Crystal Structure of the Heterodimeric Complex of the Adaptor, ClpS, with the N-domain of the AAA (cid:1) Chaperone, ClpA*

Substrate selectivity and proteolytic activity for the E. coli ATP-dependent protease, ClpAP, is modulated by an adaptor protein, ClpS. ClpS binds to ClpA, the regulatory component of the ClpAP complex. We report the crystal structure of ClpS in complex with the isolated N-terminal domain of ClpA in two different crystal forms at 2.3- and 3.3-Å resolution. The ClpS structure forms an (cid:2) / (cid:3) -sandwich and is topologically analogous to the C-terminal domain of the ribosomal protein L7/L12. ClpS contacts two surfaces on the N-terminal domain in both crystal forms; the more extensive interface was shown to be favored in solution by protease protection experiments. The N-terminal 20 residues of ClpS are not visible in the crystal structures; the removal of the first 17 residues produces ClpS (cid:4) N, which binds to the ClpA N-domain but no longer inhibits ClpA activity. A zinc binding site involving two His and one Glu residue was identified crystallographically in the N-terminal domain of ClpA. In a model of ClpS bound to hexameric ClpA, ClpS is oriented with its N terminus directed toward the distal surface of ClpA, suggesting that the N-terminal region of ClpS may affect productive substrate interactions at the apical surface or substrate entry into the ClpA translocation channel. ClpA,

ClpA, a member of the Hsp100/Clp family of molecular chaperones, promotes the ATP-dependent degradation of proteins (1,2) by forming a complex with ClpP, a proteasome-like hollow tetradecamer (3) that degrades substrates into peptides of 5-15 amino acids (4). The crystal structure of ClpA reveals an N-terminal domain (N-domain) 1 and two tandem, nonidentical AAA ϩ modules (D1 and D2) (5). The AAA ϩ superfamily (ATPases associated with various cellular activities) comprises energy-dependent molecular machines that participate in assembly and disassembly of a wide variety of macromolecular complexes (6,7). AAA ϩ proteins are composed of modular functional domains (8): one or two AAA ϩ modules that use binding and hydrolysis of ATP to couple conformational changes in the AAA ϩ protein to structural changes in associated macromolecular substrates and additional N-or C-terminal domains that appear to interact with substrates and to transduce the motion of the AAA ϩ protein into movements of or conformational changes in the substrate.
AAA ϩ proteins also interact with adaptor proteins that mediate various functions. Escherichia coli ClpX needs RssB/SprE for specific degradation of RpoS (9,10) and uses SspB to enhance degradation of SsrA-tagged proteins (11). Bacillus subtilis MecA is needed for efficient degradation by ClpCP (12). In eukaryotes, the AAA ϩ chaperone, p97, uses several adaptor proteins: p47, to promote specific membrane fusion (13); Ufd1p, for ubiquitin-dependent proteolysis (14); and Npl4, for nuclear transport (14). Such adaptors allow the activity of the p97 to be directed toward different cellular targets. Recently, a small adaptor-like protein, ClpS, has been shown to modulate activity of E. coli ClpA (15). ClpS is a 106-residue protein encoded upstream of ClpA in most bacteria (16). ClpS protects ClpA from autodegradation and appears to redirect its activity away from soluble proteins and toward aggregated proteins (15). ClpS is thus an important regulatory factor for the activity of ClpA and ClpAP.
ClpA recognizes sequences near the N or C terminus of specific substrates (17,18), unfolds, and translocates them into the degradation chamber of an associated ClpP (18 -20). Substrates apparently bind to the apical surface of ClpA and follow an axial translocation pathway into the active site of ClpP (20). Additionally, in ClpA, there appears to be a substrate binding site within the small subdomain of the D2 AAA ϩ module (21), but the role of this site is not well understood. ClpA lacking the N-domain can interact with specific substrates and promote protein degradation nearly as efficiently as intact ClpA (22). Interestingly, ClpS cannot inhibit activity of ClpA lacking its N-domain, suggesting that ClpS interacts with the N-domain of ClpA (15). 2 Together, ClpS and the N-domain of ClpA play a modulating role in substrate selection and protein unfolding and degradation activities of ClpAP.
The N-domain of ClpA connects to the D1 domain by a long flexible loop (5) and is expected to be mobile based on electron microscopy and biochemical studies (20,23). We have found that the isolated N-domain forms a tight complex with ClpS, suggesting that ClpS might affect ClpA activity by altering N-domain orientation or interfering N-domain interaction with substrates. We further report the crystal structure of the heterodimeric complex of ClpS with the N-domain and present a model of ClpS in complex with intact hexameric ClpA. The model suggests that the N-domain helps orient ClpS, allowing it to block substrate binding to the surface of ClpA.

EXPERIMENTAL PROCEDURES
Expression, Purification, and Crystallization of the ClpS-N Complex-The clpS coding region was amplified from pWPC3.1 (16) using the forward primer, TGATAACTGCATATGGGTAAAACGAACGACTG-GC, and the back primer, TTATGCCTGCAGTCAGGCTTTTTCTAGC-GTACAC. The product was cut with NdeI and PstI and inserted into a derivative of the vector pBAD33. DH5␣ cells containing the plasmid, pBAD33-clpS, were grown in Luria broth containing chloramphenicol (34 g/liter), induced with 0.2% arabinose at an A 600 of 0.7, and collected by centrifugation after 3 h. After suspension in Bis-Tris buffer (20 mM, pH 6.5) containing 10% (v/v) glycerol, cells were broken at 20,000 p.s.i. in a French pressure cell. Cell debris was removed by centrifugation at 10,000 ϫ g for 30 min. The supernatant extract was treated with 0.05% polyethyleneimine, and precipitated material was removed by centrifugation. Ammonium sulfate (40% saturation) was added to the supernatant solution, and the precipitated protein was collected by centrifugation and dissolved in buffer containing 50 mM, Tris-HCl, pH 8.5, and 10% glycerol. After dialysis against the same buffer, the protein was further purified by chromatography on a MonoQ (Amersham Biosciences) and a hydroxyapatite (Bio-Rad) column and by gel filtration on a Superdex 75 column. Purified ClpS was concentrated to 12 mg/ml and dialyzed against 20 mM Hepes, pH 7.5. Aliquots were stored at Ϫ80°C. The production of the N-domain of ClpA followed the procedure previously described (5). ClpS and the N-domain of ClpA were mixed in an equal molar ratio and run over a Superdex 75 gel filtration column. A single peak corresponding to the ClpS-N complex was collected and concentrated to 10 mg/ml for crystallization.
Crystallization experiments with the ClpS-N complex resulted in several crystal forms. Crystals in the trigonal space group P3 1 21 were obtained by vapor diffusion after 1 week at 21°C in hanging drops of 2 l of protein solution mixed with 2 l of reservoir solution consisting of 0.1 M Bis-Tris, pH 6.5, 32% glycerol, and 10 -15 mM yttrium chloride. Crystals in the space group P4 3 2 1 2 were grown at 4°C from hanging drops of 2 l of protein solution and 2 l of well solution consisting of 0.1 M Bis-Tris, pH 6.5, 32% glycerol, and 2% polyethylene glycol 4000. Both forms of crystals can be frozen successfully without further manipulation and diffracted X-rays to 2.3-Å resolution for the P3 1 21 and 3.3-Å resolution for the P4 3 2 1 2 space group. Heavy-atom derivatives for the trigonal crystal form were prepared by soaking the crystals in the crystallization solution containing 1-10 mM heavy atom compound for 5-10 min.
Crystallographic Data Collection and Structure Determination-The native x-ray diffraction data set used in structure determination for the P3 1 21 space group was collected at the beamlines BioCARS-CAT, Advanced Photon Source at the Argonne National Laboratory, and X9B of the National Synchrotron Light Source at Brookhaven National Laboratory, using Area Detector System Co. quantum-4 CCD detectors. A data set at the wavelength corresponding to the peak of the EXAFS spectrum ( ϭ 1.0057 Å) for the mercury derivative was collected to 2.8-Å resolution at the beamline BioCARS of the Advanced Photon Source. Data reduction was performed with Denzo and Scalepack under the control of HKL2000 (24). Seven heavy atom positions were determined with SnB (25) and confirmed with ShelxD (26). Phases were calculated in SIR/AS runs of MLPhare (27), and the electron density was improved by solvent flattening and 2-fold molecular averaging in DM (28). Starting with the phases computed by DM, the program RESOLVE (29) produced excellent polyalanine models that were used to identify regions for the N-domain of ClpA and ClpS. The N-domain of ClpA was built by placing a previously determined model (5) into the density; sequence assignment of ClpS and model building were done in O (30). The models of two heterodimers were refined to 2.3-Å resolution using CNS (31). Throughout the refinement, appropriate noncrystallographic symmetry (NCS) restraints were maintained. Difference Fourier maps revealed the positions of four yttrium Bis-Tris moieties attached to acidic residues and two chloride ions. The tetragonal crystal form was solved by molecular replacement using a ClpS-N heterodimeric model built using the A interface (for details, see "Results and Discussion") in rotational and translational searches with MolRep (32). One solution was obtained and subsequently refined. Details of the structure determination and refinement statistics are listed in Table I.

RESULTS AND DISCUSSION
Complexes of ClpS with ClpA Hexamers and ClpA N-domain-Recombinant ClpS (106 amino acid residues) was purified from an overexpressing E. coli strain (see "Experimental Procedures") (Fig. 1A, lane 3). The protein produced a single band upon gel electrophoresis under nondenaturing conditions (data not shown) and had an approximate molecular mass of 12 kDa by gel filtration (Fig. 1B). ClpS is thus a monodispersed monomer in solution. In agreement with an earlier report (15), ClpS formed a complex with intact ClpA; upon Superdex 200 gel filtration in the presence of ATP␥S, both proteins were found in the 15-min fraction, corresponding to a molecular mass of 500 kDa (Fig. 1A, lane 2). In contrast, when a mutant ClpA lacking the N-domain, ClpA⌬153 (22), was used, much less ClpS was recovered in the high molecular weight fraction (data not shown), suggesting that the N-domain of ClpA is needed for tight binding to ClpS.
To assay ClpS binding to the N-domain, we purified the 143-amino acid residue N-terminal fragment of ClpA, which behaves as a monomer in solution ( Fig. 1, A (lane 4) and B). Isolated ClpA N-domain had previously been shown to block the inhibitory activity of ClpS on intact ClpA, suggesting that the N-domain of ClpA binds to ClpS (15). When N-domain was mixed with ClpS and run on a Superdex 75 gel filtration column, the two proteins emerged together in a single peak with a molecular mass of about 28 kDa (Fig. 1, A (lane 5) and B), providing direct evidence of a stoichiometric heterodimeric complex of ClpS and N-domain and suggesting that the major site of interaction with ClpS is through the N-domain of ClpA. We then generated a variant of ClpS (ClpS⌬N) by removing the N-terminal 17 amino acid residues with lysylendopeptidase C (Fig. 1A, lane 6). When ClpS⌬N was mixed with ClpA, both proteins eluted from the gel filtration column in the same fraction (Fig. 1A, lane 7). Thus, ClpS lacking its N-terminal 17 amino acid residues retains the ability to bind tightly to ClpA.
Structure Solution of the Heterodimer of ClpS and ClpA N-domain (ClpS-N)-The complex of ClpS with the N-domain of ClpA, which we will call ClpS-N, was crystallized, and two crystal forms were obtained under slightly different growth conditions (see "Experimental Procedures"). The trigonal form in space group P3 1 21 diffracted X-rays to 2.3-Å resolution and had cell dimensions of a ϭ b ϭ 88.0 Å, and c ϭ 210.2 Å. The tetragonal crystals in space group P4 3 2 1 2 diffracted to 3.3-Å resolution and possessed cell dimensions of a ϭ b ϭ 91.2 Å, and c ϭ 198.6 Å. Statistics for diffraction data sets, phase determination, and final atomic models are given in Table I. Both crystal forms are unusual in having very high solvent content; the trigonal crystal form has two molecules of ClpS-N heterodimer in the asymmetric unit with the Matthews' coefficient of 4.3 (33), corresponding to a solvent volume of 71%; the tetragonal form contains just one molecule of the heterodimer in the asymmetric unit, having a surprisingly high Matthews' In the P3 1 21 crystal, each N-domain is in contact with three molecules of ClpS using three different interfaces, whereas in P4 3 2 1 2, only two ClpS molecules interact with one N-domain, using two of the three binding interfaces observed in the P3 1 21 space group. By using the two contacts observed in both crystal forms (interfaces A and C; see below), ClpS and N-domain heterodimers produce continuous helical chains in crystals ( Fig. 2A). In the trigonal crystal, the two N-domains in the asymmetric unit are related by an NCS rotation of 167°, and the two ClpS molecules are related by an NCS rotation of 172°. The atomic model refined to 2.3-Å resolution in P3 1 21 includes 456 residues, four yttrium ions, two chloride ions, and 128 water molecules. The yttrium ions, required for crystallization, bind to the third interface between the N-domain and ClpS (interface B), which is not involved in helical chain extension of the ClpS-N heterodimers. The two chloride ions were assigned based on their positively charged environments, anomalous signals, and refined B factors.
Despite the large solvent content, the ClpS-N crystals are remarkably stable and well ordered. In the trigonal crystal, six sets of ClpS-N helical chains in six layers 60°apart in orientation and perpendicular to the c axis (three sets of two chains related by the NCS symmetry) form an interconnected network. In the tetragonal crystal, four sets of ClpS-N helical chains in four layers 90°apart in orientation stack together along the 4 3 axis. These interactions provide sufficient strength and support for the crystals to accommodate their extraordinarily large solvent content.

Structures of ClpS and the N-domain of ClpA-
In the trigonal crystal, 87 out of 106 residues in ClpS can be seen in the two NCS-related heterodimers; the missing residues are from the N terminus. ClpS has a single ␣/␤ sandwich folding motif consisting of three ␣ helices (H1, H2, and H3) on one side of a three-stranded anti-parallel ␤-sheet (S1, S2, and S3) (Fig. 2, A  and B). The ␤-stand S2 contains a conserved ␤-bulge of type C, subdividing it into strands S2a and S2b (Fig. 2, B and C) (34). The molecule is cone-shaped, with the helices H1 and H2 lying flat at the base of the cone and the twisted ␤-sheet and helix H3 converging at the tip of the cone. The N-terminal residues from 21 to 26 extend out of the top of the cone, but residues 1-20 were not all visible. Secondary structure assignment of ClpS to its sequence is provided in Fig. 2B. Sequence alignment of selected ClpS homologs from prokaryotic and eukaryotic sources (Fig. 2B) shows a highly conserved N-terminal region (S1 and H1), with most conserved residues forming the hydrophobic core. The conserved hydrophilic residues are shown as the ball-and-stick models ( Fig. 2A) and are clustered at two locations that form close contacts with the N-domain (see below).
A search of the protein structure data base (35) identified only the ribosomal protein, L7/L12, with significant structural similarity to ClpS. The C-terminal domain of L7/L12 (36,37) overlaps with ClpS with an r.m.s. deviation of 1.7 Å for 63 pairs of C-␣ atoms out of 68 residues in the structure (Fig. 2C). Structure-based sequence alignment of L7 to ClpS (Fig. 2B) shows a 13% sequence identity or 22% similarity between the two sequences, with most of the conservation concentrated at the N terminus of the sequence in S1 and H1. The ␤-bulge in ClpS between strands S2a and S2b is also present in the structure of L7/L12 (Fig. 2C). There are nine pairs of charged residues in the aligned structures; in six of them, the charges are inverted (Fig. 2C). The L7/L12 ribosomal protein has two domains: an N-terminal dimerization domain and a C-terminal where ͉F PH ͉ is the structure-factor amplitude for the heavy atom derivative, ͉F P ͉ is the structure-factor amplitude for the native crystal, and ͉f H ͉ is the amplitude of calculated structure-factor contribution of heavy atoms in the derivative. The summation is overall centric or acentric reflections. The N-domain is a pseudodimer, shown with the N-terminal half in yellow and the C-terminal half in blue. The location of the Zn 2ϩ binding site is as labeled. In the ClpS ␣/␤ sandwich, the ␤-sheet is in red, and the ␣-helices are in green. Secondary structure elements as well as conserved domain, connected by a flexible loop. The influence of the C-terminal domain of L7/L12 on the elongation factor-dependent GTP hydrolysis (38,39) suggests that it might physically interact with these proteins. Indeed, in cryoelectron microscopic images, the stalk region (i.e. presumably L7/L12) can be seen in contact with a ternary EF-TU complex (40). The significant similarity in structure of ClpS to the L7/L12 may imply analogous functions of ClpS in associating with and affecting ATPase activity of ClpA. In this context, we have found that whereas ClpS has minimal effect on the ATPase activity of native ClpA, it leads to a 2-fold increase in ATPase activity of a truncated mutant consisting of the N-terminal and D1 AAA ϩ domains. 2 The structure of the ClpA N-domain consists of eight ␣-helices ( Fig. 2A). The first four-helix bundle (N-H1 to N-H4) is related to the second four-helix bundle (N-H1Ј to N-H4Ј) by a pseudo-2-fold axis and is connected by a long acidic loop (positions 66 -79) that is disordered in the full-length ClpA structure (5) and partially stabilized upon binding of ClpS. Sequence conservation between the two halves of the pseudodimer is poor except in the hydrophobic core (41) and in a few conserved hydrophilic residues distributed at both ends of helix bundles (Fig. 2B). The N-domain is apparently not related to any known protein structure in the protein structure data base (35). The details of the N-domain structure, in isolation and as part of the full-length ClpA subunit, and of the tandem repeat in the N terminus have been reported recently (5). There is a piece of isolated electron density identified on the surface hydrophobic depression between the N-domain pseudorepeating units. A tentative peptide model of the N terminus of ClpS was built (data not shown), suggesting a potential peptide binding site for the N-domain.
Heterodimeric Association of ClpS and N-domain-In the trigonal crystal, an N-domain has three ClpS neighbors, related by either noncrystallographic or crystallographic symmetry. We designate the three contacting interfaces as A, B, and C. Buried surface areas (42) upon association for A, B, and C were 1,639, 955, and 824 Å 2 , respectively. Contact B involves helices H1 and H4 of the N-domain, part of which have been shown to interact with the first ATPase domain (D1) in fulllength ClpA (5). Contact B is mediated by an yttrium ion, calling into question its significance as a physiological dimer interface. The yttrium ion is tightly capped by a Bis-Tris molecule and neutralizes three negative charges from glutamate residues of ClpS or the N-domain. The yttrium ion being coordinated by nine atoms from three ligands, it seems unlikely that the metal ions more commonly found in a biological systems would occupy this site and stabilize the B interface. In an attempt to differentiate between contacts A and C, we solved the structure of a different crystal form of the ClpS-N heterodimer (space group P4 3 2 1 2) at 3.3-Å resolution by molecular replacement (Table I). However, we found that the tetragonal crystal uses only contacts A and C for ClpS and N-domain interaction as shown in Fig. 2A. Whereas this observation further ruled out the B interface as a physiological contact, it did not help to establish either A or C as the true ClpS-N dimer.
In contact A, both halves of the N-domain pseudodimer contribute to the interface. Binding involves residues from the ends of N-domain helices N-H1, N-H2, N-H1Ј, N-H2Ј, and N-H3Ј, as well as the acidic loop between N-H4 and N-H1Ј; the contribution of ClpS to the interface comes from residues in S-H1, S-H3, S-S1, and S-S2b (Fig. 3A). There are 11 close interactions involving hydrogen bonds and salt bridges, three of which are mediated through structural water molecules contacting the main chain atoms (Table II). Five conserved hydrophilic residues in ClpS and the N-domain, N-Arg 28 , N-Thr 81 , S-Lys 84 , S-Glu 79 , and S-Tyr 28 , participate in the interaction. The N-terminal chain of ClpS extends along the N-domain parallel to helix H3Ј until the density disappears; this configuration would probably place the missing residues near the C terminus of the N-domain and away from the acidic loop connecting the two halves ( Fig. 2A).
In contact C, ClpS is bound to the N-domain via a relatively small surface area involving residues from helices N-H1Ј, N-H2Ј, and N-H4Ј of the second half of the N-domain and from the base of the ClpS cone (Fig. 3B). There are six hydrogenbonding and salt-bridging interactions, involving three conserved residues: N-Asn 107 , N-Arg 100 , and S-Asp 36 . The four helices in the C-terminal half of the N-domain form a cleft into which S-Asp 36 and S-Tyr 37 , both conserved within prokaryotes, penetrate to interact with N-Asn 107 and N-Arg 100 , respectively (Table II). The N terminus of ClpS in this configuration extends in the opposite direction of the N-domain ( Fig. 2A).
To decide which interface is more likely to be the one dominant under physiological conditions, we generated models of the three crystallographically observed S-N-S trimers as in Fig.  2A; two of the models are NCS-related and come from the trigonal crystal form, and one comes from the tetragonal crystal form. The three trimers are superimposed on their N-domains as shown in Fig. 3C. The three aligned N-domains give rise to an r.m.s. deviation of 0.44 Å for 142 aligned C-␣ atoms. The three ClpS molecules that bind to the N-domain via interface A have a slightly larger r.m.s. deviation of 0.56 Å for 86 C-␣ atoms, whereas those binding to the N-domain through the interface C show an r.m.s. deviation of 1.65 Å. The data demonstrate that the ClpS binding surface A provides a more constant and apparently more stable contact than C in the different crystal environments. Therefore, surface A appears to be more favored as the dimer interface under physiological conditions.
To identify the contact that predominates in solution, we probed exposed surfaces of the proteins by limited proteolysis. The ClpS, the N-domain of ClpA, or a mixture of the two in a 1:1 molar ratio was incubated with chymotrypsin for various times, cleavage products were separated on a SDS gel, and N-terminal sequences of the products were determined (Fig. 4). When incubated alone, ClpA-N was cut at Phe 14 , Tyr 122 , and Phe 137 , but when ClpS was present, the Tyr 122 site was protected from digestion, suggesting that ClpS blocks access to Tyr 122 . In the N-domain structure, Tyr 122 is exposed on the surface, but in the complex Tyr 122 is shielded by residues 22-24 of ClpS making the A contact. Interestingly, Phe 14 was also protected by ClpS. This residue was part of the yttrium-stabilized interface B discussed above, suggesting that the B interface may form transiently in solution. Phe 137 was not protected and is thus not involved in contact with ClpS. Analysis of the hydrophilic residues are labeled accordingly. The diagram is produced with Molscript (48), Bobscript (49), and Povray (available on the World Wide Web at www.povray.com) interfaced with GL-render (L. Esser, unpublished work). B, structure-based sequence alignment of ClpS homologs. Selected sequences of ClpS homologs from prokaryotes and eukaryotes are shown. The alignment and secondary structure elements assignment to the sequence are based on the structures of ClpS determined here and E. coli ribosomal protein L7/L12 (Protein Data Bank entry 1CTF (36)). The ␤-strands, represented as green arrows, are labeled S1, S2, and S3 sequentially as they appear in the sequence; strand S2 is subdivided into S2a and S2b. The ␣-helices, shown as wiggling curves, are labeled H1, H2, and H3. Coils are shown as straight line segments. C, superposition of the structures of ClpS and E. coli ribosomal protein L7/L12. ClpS is shown in red, and L7/L12 is shown in green. Residues in which there is a charge inversion are shown as stick models and are labeled.

FIG. 3. Interfaces between the ClpS and the N-domain.
The secondary structure elements that contribute residues to subunit interactions are labeled. Residues that form hydrogen bonds and salt bridge pairs are shown in stick models. Dashed lines are drawn between atom pairs less than 3 Å apart. Water molecules are shown as gray balls. A, stereo pair showing hydrogen bonds and salt bridges between residues from ClpS (green) and the N-domain (yellow) at interface A. B, stereo pair showing hydrogen bonds and salt bridges between residues from the ClpS (dark blue) and the N-domain (yellow) at interface C. C, ClpS bound at interface C is conformationally more flexible than that bound at interface A. Three independent N-domains (two NCS-related in P3 1 21 and one in P4 3 2 1 2) with attached ClpS molecules at both interface A and C were superimposed. The ClpS-N-domain/ClpS molecules in red/ purple/red are from the trigonal crystal; the green/light green/green molecules are NCS-related to the red/purple/red in the trigonal crystal, and the blue/cyan/blue molecules are from the tetragonal crystal. The A and C interfaces are labeled.
ClpS cleavage pattern gave reciprocal results. ClpS was cut at Tyr 28 in the absence of the N-domain and was protected in the complex (Fig. 4). Tyr 28 is involved in contact A with the N-domain. These data further support the model that the A interface is used for the ClpS-N dimer in solution and probably under physiological conditions. We cannot rule out that the B and C interfaces could also represent modes of interaction between ClpS and ClpA under some circumstances. Alternatively, the surfaces of ClpS or N-domain involved in these interfaces might have other binding functions in solution. It is not uncommon for substrate-binding sites and protein-protein interfaces to provide sites for crystal lattice contacts (43). In fact, in our structure of the full-length ClpA (5), the N-D1 interface employs the same region of the N-domain that makes interface B in the trigonal crystal form of ClpS-N. Also, the limited but rather precise fit producing contact C in the ClpS-N crystal may mimic another substrate or adaptor binding site that specifically recognizes an exposed Asp-Tyr sequence motif.
The N-domain of ClpA Contains a Zn 2ϩ Binding Site-A weak but consistent zinc signal measured by x-ray fluorescent spectra of ClpS-N crystal (data not shown) prompted an investigation of potential zinc binding property of the ClpS-N complex. A crystal of ClpS-N grown in the presence of 2 mM of ZnCl 2 diffracted X-rays to 2.25-Å resolution using a home x-ray source and was isomorphic to the trigonal crystal form. The structure was refined to an R free of 0.236 and R work of 0.214, respectively (Table I). The Fourier synthesis using the coefficients of ͉F o ͉ Ϫ ͉F c ͉ located a 14 peak in the N-domain, and the Fourier synthesis using the coefficients of ͉F ϩ ͉ Ϫ ͉F Ϫ ͉ produced a 7.5 anomalous peak overlapping the same site. A zinc ion was introduced into the ClpS-N model and was placed in the N-terminal half of the N-domain in a depression between N-H1 and N-H4 near the acidic loop connecting the two halves of the N-domain ( Fig. 2A). The Zn 2ϩ ion is tetrahedrally coordinated by three residues and a water molecule; His 20 , His 22 , and Glu 63 are 2.26, 2.15, and 2.14 Å, respectively, from the Zn 2ϩ , and the water molecule is 2.14 Å away (Fig. 5). When superimposed, the N-domains with and without Zn 2ϩ produced an r.m.s. deviation of 0.242 Å for 142 C-␣ pairs. Superposition of the N-terminal halves of the N-domain with and without Zn 2ϩ yielded an even smaller r.m.s. deviation of 0.146 Å for 48 C-␣ atom pairs. Large conformational changes are observed for the Zn 2ϩ ligands; the imidazole ring of His 22 is rotated nearly 120°, and the side chain of Glu 63 is moved more than 3 Å (Fig. 5).
The Zn 2ϩ coordination and binding residues in the N-domain is highly reminiscent of the Zn 2ϩ binding sites observed in carboxypeptidase A (44), thermolysin (45), mitochondrial processing peptidase (46), and other zinc-metalloproteases, although preliminary assays of peptidase activities in the presence of added Zn 2ϩ showed no evidence for activity, and there is lack of secondary structure resemblance. Although it is unlikely for ClpA to have an uncontrolled peptidase activity, we cannot rule out the possibility that specific cleavage may occur when the N-domain is attached to the D1 domain. Indeed, in the full-length ClpA, the Zn 2ϩ binding site is facing the D1 small domain, forming a cleft that is highly positively charged.   (Fig. 6). At the A contact, the helix N-H2, the small turn between H2Ј and H3Ј, and the acidic loop connecting the two halves of the N-domain undergo large displacement. Residues whose side chains display large motions are shown in Fig. 6. At the C contact, there is a large conformational change involving helices N-H1Ј and N-H4Ј that form the cleft where the S-Tyr 37 inserts; both helices are pushed apart to accommodate the intrusion of S-Tyr 37 . Residues Ser 93 and Gly 99 at the tip of the cleft between N-H3 and N-H4 are displaced by as much as 3 Å.
Models of Hexameric ClpAS Complex-In our recently reported crystal structure of the full-length ClpA subunit (5), the N-domain is attached to the D1 AAA ϩ module, making major contacts with the D1 small subdomain. In the hexamer model, the N-domains are located on the periphery of the D1 hexameric ring with the acidic loop between the halves of the pseudodimer projecting toward D2. To model the ClpS bound to the ClpA hexamer, we superimposed the N-domain of ClpS-N with ClpS attached at two possible interfaces (the A or the C interface) onto the N-domain of ClpA (Fig. 7). We refer to the ClpAS complex using interface A as the A model and that using interface C as the C model. In the A model, the ClpS wedges comfortably into an empty space between the D1 and D2 domain, making contacts with the transition helix connecting D1 and D2 and with the D2 small domain. The N-terminal extension of ClpS would protrude toward the apical surface of D1, making it possible that the missing N terminus of ClpS might reach the edge of the apical surface of the ClpA hexamer. As mentioned earlier, a peptide segment (8 residues) bound to the N-domain with tentative sequence assignment of the disordered N terminus of ClpS is in a position to make a connection to the ClpS in the ClpAS model, which would allow the N terminus of ClpS to extend 20 -30 Å out from the tip of the conical domain. In the C model, ClpS is bound to the N-domain projecting outward from the apical surface of the D1 ring (Fig.  7). With the N terminus protruding from ClpS, this configuration gives the complex the appearance of a jellyfish with six tentacles extending out from the substrate binding surface. It is conceivable that both the A and C models may be present when concentrations of ClpS reach a given threshold. At low ClpS concentrations, the A model would be expected to be the dominant species.
Substrate binding sites in ClpA, ClpX, and related proteins have been demonstrated at the apical surface of the D1 domain (20) and within the sensor and substrate discrimination (SSD) domains, located in ClpA D2 and ClpX D1 (21). The six D1 domains in the apical ring of ClpA enclose a cavity visible in FIG. 5. The Zn 2؉ binding site and its coordination in the Ndomain of ClpA. A stereo pair shows the Zn 2ϩ in tetrahedral coordination with four ligands in the ball-and-stick models and as labeled. The same four ligands in different conformation from the Zn 2ϩ -free N-domain are also shown as thin stick models. The helix H4 is shown as a ribbon in red, and the loop between helices H1 and H2 is shown as a blue coil.  both the electron microscopic reconstructions (23) and in the hexameric ring model based on the crystal structure of the ClpA subunit (5). Both N-terminally and C-terminally tagged substrates bind to the apical surface and are translocated along an axial pathway through the D1 cavity and eventually into ClpP (20,47). SSD domains of different substrate recognition components of ATP-dependent proteases have preferences for different substrates. For example, the SSD of ClpA binds favorably to the heat shock transcription factor, 32 , whereas that of ClpX is more specific for SsrA-tagged proteins (21). It is possible that both types of substrate binding sites play a role in the catalytic cycle of ClpA.
The ClpA SSD domain (21) corresponds to residues 600 -758, encompassing the sensor 2 motif and the small domain of D2 in the recently determined ClpA full-length structure (5). The small subdomain has an ␣/␤ sandwich topology with three helices and a mixed three-stranded ␤-sheet. The contacts between the small and large subdomains of the AAA ϩ module are affected by the nucleotide state. Thus, the SSD domain offers a site for substrate-dependent regulation of activity and might provide a means of communication with the apical surface, where bound substrate undergoes unfolding and translocation. The binding of ClpS to ClpA via the interface A places ClpS in contact with D2 and in a position to affect either substrate binding or changes in D2 induced upon substrate binding. However, this interaction is not sufficient to explain the inhibitory effect of ClpS. When the N-terminal 17 amino acids of ClpS were removed by lysylendopeptidase C treatment, the resulting ClpS⌬N could still bind to ClpA and the isolated N-domain of ClpA (Fig. 1, A and B); however, ClpS⌬N was no longer able to inhibit proteolytic activity of ClpAP. Table III shows that both casein and green fluorescent protein-SsrA degradation were unaffected by ClpS⌬N. Since ClpS⌬N contains the part predicted to interact with the D2 domain of ClpA, the inhibitory effect of the N-terminal extension of ClpS might be exerted elsewhere on ClpA.
Where is the N-terminal extension of ClpS when it is bound to ClpA hexamers? In the model of ClpS bound to ClpA hexamers using the favored A interface, the N terminus of ClpS would project out from the apical surface of the ring but be in close proximity to the surface. In this position, the ClpS N-terminal region could either sterically block the approach of protein substrates to the apical surface of ClpA or could actually compete with substrates by binding to substrate recognition sites on the apical surface. Interestingly, in the model with ClpS bound by the C interface, the ClpS N terminus would also be in a position to sterically block substrate access, but the N terminus of ClpS would be far from the apical surface and probably would not be able to make contact.
In addition to the crystallographic and protease protection data, another factor favoring the A interface model is the ability of ClpS to protect ClpAP from autodegradation (15). Although the mechanism by which ClpS exerts its protective effect is not known, ClpS in the A model makes much more extensive contact with ClpA and is in a better position to protect large areas of the most sensitive regions of ClpA, whereas the ClpS in the C model is not in an obvious position to make a difference in ClpA stability. Taking the A interface as the one favored in solution, we can propose a model in which the N terminus of ClpS blocks substrate binding by competing with the substrate protein for binding to sites on the apical surface of ClpA. Since ClpS blocks degradation of soluble substrates but has apparent activating effects on degradation of aggregated proteins, we speculate that the N terminus of ClpS may block some but not all of the substrate binding sites on ClpA. By occupying substrate-binding sites on the apical surface of ClpA, ClpS would prevent binding of soluble substrates with single recognition sites and would thus impede their unfolding and degradation. At the same time, the blocking of some but not all of the sites could have an activating effect with aggregated protein substrates, which are expected to have many exposed sites for interaction with ClpA. So far, none of the models of ClpS action takes into account the probable mobility of the N-domain of ClpA itself, which would lead to significant changes in the position of bound ClpS and in its ability to affect substrate access to ClpA or binding to the apical surface. In the future, we hope to obtain crystals of ClpS bound to hexameric ClpA in different nucleotide states that should shed further light on the mechanism of action of this adaptor protein. ClpAP proteolytic activity was measured with 0.8 g/ml ClpA and 20 g/ml ClpP in standard assay buffer (see "Experimental Procedures"). Activity in the absence of inhibitor was 25 g of casein and 3 g of GFP-SsrA/h/g of ClpA. ClpS or ClpS⌬N was added at 0.5 M to casein degradation assays and 3 M to GFP-SsrA assays. ClpS⌬N is missing 17 N-terminal amino acids.