Crystal Structure of the Cytomegalovirus DNA Polymerase Subunit UL44 in Complex with the C Terminus from the Catalytic Subunit

The human cytomegalovirus DNA polymerase is composed of a catalytic subunit, UL54, and an accessory protein, UL44, which has a structural fold similar to that of other processivity factors, including herpes simplex virus UL42 and homotrimeric sliding clamps such as proliferating cell nuclear antigen. Several specific residues in the C-terminal region of UL54 and in the “connector loop” of UL44 are required for the association of these proteins. Here, we describe the crystal structure of residues 1-290 of UL44 in complex with a peptide from the extreme C terminus of UL54, which explains this interaction at a molecular level. The UL54 peptide binds to structural elements similar to those used by UL42 and the sliding clamps to associate with their respective binding partners. However, the details of the interaction differ from those of other processivity factor-peptide complexes. Crucial residues include a three-residue hydrophobic “plug” from the UL54 peptide and Ile135 of UL44, which forms a critical intramolecular hydrophobic anchor for interactions between the connector loop and the peptide. As was the case for the unliganded UL44 structure, the UL44-peptide complex forms a head-to-head dimer that could potentially form a C-shaped clamp on DNA. However, the peptide-bound structure displays subtle differences in the relative orientation of the two subdomains of the protein, resulting in a more open clamp, which we predicted would affect its association with DNA. Indeed, filter binding assays revealed that peptide-bound UL44 binds DNA with higher affinity. Thus, interaction with the catalytic subunit appears to affect both the structure and function of UL44.

The replication of DNA requires a number of multiprotein assemblies, including a DNA polymerase that synthesizes tens of thousands of nucleotides without dissociating from the primer-template. Most replicative DNA polymerases require not only catalytic subunits, but also accessory subunits known as processivity factors to remain tethered to the template during replication. The well characterized processivity factors known as sliding clamps, which include proliferating cell nuclear antigen (PCNA) 5 from eukaryotes and archaebacteria, ␤-subunits from prokaryotes, and gp45 from T4 and RB69 bacteriophage, display no intrinsic affinity for DNA. They tether their corresponding catalytic subunits to DNA after they are loaded onto DNA as toroidal homomultimers via clamp loader complexes in ATP-dependent processes (reviewed in Ref. 1).
Although not yet rigorously proven by template challenge experiments, UL44 is believed to serve as the processivity factor for the polymerase.
The crystal structure of residues 1-290 of UL44 (UL44⌬C290) (15) showed that UL44 has a fold remarkably similar to that of other processivity factors, including HSV UL42 (16) and monomers of the sliding clamps such as PCNA (17,18), even though these proteins have no obvious sequence homology. Thus, each subunit of UL44, UL42, and PCNA consists of two topologically similar domains. The two domains share a central ␤-sheet and are connected by a long connector loop running lengthwise across the front face of the molecule.
The extreme C terminus of the HSV catalytic subunit, UL30, binds to the connector loop of UL42 in the crystal structure (16). As is the case with HSV UL42 (16,28), residues in the connector loop of UL44 are required for interaction with its cognate catalytic subunit (19). Moreover, as is the case with HSV UL30, the C terminus of UL54 is both necessary and sufficient for binding to UL44 (12). This region displays some ␣-helical propensity in solution (29), although this tendency is less than that of corresponding peptides from the C terminus of HSV UL30 (30). However, unlike the HSV UL42-UL30 interaction, which is dominated by polar contacts (16,31), the strength of the UL44-UL54 interaction appears most dependent upon several specific hydrophobic residues (12,19). The strength and specificity of the interaction between UL44 and a peptide corresponding to the C-terminal 22 residues of UL54 suggested that these two polypeptides might be crystallized as a complex. We have solved the structure of the complex of UL44⌬C290 with this C-terminal peptide from UL54. The structure provides an explanation for previous biochemical and molecular genetic studies (12,19) that investigated the physical and functional association of the two subunits. The UL44-UL54 structure differs from other "processivity fold" structures in complex with their cognate binding partners. Like the free protein, the UL44-peptide complex forms a dimeric C-shaped clamp. However, the UL44 peptide structure has a more "open" conformation, which raised the possibility that this difference could affect DNA binding. This possibility has been confirmed experimentally, suggesting that interaction with UL54 affects both the structure and function of UL44.

MATERIALS AND METHODS
Protein Purification and Peptide Synthesis-Construction of the pD15-UL44⌬C290wt (where wt is wild-type) and pD15-UL44⌬C290(I135A) plasmids, which express residues 1-290 of UL44 as glutathione S-transferase fusion proteins, as well as the expression and purification of wild-type UL44⌬C290, UL44⌬C290(I135A), and selenomethionyl-UL44⌬C290 protein samples were described previously (12,15,19). Peptides corresponding to the C-terminal 22 residues of HCMV UL54 and the C-terminal 36 residues of HSV UL30 were synthesized and purified in the Biopolymers Laboratory of the Department of Biological Chemistry and Molecular Pharmacology and resuspended in water as described (12).
Crystallization and Structure Solution-Selenomethionyl-UL44⌬C290 was concentrated to ϳ125 M in 20 mM Tris-HCl (pH 7.5), 500 mM sodium chloride, 20% glycerol, 0.1 mM EDTA, and 2 mM dithiothreitol; stored at Ϫ80°C; and thawed from frozen stocks as described (15). One l each of UL44⌬C290, UL54 peptide (ϳ300 M in water), and well solution were combined and crystallized by vapor diffusion at 22°C in hanging drops. Similar orthorhombic crystal forms were found using a well solution composed of either 2 M ammonium sulfate, 100 mM phosphate-citrate (pH 4.2), and 10 mM dithiothreitol or 2.5 M sodium chloride, 100 mM sodium acetate (pH 4.5), 200 mM lithium sulfate, and 10 mM dithiothreitol. For structure determination, the well solution containing ammonium sulfate was used. Crystals reached ϳ250 ϫ 250 ϫ 250 m in 5-7 days. The drop solution was supplemented with a cryosolution of 40% ethylene glycol plus well solution until the final concentration reached ϳ20 -25% ethylene glycol. Crystals were flash-frozen in liquid nitrogen.
Experimental phases were obtained from a multiwavelength anomalous dispersion experiment on a single crystal of selenomethionyl-UL44⌬C290 complexed with the C-terminal UL54 peptide using a fourwavelength data set collected at beamline 19-ID at the Advanced Photon Source (Argonne, IL) (see Table 1). Images were processed with DENZO and merged with SCALEPACK (32). Initial phases were determined using SOLVE (33), which located four of five possible selenium sites and reported a figure of merit of 0.65. Density modification was performed with RESOLVE (34), resulting in readily interpretable electron density maps. Atomic models were built using XtalView (35) and refined by the maximum likelihood method with REFMAC Version 5.1 (36), maintaining experimental phases as restraints throughout refinement. The anisotropic motion of the protein was modeled using TLS (translation/libration/screw motion) refinement in REFMAC. Figures were produced using PyMOL (available at www.pymol.org) or Mol-Script (38) with Raster3D (39).
Filter Binding Assay-Filter binding assays of DNA binding were performed as described (15) with minor modifications. Briefly, increasing amounts of wild-type UL44⌬C290 or UL44⌬C290(I135A) were incubated with 1 fmol of radiolabeled double-stranded 30-bp DNA in the absence or presence of 50 M peptide corresponding to the C-terminal 22 residues of HCMV UL54 or the C-terminal 36 residues of HSV UL30.

RESULTS AND DISCUSSION
Structure Determination-The complex of UL44⌬C290 and a 22-residue peptide corresponding to the C terminus of UL54 (residues 1221-1242) was crystallized in space group C222 1 with unit cell dimensions of a ϭ 91.8, b ϭ 127.6, and c ϭ 66.0 Å and with one molecule/asymmetric unit. The structure of the complex was determined by multiwavelength anomalous dispersion of one selenomethionyl crystal using SOLVE (33) and was subsequently refined to 2.5-Å resolution ( Table 1). Three regions of the UL44⌬C290 molecule that are disordered in the uncomplexed structure (15) remain disordered in the complex. These regions are the first eight residues of the N terminus, the last 19 residues of the C terminus, and an internal loop composed of residues 163-176. All of the 22-residue UL54 peptide could be positioned unambiguously in the electron density maps, except for the first two residues of the N terminus. The atomic model is reported with an R cryst and R free of 0.195 and 0.227, respectively.
General Features of the Structure-As observed in the unliganded structure (15), UL44⌬C290 adopts a fold similar to those of UL42 (16) and PCNA (17,18) when bound to the peptide (Fig. 1). The UL44⌬C290 structure is composed of two topologically similar domains (an N-terminal domain, residues 9 -128; and a C-terminal domain, residues 143-271) linked covalently by an interdomain connector loop (residues 129 -142). The ␤-strands at the edge of each domain are hydrogen-bonded with one another to create a central, nine-stranded ␤-sheet. Each UL44 molecule can be defined to have a "front" and "back" face. The connector loop crosses the front face of the central ␤-sheet, creating a potential binding site for the peptide on each side of the loop. The back face of the central ␤-sheet of UL44⌬C290 is decorated by four helices that include several basic residues, previously hypothesized to be involved in binding to the sugar phosphate backbone of DNA (15). The N-and C-terminal domains each have an additional four-or five-stranded ␤-sheet at the distal end of the molecule that lies roughly perpendicular to the large central ␤-sheet and that participates in forming multimers of UL44 (see below). The UL54 peptide begins with an extended conformation of its N-terminal six residues before forming a short helix (residues 1228 -1233), followed by a ␤-strand (residues 1235-1237) and a single helical turn (residues 1239 -1241) (Fig. 1). (Residue numbers from 1221 to 1242 correspond to the UL54 peptide, whereas all other numbers correspond to UL44.) Thus, this UL54 peptide is less ␣-helical than the analogous peptide derived from the HSV catalytic subunit, UL30 (16). This observation is consistent with circular dichroism studies demonstrating that the C-terminal peptide of HCMV UL54 is less ␣-helical in solution than similar peptides from HSV UL30 (29,30). The UL54 peptide makes several interactions with regions on the front face of UL44, which are detailed below. Altogether, peptide binding by UL44 buries ϳ1400 Å 2 of solvent-accessible surface and exhibits a shape complementarity of 0.75 (the shape complementarity statistic as defined by Lawrence and Colman (40)).
Molecular Details of the UL44-UL54 Interaction-The UL54 peptide makes two major sets of interactions with UL44. One involves a hydrogen-bonding network between the middle portion of the peptide (resi-dues 1234 -1238) and, lying along its "right" edge, the central part of the connector loop (residues 133-137) ( Fig. 2A). This includes four main chain-to-main chain hydrogen bonds, which participate in an antiparallel ␤-sheet between the connector loop and the peptide. Additional contacts with the peptide come from UL44 side chains. Thus, the ⑀-oxygen and ⑀-nitrogen of Gln 133 form hydrogen bonds with the main chain nitrogen and oxygen, respectively, of Leu 1225 . Additionally, the side chain of Asp 134 forms a salt bridge with Lys 1237 of the peptide and coordinates two water molecules with the main chain of the peptide at Leu 1225 and Leu 1227 (data not shown). The central portion of the connector loop is further stabilized by three intramolecular hydrogen bonds between the side chains of Gln 51 and Lys 60 , which extend upward from the central ␤-sheet and bind to the main chain of the connector loop (thus extending the ␤-sheet hydrogen-bonding pattern along its right edge) at Asp 134 and Val 136 , respectively ( Fig. 2A). Although the side chain of Ile 135 forms some van der Waals interactions with the peptide, including the sulfur of Cys 1241 and C-␤ of Ala 1238 , these interactions are less extensive than the hydrophobic interactions that Ile 135 makes with other side chains of UL44. Thus, the side chain of Ile 135 is positioned as a hydrophobic "anchor" underneath the hydrogen-bonding network, where it makes favorable van der Waals contacts with several important residues from the central ␤-sheet of UL44, including Val 41 , Ile 49 , and the aliphatic portions of Gln 51 and Lys 60 .
The second set of interactions involves the packing of a hydrophobic "plug" into a crevice along the central ␤-sheet of UL44 (Fig. 2, B and C). This crevice is formed principally by side chains of several residues from the central ␤-sheet and is located just to the right of the connector loop at the boundary between the N-and C-terminal domains. The hydrophobic plug of the UL54 peptide is composed of three side chains (Leu 1227 , Phe 1231 , and Tyr 1234 ). All three side chains extend toward the surface of UL44 from a segment of the main chain (residues 1228 -1234) that assumes a roughly helical conformation, despite the presence of proline residues at positions 1229 and 1233. Leu 1227 and Phe 1231 pack into pockets formed by Val 136 of the UL44 connector loop and several residues from the central ␤-sheet, including Val 53 , Val 58 , Leu 251 , and Phe 266 , as well as the aliphatic portions of Thr 268 and Lys 60 . Tyr 1234 of where R free represents 5% of the data selected randomly. d Values represent the percentage of residues in the most favored, additionally allowed, generously allowed, and disallowed regions, respectively, of the Ramachandran plot as determined by PROCHECK (37).

Relation of the Observed Interactions in the UL44 Peptide Structure to the Results of Molecular Genetic and Biochemical Experiments-Prior
to this study, experiments were initiated to identify which residues are crucial for the interaction between the two HCMV DNA polymerase subunits (12,19). Based on analogous studies of HSV DNA polymerase subunits (16,28,31), a series of mutant constructs were engineered with individual alanine substitutions in residues 129 -140 of the HCMV UL44 connector loop (19) or in the C-terminal 22 residues of HCMV UL54 (12). Each mutation was tested for its effect on the association of UL44 and UL54 and on the ability of UL44 to stimulate long chain DNA synthesis by UL54. Several UL44 mutants were also tested for their affinity for the peptide ligand that is present in the crystal structure, which corresponds to the C terminus of UL54.
In the connector loop, only alanine substitutions at UL44 residues 133-136 reduce the physical and functional interactions of UL44 with full-length UL54 and with its C terminus (19). Of these, only one substitution (I135A) completely disrupts this interaction. As noted above, the bulky side chain of Ile 135 makes extensive intramolecular interactions to form a hydrophobic anchor beneath the connector loop, along with some less extensive interactions with the peptide. Indeed, even though Ile 135 contacts the sulfur group of Cys 1241 , an alanine substitution at residue 1241 has no impact. The atomic model of the complex with the peptide suggests why the Ile 135 side chain is so critical. Removal of that side chain would leave a highly unfavorable hole in a buried hydrophobic core, implying that the I135A mutant cannot form a stable complex.
Although the central portion of the connector loop (residues 133-137) is tightly associated with residues of the underlying ␤-sheet and makes numerous contacts, the flanking sequences (positions 129 -132 and 138 -140) do not participate in sequence-specific interactions. These structural observations help to explain the results of our molecular genetic and biochemical studies, wherein single alanine substitutions at Gln 133 , Asp 134 , and Val 136 partially reduce HCMV UL44-UL54 binding, and replacement of Ile 135 abolishes binding, whereas alanine substitutions in the flanking sequences of the connector loop have no effect.
In a parallel study, we examined mutations in the C-terminal region of UL54 (12). Alanine substitution at Leu 1227 or Phe 1231 abolishes detectable binding between UL44 and UL54. A partial impairment is also observed with substitution at Arg 1224 , His 1226 , or Tyr 1234 or deletion of the two C-terminal cysteines at residues 1241 and 1242. The effects of most of these mutations are explained by interactions observed in the crystal structure. Three of these substitutions, including the two with the most severe defects, affect hydrophobic residues, viz. Leu 1227 , Phe 1231 , and Tyr 1234 , which form the hydrophobic plug (Fig. 2, B and C). Substitution at Arg 1224 or His 1226 also slightly affect binding, but the reason is less clear, as these two residues do not make noticeable interactions with UL44 in the structure. Deletion of the two C-terminal cysteines also reduces interactions of either full-length UL54 or the UL54 C-terminal peptide with UL44, reducing the affinity of the peptide for UL44 by ϳ10-fold (12). In the crystal structure, the C-terminal four residues of UL54 are positioned adjacent to the connector loop on the "left" side opposite the hydrophobic crevice (Fig. 2, B and C). These four residues help to bury a hydrophobic patch that includes the side chains of Ile 135 , Leu 43 , and Ile 49 , which would be solvent-exposed in their absence. In addition, we observed a hydrogen bond between the terminal carboxylate group of the peptide and the side chain of Thr 45 . Shortening the peptide by two residues would have positioned its charged carboxylate in a much less favorable environment, near the aliphatic side chains of Leu 43 and Pro 77 . Thus, the UL44-UL54 structure provides a clear explanation for the effects of most of the alanine substitutions and helps to explain why alanine substitutions at hydrophobic residues have the most dramatic effects in HCMV (12,19).
Comparison with the HSV UL42-UL30 Peptide Structure-The HCMV UL44-UL54 peptide structure is significantly different from the structure of HSV UL42 in complex with a peptide that corresponds to the C-terminal 36 residues of UL30 (16). In the HSV UL42-UL30 complex, the peptide adopts an ␣-␤-␣ fold (Fig. 3). As in the HCMV complex, the middle of the HSV peptide forms an antiparallel ␤-␤ interaction with the middle of the interdomain connector loop. However, these contacts are probably less important in HSV. In HSV, most of the buried surface and critical contacts involve the C-terminal 15-residue helix of the UL30 peptide, which has no structural counterpart in HCMV. It binds within an extended groove to the left of the ␤-␤ interactions (Fig.  3A). Isothermal titration calorimetry experiments are consistent with structural observations. They have demonstrated the importance of this helix in HSV, showing that a peptide corresponding to the C-terminal 18 residues of HSV UL30 binds to UL42 nearly as well as the 36-residue peptide does and that three residues from the terminal helix (His 1228 , Arg 1229 , and Phe 1231 ) are required for binding (31). All three residues make obvious contacts in the crystal structure (Fig. 3A). Note that, unlike HCMV, several of the key sequence-specific interactions in HSV are polar.
A closer structural parallel between HSV and HCMV can be drawn involving the helix that precedes the antiparallel ␤-␤ interac- tion in the sequence of the bound C-terminal peptide. Such a helix is present in both HCMV UL54 and HSV UL30 (and also in some complexes with PCNA; see below). In each case, this helix lies to the upper right of the ␤-␤ region, and its principal interactions with the processivity factor involve the side chains of hydrophobic residues from the C-terminal end of the helix, binding to a positionally conserved hydrophobic "crevice" on the surface of the processivity factor. In this analogy, Phe 1211 and Leu 1206 of HSV might be functionally similar to Phe 1231 and Tyr 1234 of HCMV (Fig. 3), although very different in detail. In HSV, the helix, a true ␣-helix, makes contact with UL42 only at its C-terminal end (The remainder of the helix projects away from the UL42 surface.) Although the existence of these hydrophobic side chain interactions in HSV might offer clues to evolutionary relationships, they are much less important for complex formation than the ones in HCMV, as they can be deleted without affecting binding affinity (31). In contrast, in HCMV, an overall helical conformation of the main chain is maintained by main chain hydrogen bonding, but it is interrupted by the presence of prolines; the large hydrophobic side chains are distributed differently on the surface (Fig. 3); and mutating any of them to alanine has a significant impact on the binding of the HCMV UL54 peptide to UL44 (12).
The Sliding Clamps Interact with Their Binding Partners via a Hydrophobic Crevice-Even though sliding clamps were originally described as processivity factors for replicative DNA polymerases, subsequent studies have demonstrated their remarkable capacity to interact with other proteins involved in DNA replication as well as proteins involved in cell cycle regulation, DNA repair, and DNA recombination (reviewed in Refs. 41 and 42). Many of the proteins that bind to PCNA encode an eight-residue PCNA-interacting protein box described as Qxxhxxaa (where x, h, and a represent any residue, large hydrophobic residues (Met/Leu/Ile), and aromatic residues (Phe/Tyr/Trp), respectively). Several structures of sliding clamps have been solved in complex with peptides or full-length proteins from these binding partners, including the p21 WAF1/CIP1 cell cycle inhibitor (17), the Fen-1 endonuclease (43)(44)(45), the clamp loader complex (46,47), and the DNA polymerase (44,48). In each complex, the binding partner binds specifically to the front face of the clamp, where hydrophobic interactions are formed between the hxxaa motif and a pocket on the right side of the connector loop. (Fig.  3C shows a representative example of the PCNA-p21 WAF1/CIP1 structure.) In general, the first, fourth, and fifth residues of the hxxaa motif lie packed together on one face of a 3 10 helix that is conserved among the PCNA-protein complexes. These side chains are involved in hydrophobic interactions with the pocket, which is formed by side chains from the connector loop and ␤-strands at the interdomain boundary of the clamp. In contrast, interactions involving residues outside of the hxxaa motif are not as well conserved. For example, the canonical PCNA-interacting protein box begins with a glutamine residue (41,42), which forms hydrogen bonds with the last ␤-strand of the clamp, but this interaction is not observed in the clamp loader structures (46,47). In addition, only the p21 WAF1/CIP1 peptide (17) and fulllength Fen-1 (45) structures have an extended ␤-␤ interaction with the interdomain connector loop analogous to UL44 and UL42 (Fig. 3). Notably, however, the position of the hydrophobic pocket to which the hxxaa motif binds is conserved among the sliding clamp structures and is similar in location to the crevice found in UL44.
A similarly positioned hydrophobic binding site is also seen in yet another multimeric processivity factor, the ␤-subunit from Escherichia coli (49). The ␤-subunit is composed of three subdomains that are similar in topology to the N-and C-terminal domains of UL44 and PCNA. The ␤-subunit contains a hydrophobic binding site that is in a position analogous to the crevice on UL44 and PCNA. This was demonstrated in structures of the ␤-subunit bound to the ␦-subunit of the clamp loader complex (50) and the translesion DNA polymerase, polymerase IV (51,52).
It is noteworthy that the position of the hydrophobic crevice is conserved among PCNA, UL44, UL42, and the ␤-subunit. In each case, the bound peptide that contacts the hydrophobic crevice assumes a roughly helical conformation, with the important hydrophobic side chains clustered on the face of the helix facing the crevice (Fig. 3). Furthermore, in each case, the hydrophobic crevice is formed principally by hydrophobic side chains (and the aliphatic portions of polar side chains) projecting outward from the ␤-sheet of the processivity factor and laterally from the connector loop. However, the molecular details that define the interactions with their binding partners are not identical. Like the p21 WAF1/CIP1 peptide, the HCMV UL54 peptide relies on a plug composed of one hydrophobic (Leu 1227 ) and two aromatic (Phe 1231 and Tyr 1234 ) residues that are buried in the hydrophobic crevice. However, these three residues do not participate in forming the hxxaa motif or the ideal 3 10 helix that is conserved among the PCNA-protein complexes. When the structurally equivalent residues of the HCMV UL44-UL54 peptide and PCNA-p21 WAF1/CIP1 peptide complexes are superimposed, both similarities and differences are observed (Fig. 4). Thus, Phe 1231 and Tyr 1234 of HCMV UL54 occupy positions that are structurally analogous to the binding sites of Met 147 and Tyr 151 of the This interaction depends largely on three hydrophobic residues from the peptide that bind to a hydrophobic crevice on UL44. The side chains of Leu 1227 and Phe 1231 of UL54 are required for the association with UL44. Although the HSV UL30 peptide has an aromatic residue (Phe 1211 ) that packs into an analogous crevice on UL42 (see inset in A), this interaction is not essential. C, like UL44, PCNA (Protein Data Bank code 1AXC) contains a hydrophobic crevice on the right side of the connector loop that binds to a peptide from its respective binding partner, p21 WAF1/CIP1 (green). Similar to the UL54 peptide, the p21 WAF1/CIP1 peptide buries three hydrophobic residues in the crevice. The connector loop from each processivity factor forms an antiparallel ␤-sheet with its respective peptide. In each inset, the processivity factor is gray; the connector loop is red; and the peptide backbone is tan. p21 WAF1/CIP1 peptide. However, HCMV Leu 1227 is bound in a spot that has no counterpart in the PCNA complex, and its main chain atoms do not participate in forming the helical arrangement that is common to all of the bound peptides. It should also be noted that, although peptide binding to the hydrophobic crevice has been structurally conserved among the multimeric processivity factors, it plays a more limited role in peptide binding to HSV UL42, where only a single aromatic side chain from the peptide is buried in the pocket. Both the similarities in folding topology and similarities in peptide binding may eventually be relevant to working out the evolutionary relationships between UL44, UL42, and PCNA.
The existence of an independent binding site on each subunit of HCMV UL44 and the sliding clamps is intriguing because it creates the potential for a number of different proteins to be bound simultaneously. For example, during DNA lesion repair, it has been hypothesized that sliding clamps function as "tool belts" in which each binding site can accommodate a different repair polymerase (53). Perhaps HCMV UL44 could similarly interact with proteins that include a domain that resembles the plug observed in UL54. Thus, if UL44 functions as a homodimeric processivity factor in complex with one copy of UL54, the second binding site could accommodate additional viral or cellular factors needed for DNA replication. Alternatively, in the absence of UL54, UL44 could bind other proteins and function as a mobile platform on DNA by sliding nonspecifically via electrostatic interactions formed between basic residues of the back face of the protein and the phosphate backbone of DNA.
The HCMV UL44-UL54 Interaction as a Drug Target-Currently available drugs against HCMV, which target the polymerase activity of UL54, are hampered by problems of pharmacokinetics, resistance, and toxicity (54). The interaction between UL44 and UL54 would make an attractive drug target because it is specific and differs in important details from the PCNA-binding partner interactions and because both proteins are essential for viral replication (3,4,55). The interaction is known to be inhibited by peptides corresponding to the C-terminal 22 residues of UL54 (29), and the interaction has been shown to be sensitive to amino acid substitutions in either of the two subunits (12,19). A similar strategy has been utilized for their counterparts in HSV (30,31,56), and small molecules have been identified that block both the HSV UL42-UL30 interaction in vitro and viral replication (57). In HCMV, the interaction appears even more amenable to small molecule inhibitors given the discrete hydrophobic crevice of UL44. Indeed, small molecules have been identified that block the HCMV UL44-UL54 peptide interaction, the UL44-UL54 interaction, and long chain DNA synthesis in vitro and that interfere with viral replication in cell-based assays (58). Details of the UL44 complex with the C terminus of UL54 may provide a basis for structure-based drug design.

Differences in Conformation of the C-shaped Clamp in the Unliganded Versus
Liganded State-We reported previously that the unliganded structure of UL44⌬C290 forms a C clamp-shaped homodimer (15), and it was verified that UL44⌬C290 dimerizes in solution (15,24). At that time, we presented a model in which the central cavity of the UL44 dimer could accommodate double-stranded DNA, and the basic back face of UL44 would make favorable electrostatic interactions with the phosphate backbone (15). These interactions would permit UL44 to tether the polymerase to the template, but also slide during replication. Consistent with this model, UL44⌬C290 also assembled as a C clampshaped homodimer when bound to the UL54 peptide (Fig. 5). In both the unliganded and peptide-bound crystal structures of UL44⌬C290 (space groups P6 1 22 and C222 1 , respectively), this head-to-head dimerization involves the short ␤-sheet at the N-terminal end of the molecule. Two copies of this sheet, related by an intermolecular 2-fold axis, are hydrogen-bonded to form an extended sheet in each dimer. In both crystal forms, this intermolecular 2-fold axis happens to coincide with a crystallographic 2-fold axis, resulting in a crystal form with one molecule/asymmetric unit. Stabilization of the homodimer interface also involves van der Waals interactions between several hydrophobic side chains. Alanine substitution of a number of these residues disrupts both dimerization in solution and DNA binding (15). When the region around the homodimer interface is compared in the two crystal structures, few appreciable differences are seen (see below). This finding confirms that the homodimer interface is stable and supports the idea that the C-shaped clamp is biologically relevant, rather than an artifact of crystallization. In both crystal structures, with and without bound peptide, model-based statistics (buried surface area and shape complementarity) ( Table 2) suggest that the monomers bind tightly to one another in head-to-head dimerization. The hydrophobic crevices of UL44 (tan) and PCNA (light blue) are in similar locations adjacent to the connector loop. However, the three hydrophobic residues of the p21 WAF1/CIP1 (backbone (red) and side chains (cyan)) and UL54 (backbone (yellow) and side chains (magenta)) peptides, which are critical for binding, are not structurally similar to one another.  Table 2) are included as landmarks to make molecular motions (pseudo-torsions) more obvious. A and C are orthogonal to B and D. The leftmost monomer from each head-to-head dimer has been superimposed in E so that positional differences in the rightmost monomers are emphasized.
Despite these structural similarities at the head-to-head homodimer interface, the C-shaped clamp exhibits a surprisingly large increase in the diameter of the cavity from ϳ28 to 40 Å (Fig. 5). The program DYNDOM (59), which is designed to detect domain motions automatically, confirmed that it is valid to regard the dimer as being composed of two distinct domains. Root mean square differences between main chain atoms of the superimposed dimers (3.00 Å) are large compared with the differences within monomers (1.44 Å), although no specific mechanical hinge points can be identified in the dimer to explain the motion. Consistent with this idea, the differences between structures increase gradually as larger and larger groups of atoms are considered (data not shown). The opening and closing motion of the UL44 dimer can be understood as analogous to the conformational change in a lock washer. Thus, as the washer becomes flatter, the ends get closer. Correspondingly, we can measure a twisting motion of up to 12°around an axis lying along the extended head-to-head ␤-sheet. To aid in calculating and visualizing the mechanical motions in the UL44 dimer, a few specific "proxy" ␣-carbons were chosen whose positions span the various ␤-sheets ( Fig. 5 and Table 2).
Peptide Binding Changes the Conformation of the Connector Loop-In the absence of bound peptide, the connector loop assumes a significantly different conformation (Fig. 6). The side chains of Gln 51 and Leu 60 serve as convenient reference points for understanding the change, as they are hydrogen-bonded to different main chain oxygen atoms of the connector loop in each of the two structures. With the UL54 peptide bound, these two side chains are hydrogen-bonded to the peptide oxygens of residues 134 and 136 (and the Ile 135 side chain lies on the inward-facing surface of the two-stranded ␤-sheet and is buried). In contrast, in the crystal structure without the UL54 peptide, the main chain oxygens of residues 133 and 135 are hydrogen-bonded instead (and correspondingly, this middle portion of the connector loop, still in an extended conformation, is flipped over, exposing Ile 135 ) (Fig. 6B). To accommodate this one-residue translocation of the "middle" of the connector, the flanking loop sequence 129 -132 must become more taut (in the peptide-free structure), whereas the flanking loop sequence 138 -142 must loop outward to a greater extent. Note that these two flanking areas represent the regions of greatest localized conformational difference between the two crystal structures, with ␣-carbon shifts exceeding 6 Å.
The altered conformation is explained by the structure: when the peptide is bound, the position of the ␤-strand of the peptide is constrained by the position of its hydrophobic plug. In the complex, the register of the connector loop (Fig. 6A) is dictated by its ability to form the ␤-sheet hydrogen-bonding pattern with the peptide while simultaneously burying the side chains of Ile 135 ( Fig. 2A) and Val 136 (Fig. 2B).
In the presence of the UL54 peptide, the connector loop forms many more specific interactions (Fig. 6). We therefore hypothesize that the binding of the UL54 peptide induces increased order in both the connector loop and the ligand. Indeed, such a dramatic increase in the number of interatomic contacts is consistent with tight binding of the peptide. Hydrogen bonds between the connector loop and the peptide are omitted for clarity (see "Results and Discussion"). B, in the absence of peptide, the side chain of Ile 135 is exposed and in a position similar to that of Val 136 in the complex. Translation of the main chain causes the side chains of Lys 60 and Gln 51 , projecting upwards from the central ␤-sheet, to be hydrogen-bonded to different main chain oxygen atoms in the two cases. As previously reported for the unliganded UL44 structure (15). b Shape complementarity is defined by the shape complementarity statistic (40). c Residues denoted with primes belong to the symmetry-related monomer lying across the head-to-head dimer interface. Ile 80 is a residue from one edge of the central ␤-sheet, adjacent to the amino (head-to-head) dimer contact area. Phe 211 is a residue from the outer edge of the central ␤-sheet, adjacent to the carboxyl (tail-to-tail) dimer contact area. Thr 196 belongs to the central ␤-sheet at the C-terminal end, is close to the tail-to-tail contact site, and faces the putative DNA-binding cavity. Observe that even when the symmetry-related Ile 80 Ј is included, the monomer is a good approximation to a single rigid body. In contrast, when the Ile 80 -Ile 80 Ј axis is used as pivot, the pseudo-torsion between symmetry-related copies of the central sheet varies by 8 -12°, which is largely responsible for the 12-Å change in the width of the central cavity of the dimer.
Peptide Binding Increases the Affinity of UL44 for DNA-We wished to rule out the possibility that opening and closing of the C-shaped clamp results from differences in crystal-packing interactions. Those changes that affect the back face of UL44 and the cavity of the homodimer would be predicted to affect the DNA binding properties of UL44. We therefore used a previously described filter binding assay (15) to compare the interaction of unliganded versus peptide-bound UL44⌬C290 with DNA (Fig. 7). Peptide-bound UL44⌬C290 exhibited a 4-fold lower apparent K d for DNA than did unliganded UL44⌬C290. As a control, we also tested a UL44⌬C290 mutant (I135A) that does not detectably bind this peptide by isothermal titration calorimetry (19). In the presence of the UL54 peptide (Fig. 7) or its absence, 6 UL44⌬C290(I135A) bound DNA with the same affinity as wild-type UL44⌬C290 did in the absence of peptide. As an additional control, we tested a peptide corresponding to the C terminus of HSV UL30, and it did not affect the binding of wild-type UL44⌬C290 to DNA (Fig. 7). We conclude that the binding of the UL54 peptide to UL44 alters its conformation to increase its affinity for DNA.
The observed increase in affinity could be attributable to an increase in the on-rate, which might result from an opening of the binding site, and/or a decrease in the off-rate, which would be consistent with an improved complementarity between the protein and DNA in the complex. It should be noted, however, that the increased affinity of UL44 for DNA was observed under conditions in which two peptides bind per UL44 dimer. It is not clear whether both monomers would be occupied in a UL44-UL54 complex in vivo, or if binding a single peptide would be sufficient to propagate the change to the other monomer in the dimer pair. Regardless, these results suggest that conformational changes in UL44 upon UL54 binding may play a biological role in regulating the affinity of the dimer for duplex DNA. One simple possibility is that the conformational change simply increases affinity, thereby promoting processivity. An alternative possibility is that one conformation of UL44 might be shaped to bind classic B-form DNA better in the absence of UL54. During DNA replication, however, UL44 may have to adapt to DNA that is less like the classic B-form due to distortions caused by the HCMV DNA polymerase during translocation on the DNA template.
UL44 May Be Capable of Forming Higher Order Aggregates-The crystal structure of the complex with the peptide also points out a way that UL44 might form higher order aggregates when bound to DNA. Thus, in the crystal structure of the complex with the peptide, crystal packing either creates or stabilizes a second 2-fold interface. This additional interface involves extended ␤-interactions between the fourstranded ␤-sheets at the C-terminal end of the molecules (tail end), rather than the five-stranded ␤-sheets at the N-terminal ends (head end). With both interfaces forming extended ␤-sheets, a continuous helical arrangement of C clamp-shaped dimers is created in the crystal with the peptide bound, having an inner diameter of ϳ45 Å. This second interface appears to be significantly weaker than the head-to-head interaction, as it lacks the intricate interdigitation of aromatic and aliphatic side chains that stabilizes the latter (15); it buries approximately threefourths as much total surface area (see Table 2); shape complementarity is less ( Table 2); and no higher order aggregates are detected in solution (15). Nevertheless, the second dimer interface still exhibits a greater degree of structural complementarity than what is usually required for a macromolecular crystal-packing contact.
A recent electron microscopy study indicated that the UL44 homolog from the ␥-herpesvirus Epstein-Barr virus assembles in a higher ordered structure that is reminiscent of a lock washer or ring-like arrangement (60). Furthermore, the inner diameter of the rings in these electron microscope images have widths similar to that of the central cavity of the infinite "lock washer" helical arrangement of dimers observed in the UL44-UL54 crystal lattice. Although the putative biological function of these higher order assemblies is unknown, UL44 and other herpesvirus processivity factors are present in excess relative to their catalytic subunits, and these "spare" subunits may be associated with DNA. Therefore, future studies should keep in mind that, when multiple copies of the C clamp-shaped dimer are bound to the same DNA duplex or in the vicinity of the same replication fork, they may have a tendency to associate along the weaker interface and form higher ordered complexes as a result.