noncleavable’’ cleavage site (Luo et al., 2001) must accompany at least two of the molecules whose catalytic sites are involved in cooperative contacts in this model. Accordingly, such noncleavable sites were built using 1jhh for two subunits and cleavable sites built using 1jhe for the two other subunits within the tetramer. 2.4. Relative juxtapositioning of the two domains (1lwq) The entire assembly of the tetramer together with the attached cleavage sites (region 102–236) after annealing and energy refinement was placed symmetrically relative to the assembly of the four N-domains bound to the operator DNA exactly as in entries 1gfx and 1j5g, resembling Fig. 6 of Bell and Lewis (2001). The connection between 92 and 102 was chosen so that the 92–102 distance was almost equal in both subunits comprising a dimer. This assembly lacks the 2-fold axes relating the two subunits within the dimers but the 2-fold of the tetramer coincides with the pseudo 2-fold of the N-domain operator assembly.
All the later models are in general agreement with various kinds of data enumerated below: 1. Circular dichroism measurements. These showed (Ghosh and Chattopadhyaya, 2001) that k repressorÕs carboxy-terminal domain (132–236) is largely made up of b-sheet. Fragment 93–236 contains more helix than 112–236, which in turn contains more helix than 132– 236, as shown by their CD spectra. This agrees with our unified secondary structure prediction (Fig. 1b). 2. FTIR measurements. The 93–236 fragment spectrum (Ghosh and Chattopadhyaya, 2001) shows a-helix content as evidenced by the peak at 1656 cm1 in the deconvoluted spectrum whereas the 132–236 fragment shows little or no a-helix content from the absence of any peak at 1656 cm1 in the deconvoluted spectrum (Surewicz et al., 1993). 3. None of the three Cys reactive in the folded structure. Cys 180, Cys 215, and Cys 219 thiols did not react with the reagent DTNB either in the intact repressor (Banik et al., 1992) or in fragment 132–236 (Ghosh and Chattopadhyaya, 2001), though protease cleaved repressor showed near full reaction. Since at the 10 lM initial concentration of 132–236, some monomer exists, the dimerization did not prevent the reaction. Thus, neither are the cysteine thiols on the dimerization surface nor the N-domain are protecting them; it is a result of the protein fold. Thiol hydrogens cover the sulfur atoms in 1gfx and 1j5g. However, in the crystal structures Cys 215 and 219 are seen to be disulfide linked. 4. The activation energy of the autocleavage reaction and its consequence. We measured the free energy of activation ðEa ¼ DH þ RT Þ to be 22.2 kcal/mol at 300 K (Ghosh and Chattopadhyaya, 2001). For this value of Ea the Arrhenius exponential factor is 7 1017 , hence the preexponential factor A is 1:9 10þ10 s1 for our rate constant and 2 10þ11 s1 assuming the previous rate constant (Slilaty et al., 1986). Since the preexponential Arrhenius factor represents the number of collisions per second, such a high number of collisions requires that the catalytic groups are quite close to the substrate Ala-Gly. 5. Electron microscopy data, hydrodynamic parameters, hydration factors. Electron microscopy data (Brack and Pirrotta, 1975) suggested monomeric dimensions of , consistent with our model. It is a prolate 45 65 5 A ellipsoid of axial ratio 1.5, (f / f o ) of 1.15 as hydrodynamic parameter (Burz et al., 1994a), and a hydration factor of 0.30 g/g of protein. The wild-type repressor is exceptionally compact and apparent Stokes radius increases for all seven single-site mutations (Burz et al., 1994a).
R. Chattopadhyaya, K. Ghosh / Journal of Structural Biology 141 (2003) 103–114
6. C-domain contributes to operator affinity. The importance of C-domain dimerization is indicated by the loss of operator affinity following separation of the two domains and this is the mechanism utilized by RecA in switching from lysogeny to lytic state. The dissociation constant of the intact repressor dimer is 3 1013 M and of the N-domain, 3 108 M (Burz and Ackers, 1994b). 7. The hinge region is ordered and part of the Cdomain. The so-called ‘‘hinge’’ region (93–131) is partially ordered (Johnson et al., 1981; Banik et al., 1992; Ghosh and Chattopadhyaya, 2001), and associates with and stabilizes the remainder (132–236) of the carboxyterminal domain. In a study (De Anda et al., 1983) with the P22 repressor, fragment AB remains associated with the rest of the domain (fragment BC) after proteolysis; they can be separated only under mildly denaturing conditions. 8. NMR studies. These showed that the k repressor contains two domains, the C-domain contains strong dimer and higher order contacts, the two domains are loosely tethered by a linker peptide; and proteolysis does not perturb the structure of either domain (Weiss et al., 1987; Sauer et al., 1990). The NMR studies were carried out in the absence of DNA, whereas the model proposed by us is the operator-bound form of the repressors, which have considerably reduced flexibility of the linker peptide as with the NF-jB homodimer bound to DNA. Further, it has been argued that the NMR studies performed at 5 mM concentration are likely to reflect properties of higher oligomers and a fluorescence anisotropy study concluded that the hinge is not that flexible (Banik et al., 1993). 9. Crystal structure of N-domain. Our models are automatically compatible with the crystal structure of k repressor N-domain with operator DNA (Beamer and Pabo, 1992), since the C-domain assemblies were docked on this crystal structure and linked via residues 93–108 (1gfx) or 93–104 (1j5g) or 93–103 (1lwq). 10. a5 –a05 interaction persists in intact k repressor dimer. The last helix of the N-domain, a5 , stabilizes the dimer (Sauer et al., 1990; Kombo et al., 1996); this interaction is almost fully preserved in our intact repressor models. Covalent bond formation was observed between Cys 88 residues in a YC88 mutant (Sauer et al., 1990). In addition, IS84 mutation reduces operator affinity 100fold. 11. Proteolytic cleavage sites in loop regions. Papain is observed to cleave k repressor at 92–93, then at 121–122 and lastly at 131–132. The first two are in loop regions (Fig. 1) and cleavage at the third happens only at a later stage. 12. Potential glycosylation site must be on the surface. Glycosylation site is predicted for sequences Asn-X-Thr and Asn-X-Ser, and such a site exists at 216–218 in k repressor. Both Asn 216 and Ser 218 side chains are on the surface in all our models. A prokaryotic protein
107
never shows glycosylation, but protein structures are largely determined by sequence and not by presence or absence of glycosylation, hence these side chains, namely Asn 216 and Ser 218 are expected to protrude on the surface, as seen in our models. 13. Correlations with various autocleavage data. We have used several autocleavage data to build our model: Ser 149 and Lys 192, the catalytic residues (Sauer et al., 1990; Slilaty and Little, 1987; Little, 1984), are close to the substrate Ala-Gly. Thr 122, Asp 125, Glu 127, and Ala 152 when mutated resulted in ind mutants resistant to RecA-mediated cleavage (Gimble and Sauer, 1986). Mutations at these residues seem to affect the binding of activated RecA to the repressor and hence must themselves be surface residues, as in our models. PT158 and EK233 mutations show easier RecA-mediated cleavage as they favor the monomeric state (Sauer et al., 1990). In our models both Pro 158 and Glu 233 are near/part of the dimeric interface, thus explaining how mutations at these positions may affect the dimerization. Gly 185 and Phe 189 mutants remove RecA-mediated cleavage and the region 185–189 is involved in RecA binding—it becomes more accessible in the monomeric state (Cohen et al., 1981; Sauer et al., 1990). It was also shown that RecA does not cleave a covalent repressor dimer having a YC88 mutant (Gimble and Sauer, 1989). Since we have built a dimeric model bound to the operators, we hypothesize that in the operator-bound state, RecAmediated cleavage will not occur as the region 185–189 is partly covered by the cleavage site region. In the monomeric state, more flexibility in the cleavage site region will lead to the requisite exposure of the 185–189 RecA-binding region. In 1gfx, 1j5g, region 185–189 is partly covered by 115–119 and to a lesser extent by 123– 126. In 1lwq, one-half of the subunits are so covered by 115–119, the other half being part of the cooperative interface. 14. Surface residues from antirepressor protein data. Antirepressor protein (Ant) of phage P22 is able to inactivate k repressor by a mechanism not involving the cleavage of the latter (Sauer et al., 1990). Ant recognizes portions of the C-domain; it can bind repressor dimers and monomers equally well. EK144, GD147, GN147, and GV147 mutants are all Ant resistant. These data agree with our models since Glu 144 and Gly 147 are surface residues but away from the dimerization interface. The mechanism of Ant-mediated inactivation is not clear, but Ant binding blocks binding to operator DNA. 15. Accessibility of Trp residues. One and possibly two of the three Trp residues are highly exposed and can be quenched readily at low concentrations of acrylamide; one Trp is shielded even at high acrylamide but becomes exposed at 3 M urea (Banik et al., 1992). In 1gfx and 1j5g, Trp 230 is the most exposed (Bandyopadhyay et al., 1995), Trp 142 is the most shielded, and Trp 129 is
108
R. Chattopadhyaya, K. Ghosh / Journal of Structural Biology 141 (2003) 103–114
intermediate in accessibility. Trp 129 is part of a7 that melts at 3 M urea, as helicity is removed (Banik et al., 1992). 16. Dimerization through side chains of the most hydrophobic regions. We initially predicted these regions to be, b3 ; b8 ; b4 , and b2 in decreasing order of hydrophobicity, though b4 is not used in our comparative model. A study on the P22 and 434 repressors (Donner et al., 1997) showed that 434 residues 169–209 (193–233 of k) and 99–174 (122–198 of k) contain such regions. 17. Compatibility with thermodynamic data of dimerization. Thermodynamic data showed hydrophobic association, charge neutralization, and/or specific dehydration events occur with dimer formation and Asp/ Glu residues are part of the interface (Koblan and Ackers, 1991a), consistent with our models 1gfx and 1j5g. 18. Compatibility of thermodynamic data with cooperative surface proposed. Thermodynamic data for dimer-dimer association showed no effect of salt concentration on the association (Koblan and Ackers, 1991b), ruling out involvement of charged side chains; neutral polar residues and hydrophobic interactions are indicated, as in our models. 19. Surface residues from cooperativity data. Whipple et al. (1994) predicted that Asn 148, Ser 149, Ser 159, Glu 188, Lys 192, Arg 196, Asp 197, Ser 198, Gly 199, Phe 202, Tyr 210, Met 212, Ser 228, and Thr 234 are surface residues in k repressor. In our comparative models, Asn 148, Ser 149, Tyr 210, and Lys 192, ‘‘blue patch’’ residues inferred from the crystal (Bell et al., 2000; Whipple et al., 1998), are buried since the Ala-Gly covers the catalytic site. 20. Dimerization mutants. Pro 158 and Ser 159 are near the dimerization interface in the comparative models (1gfx, 1j5g). Ser 159, conserved among the three phage repressors, or Ser 228, replaced by Ala in P22, 434, and LexA (Sauer et al., 1982), when mutated to Asn weakens dimerization (Burz et al., 1994a). Differences in hydrogen-bonding properties, or polarity, do not explain the difference observed due to these SN mutations. In both models, Ser 228 is not at the dimerization surface but nearby; an introduction of a large side chain here disrupts optimal dimerization. The changes in Stokes radius observed (Burz et al., 1994a) result from this, being more pronounced for the SR228 mutation. The EK233 and TK234 mutations also weaken dimerization (Burz et al., 1994a; Sauer et al., 1990). Thr 234 is not required for dimerization, since the 434 repressor terminates at k position 233 but it can still dimerize. Introduction of lysine affects dimerization probably due to repulsion from positively charged side chains nearby. Positions 233 and 234 are close to the interface in our comparative models. 21. Immunological analysis of k repressor. Antibodies generated against three peptides corresponding to resi-
dues 107–120, 143–157 and 181–195 of k repressor (Sussman and Alexander, 1989) showed the highest reactivity towards these immunogenic peptides and no cross-reaction, and it was concluded that all three regions in the repressor are folded. Region 107–120 was inferred to be folded under physiological conditions, becoming more exposed under denaturing conditions. Further, it was concluded that these regions are located on the same side of the C-domain (Sussman and Alexander, 1989). This observation is satisfied by our comparative models since region 110–120 is sandwiched between b4 –b5 and b2 in the former. 22. Intertwined linker in the dimer? The linker region is used by each monomer to wind about the other monomer in 1gfx, thus explaining why repressor binding to DNA is cooperative as it can effectively bind to DNA only as a dimer (Ptashne, 1992); it cannot leave the operator as a monomer unless the linker region is cut or unwound. Normally it would have to dissociate as a dimer (order of addition experiment, pages 81–83 of Ptashne, 1992). The situation is different for LexA repressor binding to its operator, where a monomer binds independently of the other monomer (Kim and Little, 1992). The k repressor has a different relative disposition of the C-terminal end of its N-domains at an operator site compared to LexA (Chattopadhyaya et al., 2000), allowing a swap model as in 1j5g; this also may result in such an observation without requiring intertwining. Intertwining was difficult in the case of 1j5g and 1lwq due to steric clashes. Comparative model. Fragment 112–236 of k repressor shares 26% sequence identity with UmuD0 (Peat et al., 1996). Both template searches in comparative modeling using the SCOP database (Murzin et al., 1995) and fold recognition (Fischer and Eisenberg, 1996) identified the UmuD0 as the correct answer, and various biochemical and biophysical data have been shown to be in agreement with this comparative model. However, it does not show Asn 148, Ser 149, Lys 192, and Tyr 210 as surface residues (Whipple et al., 1994). The dimerization of the C-domain in our models (1gfx, 1j5g) is in agreement with residues comprising the dimerization surface in P22 and 434 repressors (Donner et al., 1997) and other mutant data (Burz et al., 1994a). The UmuD0 dimer observed in the crystal (Peat et al., 1996) includes a salt bridge between Lys 55 and Glu 93, but as the corresponding k residues are Glu 144 and Glu 188, such a salt bridge is not possible for k. Further, Glu 144 is important for Ant repressor binding and since Ant can bind both k monomers and dimers equally well, Glu 144 is unlikely to be a part of the dimerization surface. Our dimerization is not the same as the UmuD0 dimerization in solution which is exclusively through the C-terminal b-strand (Ferentz et al., 1997), as in our model side chains from b2 ; b3 , and b8 participate.
R. Chattopadhyaya, K. Ghosh / Journal of Structural Biology 141 (2003) 103–114
Dimerization and cooperativity in our tetramer models. Though our monomer (in 1gfx) is nearly identical to the recent crystal structure of the 132–236 fragment (Bell et al., 2000; pdb code 1f39), with an RMS deviation of using all backbone atoms in regions 139–152, 1.24 A 159–184, 188–197, 201–215, and 218–230 for alignment (Fig. 2), we propose different dimerization and cooperative surfaces in 1gfx and 1j5g compared to those seen in the crystal. In the crystal dimers found in 1f39, 1kca, and 1lwq, Trp 230 is completely buried in the dimerization interface (Bell et al., 2000) but solution data showed that Trp 230 in the intact repressor is accessible in the dimeric form and becomes inaccessible only in the tetramer (Bandyopadhyay et al., 1995). In the FC235 mutant, the cysteine at position 235 is reactive to DTNB (Bandyopadhyay et al., 1995), whereas in the crystal dimer Phe 235 is bound to a hydrophobic pocket from the other subunit and buried (Bell et al., 2000). Further, the dimer-dimer interfaces of the intact repressor are different in operator-bound and free states (Banik et al., 1993; Bandyopadhyay et al., 1996). In the crystal, the Tyr 210 side chain is exposed as part of the ‘‘blue patches’’ (Bell et al., 2000), half of which remain exposed even after the crystalline tetramer formation. Thus the crystalline tetramer (Bell et al., 2000) does not explain why the Cys 210 side chain in the YC210 mutant shows little reaction with DTNB at 6.8 lM protein (Deb et al., 1998). In 1gfx, 1j5g the Ala-Gly covers the Tyr 210 side
Fig. 2. Stereo view of a Ca superposition of residues 109–236 of comparative model (green, 1gfx) with 136–236 of crystal structure (orange, 1f39). Main-chain atoms in 78 residues in regions 138–152, 159–184, 188–197, 201–216 and 218–230 were used to obtain this op between the timal superposition having an RMS deviation of 1.24 A two structures. The extra region 109–135 not determined in the crystallographic study is closer to the viewer, the Ala111-Gly112 placed as close to the catalytic site as possible. Hydrophobic side chains in 109– 135 interact with surface hydrophobic residues in the rest of the Cdomain, and the two a-helices in this region are also apparent. In 1j5g, an improved version of the comparative model, residues 136–236 have nearly the same structure as 1gfx, but residues 104–118 are based on the cleavage site region in the LexA available in 1jhe and residues 150– 160 are taken from 1f39, shown in Fig. 4.
109
chain in all the subunits, rendering it buried in the intact repressor, consistent with the DTNB data (Deb et al., 1998). In 1lwq these data are satisfied since half the catalytic sites are involved in cooperative contacts while the other half are covered by their own cleavage sites. The crystal structure of fragment 132–236 excludes region 93–131; their model (Bell et al., 2000) ignores 93–131 and consequently the RecA mediated cleavage mutants at positions 122, 125, and 127 (Sauer et al., 1990). In our comparative model, region 122–127 is close to the other RecA binding site at region 185–189 (Sauer et al., 1990) as a result of the Ala-Gly covering the catalytic site. We propose that only the ‘‘red patch’’ (Bell et al., 2000) on one surface of the monomer including side chains from residues 197–200, 202, 204, 211, 212, and 214 are utilized for the dimer-dimer interactions in vivo (Fig. 3b). The lack of cooperativity in the ‘‘blue patch’’ mutants (Bell et al., 2000) observed in the genetic data can be explained by an indirect effect these mutations cause in the conformation of residues near the Ala-Gly and hence the linker region as a whole. The juxtaposition of the red and blue patches comprising several ion pairs and hydrogen bonds (Bell et al., 2000) is unlikely for the OR1 –OR2 cooperativity since the cooperative interaction was found to be unaffected by KCl concentration (Koblan and Ackers, 1991b). Salt would compete with the ion pairs, especially since the crystal dimer-dimer association surface has gaps (Bell et al., 2000), but no effect of salt was observed (Koblan and Ackers, 1991b). The contacts between blue and red patches is probably a result of salting out since 2:2 M salt was used for the crystallization (Bell et al., 2000). Like the crystal dimer, our model includes several residues like Pro 158, Ser 159, Ser 228, Trp 230, Glu 233, Thr 234, and Phe 235 at or near the dimerization surface, as described in point 20 above. However, it also includes some hydrophobic side chains from b2 ; b3 , and b8 , i.e., Ala 140, Trp 142, Leu 165, Leu 167, Ile 226, Ala 227, and the neutral polar side chain of Gln 229 in agreement with a study that residues from several regions of the protein are involved in dimerization (Donner et al., 1997). Several negatively charged side chains like Asp 169, Glu 171, Glu 175, and Glu 233 are near the dimerization surface (shown but not labeled in Fig. 3a) as predicted from the pH dependence of dimerization (Koblan and Ackers, 1991a). Our dimerization model explains why papain cannot cleave the operator-bound repressor dimer (Ghosh and Chattopadhyaya, 2001) as 92–93 is buried in the operatorbound state; the subsequent cleavages at 121–122 and 131–132 also do not occur since the cleavage at 121–122 needs the first cleavage as a prerequisite and 131–132 is part of a7 in the operator-bound state. Further, the dimerization of the fragment 93–236 seems to differ in the wt compared to SN228 significantly as assessed by the observed rate of disappearance of 93–236 (Ghosh and
110
R. Chattopadhyaya, K. Ghosh / Journal of Structural Biology 141 (2003) 103–114
Fig. 3. Comparative model 1gfx built using residues 109–236 as shown in Fig. 2. (a) Ca-trace of our dimer in stereo shows that side chains from b2 ; b3 , and b8 comprise the dimeric interface, some residues are labeled in molecule B to the right; the vertical 2-fold pseudosymmetry axis relates the two Cdomains as well as the N-domains resulting in an intertwined dimer; papain cleavage, autocleavage sites, catalytic residues, and those implicated in RecA binding, ant binding, antibody binding are marked in molecule A to the left. (b) Two dimers bound to adjacent sites separated by a severely untwisted and bent 7-bp spacer found between OR1 (top) and OR2 (bottom). Regions 151–157 (weakly) and 194–218 participate in dimer-dimer contacts. A third pseudo-two fold bisecting the ones at OR1 and OR2 centers is formed, roughly horizontal here. Molecules A through D are from top to bottom; C-domains of molecules B and C are involved in dimer-dimer contacts.
Chattopadhyaya, 2001), assuming that digestion requires dissociation, while dimerization of fragment 132– 236 is not that different in wt and SN228 (Ghosh and
Chattopadhyaya, 2001). This suggests that the dimer of 132–236 does not reflect intact repressor dimerization but the 93–236 dimer does. This has been verified in the
R. Chattopadhyaya, K. Ghosh / Journal of Structural Biology 141 (2003) 103–114
111
Fig. 4. Improved comparative model 1j5g built using residues 135–236 from Fig. 3, with replacement of the region 150–160 from 1f39, and region 104–118 using the cleavage site region as in 1jhe. (a) Ca-trace of the 1j5g dimer in stereo shows that the positions of the C-domains are largely identical to the previous model 1gfx presented in Fig. 3 relative the N-domains, hence the dimeric interface. However the N–C connectivity is different as this is a swap model with the C-domain of molecule B sitting atop the N-domain of molecule A and vice versa, unlike in 1gfx. The vertical 2-fold pseudosymmetry axis relates the two C-domains as well as the N-domains but this dimer is not as intertwined as the one in Fig. 3. Papain cleavage, autocleavage sites, catalytic residues, and those implicated in RecA binding, ant binding, antibody binding marked in the C-domain to the left, whereas residues near the dimeric interface are labelled in the C-domain to the right. (b) Two dimers bound to adjacent sites separated by a severely untwisted and bent 7 bp spacer found between OR1 (top) and OR2 (bottom). Regions 151–157 (weakly) and 194–218 participate in dimer-dimer contacts. A third pseudo 2-fold bisecting the ones at OR1 and OR2 centers is formed, roughly horizontal here. N-domains of molecules A through D are from top to bottom; C-domains of molecules A and D are involved in dimer–dimer contacts.
112
R. Chattopadhyaya, K. Ghosh / Journal of Structural Biology 141 (2003) 103–114
Fig. 5. Model 1lwq based on 1kca for residues 140–236, on 1jhe or 1jhh for region 102–123. (a) The dimer shown here in stereo is unlike those in Figs. 3 and 4 in respect of the 2-fold relating the two C-domains does not coincide with the 2-fold relating the N-domains. Molecule A has its N-domain to the left and the C-domain toward the viewer, while molecule B has its N-domain to the right and C-domain away from the viewer. Ser 149 in molecule B cannot have the cleavage site region very close, as it is involved in the cooperative interface (not shown in part a), whereas Ser 149 in molecule A can have the Ôcleavable formÕ of the cleavage site. Not only the cleavage site region but also the linker region 93–101 are different for the two molecules within a dimer, due to this lack of symmetry. (b) Full tetramer and operator DNA shown for model 1lwq, with the N-domain and DNA coordinates being virtually identical to those for 1gfx and 1j5g.
R. Chattopadhyaya, K. Ghosh / Journal of Structural Biology 141 (2003) 103–114
case of the homologous LexA structure (Luo et al., 2001) where part of the linker region is also involved in the dimeric interface. Also, there is no evidence for intermolecular b-sheet formation, which shows a sharp peak at 1617 cm1 or any a-helix in our FTIR data of 132–236 (Ghosh and Chattopadhyaya, 2001). Four different C-domain temperature sensitive mutants isolated for k included FI141, PL153, NT207, and KE224 (Jana et al., 1999). Of these, 207 is part of the red patch hence a cooperativity mutant, both 141 and 224 side chains are close to the dimerization surface, and 153 is part of a loop touching both dimerization and dimerdimer surfaces (Fig. 3b), part of the insert absent in the SOS proteins (Peat et al., 1996). Hydroxyl radical footprinting implied a stable bend between the OR1 and OR2 sites of 18° (Strahs and Brenowitz, 1994). In our model (1gfx) there is a bend in the helix axis of 15 ° as a result of repressor binding between the two operator sites (Fig. 3b), in addition to the 16° bend observed along each operator (Beamer and Pabo, 1992). A significant change in the bend between the two operator sites is the untwisting of the DNA to about 25–27° twist angle per step and a consequent increase in rise per step. The reduction in twist angles in this region brings the two operator dyads closer to each other. With a normal twist angle of 35 °, 24 base steps between the centers of the dyads would skew these dyads by about (24–21) 35 or 105 ° and cooperative contacts could not be observed. The untwisting in the intervening region (by 55 °) results in a reduced skew angle of 50° in our model between the dyads, permitting the cooperative contacts, looking down the DNA helix. This is the reason why much less DNA distortion is observed for OR2 –OR3 compared to OR1 –OR2 upon repressor binding (Deb et al., 2000), since the dyad axes in the OR2 and OR3 centers are separated by 23 base steps and hence skewed by ð23–21Þ 35 or 70° in normal DNA, requiring un untwisting by only 20° to achieve a similar skew angle as in OR1 –OR2 and similar cooperative contacts. It is remarkable that the OL1 –OL2 also has 24 base steps separating the dyad centers like OR1 –OR2 ; OL2 –OL3 has only 20 base steps and will need an overtwisted spacer to achieve the same cooperative interaction, or, it may involve a different cooperative interface (see Figs. 4 and 5).
Acknowledgments The authors thank Drs. S. Bandyopadhyay and S. Roy for useful discussions, and C. Mukhopadhyay for help regarding various computer programs. The authors also thank the Distributive Information Centre of Bose Institute for the modeling facility supported by DBT, Government of India.
113
References Aggarwal, A.K., Rodgers, D.W., Drottar, M., Ptashne, M., Harrison, S.C., 1988. Recognition of phage 434: A view at high resolution. Science 242, 899–907. Anderson, J., Ptashne, M., Harrison, S.C., 1984. Cocrystals of the DNA-binding domain of phage 434 repressor and a synthetic phage operator. Proc. Natl. Acad. Sci. USA 81, 1307–1311. Bandyopadhyay, S., Banik, U., Bhattacharyya, B., Mandal, N.C., Roy, S., 1995. Role of the C-terminal tail region in the selfassembly of k-repressor. Biochemistry 34, 5090–5097. Bandyopadhyay, S., Mukhopadhyay, S., Roy, S., 1996. Dimer–dimer interfaces of the k-repressor are different in liganded and free states. Biochemistry 35, 5033–5040. Banik, U., Mandal, N.C., Bhattacharyya, B., Roy, S., 1992. Multiphasic denaturation of the k repressor by urea and its implications for the repressor structure. Eur. J. Biochem. 206, 15–21. Banik, U., Mandal, N.C., Bhattacharyya, R., Roy, S., 1993. A fluorescence anisotropy study of tetramer–dimer equilibrium of k repressor and its implication for function. J. Biol. Chem. 268, 3938–3943. crystal structure of the k Beamer, L.J., Pabo, C.O., 1992. Refined 1.8 A repressor–operator complex. J. Mol. Biol. 227, 177–196. Bell, C.E., Frescura, P., Hochschild, A., Lewis, M., 2000. Crystal structure of the k repressor C-terminal domain provides a model for cooperative operator binding. Cell 101, 801–811. Bell, C.E., Lewis, M., 2001. Crystal structure of the k repressor Cterminal domain octamer. J. Mol. Biol. 314, 1127–1136. Biosym/Molecular Simulations, 1995. Insight II User Guide Release 95.0, San Diego, CA. Brack, C., Pirrotta, V., 1975. Electron microscopic study of the repressor of bacteriophage k and its interaction with operator DNA. J. Mol. Biol. 96, 139–152. Burz, D.S., Beckett, D., Benson, N., Ackers, G., 1994a. Self-Assembly of bacteriophage k cI repressor: effects of single-site mutations on the monomer-dimer equilibrium. Biochemistry 33, 8399–8405. Burz, D.S., Ackers, G.K., 1994b. Single-site mutations in the Cterminal domain of bacteriophage k cI repressor alter cooperative interactions between dimers adjacently bound to OR . Biochemistry 33, 8406–8416. Chattopadhyaya, R., Ghosh, K., Namboodiri, V.M.H., 2000. Model of a lexA repressor dimer bound to recA operator. J. Biomol. Struct. Dyn. 18, 181–197. Chothia, C., Lesk, A.M., 1986. The relation between the divergence of sequence and structure in proteins. EMBO J. 5, 823–826. Chou, P.Y., Fasman, G.D., 1978. Empirical predictions of protein conformation. Ann. Rev. Biochem. 47, 251–276. Cohen, S., Knoll, B.J., Little, J.W., Mount, D.W., 1981. Preferential cleavage of phage k repressor monomers by RecA protease. Nature 294, 182–184. De Anda, J.D., Poteete, A.R., Sauer, R.T., 1983. P22 c2 repressor: Domain structure and function. J. Biol. Chem. 258, 10536–10542. Deb, S., Bandyopadhyay, S., Roy, S., 1998. Spectroscopic study of Y210C k-repressor: implications for cooperative interaction. Protein Eng. 11, 481–487. Deb, S., Bandyopadhyay, S., Roy, S., 2000. DNA sequence dependent and independent conformational changes in multipartite operator recognition by k-repressor. Biochemistry 39, 3377–3383. Donner, A.L., Carlson, P.A., Koudelka, G.B., 1997. Dimerization specificity of P22 and 434 repressors is determined by multiple polypeptide segments. J. Bacteriol. 179, 1253–1261. Ferentz, A.E., Opperman, T., Walker, G.C., Wagner, G., 1997. Dimerization of the UmuD0 protein in solution and its implications for regulation of SOS mutagenesis. Nature Struct. Biol. 4, 979–983. Fischer, D., Eisenberg, D., 1996. Protein fold recognition using sequence-derived predictions. Protein Sci. 5, 947–955.
114
R. Chattopadhyaya, K. Ghosh / Journal of Structural Biology 141 (2003) 103–114
Garnier, J., Osguthorpe, D.J., Robson, B., 1978. Analysis of the accuracy and implications of simple methods for predicting secondary structure of globular proteins. J. Mol. Biol. 120, 97–120. Ghosh, K., Chattopadhyaya, R., 2001. Papain does not cleave operator-bound lambda repressor: structural characterization of the carboxy terminal domain and the hinge. J. Biomol. Struct. Dyn. 18, 557–567. Gimble, F.S., Sauer, R.T., 1986. k repressor inactivation: properties of purified ind proteins in the autodigestion and RecA-mediated cleavage reactions. J. Mol. Biol. 192, 39–47. Gimble, F.S., Sauer, R.T., 1989. k repressor mutants that are better substrates for RecA-mediated cleavage. J. Mol. Biol. 206, 29–39. Jana, N.K., Roy, S., Bhattacharyya, B., Mandal, N.C., 1999. Amino acid changes in the repressor of bacteriophage lambda due to temperature-sensitive mutations in its cI gene and the structure of a highly temperature-sensitive mutant repressor. Protein Eng. 12, 225–233. Johnson, A.D., Poteete, A.R., Lauer, G., Sauer, R.T., Ackers, G.K., Ptashne, M., 1981. k repressor and cro—components of an efficient molecular switch. Nature 294, 217–223. Kim, B., Little, J.W., 1992. Dimerization of a specific DNA-binding protein on the DNA. Science 255, 203–206. Koblan, K.S., Ackers, G.K., 1991a. Energetics of subunit dimerization in bacteriophage k cI repressor: linkage to protons, temperature and KCl. Biochemistry 30, 7817–7821. Koblan, K.S., Ackers, G.K., 1991b. Cooperative protein–DNA interactions: effects of KCl on k cI Binding to OR . Biochemistry 30, 7822–7827. Kombo, D.C., Nemethy, G., Gibson, K.D., Rachovsky, S., Scheraga, H.A., 1996. Computer-aided discrimination between active and inactive mutants of the N-terminal domain of the bacteriophage k repressor. J. Mol. Biol. 256, 517–532. Little, J.W., Mount, D.W., 1982. The SOS Regulatory System of Escherichia coli. Cell 29, 11–22. Little, J.W., 1984. Autodigestion of lexA and phage k repressors. Proc. Natl. Acad. Sci. USA 81, 1375–1379. Luo, Y., Pfuetzner, R.A., Mosimann, S., Paetzel, M., Frey, E.A., Cherney, M., Kim, B., Little, J.W., Strynadka, N.C.J., 2001. Crystal structure of LexA: a conformational switch for regulation of self-cleavage. Cell 106, 585–594. Moult, J., Hubbard, T., Fidelis, K., Pederson, J.T., 1999. Critical assessment of methods of protein structure prediction (CASP): Round III. Proteins Suppl. 3, 1–6. Murzin, A.G., Brenner, S.E., Hubbard, T., Chothia, C., 1995. SCOP: a structural classification of proteins database for the investigation of sequences and structures. J. Mol. Biol. 247, 536–540.
Pabo, C.O., Sauer, R.T., Sturtevant, J.M., Ptashne, M., 1979. The k repressor contains two domains. Proc. Natl. Acad. Sci. USA 76, 1608–1612. Pabo, C.O., Lewis, M., 1982. The operator-binding domain of k repressor: structure and DNA recognition. Nature 298, 443–447. Peat, T.S., Frank, E.G., McDonald, J.P., Levine, A.S., Woodgate, R., Hendrickson, W.A., 1996. Structure of the UmuD0 protein and its regulation in response to DNA damage. Nature 380, 727–730. Ptashne, M., 1992. A genetic switch phage kand higher organisms. Cell Press & Blackwell, Cambridge, MA. Sauer, R.T., Yocum, R.R., Doolittle, R.F., Lewis, M., Pabo, C.O., 1982. Homology among DNA-binding proteins suggests use of a conserved super-secondary structure. Nature 298, 447–451. Sauer, R.T., Jordan, S.R., Pabo, C.O., 1990. k Repressor: a model system for understanding protein–DNA interactions and protein stability. Adv. Prot. Chem. 40, 1–61. Shenkin, P.S., Yarmush, D.L., Fine, R.M., Wang, H., Levinthal, C., 1987. Predicting antibody hypervariable loop conformation. Biopolymers 26, 2053–2085. Shimon, L.J.W., Harrison, S.C., 1993. The phage 434 OR 2/R1-69 resolution. J. Mol. Biol. 232, 826–838. complex at 2.5 A Slilaty, S.N., Rupley, J.A., Little, J.W., 1986. Intramolecular clevage of LexA and phage k repressors: Dependence of kinetics on repressor concentration, pH, temperature, and solvent. Biochemistry 25, 6866–6875. Slilaty, S.N., Little, J.W., 1987. Lysine-156 and serine-119 are required for LexA repressor cleavage: a possible mechanism. Proc. Natl. Acad. Sci. USA 84, 3987–3991. Strahs, D., Brenowitz, M., 1994. DNA conformational changes associated with the cooperative binding of cI-repressor of bacteriophage k to OR . J. Mol. Biol. 244, 494–510. Surewicz, W.K., Mantsch, H.H., Chapman, D., 1993. Determination of protein secondary structure by Fourier transform infrared spectroscopy: a critical assessment. Biochemistry 32, 389–394. Sussman, R., Alexander, H.B., 1989. Structural analysis of the carboxy terminus of bacteriophage lambda repressor deptermined by antipeptide antibodies. J. Bacteriol. 171, 1235–1244. Weiss, M.A., Karplus, M., Sauer, R.T., 1987. Quarternary structure and function in phage k repressor: 1 H- NMR studies of genetically altered proteins. J. Biomol. Struct. Dyn. 5, 539–556. Whipple, F.W., Kuldell, N.H., Cheatham, L.A., Hochschild, A., 1994. Specificity determinants for the interaction of k repressor and P22 repressor dimers. Genes Dev. 8, 1212–1223. Whipple, F.W., Hou, E.F., Hochschild, A., 1998. Amino-acid-amino acid contacts at the cooperativity interface of the bacteriophage k and P22 repressors. Genes Dev. 12, 2791–2802.
Structural Biology Journal of Structural Biology 141 (2003) 103–114 www.elsevier.com/locate/yjsbi
A comparative three-dimensional model of the carboxy-terminal domain of the lambda repressor and its use to build intact repressor tetramer models bound to adjacent operator sitesq Rajagopal Chattopadhyaya* and Kaushik Ghosh Department of Biochemistry, Bose Institute, P-1/12, C.I.T. Scheme VII M, Calcutta 700054, India Received 1 July 2002, and in revised form 21 November 2002
Abstract A model for residues 93–236 of the k repressor (1gfx) was predicted, based on the UmuD0 crystal structure, as part of four intact repressor molecules bound to two adjacent operator sites. The structure of region 136–230 in 1gfx was found to be nearly identical to the independently determined crystal structure of the 132–236 fragment, 1f39, released later by the PDB. Later, two more tetrameric models of the k repressor tetramer bound to two adjacent operator sites were constructed by us; in one of these, 1j5g, the N-domain and C-domain coordinates and hence monomer-monomer and dimer-dimer interactions are almost the same as in 1gfx, but the structure of the linker region is partly based on the linker region of the LexA dimer in 1jhe; in the other, 1lwq, the crystalline tetramer for region 140–236 has been coopted from the crystal structure deposited in 1kca, the operator DNA and N-domain coordinates of which are same as those in 1gfx and 1j5g, but the linker region is partly based on the LexA dimer structures 1jhe and 1jhh. Monomer-monomer interactions at the same operator site are stabilized by exposed hydrophobic side chains in b-strands while cooperative interactions are mostly confined to b6 and some adjacent residues in both 1gfx and 1j5g. Mutational data, existence of a twofold axis relating two C-domains within a dimer, and minimization of DNA distortion between adjacent operator sites allow us to roughly position the C-domain with respect to the N-domain for both 1gfx and 1j5g. The study correlates these models with functional, biochemical, biophysical, and immunological data on the repressor in the literature. The oligomerization mode observed in the crystal structure of 132–236 may not exist in the intact repressor bound to the operator since it is shown to contradict several published biochemical data on the intact repressor. Ó 2003 Elsevier Science (USA). All rights reserved. Keywords: Lambda repressor cI; Gene regulation; Protein structure prediction; Structure–function relationship
1. Introduction A common important feature of gene expression in each of the bacteriophages k, P22, and 434 is the use of a negative regulatory protein called repressor (Ptashne,
q PDB accession numbers for the k repressor first comparative model in RCSB001477 (1gfx) was processed on June 22, 2000, the second model in RCSB001635 (1j5g) processed on May 3, 2002, and third comparative model in RCSB016355 processed May 30, 2002 (1lwq); our related model of LexA repressor in RCSB000408 (1qaa) was processed on February 2, 1999. * Corresponding author. Fax: +91-33-234-3886. E-mail address: [email protected] (R. Chattopadhyaya).
1992; Johnson et al., 1981), important for maintaining lysogeny. The repressors of k, P22, and 434 consist of 236, 216, and 209 residues, respectively, folded into two structural domains, separable by protease digestion (Pabo et al., 1979; De Anda et al., 1983; Anderson et al., 1984). For controlling expression, each repressor molecule forms a dimer that can bind to two sets of three homologous, partially symmetric binding sites (OR,OL) known as operator sites (Johnson et al., 1981) through the amino terminal domains (N-domains), while the carboxy-terminal domains (C-domains) are involved in dimer formation and dimer-dimer interactions crucial for its function. The C-domains also contain both the catalytic and the cleavage sites for autocleavage assisted
1047-8477/03/$ - see front matter Ó 2003 Elsevier Science (USA). All rights reserved. doi:10.1016/S1047-8477(02)00627-5
104
R. Chattopadhyaya, K. Ghosh / Journal of Structural Biology 141 (2003) 103–114
by activated RecA of the host leading to the lytic pathway, triggered by host DNA damage. Crystal structures of all the intact repressor proteins remain hitherto undetermined. Those of the N-domains of k (residues 1–92) (Pabo and Lewis, 1982; Beamer and Pabo, 1992) and 434 (residues 1–69) (Aggarwal et al., 1988; Shimon and Harrison, 1993) have been elucidated. Recently, the crystal structure of fragment 132–236 has been reported (Bell et al., 2000). Independently, we have predicted the three-dimensional structure of fragment 93–236 and combined it with the crystal structure of the N-domain bound to operator DNA (Beamer and Pabo, 1992). The structural classification of protein database available on the web at http://scop.mrc-lmb.cam.ac.uk/ scop/ (Murzin et al., 1995) contains a description of the evolutionary and structural relation of proteins with experimentally determined structures. The protein structure prediction field has received intense interest and high level of participation lately as evidenced by three meetings on the critical assessment of techniques for protein structure prediction in 1994, 1996, and 1998 (Moult et al., 1999). The methodology of protein structure predictions can be of three broad kinds: comparative or homology modeling, fold recognition, and ab initio rediction (Moult et al., 1999). We have used all three methods for predicting the three-dimensional structure of the repressor C-domain. Though the ab initio model of the C-domain was initially used to build a tetramer model bound to two adjacent operator sites, the C-domain in the comparative model 1gfx was found to be almost identical with the subsequently released crystal structure in 1f39. Another model, 1j5g, with a different structure of the cleavage site based on a LexA dimer crystal structure, 1jhe, was constructed using the comparative model as previously deposited in 1gfx. Finally, a crystal structure of the tetramer for region 140–236 taken from 1kca was used to build a third model 1lwq of the lambda repressor tetramer bound to two adjacent operator sites similar to the one partly proposed in Fig. 6 of Bell and Lewis (2001). The present paper analyzes to what extent each of these alternative models for the tetramer satisfies the biochemical and biophysical data accumulated in the literature. The methods here are similar to, but not identical with, those adopted recently for building a model of the LexA dimer bound to the recA operator (Chattopadhyaya et al., 2000).
2. Methods Carboxy-terminal sequence similarities among the repressors. The bacteriophage repressors of k, P22, and 434 share extensive sequence identity in their C-domains (Sauer et al., 1982). In pairwise comparisons identities
are significant (434 and P22, 82/139 or 59%, P22 and k, 62/138 or 45%; 434 and k, 48/139 or 35%). For the three phage repressors, 44 positions are identical and including LexA, an SOS protein (Little and Mount, 1982), 20 positions are still identical (Sauer et al., 1982). In addition, both phage and SOS repressors like LexA, UmuD, and mucA are cleaved by RecA protein at identical AlaGly sequences in a reaction involving RecA binding to the respective C-domains (Slilaty et al., 1986; Sauer et al., 1982; Little and Mount, 1982). These similarities prompted us to hypothesize that the three phage repressors possess C-domains whose three-dimensional structures are more or less identical (Chothia and Lesk, 1986). Phased secondary structure prediction. Secondary structure was predicted for each of the phage primary sequences using the Chou and Fasman (1978) and Garnier et al. (1978) methods independently. To find the common secondary structure of the phage repressors, we emphasized the phased nature of the individual aligned amino acid probabilities (Pa ; Pb ; Pt ) with those of other members of the family assuming a multiple alignment (Sauer et al., 1982) and their magnitudes; the correct secondary structure in any region showed this phasing. In Fig. 1 our secondary structure prediction is seen to match quite well with that found in the UmuD0 crystal structure which was used as a template in our comparative modeling. Template search for comparative modeling. Using the primary sequence of residues 93–236 of the k cI repressor as a query sequence, the structural classification database (Murzin et al., 1995) was searched for its predicted similarities with experimentally solved structures through the website http://stash.mrc-lmb.cam.ac.uk/ PDB_ISL/. Template search for fold recognition. Using the primary sequence of residues 93–236 of the k repressor, a fold recognition program using comparison of predicted secondary structure strings with those of known folds (Fischer and Eisenberg, 1996) was run at the UCLA fold recognition server http://www.mbi.ucla.edu/people/ fischer/BENCH/benchmark1.html. Comparative model building for region 138–230 (1gfx, 1j5g). Both template searches for comparative modeling and fold recognition described above yielded the UmuD0 crystal structure (Peat et al., 1996) as the correct answer. For building the k repressor model the amino acid alignment in Peat et al. (1996) was used as a starting point for UmuD0 residues 49–136 or k residues 138–230. The loop regions 153–158, 185–188, and 196–200 of k repressor had insertions and a deletion at 216–218 with respect to UmuD0 (Fig. 1b), and hence were built using a procedure for generating loops de novo within the Homology module of InsightII (Shenkin et al., 1987; Biosym/MSI, 1995), choosing the best loop by the criterion of lack of steric clash with the rest of the protein
R. Chattopadhyaya, K. Ghosh / Journal of Structural Biology 141 (2003) 103–114
105
Fig. 1. (a) Pairwise amino acid sequence alignment of k repressor residues 102–135 with residues 75–105 of LexA. The LexA crystal structure (Luo et al., 2001) was used only to build region 104–118 in our revised comparative model in 1j5g. (b) Pairwise amino acid sequence alignment of k repressor residues 100–236 with UmuD0 residues 25–139 used in building the comparative model with our phased secondary structure prediction versus that found in the crystal structure (Peat et al., 1996). Identical residues (6 of 31 LexA positions, 31 of 115 UmuD0 positions) are in bold, and stretches of the parent structures, a total of 78 residues, used to build comparative models 1gfx and 1j5g are underlined. Secondary structure nomenclature in Peat et al. (1996) also given for comparison in the lowest line.
structure and subsequent energy refinement within the Discover module of InsightII. The alignment given in Peat et al. (1996) for the region 214–220 between b7 and b8 cannot be correct since that would expose both Cys 215 and Cys 219 to the surface. In our alignment (Fig. 1b) for this region, Cys 215 aligns with position 119 of UmuD0 , rather than 120. This loop is shorter in k compared to UmuD0 by two residues and Cys 219 is also partially hidden as a result, aligned with UmuD0 residue 125. The structure of region 150–160 from 1f39 has been coopted in 1j5g, otherwise the C-domain in 1j5g is the same as in 1gfx. Cleavage site in 1gfx. The conformation of the region 32–48 in UmuD0 was ignored and not used for model building, since from our autocleavage experiments it was concluded that the substrate Ala-Gly must be close to the catalytic side chains of Ser 149 and Lys 192 due to the large value of the preexponential factor (see Results and discussion) and the binding of fragment AB with BC for P22 c2 repressor (De Anda et al., 1983). from the Ser 149 Keeping the Ala-Gly at about 7–8 A the region 109–137 was modeled based on the secondary structure prediction including loops and two brief ahelices whose apolar sides pointed toward the rest of the C-domain. Relative juxtapositioning of the two domains (1gfx). For building the intact repressor, the N-domain–C-domain steric relationship was not decided directly. A dimer consisting of two C-domains (109–236) related by a
2-fold axis was constructed first such that hydrophobicity considerations, thermodynamics of dimerization and biochemical data were satisfied (see points 16,17, and 20 under Results and discussion). Later, the C–C dimer 2-fold was coincided with that of the N–N dimer; while keeping the N-domains and the operator DNA fixed, the C–C dimer can be rotated as a unit (Chattopadhyaya et al., 2000) about the 2-fold, or translated parallel to the 2-fold (i.e., two degrees of freedom). By trial and error, the C-domain dimer was placed on the N-domain dimer such that correct dimer–dimer contacts could be maintained and distortion in the DNA structure minimized. We assumed that the 2-fold relating the C-domains within a dimer coincides with the 2-fold relating the N-domains in the same dimer. The dimensions of the repressor dimer obtained from electron microscopy (Brack and Pirrotta, 1975) places an upper limit to the separation of the C–C dimer from the N–N dimer, while steric clashes placed at a lower limit. After the Cdomains were properly placed with respect to the Ndomains, residues 93–108 were used to join the two domains. Two operator duplexes with two bound intact repressor molecules were placed by trial and error on a graphics terminal such that the regions in the C-domain implicated in cooperative binding (see points 18 and 19 under Results and discussion) were in contact and the DNA could be made continuous using the sequence separating the OR1 and OR2 sites with minimal distortion from ideal B-DNA.
106
R. Chattopadhyaya, K. Ghosh / Journal of Structural Biology 141 (2003) 103–114
2.1. Cleavage site in 1j5g
3. Results and discussion
Since in our autocleavage experiments for the k repressor bound to operator DNA (A. Pal and R. Chattopadhyaya, unpublished, 2001), we find cleavage, we used structure 1jhe of the LexA dimer (Luo et al., 2001) containing the cleavable form of the cleavage site for region 105–118. This region has a different structure compared to the corresponding region in 1gfx. There is a helix in region 125–132 in this model in agreement with our secondary structure prediction, CD and FTIR data. Only 6 out of 34 amino acids in this region are identical between LexA and k cI repressors (Fig. 1a); in addition, cI has 3 insertions compared to LexA—hence it is quite possible that these extra residues will be accommodated as a helix whereas a helix does not exist in LexA at the corresponding location (Luo et al., 2001).
3.1. Comparative and crystal structure-based models
2.2. Relative juxtapositioning of the two domains (1j5g) The two domains and operator DNA are almost unchanged from 1gfx though the C-domains have been renamed. The N–C connection is different in two ways, one being the structure of the cleavage site as described above and the other being a swap type N–C connection unlike in 1gfx. As a result of this connection, if ABCD is the order of four molecules on the operator DNA, molecules A and D are involved in the cooperative contacts, whereas in 1gfx, molecules B and C were involved in the cooperative contacts. 2.3. Cleavage site in 1lwq Since region 140–236 of each molecule forming a tetramer is taken from the crystal structure in 1kca, it was found that the ‘‘noncleavable’’ cleavage site (Luo et al., 2001) must accompany at least two of the molecules whose catalytic sites are involved in cooperative contacts in this model. Accordingly, such noncleavable sites were built using 1jhh for two subunits and cleavable sites built using 1jhe for the two other subunits within the tetramer. 2.4. Relative juxtapositioning of the two domains (1lwq) The entire assembly of the tetramer together with the attached cleavage sites (region 102–236) after annealing and energy refinement was placed symmetrically relative to the assembly of the four N-domains bound to the operator DNA exactly as in entries 1gfx and 1j5g, resembling Fig. 6 of Bell and Lewis (2001). The connection between 92 and 102 was chosen so that the 92–102 distance was almost equal in both subunits comprising a dimer. This assembly lacks the 2-fold axes relating the two subunits within the dimers but the 2-fold of the tetramer coincides with the pseudo 2-fold of the N-domain operator assembly.
All the later models are in general agreement with various kinds of data enumerated below: 1. Circular dichroism measurements. These showed (Ghosh and Chattopadhyaya, 2001) that k repressorÕs carboxy-terminal domain (132–236) is largely made up of b-sheet. Fragment 93–236 contains more helix than 112–236, which in turn contains more helix than 132– 236, as shown by their CD spectra. This agrees with our unified secondary structure prediction (Fig. 1b). 2. FTIR measurements. The 93–236 fragment spectrum (Ghosh and Chattopadhyaya, 2001) shows a-helix content as evidenced by the peak at 1656 cm1 in the deconvoluted spectrum whereas the 132–236 fragment shows little or no a-helix content from the absence of any peak at 1656 cm1 in the deconvoluted spectrum (Surewicz et al., 1993). 3. None of the three Cys reactive in the folded structure. Cys 180, Cys 215, and Cys 219 thiols did not react with the reagent DTNB either in the intact repressor (Banik et al., 1992) or in fragment 132–236 (Ghosh and Chattopadhyaya, 2001), though protease cleaved repressor showed near full reaction. Since at the 10 lM initial concentration of 132–236, some monomer exists, the dimerization did not prevent the reaction. Thus, neither are the cysteine thiols on the dimerization surface nor the N-domain are protecting them; it is a result of the protein fold. Thiol hydrogens cover the sulfur atoms in 1gfx and 1j5g. However, in the crystal structures Cys 215 and 219 are seen to be disulfide linked. 4. The activation energy of the autocleavage reaction and its consequence. We measured the free energy of activation ðEa ¼ DH þ RT Þ to be 22.2 kcal/mol at 300 K (Ghosh and Chattopadhyaya, 2001). For this value of Ea the Arrhenius exponential factor is 7 1017 , hence the preexponential factor A is 1:9 10þ10 s1 for our rate constant and 2 10þ11 s1 assuming the previous rate constant (Slilaty et al., 1986). Since the preexponential Arrhenius factor represents the number of collisions per second, such a high number of collisions requires that the catalytic groups are quite close to the substrate Ala-Gly. 5. Electron microscopy data, hydrodynamic parameters, hydration factors. Electron microscopy data (Brack and Pirrotta, 1975) suggested monomeric dimensions of , consistent with our model. It is a prolate 45 65 5 A ellipsoid of axial ratio 1.5, (f / f o ) of 1.15 as hydrodynamic parameter (Burz et al., 1994a), and a hydration factor of 0.30 g/g of protein. The wild-type repressor is exceptionally compact and apparent Stokes radius increases for all seven single-site mutations (Burz et al., 1994a).
R. Chattopadhyaya, K. Ghosh / Journal of Structural Biology 141 (2003) 103–114
6. C-domain contributes to operator affinity. The importance of C-domain dimerization is indicated by the loss of operator affinity following separation of the two domains and this is the mechanism utilized by RecA in switching from lysogeny to lytic state. The dissociation constant of the intact repressor dimer is 3 1013 M and of the N-domain, 3 108 M (Burz and Ackers, 1994b). 7. The hinge region is ordered and part of the Cdomain. The so-called ‘‘hinge’’ region (93–131) is partially ordered (Johnson et al., 1981; Banik et al., 1992; Ghosh and Chattopadhyaya, 2001), and associates with and stabilizes the remainder (132–236) of the carboxyterminal domain. In a study (De Anda et al., 1983) with the P22 repressor, fragment AB remains associated with the rest of the domain (fragment BC) after proteolysis; they can be separated only under mildly denaturing conditions. 8. NMR studies. These showed that the k repressor contains two domains, the C-domain contains strong dimer and higher order contacts, the two domains are loosely tethered by a linker peptide; and proteolysis does not perturb the structure of either domain (Weiss et al., 1987; Sauer et al., 1990). The NMR studies were carried out in the absence of DNA, whereas the model proposed by us is the operator-bound form of the repressors, which have considerably reduced flexibility of the linker peptide as with the NF-jB homodimer bound to DNA. Further, it has been argued that the NMR studies performed at 5 mM concentration are likely to reflect properties of higher oligomers and a fluorescence anisotropy study concluded that the hinge is not that flexible (Banik et al., 1993). 9. Crystal structure of N-domain. Our models are automatically compatible with the crystal structure of k repressor N-domain with operator DNA (Beamer and Pabo, 1992), since the C-domain assemblies were docked on this crystal structure and linked via residues 93–108 (1gfx) or 93–104 (1j5g) or 93–103 (1lwq). 10. a5 –a05 interaction persists in intact k repressor dimer. The last helix of the N-domain, a5 , stabilizes the dimer (Sauer et al., 1990; Kombo et al., 1996); this interaction is almost fully preserved in our intact repressor models. Covalent bond formation was observed between Cys 88 residues in a YC88 mutant (Sauer et al., 1990). In addition, IS84 mutation reduces operator affinity 100fold. 11. Proteolytic cleavage sites in loop regions. Papain is observed to cleave k repressor at 92–93, then at 121–122 and lastly at 131–132. The first two are in loop regions (Fig. 1) and cleavage at the third happens only at a later stage. 12. Potential glycosylation site must be on the surface. Glycosylation site is predicted for sequences Asn-X-Thr and Asn-X-Ser, and such a site exists at 216–218 in k repressor. Both Asn 216 and Ser 218 side chains are on the surface in all our models. A prokaryotic protein
107
never shows glycosylation, but protein structures are largely determined by sequence and not by presence or absence of glycosylation, hence these side chains, namely Asn 216 and Ser 218 are expected to protrude on the surface, as seen in our models. 13. Correlations with various autocleavage data. We have used several autocleavage data to build our model: Ser 149 and Lys 192, the catalytic residues (Sauer et al., 1990; Slilaty and Little, 1987; Little, 1984), are close to the substrate Ala-Gly. Thr 122, Asp 125, Glu 127, and Ala 152 when mutated resulted in ind mutants resistant to RecA-mediated cleavage (Gimble and Sauer, 1986). Mutations at these residues seem to affect the binding of activated RecA to the repressor and hence must themselves be surface residues, as in our models. PT158 and EK233 mutations show easier RecA-mediated cleavage as they favor the monomeric state (Sauer et al., 1990). In our models both Pro 158 and Glu 233 are near/part of the dimeric interface, thus explaining how mutations at these positions may affect the dimerization. Gly 185 and Phe 189 mutants remove RecA-mediated cleavage and the region 185–189 is involved in RecA binding—it becomes more accessible in the monomeric state (Cohen et al., 1981; Sauer et al., 1990). It was also shown that RecA does not cleave a covalent repressor dimer having a YC88 mutant (Gimble and Sauer, 1989). Since we have built a dimeric model bound to the operators, we hypothesize that in the operator-bound state, RecAmediated cleavage will not occur as the region 185–189 is partly covered by the cleavage site region. In the monomeric state, more flexibility in the cleavage site region will lead to the requisite exposure of the 185–189 RecA-binding region. In 1gfx, 1j5g, region 185–189 is partly covered by 115–119 and to a lesser extent by 123– 126. In 1lwq, one-half of the subunits are so covered by 115–119, the other half being part of the cooperative interface. 14. Surface residues from antirepressor protein data. Antirepressor protein (Ant) of phage P22 is able to inactivate k repressor by a mechanism not involving the cleavage of the latter (Sauer et al., 1990). Ant recognizes portions of the C-domain; it can bind repressor dimers and monomers equally well. EK144, GD147, GN147, and GV147 mutants are all Ant resistant. These data agree with our models since Glu 144 and Gly 147 are surface residues but away from the dimerization interface. The mechanism of Ant-mediated inactivation is not clear, but Ant binding blocks binding to operator DNA. 15. Accessibility of Trp residues. One and possibly two of the three Trp residues are highly exposed and can be quenched readily at low concentrations of acrylamide; one Trp is shielded even at high acrylamide but becomes exposed at 3 M urea (Banik et al., 1992). In 1gfx and 1j5g, Trp 230 is the most exposed (Bandyopadhyay et al., 1995), Trp 142 is the most shielded, and Trp 129 is
108
R. Chattopadhyaya, K. Ghosh / Journal of Structural Biology 141 (2003) 103–114
intermediate in accessibility. Trp 129 is part of a7 that melts at 3 M urea, as helicity is removed (Banik et al., 1992). 16. Dimerization through side chains of the most hydrophobic regions. We initially predicted these regions to be, b3 ; b8 ; b4 , and b2 in decreasing order of hydrophobicity, though b4 is not used in our comparative model. A study on the P22 and 434 repressors (Donner et al., 1997) showed that 434 residues 169–209 (193–233 of k) and 99–174 (122–198 of k) contain such regions. 17. Compatibility with thermodynamic data of dimerization. Thermodynamic data showed hydrophobic association, charge neutralization, and/or specific dehydration events occur with dimer formation and Asp/ Glu residues are part of the interface (Koblan and Ackers, 1991a), consistent with our models 1gfx and 1j5g. 18. Compatibility of thermodynamic data with cooperative surface proposed. Thermodynamic data for dimer-dimer association showed no effect of salt concentration on the association (Koblan and Ackers, 1991b), ruling out involvement of charged side chains; neutral polar residues and hydrophobic interactions are indicated, as in our models. 19. Surface residues from cooperativity data. Whipple et al. (1994) predicted that Asn 148, Ser 149, Ser 159, Glu 188, Lys 192, Arg 196, Asp 197, Ser 198, Gly 199, Phe 202, Tyr 210, Met 212, Ser 228, and Thr 234 are surface residues in k repressor. In our comparative models, Asn 148, Ser 149, Tyr 210, and Lys 192, ‘‘blue patch’’ residues inferred from the crystal (Bell et al., 2000; Whipple et al., 1998), are buried since the Ala-Gly covers the catalytic site. 20. Dimerization mutants. Pro 158 and Ser 159 are near the dimerization interface in the comparative models (1gfx, 1j5g). Ser 159, conserved among the three phage repressors, or Ser 228, replaced by Ala in P22, 434, and LexA (Sauer et al., 1982), when mutated to Asn weakens dimerization (Burz et al., 1994a). Differences in hydrogen-bonding properties, or polarity, do not explain the difference observed due to these SN mutations. In both models, Ser 228 is not at the dimerization surface but nearby; an introduction of a large side chain here disrupts optimal dimerization. The changes in Stokes radius observed (Burz et al., 1994a) result from this, being more pronounced for the SR228 mutation. The EK233 and TK234 mutations also weaken dimerization (Burz et al., 1994a; Sauer et al., 1990). Thr 234 is not required for dimerization, since the 434 repressor terminates at k position 233 but it can still dimerize. Introduction of lysine affects dimerization probably due to repulsion from positively charged side chains nearby. Positions 233 and 234 are close to the interface in our comparative models. 21. Immunological analysis of k repressor. Antibodies generated against three peptides corresponding to resi-
dues 107–120, 143–157 and 181–195 of k repressor (Sussman and Alexander, 1989) showed the highest reactivity towards these immunogenic peptides and no cross-reaction, and it was concluded that all three regions in the repressor are folded. Region 107–120 was inferred to be folded under physiological conditions, becoming more exposed under denaturing conditions. Further, it was concluded that these regions are located on the same side of the C-domain (Sussman and Alexander, 1989). This observation is satisfied by our comparative models since region 110–120 is sandwiched between b4 –b5 and b2 in the former. 22. Intertwined linker in the dimer? The linker region is used by each monomer to wind about the other monomer in 1gfx, thus explaining why repressor binding to DNA is cooperative as it can effectively bind to DNA only as a dimer (Ptashne, 1992); it cannot leave the operator as a monomer unless the linker region is cut or unwound. Normally it would have to dissociate as a dimer (order of addition experiment, pages 81–83 of Ptashne, 1992). The situation is different for LexA repressor binding to its operator, where a monomer binds independently of the other monomer (Kim and Little, 1992). The k repressor has a different relative disposition of the C-terminal end of its N-domains at an operator site compared to LexA (Chattopadhyaya et al., 2000), allowing a swap model as in 1j5g; this also may result in such an observation without requiring intertwining. Intertwining was difficult in the case of 1j5g and 1lwq due to steric clashes. Comparative model. Fragment 112–236 of k repressor shares 26% sequence identity with UmuD0 (Peat et al., 1996). Both template searches in comparative modeling using the SCOP database (Murzin et al., 1995) and fold recognition (Fischer and Eisenberg, 1996) identified the UmuD0 as the correct answer, and various biochemical and biophysical data have been shown to be in agreement with this comparative model. However, it does not show Asn 148, Ser 149, Lys 192, and Tyr 210 as surface residues (Whipple et al., 1994). The dimerization of the C-domain in our models (1gfx, 1j5g) is in agreement with residues comprising the dimerization surface in P22 and 434 repressors (Donner et al., 1997) and other mutant data (Burz et al., 1994a). The UmuD0 dimer observed in the crystal (Peat et al., 1996) includes a salt bridge between Lys 55 and Glu 93, but as the corresponding k residues are Glu 144 and Glu 188, such a salt bridge is not possible for k. Further, Glu 144 is important for Ant repressor binding and since Ant can bind both k monomers and dimers equally well, Glu 144 is unlikely to be a part of the dimerization surface. Our dimerization is not the same as the UmuD0 dimerization in solution which is exclusively through the C-terminal b-strand (Ferentz et al., 1997), as in our model side chains from b2 ; b3 , and b8 participate.
R. Chattopadhyaya, K. Ghosh / Journal of Structural Biology 141 (2003) 103–114
Dimerization and cooperativity in our tetramer models. Though our monomer (in 1gfx) is nearly identical to the recent crystal structure of the 132–236 fragment (Bell et al., 2000; pdb code 1f39), with an RMS deviation of using all backbone atoms in regions 139–152, 1.24 A 159–184, 188–197, 201–215, and 218–230 for alignment (Fig. 2), we propose different dimerization and cooperative surfaces in 1gfx and 1j5g compared to those seen in the crystal. In the crystal dimers found in 1f39, 1kca, and 1lwq, Trp 230 is completely buried in the dimerization interface (Bell et al., 2000) but solution data showed that Trp 230 in the intact repressor is accessible in the dimeric form and becomes inaccessible only in the tetramer (Bandyopadhyay et al., 1995). In the FC235 mutant, the cysteine at position 235 is reactive to DTNB (Bandyopadhyay et al., 1995), whereas in the crystal dimer Phe 235 is bound to a hydrophobic pocket from the other subunit and buried (Bell et al., 2000). Further, the dimer-dimer interfaces of the intact repressor are different in operator-bound and free states (Banik et al., 1993; Bandyopadhyay et al., 1996). In the crystal, the Tyr 210 side chain is exposed as part of the ‘‘blue patches’’ (Bell et al., 2000), half of which remain exposed even after the crystalline tetramer formation. Thus the crystalline tetramer (Bell et al., 2000) does not explain why the Cys 210 side chain in the YC210 mutant shows little reaction with DTNB at 6.8 lM protein (Deb et al., 1998). In 1gfx, 1j5g the Ala-Gly covers the Tyr 210 side
Fig. 2. Stereo view of a Ca superposition of residues 109–236 of comparative model (green, 1gfx) with 136–236 of crystal structure (orange, 1f39). Main-chain atoms in 78 residues in regions 138–152, 159–184, 188–197, 201–216 and 218–230 were used to obtain this op between the timal superposition having an RMS deviation of 1.24 A two structures. The extra region 109–135 not determined in the crystallographic study is closer to the viewer, the Ala111-Gly112 placed as close to the catalytic site as possible. Hydrophobic side chains in 109– 135 interact with surface hydrophobic residues in the rest of the Cdomain, and the two a-helices in this region are also apparent. In 1j5g, an improved version of the comparative model, residues 136–236 have nearly the same structure as 1gfx, but residues 104–118 are based on the cleavage site region in the LexA available in 1jhe and residues 150– 160 are taken from 1f39, shown in Fig. 4.
109
chain in all the subunits, rendering it buried in the intact repressor, consistent with the DTNB data (Deb et al., 1998). In 1lwq these data are satisfied since half the catalytic sites are involved in cooperative contacts while the other half are covered by their own cleavage sites. The crystal structure of fragment 132–236 excludes region 93–131; their model (Bell et al., 2000) ignores 93–131 and consequently the RecA mediated cleavage mutants at positions 122, 125, and 127 (Sauer et al., 1990). In our comparative model, region 122–127 is close to the other RecA binding site at region 185–189 (Sauer et al., 1990) as a result of the Ala-Gly covering the catalytic site. We propose that only the ‘‘red patch’’ (Bell et al., 2000) on one surface of the monomer including side chains from residues 197–200, 202, 204, 211, 212, and 214 are utilized for the dimer-dimer interactions in vivo (Fig. 3b). The lack of cooperativity in the ‘‘blue patch’’ mutants (Bell et al., 2000) observed in the genetic data can be explained by an indirect effect these mutations cause in the conformation of residues near the Ala-Gly and hence the linker region as a whole. The juxtaposition of the red and blue patches comprising several ion pairs and hydrogen bonds (Bell et al., 2000) is unlikely for the OR1 –OR2 cooperativity since the cooperative interaction was found to be unaffected by KCl concentration (Koblan and Ackers, 1991b). Salt would compete with the ion pairs, especially since the crystal dimer-dimer association surface has gaps (Bell et al., 2000), but no effect of salt was observed (Koblan and Ackers, 1991b). The contacts between blue and red patches is probably a result of salting out since 2:2 M salt was used for the crystallization (Bell et al., 2000). Like the crystal dimer, our model includes several residues like Pro 158, Ser 159, Ser 228, Trp 230, Glu 233, Thr 234, and Phe 235 at or near the dimerization surface, as described in point 20 above. However, it also includes some hydrophobic side chains from b2 ; b3 , and b8 , i.e., Ala 140, Trp 142, Leu 165, Leu 167, Ile 226, Ala 227, and the neutral polar side chain of Gln 229 in agreement with a study that residues from several regions of the protein are involved in dimerization (Donner et al., 1997). Several negatively charged side chains like Asp 169, Glu 171, Glu 175, and Glu 233 are near the dimerization surface (shown but not labeled in Fig. 3a) as predicted from the pH dependence of dimerization (Koblan and Ackers, 1991a). Our dimerization model explains why papain cannot cleave the operator-bound repressor dimer (Ghosh and Chattopadhyaya, 2001) as 92–93 is buried in the operatorbound state; the subsequent cleavages at 121–122 and 131–132 also do not occur since the cleavage at 121–122 needs the first cleavage as a prerequisite and 131–132 is part of a7 in the operator-bound state. Further, the dimerization of the fragment 93–236 seems to differ in the wt compared to SN228 significantly as assessed by the observed rate of disappearance of 93–236 (Ghosh and
110
R. Chattopadhyaya, K. Ghosh / Journal of Structural Biology 141 (2003) 103–114
Fig. 3. Comparative model 1gfx built using residues 109–236 as shown in Fig. 2. (a) Ca-trace of our dimer in stereo shows that side chains from b2 ; b3 , and b8 comprise the dimeric interface, some residues are labeled in molecule B to the right; the vertical 2-fold pseudosymmetry axis relates the two Cdomains as well as the N-domains resulting in an intertwined dimer; papain cleavage, autocleavage sites, catalytic residues, and those implicated in RecA binding, ant binding, antibody binding are marked in molecule A to the left. (b) Two dimers bound to adjacent sites separated by a severely untwisted and bent 7-bp spacer found between OR1 (top) and OR2 (bottom). Regions 151–157 (weakly) and 194–218 participate in dimer-dimer contacts. A third pseudo-two fold bisecting the ones at OR1 and OR2 centers is formed, roughly horizontal here. Molecules A through D are from top to bottom; C-domains of molecules B and C are involved in dimer-dimer contacts.
Chattopadhyaya, 2001), assuming that digestion requires dissociation, while dimerization of fragment 132– 236 is not that different in wt and SN228 (Ghosh and
Chattopadhyaya, 2001). This suggests that the dimer of 132–236 does not reflect intact repressor dimerization but the 93–236 dimer does. This has been verified in the
R. Chattopadhyaya, K. Ghosh / Journal of Structural Biology 141 (2003) 103–114
111
Fig. 4. Improved comparative model 1j5g built using residues 135–236 from Fig. 3, with replacement of the region 150–160 from 1f39, and region 104–118 using the cleavage site region as in 1jhe. (a) Ca-trace of the 1j5g dimer in stereo shows that the positions of the C-domains are largely identical to the previous model 1gfx presented in Fig. 3 relative the N-domains, hence the dimeric interface. However the N–C connectivity is different as this is a swap model with the C-domain of molecule B sitting atop the N-domain of molecule A and vice versa, unlike in 1gfx. The vertical 2-fold pseudosymmetry axis relates the two C-domains as well as the N-domains but this dimer is not as intertwined as the one in Fig. 3. Papain cleavage, autocleavage sites, catalytic residues, and those implicated in RecA binding, ant binding, antibody binding marked in the C-domain to the left, whereas residues near the dimeric interface are labelled in the C-domain to the right. (b) Two dimers bound to adjacent sites separated by a severely untwisted and bent 7 bp spacer found between OR1 (top) and OR2 (bottom). Regions 151–157 (weakly) and 194–218 participate in dimer-dimer contacts. A third pseudo 2-fold bisecting the ones at OR1 and OR2 centers is formed, roughly horizontal here. N-domains of molecules A through D are from top to bottom; C-domains of molecules A and D are involved in dimer–dimer contacts.
112
R. Chattopadhyaya, K. Ghosh / Journal of Structural Biology 141 (2003) 103–114
Fig. 5. Model 1lwq based on 1kca for residues 140–236, on 1jhe or 1jhh for region 102–123. (a) The dimer shown here in stereo is unlike those in Figs. 3 and 4 in respect of the 2-fold relating the two C-domains does not coincide with the 2-fold relating the N-domains. Molecule A has its N-domain to the left and the C-domain toward the viewer, while molecule B has its N-domain to the right and C-domain away from the viewer. Ser 149 in molecule B cannot have the cleavage site region very close, as it is involved in the cooperative interface (not shown in part a), whereas Ser 149 in molecule A can have the Ôcleavable formÕ of the cleavage site. Not only the cleavage site region but also the linker region 93–101 are different for the two molecules within a dimer, due to this lack of symmetry. (b) Full tetramer and operator DNA shown for model 1lwq, with the N-domain and DNA coordinates being virtually identical to those for 1gfx and 1j5g.
R. Chattopadhyaya, K. Ghosh / Journal of Structural Biology 141 (2003) 103–114
case of the homologous LexA structure (Luo et al., 2001) where part of the linker region is also involved in the dimeric interface. Also, there is no evidence for intermolecular b-sheet formation, which shows a sharp peak at 1617 cm1 or any a-helix in our FTIR data of 132–236 (Ghosh and Chattopadhyaya, 2001). Four different C-domain temperature sensitive mutants isolated for k included FI141, PL153, NT207, and KE224 (Jana et al., 1999). Of these, 207 is part of the red patch hence a cooperativity mutant, both 141 and 224 side chains are close to the dimerization surface, and 153 is part of a loop touching both dimerization and dimerdimer surfaces (Fig. 3b), part of the insert absent in the SOS proteins (Peat et al., 1996). Hydroxyl radical footprinting implied a stable bend between the OR1 and OR2 sites of 18° (Strahs and Brenowitz, 1994). In our model (1gfx) there is a bend in the helix axis of 15 ° as a result of repressor binding between the two operator sites (Fig. 3b), in addition to the 16° bend observed along each operator (Beamer and Pabo, 1992). A significant change in the bend between the two operator sites is the untwisting of the DNA to about 25–27° twist angle per step and a consequent increase in rise per step. The reduction in twist angles in this region brings the two operator dyads closer to each other. With a normal twist angle of 35 °, 24 base steps between the centers of the dyads would skew these dyads by about (24–21) 35 or 105 ° and cooperative contacts could not be observed. The untwisting in the intervening region (by 55 °) results in a reduced skew angle of 50° in our model between the dyads, permitting the cooperative contacts, looking down the DNA helix. This is the reason why much less DNA distortion is observed for OR2 –OR3 compared to OR1 –OR2 upon repressor binding (Deb et al., 2000), since the dyad axes in the OR2 and OR3 centers are separated by 23 base steps and hence skewed by ð23–21Þ 35 or 70° in normal DNA, requiring un untwisting by only 20° to achieve a similar skew angle as in OR1 –OR2 and similar cooperative contacts. It is remarkable that the OL1 –OL2 also has 24 base steps separating the dyad centers like OR1 –OR2 ; OL2 –OL3 has only 20 base steps and will need an overtwisted spacer to achieve the same cooperative interaction, or, it may involve a different cooperative interface (see Figs. 4 and 5).
Acknowledgments The authors thank Drs. S. Bandyopadhyay and S. Roy for useful discussions, and C. Mukhopadhyay for help regarding various computer programs. The authors also thank the Distributive Information Centre of Bose Institute for the modeling facility supported by DBT, Government of India.
113
References Aggarwal, A.K., Rodgers, D.W., Drottar, M., Ptashne, M., Harrison, S.C., 1988. Recognition of phage 434: A view at high resolution. Science 242, 899–907. Anderson, J., Ptashne, M., Harrison, S.C., 1984. Cocrystals of the DNA-binding domain of phage 434 repressor and a synthetic phage operator. Proc. Natl. Acad. Sci. USA 81, 1307–1311. Bandyopadhyay, S., Banik, U., Bhattacharyya, B., Mandal, N.C., Roy, S., 1995. Role of the C-terminal tail region in the selfassembly of k-repressor. Biochemistry 34, 5090–5097. Bandyopadhyay, S., Mukhopadhyay, S., Roy, S., 1996. Dimer–dimer interfaces of the k-repressor are different in liganded and free states. Biochemistry 35, 5033–5040. Banik, U., Mandal, N.C., Bhattacharyya, B., Roy, S., 1992. Multiphasic denaturation of the k repressor by urea and its implications for the repressor structure. Eur. J. Biochem. 206, 15–21. Banik, U., Mandal, N.C., Bhattacharyya, R., Roy, S., 1993. A fluorescence anisotropy study of tetramer–dimer equilibrium of k repressor and its implication for function. J. Biol. Chem. 268, 3938–3943. crystal structure of the k Beamer, L.J., Pabo, C.O., 1992. Refined 1.8 A repressor–operator complex. J. Mol. Biol. 227, 177–196. Bell, C.E., Frescura, P., Hochschild, A., Lewis, M., 2000. Crystal structure of the k repressor C-terminal domain provides a model for cooperative operator binding. Cell 101, 801–811. Bell, C.E., Lewis, M., 2001. Crystal structure of the k repressor Cterminal domain octamer. J. Mol. Biol. 314, 1127–1136. Biosym/Molecular Simulations, 1995. Insight II User Guide Release 95.0, San Diego, CA. Brack, C., Pirrotta, V., 1975. Electron microscopic study of the repressor of bacteriophage k and its interaction with operator DNA. J. Mol. Biol. 96, 139–152. Burz, D.S., Beckett, D., Benson, N., Ackers, G., 1994a. Self-Assembly of bacteriophage k cI repressor: effects of single-site mutations on the monomer-dimer equilibrium. Biochemistry 33, 8399–8405. Burz, D.S., Ackers, G.K., 1994b. Single-site mutations in the Cterminal domain of bacteriophage k cI repressor alter cooperative interactions between dimers adjacently bound to OR . Biochemistry 33, 8406–8416. Chattopadhyaya, R., Ghosh, K., Namboodiri, V.M.H., 2000. Model of a lexA repressor dimer bound to recA operator. J. Biomol. Struct. Dyn. 18, 181–197. Chothia, C., Lesk, A.M., 1986. The relation between the divergence of sequence and structure in proteins. EMBO J. 5, 823–826. Chou, P.Y., Fasman, G.D., 1978. Empirical predictions of protein conformation. Ann. Rev. Biochem. 47, 251–276. Cohen, S., Knoll, B.J., Little, J.W., Mount, D.W., 1981. Preferential cleavage of phage k repressor monomers by RecA protease. Nature 294, 182–184. De Anda, J.D., Poteete, A.R., Sauer, R.T., 1983. P22 c2 repressor: Domain structure and function. J. Biol. Chem. 258, 10536–10542. Deb, S., Bandyopadhyay, S., Roy, S., 1998. Spectroscopic study of Y210C k-repressor: implications for cooperative interaction. Protein Eng. 11, 481–487. Deb, S., Bandyopadhyay, S., Roy, S., 2000. DNA sequence dependent and independent conformational changes in multipartite operator recognition by k-repressor. Biochemistry 39, 3377–3383. Donner, A.L., Carlson, P.A., Koudelka, G.B., 1997. Dimerization specificity of P22 and 434 repressors is determined by multiple polypeptide segments. J. Bacteriol. 179, 1253–1261. Ferentz, A.E., Opperman, T., Walker, G.C., Wagner, G., 1997. Dimerization of the UmuD0 protein in solution and its implications for regulation of SOS mutagenesis. Nature Struct. Biol. 4, 979–983. Fischer, D., Eisenberg, D., 1996. Protein fold recognition using sequence-derived predictions. Protein Sci. 5, 947–955.
114
R. Chattopadhyaya, K. Ghosh / Journal of Structural Biology 141 (2003) 103–114
Garnier, J., Osguthorpe, D.J., Robson, B., 1978. Analysis of the accuracy and implications of simple methods for predicting secondary structure of globular proteins. J. Mol. Biol. 120, 97–120. Ghosh, K., Chattopadhyaya, R., 2001. Papain does not cleave operator-bound lambda repressor: structural characterization of the carboxy terminal domain and the hinge. J. Biomol. Struct. Dyn. 18, 557–567. Gimble, F.S., Sauer, R.T., 1986. k repressor inactivation: properties of purified ind proteins in the autodigestion and RecA-mediated cleavage reactions. J. Mol. Biol. 192, 39–47. Gimble, F.S., Sauer, R.T., 1989. k repressor mutants that are better substrates for RecA-mediated cleavage. J. Mol. Biol. 206, 29–39. Jana, N.K., Roy, S., Bhattacharyya, B., Mandal, N.C., 1999. Amino acid changes in the repressor of bacteriophage lambda due to temperature-sensitive mutations in its cI gene and the structure of a highly temperature-sensitive mutant repressor. Protein Eng. 12, 225–233. Johnson, A.D., Poteete, A.R., Lauer, G., Sauer, R.T., Ackers, G.K., Ptashne, M., 1981. k repressor and cro—components of an efficient molecular switch. Nature 294, 217–223. Kim, B., Little, J.W., 1992. Dimerization of a specific DNA-binding protein on the DNA. Science 255, 203–206. Koblan, K.S., Ackers, G.K., 1991a. Energetics of subunit dimerization in bacteriophage k cI repressor: linkage to protons, temperature and KCl. Biochemistry 30, 7817–7821. Koblan, K.S., Ackers, G.K., 1991b. Cooperative protein–DNA interactions: effects of KCl on k cI Binding to OR . Biochemistry 30, 7822–7827. Kombo, D.C., Nemethy, G., Gibson, K.D., Rachovsky, S., Scheraga, H.A., 1996. Computer-aided discrimination between active and inactive mutants of the N-terminal domain of the bacteriophage k repressor. J. Mol. Biol. 256, 517–532. Little, J.W., Mount, D.W., 1982. The SOS Regulatory System of Escherichia coli. Cell 29, 11–22. Little, J.W., 1984. Autodigestion of lexA and phage k repressors. Proc. Natl. Acad. Sci. USA 81, 1375–1379. Luo, Y., Pfuetzner, R.A., Mosimann, S., Paetzel, M., Frey, E.A., Cherney, M., Kim, B., Little, J.W., Strynadka, N.C.J., 2001. Crystal structure of LexA: a conformational switch for regulation of self-cleavage. Cell 106, 585–594. Moult, J., Hubbard, T., Fidelis, K., Pederson, J.T., 1999. Critical assessment of methods of protein structure prediction (CASP): Round III. Proteins Suppl. 3, 1–6. Murzin, A.G., Brenner, S.E., Hubbard, T., Chothia, C., 1995. SCOP: a structural classification of proteins database for the investigation of sequences and structures. J. Mol. Biol. 247, 536–540.
Pabo, C.O., Sauer, R.T., Sturtevant, J.M., Ptashne, M., 1979. The k repressor contains two domains. Proc. Natl. Acad. Sci. USA 76, 1608–1612. Pabo, C.O., Lewis, M., 1982. The operator-binding domain of k repressor: structure and DNA recognition. Nature 298, 443–447. Peat, T.S., Frank, E.G., McDonald, J.P., Levine, A.S., Woodgate, R., Hendrickson, W.A., 1996. Structure of the UmuD0 protein and its regulation in response to DNA damage. Nature 380, 727–730. Ptashne, M., 1992. A genetic switch phage kand higher organisms. Cell Press & Blackwell, Cambridge, MA. Sauer, R.T., Yocum, R.R., Doolittle, R.F., Lewis, M., Pabo, C.O., 1982. Homology among DNA-binding proteins suggests use of a conserved super-secondary structure. Nature 298, 447–451. Sauer, R.T., Jordan, S.R., Pabo, C.O., 1990. k Repressor: a model system for understanding protein–DNA interactions and protein stability. Adv. Prot. Chem. 40, 1–61. Shenkin, P.S., Yarmush, D.L., Fine, R.M., Wang, H., Levinthal, C., 1987. Predicting antibody hypervariable loop conformation. Biopolymers 26, 2053–2085. Shimon, L.J.W., Harrison, S.C., 1993. The phage 434 OR 2/R1-69 resolution. J. Mol. Biol. 232, 826–838. complex at 2.5 A Slilaty, S.N., Rupley, J.A., Little, J.W., 1986. Intramolecular clevage of LexA and phage k repressors: Dependence of kinetics on repressor concentration, pH, temperature, and solvent. Biochemistry 25, 6866–6875. Slilaty, S.N., Little, J.W., 1987. Lysine-156 and serine-119 are required for LexA repressor cleavage: a possible mechanism. Proc. Natl. Acad. Sci. USA 84, 3987–3991. Strahs, D., Brenowitz, M., 1994. DNA conformational changes associated with the cooperative binding of cI-repressor of bacteriophage k to OR . J. Mol. Biol. 244, 494–510. Surewicz, W.K., Mantsch, H.H., Chapman, D., 1993. Determination of protein secondary structure by Fourier transform infrared spectroscopy: a critical assessment. Biochemistry 32, 389–394. Sussman, R., Alexander, H.B., 1989. Structural analysis of the carboxy terminus of bacteriophage lambda repressor deptermined by antipeptide antibodies. J. Bacteriol. 171, 1235–1244. Weiss, M.A., Karplus, M., Sauer, R.T., 1987. Quarternary structure and function in phage k repressor: 1 H- NMR studies of genetically altered proteins. J. Biomol. Struct. Dyn. 5, 539–556. Whipple, F.W., Kuldell, N.H., Cheatham, L.A., Hochschild, A., 1994. Specificity determinants for the interaction of k repressor and P22 repressor dimers. Genes Dev. 8, 1212–1223. Whipple, F.W., Hou, E.F., Hochschild, A., 1998. Amino-acid-amino acid contacts at the cooperativity interface of the bacteriophage k and P22 repressors. Genes Dev. 12, 2791–2802.