Crystal structure of the human Polϵ B-subunit in complex with the C-terminal domain of the catalytic subunit

The eukaryotic B-family DNA polymerases include four members: Polα, Polδ, Polϵ, and Polζ, which share common architectural features, such as the exonuclease/polymerase and C-terminal domains (CTDs) of catalytic subunits bound to indispensable B-subunits, which serve as scaffolds that mediate interactions with other components of the replication machinery. Crystal structures for the B-subunits of Polα and Polδ/Polζ have been reported: the former within the primosome and separately with CTD and the latter with the N-terminal domain of the C-subunit. Here we present the crystal structure of the human Polϵ B-subunit (p59) in complex with CTD of the catalytic subunit (p261C). The structure revealed a well defined electron density for p261C and the phosphodiesterase and oligonucleotide/oligosaccharide-binding domains of p59. However, electron density was missing for the p59 N-terminal domain and for the linker connecting it to the phosphodiesterase domain. Similar to Polα, p261C of Polϵ contains a three-helix bundle in the middle and zinc-binding modules on each side. Intersubunit interactions involving 11 hydrogen bonds and numerous hydrophobic contacts account for stable complex formation with a buried surface area of 3094 Å2. Comparative structural analysis of p59–p261C with the corresponding Polα complex revealed significant differences between the B-subunits and CTDs, as well as their interaction interfaces. The B-subunit of Polδ/Polζ also substantially differs from B-subunits of either Polα or Polϵ. This work provides a structural basis to explain biochemical and genetic data on the importance of B-subunit integrity in replisome function in vivo.

In eukaryotes, the bulk of replication is performed by the members of the B-family DNA polymerases Pol␦ and Pol⑀, 3 which participate in lagging and leading DNA strand synthesis, respectively (1)(2)(3). The other members of this family are Pol␣ and Pol. Pol␣ extends RNA primers with a short patch of deoxyribonucleotides, and then synthesis is switching to Pol␦ and Pol⑀ (4 -6). Pol is one of the key players in replication of damaged DNA (7)(8)(9). The common architectural features of all four polymerases are the catalytic subunits with exonuclease/polymerase module(s) and the C-terminal domain (CTD) and the B-subunits (10). Historically, subunits in different organisms were named differently, and the current nomenclature of polymerase subunits is shown in Table 1. The B-subunits stabilize the catalytic subunit (11,12), act as scaffolding proteins to mediate interactions with other components of the replication machinery (13)(14)(15)(16), and regulate DNA-synthesizing activity in a cell cycle-dependent manner, during checkpoints, and telomere maintenance (17)(18)(19)(20)(21).
Human Pol⑀ consists of four subunits (Table 1): the catalytic subunit (p261) and the auxiliary subunits (the second subunit, p59, also referred to as the B-subunit, and the third and fourth subunits, p17 and p12, respectively) (10). p261 is a bi-lobe protein comprised of two tandem exonuclease/polymerase domains, the first of which is active, whereas the second is inactive and, along with CTD, plays an essential structural role (16,(22)(23)(24). CTD interacts with p59 and has eight conservative cysteines coordinating two zinc ions (25,26). In yeast, deletion of CTD abolishes interaction between the catalytic and B-subunit and is lethal (23,27). The integrity of the CTD-B-subunit complex is important for stability of the genome (28,29). Auxiliary subunits play a regulatory role and guide Pol⑀ to different biological pathways. The B-subunit integrates Pol⑀ into the eukaryotic replisome, and in yeast, its N-terminal helical domain is essential for assembly of the Cdc45-MCM-GINS helicase (15). The small subunits p17 and p12 interact with each other through histone-fold motifs similar to those found in histones H2A and H2B (30). As a heterodimer, they bind p261 at the second, inactivated exonuclease/polymerase domain and increase Pol⑀ affinity to double-stranded DNA (31, 32). Small subunits are not essential for cell growth, but their deletion in yeast destabilizes the replication fork and elevates spontaneous mutagenesis (33).
The crystal structure of the human Pol␦ B-subunit (p50) in complex with the N-terminal domain of the C-subunit (p66 N ) provided the first view of the three-dimensional organization of B-subunits (34). The docking surface of Pol␦ CTD on the B-subunit has been predicted (25) using structural and mutational analyses (34,35). Moreover, structural and biochemical studies led to a discovery that the CTD of human Pol is capable of binding the p50 -p66 complex of human Pol␦ in the same manner as the CTD of Pol␦ (25), meaning that both the Pol and Pol␦ catalytic subunits form a complex with p50 -p66. These studies led to experiments that confirmed this discovery in yeast (9). The crystal structures of the yeast and human Pol␣ B-subunits in complex with CTDs of the catalytic subunits have also been reported as separate complexes (36,37), as well as within the human primosome (38). Until now, the structure of the Pol⑀ B-subunit was the only one unsolved among the human B-family DNA polymerases. Here we report the crystallization, structure determination, and comparative structural analysis of the full-length human Pol⑀ B-subunit (p59) in complex with the CTD of the catalytic subunit (p261 C ).

Overall structure of the human Pol⑀ CTD-B-subunit complex (p59 -p261 C )
The crystal of the p59 -p261 C complex contains two independent molecules in an asymmetric unit with root mean square deviations (RMSD) of 0.25 Å for 552 superimposed ␣-carbons. Each molecule is comprised of p261 C (residues 2142-2286) and full-length p59 (residues 1-527) (Fig. 1A). The density was absent for the N-terminal residues 1-83 and for the loop region 146 -162 in both p59 subunits, as well as for the Nand C-terminal tails (2142-2149 and 2174 -2182) and loop 2283-2286 in both p261 C domains. The molecule has a slablike shape with overall dimensions of 79 Å ϫ 75 Å ϫ 53 Å (Fig.  1B).

B-subunit
The p59 subunit is comprised of three domains: an N-terminal domain (NTD), residues 1-75; a phosphodiesterase-like (PDE) domain, residues 84 -121 and 245-527; and an oligonucleotide/oligosaccharide-binding (OB) domain, residues 122-244 (Fig. 1). The absence of the density for residues 1-83 in the crystal suggests that the NTD is connected to the rest of the molecule with a flexible linker and lacks the stabilizing interdomain interactions with the PDE, the OB, and p261 C . The NMR structure of the NTD revealed a globular four-helical bundle that resembles the C domain of AAAϩ proteins (39). Because the residues 1-75 are well folded in solution, the flexible linker connecting the NTD with the rest of p59 should be within residues 76 QSVDETIE 83 (Fig. 2). The PDE domain is comprised of a central two-layer ␤-sheet (layer 1, 1␤ 19 2␤ 20 2␤ 18 2␤ 17 1␤ 16 2␤ 3 , and layer 2, 2␤ 22 1␤ 21 2␤ 12 2␤ 13 2␤ 14 2␤ 15 ). Helices ␣ 7 , ␣ 13 , ␣ 14 , and 3 10 -2 are flanking layer 1, helices ␣ 9 -␣ 12 are flanking layer 2, and helix ␣ 8 is located on the side of both layers. The PDE is inactive because it lacks the conserved histidine and aspartate residues involved in metal coordination and catalysis. The OB domain contains a ␤-barrel (1␤ 6 2␤ 7 1␤ 8 ␤ 9 1␤ 11 2␤ 10 ), a helix ␣ 5 on the side, and a helix ␣ 6 on the bottom of the barrel. The majority of residues in the loop linking helices ␣ 5 and ␣ 6 are disordered. One of the strands in the barrel is divided by an inserted loop into two strands: ␤ 8 and ␤ 9 . Contrary to the NTD, the PDE and OB domains are well stabilized relative to each other. The interaction between them is dominated by polar contacts with 22 well defined interdomain hydrogen bonds.

Interaction between p59 and p261 C
The p59 -p261 C interface has an elongated shape with complementary surfaces that bury the surface area of 3094 Å 2 . The subunits are tightly associated by 11 intersubunit hydrogen bonds and by extended hydrophobic areas (Figs. 3 and 4). The N-terminal part of the helix ␣ 2 and the entire helices ␣ 1 and ␣ 3 of the three-helix bundle of p261 C interact mainly with the loops of 3 10 -1 -␣ 5 , ␤ 14 -␣ 12 , and ␣ 14 -3 10 -2 of the p59 PDE domain. The most prominent feature of this interaction is the wedging of the p59 Phe 120 side chain deep into the hydrophobic core of the p261 C three-helix bundle (Fig. 4). In addition, the ␤-sheet and the loop ␤ 6 -␤ 7 of p261 C Zn2 are engaged in interactions with the OB domain of p59, where the p261 C Met 2232 side chain is inserted into the hydrophobic depression between the helix ␣ 5 and the ␤-barrel of the OB domain. The modular architecture of p261 C suggests the possibility of independent folding and p59 binding for Zn2 and the three-helix bundle. This structural feature may explain the effect of small deletions in the yeast ortholog of p261 C on its affinity to the B-subunit. It was demonstrated by a two-hybrid assay that partial disruption of either Zn2 or the ␣-helices weakened intersubunit interaction but did not completely abolish it (27). Only the combined deletion of Zn2 and helices ␣2 and ␣3 completely eradicated the association between the catalytic and B-subunits; probably, in this case, the folding of the helix ␣1, as well as its interaction with the B-subunit were affected in the absence of helices ␣2 and ␣3. Unlike Zn2, the Zn1 module does not participate in intersubunit interaction, but it is important for the folding of a three-helix bundle, which explains the defect in cell growth upon partial deletion of Zn1 in CTD of yeast Pol⑀ (23,27). In contrast to the Zn2 module and the helical bundle of CTD,

Crystal structure of human Pol⑀ p59 -p261 C complex
the OB and PDE domains of the B-subunit are tightly associated, so a defect in either of these domains will affect the integrity of the other one and should completely abrogate the intersubunit interaction. Consistently, all truncation variants of Dpb2 were unable to interact with the catalytic subunit (40).

Comparison of the B-subunits of Pol␣, Pol␦, Pol⑀, and Pol
The crystal structures of B-subunit complexes from all human B-family DNA polymerases, including CTDs of catalytic subunits with B-subunits (p70 -p180 C of Pol␣ and p59 -p261 C of Pol⑀) and the B-subunit with a part of the C-subunit (p50 -p66 N of Pol␦ or Pol), are available now for comparison (Fig. 5).
The common features of the B-subunits are the presence of a PDE domain with a two-layer ␤-sheet and an OB domain with a ␤-barrel, the relative positioning of these two domains, and the surface for docking of the CTD (34,36,37). In addition, B-subunits contain some equivalents of helices ␣ 3 , ␣ 4 , ␣ 5 , ␣ 9 , and ␣ 10 (p59 secondary structure nomenclature is used in all compari-sons). Despite the conservation of core elements, the B-subunits exhibit significant shifts in the positions of helices and the structures of loops. As a result, the RMSD between p59 and p70 is 2.9 Å 2 for 352 matching residues of 427. The similarity between p50 and either p59 (2.9 Å 2 for 329 residues) or p70 (3.1 Å 2 for 341 residues) is low as well. Among the notable differences between p59 and p70 or p50 are the topologies of residues N-and C-terminal to the OB domain. The ␤-hairpin ␤ 2 ␤ 3 and helix 3 10 -1 from residues N-terminal to the OB domain, together with helix ␣ 7 and coil ␣ 7 -␣ 8 from residues C-terminal to the OB domain, form a bulky addition on the side of p59 (Fig.  5). This provides p59 a relatively globular shape compared with p70. The bulky addition of p59 is more prominent compared with p50, which also lacks the adjacent helix ␣ 8 (Fig. 5). The unique features of p50 are the interaction with p66 N and the absence of a globular NTD linked with a PDE domain via a flexible linker in p59 and p70. The docking of p66 N is performed mainly by p50 helices ␣9, ␣10, an ␣-helix that is inserted into the loop ␣ 9 -␤ 13 , and the N-terminal residues that are folded as an ␣-helix, an extended coil, and a small ␤-strand. In addition, p50 has an elongated strand ␤ 22 that The nitrogens, oxygens, and sulfurs are colored blue, red, and yellow, respectively.

Crystal structure of human Pol⑀ p59 -p261 C complex
interacts with a ␤-sheet of p66 N , resulting in a continuous nine-strand ␤-sheet (Fig. 5).

Comparison of CTDs
CTDs have significant differences in their size, structure, and sequence. For example, p125 C of Pol␦ and p353 C of Pol are 107 and 100 amino acids (aa) long, respectively, and significantly shorter than p180 C (195 aa) of Pol␣ and p261 C of Pol⑀ (144 aa). Both p261 C and p180 C contain a three-helix bundle with a Zn1 and a Zn2 on either side (Fig. 5). The aa identity between p125 C and p353 C is 18%, whereas the sequence identities for all other pairs are marginal and comparable with randomly selected sequences. Secondary structure predictions for p125 C and p353 C show that both contain a short (10-aa) ␣-helix N-terminal to Zn1 and two longer ␣ helices (28 and 22 aa) (Fig. 6B) that supposedly fulfill the role of ␣ 1 and ␣ 2 in p261 C and p180 C (Fig.  6A). In contrast to p180 and p261, p125 and p353 coordinate a [4Fe-4S] cluster at the second metal-binding site of their CTDs (25,41). Interestingly, two of the [4Fe-4S] coordinating cysteines are predicted to be from the C-terminal portion of helix ␣ 2 , and both the Zn1 and [4Fe-4S]-binding modules contain short loops (10 aa or less) and no ␤-strands. Therefore, they acquire a compact structure that is different from the corresponding Zn1 and Zn2 modules of p180 C and p261 C . At present, only the latter two domains are available for detailed comparison. Their superimposition indicates the conservation of helix ␣ 1 and significant differences in the length, conformation, and relative positions of helices ␣ 2 and ␣ 3 (Fig. 7). The prominent feature of p180 C is a longer helix ␣3 and two helices inserted between the helices ␣ 2 and ␣ 3 that stabilize an elongated loop in the Zn1. The corresponding loop is disordered in p261 C . Additionally, p180 C contains a C-terminal extension that forms a ␤-strand and an ␣-helix upon interaction with the large subunit of primase (38,42).
Structural insight into the Pol⑀ p59 -p261 C subcomplex confirms our previous observation that the Zn2 module of human Pol⑀ requires zinc for specific docking on p59 (25), which is important for the formation of a stable complex between the catalytic and the B-subunit. It is shown in Ref. 25 that p261 C expressed and purified alone contains a substantial level of iron, whereas the pure and stoichiometric p59 -p261 C complex is iron-free. This indicates that the Zn2 module of p261 C is able to coordinate the iron-sulfur cluster, but this prevents specific interaction with p59. According to the structure of [4Fe-4S] cluster-containing proteins, for example, human primase (43), coordination of the cluster by p261 C would require the distances between zinc-binding cysteines to be increased by ϳ3 Å, which would definitely change the Zn2 structure and result in steric hindrance with the OB domain. These peculiarities explain why an iron-sulfur cluster was seen in partially purified samples of yeast Pol⑀ and its CTD (41). In contrast to the other B-family DNA polymerases, an additional zinc-binding module is located in the catalytic domain of Pol⑀ (44), and the iron- Crystal structure of human Pol⑀ p59 -p261 C complex sulfur cluster can be misincorporated at this position as well (45). Thus, all structural data support our previous conclusion (25) that Pol⑀ and Pol␣ require only zinc for correct assembly (5,44). Moreover, concentrated and very pure samples of human Pol␣ and Pol⑀, obtained in our laboratory, were ironfree and fully functional (46,47), indicating that their operation in eukaryotic cells is not dependent on the iron-sulfur cluster assembly machinery.   p353 C (B). Blue-and green-shaded boxes depict the ␣-helices and ␤-strands, respectively. The gray-shaded boxes indicate the residues that are disordered in the crystal structures of p261 C and p180 C . The secondary structure elements are from the crystal structures of p261 C and p180 C (A) and from Phyre (57) predictions for p125 C and p353 C (B). The conserved residues are highlighted in red. Magenta-shaded boxes depict the zinc-coordinating cysteines. Yellow bars indicate the sequences comprising the zinc-binding modules Zn1 and Zn2. The red bar indicates the sequences implicated in iron-sulfur cluster binding.

Crystal structure of human Pol⑀ p59 -p261 C complex Interactions between CTD and the B-subunit
The structures of Pol␣ p70 -p180 C (36) and Pol⑀ p59 -p261 C (Fig. 1B) revealed the common features as well as significant differences in the mode of intersubunit interactions. In both molecules, Zn2 mainly interacts with the OB domain, whereas the helical bundle of CTD associates with the PDE domain of the B-subunit. In both cases, the contact area between Zn2 and the B-subunit composes ϳ30% of the total interaction interface between the two subunits. With regard to the differences, it is worth noting the predominance of hydrophobic contacts at the CTD-B-subunit interaction interface and the reduction of the buried surface area by 30% in the case of Pol⑀ (Figs. 3 and 4). Superimposition of the B-subunits in p70 -p180 C and p59 -p261 C shows a significant shift in position of p261 C relative to p180 C with an average displacement of their central three-helix bundles exceeding 6 Å (Fig. 8).
The structure of p50 with either p125 C of Pol␦ or p353 C of Pol has not been reported yet. The mapping of the p50 surface area responsible for the docking of p125 C was performed with data obtained for the fission yeast ortholog using random and directed mutagenesis techniques (35). This revealed that the p50 residues responsible for p125 C interaction are disordered. It was proposed and confirmed that disordered loops provide the ability of p50 to adopt the binding of both p125 C and p353 C despite their low amino acid sequence identity (25). Unlike p50, p70 of Pol␣ and p59 of Pol⑀ accept only one CTD: p180 C or p261 C , respectively.  Both CTD-B-subunit complexes were aligned by superposition of the B-subunits; p70 is omitted for clarity. The p59 -p261 C domains are colored according to Fig. 1, and p180 C is colored gray.

Crystal structure of human Pol⑀ p59 -p261 C complex The p59 -p261 C structure helps to explain phenotypes of the yeast Dpb2 mutants
A set of temperature-sensitive Dpb2 alleles was previously generated by using direct and random mutagenesis techniques ( Table 2). Some Dpb2 mutants exhibited a mutator phenotype comparable to that of mismatch-repair-defective strains (msh6), indicating that defects in the B-subunit of the human Pol⑀ gene can induce genomic instability. The structure of p59 -p261 C provides a rationale for explanation of the effects of amino acid changes in the yeast Pol⑀ B-subunit, characterized by weakened interaction with the catalytic subunit and the GINS complex, elevated spontaneous mutagenesis, and the affected Nrm1 branch of the DNA replication checkpoint (11,21,28,29,40,48).
High secondary structure conservation between p59 -p261 C and its yeast ortholog (Fig. 9) implies similar three-dimensional organization, allowing for mapping the residues changed in Dpb2 of Saccharomyces cerevisiae on the structure of p59 and for predicting the effect of these mutations on the integrity of Dpb2 and its interaction with the catalytic subunit ( Table 2). Structural analysis of 38 mutations revealed that two residues, Ser 453 and Pro 621 , are located at the CTD-B-subunit interface. According to modeling, Asn 333 of p59, which corresponds to Ser 453 of Dpb2, is located at the edge of the intersubunit interface and does not participate in the interaction with p261 C (Fig.  10A); there is also low probability that the hydroxyl of Dpb2 Ser 453 makes a hydrogen bond with any surrounding residue. It is likely that the appearance of a cysteine at this position does not affect the B-subunit folding in yeast and human Pol⑀ but can result in moderate steric hindrance with Val 2140 of Pol2 C , which corresponds to Ser 2197 of p261 C (Fig. 9B). In the case of the Dpb2-112 mutant, Pro 460 of p59, which corresponds to Pro 621 of Dpb2, participates in hydrophobic interactions with three amino acids of p261 C : Met 2212 , Thr 2215 , and Leu 2216 (Fig.  10B), which correspond to Glu 2155 , Leu 2158 , and Ile 2159 of Pol2 C , respectively (Fig. 9B). Therefore, the replacement of proline by serine at this position affects the hydrophobic pocket at the CTD-B-subunit interface. Pro 460 also plays a structural role for the coil ␣ 14 -3 10 -2 , and its substitution may affect the position of amino acids Leu 456 , Tyr 457 , Tyr 462 , and Trp 463 , which interact with p261 C (Fig. 9A). Thus, P621S substitution should have stronger effect on intersubunit interaction than S453C, which is well correlated with more pronounced effects on viability, affinity to Pol2, and spontaneous mutagenesis in cells expressing Dpb2-112 mutant versus Dpb2-109 (Table 2). Interestingly, several other Dpb2 mutants do not have amino acid substitutions, which can directly affect the association with a catalytic subunit, but some of them have a phenotype similar to the Dpb2-112 mutant ( Table 2). It is likely that Dpb2 folding is affected in these mutants. All temperature-sensitive mutants listed in Table 2 have a minor to moderate effect on B-subunit integrity. Two alleles with predicted minor effect on Dpb2 folding (dpb2-102 and dpb2-111) have phenotypes close to the wild type.
We propose that mutations affecting hydrophobic pockets or resulting in steric hindrance can lead to local disturbance of the protein fold, and the effect on association of subunits should be stronger as the distance between the mutated residue and the interaction interface becomes shorter. The exact effect of amino acid substitution on the Dpb2 structure depends on many factors, such as location, involvement in interactions with other residues, degree of steric hindrance, and how the molecule can tolerate this change. The contribution of some of these factors, as well as the cumulative effect of several mutations on Dpb2 integrity, is difficult to estimate accurately, especially from the structure of a homologous but diverged protein.

Table 2 Correlation between predicted effect on protein integrity and phenotypes of Dpb2 mutants
The data for cell growth, mutability, and interaction between the catalytic and B-subunits of yeast Pol⑀ were obtained from Refs. 11, 28, and 40. Amino acid changes resulting in steric hindrance are indicated by an asterisk. Mutations resulting in hydrophobic pocket disturbance caused by hydrophobic amino acid replacement by a charged or significantly smaller residue are shown in bold type. p59 residues near the intersubunit interaction interface (when the distance to the closest p261 C residue is Ͻ10 Å) are underlined; p59 resides at the interaction interface are underlined and italicized. It was assumed that mutations resulting in steric hindrance or affecting a hydrophobic pocket have a moderate effect on B subunit integrity; otherwise, they have a minor effect. K171 (N87)E, S182 (V98)P*, L284 (V183)P* Ϫ 0.7 8 ϩϩ a Alleles dpb2-105 and dpb2-106 containing more than 11 mutations are not listed. b Normal and slow cell growth at elevated temperature is designated as ϩ and Ϯ, respectively; Ϫ indicates very slow or no cell growth. c Binding experiments were conducted using the yeast two-hybrid system; ␤-galactosidase activity measured at 23°C was taken for 100% in the case of a wild-type protein.
d The frequency of forward mutations was measured at the CAN1 locus (the numbers indicate the fold increase compared with a wild-type protein); the DPB2 variants were integrated into the DPB2 chromosomal locus (⌬I(-2)I-7BYUNI300). e Integrity loss is a degree of fold disturbance defined as minor (ϩ) or moderate (ϩϩ). f In one study, cells demonstrated normal growth at 37°C when the allele dpb2-1 was located on a centromeric plasmid, but the authors were unable to obtain viable integrants of this allele on the chromosome (28). In an earlier study, cells with a chromosomal copy of dpb2-1 stopped division within 4 h after a temperature shift up to 38°C and exhibited slow growth even at 24°C (11). g The values in parentheses are obtained from experiments where DPB2 variants are located on a centromeric plasmid.

Crystal structure of human Pol⑀ p59 -p261 C complex
Despite these complications, it was possible to make unambiguous structural predictions for several Dbp2 mutants. For example, the Dpb2-113 mutant has a strongest mutator effect coinciding with disrupted binding to the catalytic subunit ( Table 2). This mutant has substitution of two residues located on ␤4 and ␤6 strands to proline, which results in steric hindrance with a main-chain oxygen of an amino acid located on the neighbor strand. For example, Val 183 of p59, corresponding to Leu 284 of Dpb2, is located in the hydrophobic pocket inside of the ␤-barrel of the OB domain (Fig. 11A). Steric hindrance between proline in this position and Ala 226 will result in disruption of at least two hydrogen bonds and local disturbance of the ␤-barrel nearby the CTD-B-subunit interface. These two proline substitutions in the Dpb2-113 mutant have an additive effect, because the Dpb2-101 mutant with only one proline substitution shows mild phenotype (Table 2). It is likely that only these two proline substitutions define the phenotype of the Dpb2-107 mutant, which has four additional mutations with predicted minor effects on protein integrity. The other allele with a severe defect in intersubunit association is a dpb2-110, which has three mutations with a predicted moderate effect on the protein fold. Probably, the L285W change is the main determinant of the phenotype of this allele. The conservative Leu 285 is located inside of a hydrophobic pocket between the ␣-helix The ␣-helices and ␤-strands are depicted by green-and blue-shaded boxes, respectively. The gray-shaded boxes indicate the residues that are disordered in the crystal structure of p59 -p261 C . The conserved residues are shown in red. The blue, pink, and black circles indicate amino acids involved in intersubunit hydrophobic, hydrophilic, and both types of interactions, respectively. The Dpb2 residues listed in Table 2 are underlined. The fold prediction for the yeast Pol⑀ subunits was performed using Phyre (57).

Crystal structure of human Pol⑀ p59 -p261 C complex
and the ␤-barrel of the OB domain, close to the intersubunit interface. Modeling of a bulky tryptophan at this position reveals strong steric hindrance with surrounding hydrophobic residues, which will disturb the OB domain, as well as its interaction with the PDE domain and CTD (Fig. 11B).
The most interesting example of the influence of B-subunit integrity on the replisome function in vivo is a dpb2-200 mutation, which results in lethality at all temperatures and corresponds to deletion of the last six residues (QEEIYI) from the C terminus of the yeast Pol⑀ B-subunit and to substitution of conservative Pro 677 by serine (49). Mapping of these mutations on the p59 structure led to the conclusion that this change induces severe destabilization of the protein fold (Figs. 9A and 12). The C-terminal deletion disrupts the last ␤-strand, which is a part of a central two-layer ␤-sheet of the PDE domain. Moreover, Pro 514 and Leu 524 of p59, which correspond to Pro 677 and Ile 687 of Dpb2, participate in the formation of two hydrophobic pockets tethering different parts of the PDE domain (Fig. 12).
These observations support the idea that the effect of Dpb2 mutations on genomic stability is due to disturbance of replication initiation and elongation because of affected polymerase integrity and tethering to the replisome (29). Indeed, mutations of Psf1, which is a component of the GINS complex and interacts with NTD of the Pol⑀ B-subunit, also increase spontaneous mutagenesis (48). Probably, in Dbp2 mutants with compromised binding to the catalytic subunit, NTD is folded correctly and supports CMG assembling (15) if solubility of the B-subunit and its transport to the nucleus are not significantly affected.

Conclusions
The B-subunits of B-family DNA polymerases have evolved as the hubs for a variety of regulatory protein-protein interactions. The analysis of their crystal structures helps to evaluate the possible mechanisms for the involvement of such hubs during various stages of DNA replication. Recent cryo-EM struc-  Crystal structure of human Pol⑀ p59 -p261 C complex ture of the yeast CMG-Pol⑀-Ctf4-Pol␣ complex indicates that Pol⑀ interacts directly with the CMG helicase, whereas Pol␣ is located at a distance from the CMG and has more mobility (50). Contrary to the short NTD-PDE linker in human Pol⑀, the corresponding linker in human and yeast Pol␣ can be extended to Ͼ200 Å, enabling the primosome to reach the priming sites on both the lagging strand and distantly located leading strand (51). Although the role of p70 NTD in Pol␣ tethering to the replisome has not been ascertained thus far, the potential replisome-anchoring site on the N terminus of the catalytic subunit is connected with the rest of the molecule by a 190-aa linker (5). The NTD-PDE linker in Dpb2 is longer than in p59, but it might be structured and compact, because secondary structure prediction shows the presence of two ␣-helices (Fig. 9A).
The crystal structure of p59 -p261 C as provided here is the first high-resolution three-dimensional view of the Pol⑀ CTD-B-subunit complex. Together with the structure of the polymerase/exonuclease domain of a yeast ortholog (44), this is the only high-resolution structural information for eukaryotic Pol⑀ available to date. The structures of small subunits (p17, p12) and a catalytically inactive polymerase domain between the catalytic core and p261 C are required to build an accurate model of the whole four-subunit complex including the data from electron microscopy studies of Pol⑀ or its complex with CMG. Unfortunately, the current resolution of electron microscopy data for Pol⑀ is not sufficient for unambiguous docking of the better-resolved crystal structures of its parts (16,24,50). Further research will be required to evaluate the function of p59 and to find the exact position of p59 -p261 C relative to the catalytic core and other domains of Pol⑀ in the replicating complex. It will also be interesting to see the detailed mechanism of p59 -p261 C participation in recruitment of Pol⑀ to the CMG helicase.

Protein expression and purification
Cloning, expression, and purification to homogeneity of the full-length human Pol⑀ B-subunit (p59; residues 1-527) in complex with the C-terminal domain of the catalytic subunit (p261 C ; residues 2142-2286) have been described elsewhere (25). The peak fractions were combined and concentrated to 11 mg ml Ϫ1 in buffer containing 10 mM Tris-HCl, pH 7.7, and 1 mM DTT. The concentrated protein solution was frozen in liquid nitrogen and stored at 193 K until use.

Crystallization
The aliquots of protein solution were defrosted and centrifuged, and the absence of aggregation was verified with dynamic light scattering. Initial crystallization screening was performed by the sitting-drop vapor diffusion method at 295 K by mixing 1 l of protein solution with 1 l of reservoir solution. The reservoirs contained 2-fold diluted Crystal Screen and Crystal Screen 2 (Hampton Research) solutions with the addition of 2 mM Tris(2-carboxyethyl)phosphine, pH 7.5. Clusters of crystals appeared in the 32nd condition of Crystal Screen. The optimized reservoir solution with 50 mM sodium citrate, pH 5.6, 0.70 M ammonium sulfate, and 2 mM Tris(2-carboxyethyl)phosphine resulted in well shaped single crystals (Fig. 13) growing at 295 K in 3-5 days.

Data collection
For diffraction data collection, the crystals were soaked in a cryoprotectant solution (50 mM sodium citrate, pH 5.6, 1 M ammonium sulfate, and 32% (v/v) glycerol) for a few seconds, scooped in a nylon-fiber loop, and flash-cooled in a dry nitrogen stream at 100 K. Diffraction data sets were collected using synchrotron X-rays on the Argonne National Laboratory Advanced Photon source beamline 24ID-C equipped with a Pilatus 6M detector. All intensity data were indexed, integrated, and scaled with DENZO and SCALEPACK from the HKL-2000 program package (52). The crystals belong to tetragonal space group P2 1 2 1 2 and diffract up to 2.35 Å resolution. The crystal parameters and data processing statistics of the best data set are summarized in Table 3.

Crystal structure determination
The crystal structure of the p59 -p261 C complex was determined by the single-wavelength anomalous dispersion method. The positions of four zinc ions were located from anomalous difference Patterson maps, indicating the presence of two inde- Crystal structure of human Pol⑀ p59 -p261 C complex pendent p59 -p261 C molecules in asymmetric unit. Initial automated model building performed with Phenix (53) revealed about 65% of correctly built structure. The rest of the crystallographic computing was performed with CNS version 1.1 (54) using phases from the automatically built partial model. Non-crystallographic symmetry restraints and density modification with 2-fold averaging were applied during the initial stages of model refinement and map calculations for manual model building with Turbo-Frodo. Application of zonal scaling (55) slightly improved the quality of the electron density maps. After the addition of solvent molecules, the model was refined to an R cryst of 22.2% and an R free of 26.28% at the highest resolution of 2.35 Å. The final refinement statistics are provided in Table 3. The figures containing molecular structures were prepared using the PyMOL Molecular Graphics System (version 1.8; Schrödinger) (56); amino acids were modified using the mutagenesis tool and the library of backbone-dependent rotamers included in this software.   Crystal structure of human Pol⑀ p59 -p261 C complex