The Family X DNA Polymerase from Deinococcus radiodurans Adopts a Non-standard Extended Conformation

Deinococcus radiodurans is an extraordinarily radioresistant bacterium that is able to repair hundreds of radiation-induced double-stranded DNA breaks. One of the players in this pathway is an X family DNA polymerase (PolXDr). Deletion of PolXDr has been shown to decrease the rate of repair of double-stranded DNA breaks and increase cell sensitivity to gamma-rays. A 3′→5′ exonuclease activity that stops cutting close to DNA loops has also been demonstrated. The present crystal structure of PolXDr solved at 2.46-Å resolution reveals that PolXDr has a novel extended conformation in stark contrast to the closed “right hand” conformation commonly observed for DNA polymerases. This extended conformation is stabilized by the C-terminal PHP domain, whose putative nuclease active site is obstructed by its interaction with the polymerase domain. The overall conformation and the presence of non standard residues in the active site of the polymerase X domain makes PolXDr the founding member of a novel class of polymerases involved in DNA repair but whose detailed mode of action still remains enigmatic.

DNA replication and repair are functions that are of vital importance for the maintenance of cellular life. These functions are carried out by various DNA replicating engines, most of them acting as multiprotein complexes. Deinococcus radiodurans, a Gram-positive bacterium, is characterized by an extraordinary resistance to ionizing radiation and desiccation. After radiation induced cutting of its 3.28-megabase genome into hundreds of small fragments, it is capable of reassembling it completely (1). Different hypotheses have been suggested to explain this radioresistance. A recently proposed mechanism involves the creation of long linear DNA intermediates by an extended synthesis-dependent strand annealing process, where overlapping chromosomal fragments are used both as primers and as templates for synthesis of complementary single strands (2). Recircularization of chromosomes would be assured by homologous recombination. Although DNA polymerase I is one of the main enzymes involved in this process, it was shown that other proteins affect double strand break repair efficiency in D. radiodurans. One of these is an X family DNA polymerase (PolX Dr ) 5 (3). Cells devoid of PolX Dr protein show increased sensitivity to ␥-irradiation and a longer delay in the restoration of an intact genome after irradiation. It was therefore proposed that PolX Dr has an important role in double strand break repair in D. radiodurans. The contribution of PolX Dr may become essential for instance when damage gets too important or, alternatively, it may act in different repair pathways from polymerase I. Indeed, some of the X DNA polymerases, such as Saccharomyces cerevisiae Pol4 and human polymerase (4) have been proposed to play important roles in different DNA repair processes, including non-homologous end-joining (5). It was shown that PolX Dr also has strong 3Ј35Ј exonuclease activity that is stimulated by Mn 2ϩ (6). This activity is associated with proofreading mechanisms in other polymerase families and encoded by protein domains or subunits distinct from the polymerase catalytic domain (7). Curiously the exonuclease activity of PolX Dr is modulated upon encounter of a stem-loop structure. The combination of both activities leads to the hypothesis that PolX Dr might be involved in DNA repair, potentially non-homologous end-joining, by processing damaged DNA or repair intermediates, thus generating substrates for other repair proteins (6). Very recently an orthologue of PolX from Bacillus subtilis was characterized. It was shown that PolX Bs is a template-directed DNA polymerase acting on DNA gaps with a downstream 5Ј phosphate group, suggesting it may play a role in base excision repair (8).
DNA polymerases all combine a catalytic palm domain, a thumb domain, binding double-stranded DNA, and a finger domain that fixes the incoming nucleotide. The polymerase domain of the X family belongs to the Pol␤-like nucleotidyltransferase superfamily, sharing ϳ25% amino acid identity with the DNA polymerase domains of Pol, Pol4, and Pol␤. PolX Dr has a second domain at the C terminus called PHP, with strong sequence identity with the histidinol phosphatase involved in histidine transport in bacteria. Due to its similarity to histidinol phosphatase and the presence of a trinuclear zinc site, the PolX Dr PHP domain is thought to function as phosphoesterase (9). In the context of DNA polymerases, this activity might be responsible for the degradation of pyrophosphate, thus driving the polymerization reaction, or contributes to a nuclease reaction that would be involved in proofreading the newly synthesized strand. The deletion of the PHP domain also had a negative effect on survival of ␥-irradiated cells suggesting that this domain possesses a function in DNA repair. Unexpectedly, deletion of the PHP domain destroys structure modulated but not the general 3Ј35Ј exonuclease activity (6). No activity could be demonstrated for the PHP domain alone.
In this report we present the crystal structure of PolX Dr at 2.46-Å resolution. Surprisingly, PolX Dr adopts a stretched out conformation instead of the commonly observed closed right hand conformation. In the active site of the polymerase catalytic domain, the two universally conserved aspartates are replaced by two glutamates, whereas the active site of the PHP domain is obstructed by its interaction with the polymerase domain.

EXPERIMENTAL PROCEDURES
Expression and Purification-PolX Dr was expressed as previously described (3). The induced cells were harvested and resuspended in buffer A (20 mM Tris-HCl, pH 7.5, 500 mM NaCl, 10 mM imidazole, and a mixture of anti-proteases (Complete, Roche Applied Science)). After sonication and centrifugation, the supernatant was heated 10 min at 60°C followed by 10 min cooling on ice. After a second centrifugation, the supernatant was loaded onto a nickel-affinity column (nickel-nitrilotriacetic acid-agarose, Qiagen). The bound proteins were eluted by buffer B (20 mM Tris-HCl, pH 7.5, 150 mM NaCl, 150 mM imidazole). The eluate was loaded onto a 5-ml HiTrap heparin column (Amersham Biosciences) pre-equilibrated with buffer B without imidazole. PolX Dr was eluted by a linear gradient from 150 mM to 1 M NaCl on buffer B using an Ä KTA Purifier System (Amersham Biosciences). Column fractions were analyzed by SDS-PAGE, and those containing PolX Dr were identified and pooled diluted to 50 mM NaCl and concentrated. The yield from 1 liter of culture was 1 mg of PolX Dr . The homogeneity and integrity of the protein were finally checked by mass spectrometry (Voyager, PerkinElmer Life Sciences).
Crystallization, Data Collection, Structure Solution, and Refinement-Native protein (4 mg/ml) crystallized at 293 K by the hanging drop, vapor-diffusion method from 1:1 l drops of protein and precipitant containing 0.1 M NaAc, pH 5, 0.2 M NaCl, 16% 2-methyl-2,4-pentanediol. Plate-like crystals were transferred in the mother liquor containing 30% 2-methyl-2,4pentanediol and 10 mM HgCl prior to flash-cooling in liquid nitrogen. X-ray diffraction data from a crystal of the native protein was collected on beamline ID23-1 at the European Synchrotron Radiation Facility (Grenoble, France). Data processing was carried out with MOSFLM, XDS, and SCALA (10,11). The crystals diffracted to a maximum resolution of 2.46 Å and belong to space group P2 1 with two molecules per asymmetric unit, corresponding to a 50% solvent content. The cell parameters and data collection statistics are reported in Table 1.
The Hyss sub-structure solution module from Phenix was used to find the anomalous scattering sites (12). Refinement and phasing was performed with Sharp (13), solvent flattening with Solomon and resolve (14). An initial model was built in the maps with resolve and manually completed using O. The structure was refined with the refinement module of Phenix (15), using NCS constraints between the two polypeptide chains. The final model contains residues 8 -564 and 6 -564 for each polypeptide chains respectively, three zinc atoms bound to the PHP domain and one mercury atom bound to C237 in each chain. Coordinates have been deposited (code 2w9m).
Small Angle X-ray Scattering-SAXS data were collected at the Swing beamline at SOLEIL, Saclay. The sample to detector distance was 1.93 m covering the range of scattering vector (q) from 0.007Å Ϫ1 to 0.35Å Ϫ1 (q ϭ 4 sin /, with 2 being the scattering angle and ϭ 1.36 Å the wavelength of the x-rays). The detector used was a charge-coupled device camera from Aviex. Twenty-five successive frames of 0.75-s exposure each separated by a 1-s pause were recorded for each protein solution and buffer alike. During exposure, the solution was contained in a quartz capillary 1.5 mm in diameter under vacuum. The solution was continuously pushed into the beam using an automated injection system. In these conditions no radiation damage was detected.
The samples were prepared at concentrations of 1.1, 4.4, and 10.7 mg/ml in 20 mM Tris-HCl, pH 7.5, 150 mM NaCl, 5% (v/v) glycerol. SAXS data were normalized to the intensity of the incident beam, averaged and background-subtracted using the program package PRIMUS (16). The curves recorded at the three concentrations were perfectly identical and therefore free from intermolecular interactions. Intensities were scaled using the scattering by water. The radius of gyration and the intensity at the origin were calculated using the Guinier law over the angular range qR g Ͻ 1.3 (17). The pair distribution function P(r) was determined using the indirect Fourier transform approach as implemented in the GNOM program (18). Alternative estimates of the radius of gyration and the intensity at the origin were derived from the calculated P(r). Scattering patterns from crystal structures were calculated using the program CRYSOL (19). Missing parts in the crystal structure, comprising 11 N-terminal residues and 5 C-terminal residues, were modeled using MODLOOP (20). The closed conformation was built by fitting the fingers and 8-kDa domains in the same orientation as the Pol␤ protein. To adjust the calculated scattering curve in the experimental one, PolX Dr was split into three fragments comprising, respectively, the 11 N-terminal residues, the residues from Asp 6 to Arg 158 , and the 412 C-terminal residues. The relative position of both domains was refined using the rigid-body modeling program SASREF (21), which uses simulated annealing to find an optimal configuration of the domains by fitting the SAXS curve. The connectivity between the two fragments was preserved during refinement by keeping the distance between the C␣ atoms of Arg 158 and Gln 159 and of Pro 5 and Asp 6 shorter than 4 Å. An ultimate adjustment was performed using the program CRYSOL (19). The goodness of fit was characterized by the following parameter, where N is the number of experimental points, c is a scaling factor, and I calc (q j ) and (q j ) are the calculated intensity and the experimental error at the scattering vector q j , respectively.

RESULTS AND DISCUSSION
Structure Determination and Overall Architecture-The crystals of PolX Dr belong to space group P2 1 with two molecules per asymmetric unit. PolX Dr is composed of two distinct parts: the polymerase domain, itself composed of subdomains, and the PHP domain. Sequence homology of the PHP domain of PolX with YcdX from Escherichia coli whose structure is known (PDB code 1M65) suggested the presence of a trinuclear zinc metal binding site. Although no zinc was added to the purification buffers of the recombinant protein, an energy scan at the zinc absorption edge showed the presence of zinc in the crystal, and a 2.46-Å dataset was recorded at the zinc absorption edge. For this crystal, the cryoprotectant solution had been supplemented with 10 mM HgCl 2 to aid phasing. Six zinc sites and two mercury sites were found from the anomalous differences. Visual inspection of the zinc heavy atom sites revealed that two sets of three sites had a spatial arrangement similar to the zinc sites in the YcdX PHP domain. Phasing led to poor but interpretable electron density maps, where two models of the PHP domain could be placed in the electron density by superposition of the zinc sites. Models of the subdomains of Pol were also placed in the density, which provided an initial model used for iterative rebuilding. The structure of PolX Dr was fully refined and rebuilt at 2.46-Å resolution.
The two polypeptide chains present in the asymmetric unit superpose perfectly with the exception of the 8-kDa N-terminal sub-domain, which adopts a slightly different orientation in both copies (see below). Apart from this slight difference, the polymerase domain and the PHP domain have the same relative orientations in both polypeptide chains. This is consistent with the idea that PolX Dr was trapped in a stable conformation that is not a crystallization artifact.
Structure of the Polymerase Subdomains-The PolX Dr polymerase domain shares the modular organization of family X polymerases: an X family-specific 8-kDa domain and the generic polymerase fingers, with the palm and thumb subdomains. Overall the PolX Dr structure is most similar to the structure of Pol (22), although the overall relative orientation of the different domains is dramatically different. Each subdomain will therefore be compared separately before considering the global conformation. These domains will be referred to in the initial notation of Pol and Pol␤, which was later revised for homogeneity with other polymerases.
The thumb domain of PolX Dr (Fig. 1, A and B, dark green) is virtually identical to the Pol domain with the exception of the loop between ␤7 and ␤8 (Pol numbering), which is approximately ten amino acids shorter in PolX Dr . The PolX Dr palm domain has the ␤-nucleotide transferase fold, common to all the family X polymerases (23) and related to the prokaryotic replicative PolIII ␣-subunit (Fig. 1, A and B, light green). Palm domains harbor the carboxylates involved in metal binding and polymerization. The loop between ␤4 and ␤5, which is around ten amino acids long in Pol, Pol␤, and the African Swine Fever Virus PolX ASFV , is reduced in PolX Dr to a three-amino acid ␤-turn. This difference is important because a longer loop would clash with the fingers subdomain of PolX Dr , which has a totally different position relative to the palm domain compared with the other family X DNA polymerases (see further). It is therefore probable that the observed domain configuration is unique to PolX Dr .
The finger domain is composed of four helices (F-I) (Fig. 1, A  and B, light blue). Helices F and G form a helix-turn-helix motif that binds the primer strand backbone in Pol. The most noteworthy difference is the I helix that provides the link between the catalytic and the fingers domain. In Pol, a kink in this helix induces a turn of ϳ70 degrees and is sometimes described as two helices (I and IЈ) (Fig. 1D). In PolX Dr , helix I and IЈ form a unique, long, and straight helix. The absence of this kink in PolX Dr profoundly modifies the standard relative orientation of the fingers and catalytic domains (Fig. 1D), creating a unique configuration of the polymerase domain as a whole (see below).
The 8-kDa domain is composed of 5 helices (A-E) (Fig. 1, A  and B, dark blue). In Pol and Pol␤, this domain possesses DNA binding and recognition activity as well as a metal-independent 5Ј-deoxyribose 5-phosphate-lyase activity. However, despite an overall identical structure, there are important differences in the PolX Dr and Pol 8-kDa domains. In Pol and Pol␤, the 5Ј-deoxyribose 5-phosphate-lyase uses the N⑀ of a lysine (Lys 312 and Lys 72 , respectively) as the sole Schiff base nucleophile. This lysine is located in a lysine-rich pocket that contains a 5Ј-phosphate binding site. In order for the lysine to act as a nucleophile, it must be deprotonated and hence the pK a of its ⑀-amino group must be lowered. It is believed that the surrounding residues, notably the lysines lining the catalytic pocket, are involved in this process. In PolX Dr , the catalytic lysine is replaced by a glutamate (Glu 72 ). The only conserved lysine in PolX Dr is Lys 33 , which corresponds to Arg 275 or Lys 35 in Pol and Pol␤, respectively, whereas the Pol␤ conserved lysines Lys 60 , Lys 68 , and Lys 72 (Lys 275 , Lys 307 , and Lys 312 in Pol) are replaced by non-charged residues. In the complex of Pol (Pol␤) with gapped DNA, Lys 307 (Lys 68 ) provides the main contribution to 5Ј-phosphate binding of the downstream primer. Although this conserved lysine is absent in PolX Dr , Lys 64 , and Lys 67 are in the vicinity of the hypothetical DNA binding site and could be involved in DNA binding. The conserved tyrosines Tyr 279 (Pol) and Tyr 39 (Pol␤), which are also implicated in phosphate binding, superpose perfectly with Tyr 37 in PolX Dr .
As a result, the 8-kDa domain in PolX Dr seems to have lost many characteristics of a 5Ј-phosphate binding site. This questions the existence of 5Ј-deoxyribose 5-phosphate-lyase activity of PolX Dr , which has not yet been documented. In any case, all these differences must contribute to a different DNA binding affinity and selectivity, which could contribute to the selection of special sites on the DNA damaged by irradiation or desiccation.
Structure of the PolX Dr PHP Domain-PolX Dr has a PHP domain at its C-terminal end (Fig. 1, B and C, red), which is absent in the other family X polymerases such as Pol and Pol␤ but present in the replicative polymerase PolIII␣ subunits (9,24,25). The C-terminal PHP domain of PolX Dr is structurally homologous to other members of the PHP family. This domain consists of a (␣/␤) 7 seven-stranded ␤ barrel closed on one side of the ␤-barrel by a C-terminal helix. The PolX Dr PHP domain superposes well with the YcdX protein from E. coli (root mean square deviation of 1.80 Å for 200 residues for 18% sequence identity) and with the PHP domain of the PolIII ␣-subunit (root mean square deviation of 2.80 Å for 136 residues for 15% sequence identity).
The PHP domain is attached to the thumb domain by a structured 30-amino acid linker, which contains a small helical region (residues 296 -326) (Fig. 1, yellow). This linker packs along the ␣6 and ␣7 helices of the PHP domain and the N and M helices of the thumb domain through hydrophobic interactions, hydrogen bonds, and salt bridges (Arg 299 -Asp 309 , Glu 302 -Arg 534 , Arg 304 -Asp 287 , and Glu 305 -Arg 190 ).
Three zinc ions were unambiguously identified in the PolX Dr PHP. In contrast to PolX Dr , in the structures of YcdX or PolIII␣ PHP domains, one or two zincs are missing in the absence of added zinc. This might arise from the fact that the exit of the putative PHP-PolX Dr catalytic site is obstructed (see below) preventing the loss the zinc atoms during purification. The three zincs are coordinated by the conserved residues of the PHP family (His 332 , His 334 , His 364 , His 428 , His 456 , His 518 , Asp 339 , Asp 401 , and Asp 516 ).
Domain Organization-In the structures of Pol␤ and Pol, domains are arranged similarly to a closed right hand, which envelops both DNA and the incoming nucleotides (26,27). Substrate binding is often accompanied by structural rearrangements of the polymerase subdomains. The structure of human Pol␤, solved in complex with gapped DNA in the presence or absence of dNTP, has revealed an open and closed conformation of the enzyme. The 8-kDa domain shows significant movement during DNA binding, while the conformational change upon dNTP binding involves a slight reorientation of the thumb subdomain (28,29). On the contrary, the structure of the Pol family X Polymerase has shown that it does not undergo large subdomain movements during catalysis (22). The configuration of the PolX Dr domains is radically different from Pol and Pol␤ in the open or closed conformation. The finger, palm, and thumb domains in PolX Dr do not create the familiar closed right hand but, on the contrary, adopt a completely extended conformation (Fig. 1C). This suggests that the thumb and palm domain on one hand and the fingers and 8-kDa domains on the other have moved as "rigid" bodies. It is interesting to note that the PolX Dr conformation does not result from a simple opening of the Pol "hand" as classically seen in the small movement of Pol upon DNA binding: the finger and 8-kDa N-domains in PolX Dr are swung out by 90°c ompared with the Pol conformation around a hinge region centered on Arg 158 , situated in the ␣I-␣J loop between the palm and fingers domain (Fig. 1D). This implies that the putative DNA binding region of the 8-kDa domain and of the palm catalytic site are on opposite sides of the protein surface (Fig.  1B). The PHP domain binds across the length of the polymerase domain and makes interactions with all four polymerase domains. The PHP domain might be involved in stabilizing the observed stretched conformation of PolX Dr by binding at the interface between palm and fingers (Fig. 1, B and C). However, the fingers and 8-kDa domain can be modeled in the same conformation, because they adopt in Pol␤ without severely clashing with the PHP domain. There is no steric hindrance preventing the polymerase to flip into a closed conformation, which would also free the entrance of the PHP domain active site. In the PolIII ␣-subunit the PHP domain wedges between the thumb and palm domain, but these interactions do not involve the putative active site region (25). The function of the PolIII PHP domain is also unknown.
Thermal parameter analysis of the structure using the TLS Motion Determination server indicated that the 8-kDa domain was the most obvious candidate for defining a TLS group for refinement (see Fig. 3A). This confirms the hypothesis that the 8-kDa domain is slightly flexible in one copy of PolX in the crystal structure. Further division of the structure in TLS groups identified the fingers, palm, thumb, and PHP domains. However, inclusion of up to five TLS domains for the modeling of atomic displacement parameters during refinement did not yield any improvement in the refinement statistics.
The linker and PHP domain make extensive interactions with all other subdomains. This interaction buries in total 2000 Å 2 and involves 7 salt bridges and 7 hydrogen bonds. Surprisingly the main interaction site is situated at the top of the PHP ␤-barrel, which coincides with the putative catalytic site containing the trinuclear zinc. The cavity of the PHP domain containing the zinc ions is occluded by the hinge between helix E (8 kDa) and F (fingers). The PHP ␤5␣5 loop makes extensive contacts in the groove formed between the 8-kDa and fingers domains (Fig. 1C). This loop could act as a latch that locks the 8-kDa/fingers domain in the stretched configuration.
As a result of these tight interactions, the access to the putative active site of the PHP domain is completely blocked. It cannot be excluded that binding of an appropriate substrate could provoke a substantial conformational change which would expose the PHP active site.
Catalytic Mechanism-It was shown for Pol␤ that the binding of the incoming nucleotide closes the active site, bringing into place the catalytic groups. These changes are characterized by rotations of thumb helices and closing in of the 8-kDa domain. Pol, however, is in a closed state before nucleotide binding, but dNTP binding shifts the DNA template strand in the active site creating crucial interactions with the minor groove of DNA. In view of the radically different relative configuration of the PolX Dr domains compared with other family X members, it remains an open question whether the same mechanism is at work here. It is interesting to note that the African swine fever PolX ASFV altogether lacks the finger and 8-kDa domains but is a genuine polymerase (30). Therefore the role of the positioning or even the presence of the individual subdomains on polymerase activities has not definitely been settled.
Structural analysis and sequence analysis strongly suggest that the phosphoryl transfer reaction of all polymerases is catalyzed by a two metal ion mechanism (31). The structure of Pol␤ in complex with gapped DNA and dideoxy-CTP showed that the two metal ions are bound by three carboxylates of the palm domain (32). Superposition of the PolIII␣ and Pol␤ palm domains aligns all three catalytic aspartates. Strikingly, superposition of these palm domains with the PolX Dr palm domain reveals the replacement of two of these metal binding aspartates (for instance Asp 429 and Asp 490 in Pol numbering) by two glutamate residues Glu 199 and Glu 234 (Fig. 2A). The third aspartate (Asp 427 in Pol numbering), which bridges the two catalytic metal ions is absent in PolX Dr . No other obvious residue is present in the vicinity to complete the coordination of the metal ions in PolX Dr . In the polymerases of family C and X, the loop between strands ␤1 and ␤2 is part of the incoming nucleotide binding pocket and contains a conserved glycine-serine/ glycine motif (Gly 179 and Ser 180 in Pol␤) involved in hydrogen bonding to the ␤ and ␥ phosphates of the nucleotide. In PolX Dr , the glycine is conserved (Gly 186 ), but the serine/glycine is replaced by an aspartate (Asp 187 ). The presence of an aspartate in this position is surprising and probably leads to important changes in the binding mode of the incoming nucleotide. However, the two Pol␤ arginines Arg 149 and Arg 183 , which are involved in binding the phosphates of the incoming nucleotide, are conserved in PolX Dr (Arg 160 and Arg 190 ), whereas the Gly 190 , which is hydrogen bonded to the ␥-phosphate is replaced by Arg 196 . The positively charged residues that bind the 3Ј-phosphate in the primer strand in Pol, Pol␤, and PolIII (Arg 254 , Arg 488 , and Lys 616 , respectively) are replaced in PolX Dr by a proline (Pro 232 ) but possibly Arg 196 could take over this role.
Clearly, because the polymerase activity of PolX Dr is strongly stimulated by Mn 2ϩ , we expected a participation of this metal in the reaction mechanism. The supposed metal binding site in the palm, however, deviates from what is observed in other members of the family X polymerases. This suggests that the active site may be organized differently in PolX Dr and/or that a variant of the canonical two-metal catalytic step is operating in PolX Dr . Complexes with nucleotides and metals are needed to settle these questions.
Exonuclease Activity-Many DNA polymerases contain a 3Ј35Ј exonuclease activity that is part of a proofreading mechanism (Pol␥, -␦, and -⑀) while it is lacking in others (Pol␣ and -␤ and the eukaryotic polymerases of the Y family). In some polymerases, this exonuclease is embedded in a separate domain (Pol␦ and -⑀) while in others it is associated to a separate subunit of the replicative complex (e.g. E. coli PolIII). It has recently been shown that the PHP domain of Thermus thermophilus replicase has 3Ј35Ј exonuclease activity that may act as an additional proofreading exonuclease (33). Surprisingly, the strong Mn 2ϩ -dependent 3Ј35Ј exonuclease activity of PolX Dr is located in the polymerase domain, because deletion of the PHP domain does not abolish canonical 3Ј35Ј exonuclease activity. The stretched conformation of PolX Dr further shows that the putative active site of the PHP domain is obstructed by its interaction with the polymerase domain. Therefore the PHP domain cannot be active as a nuclease in this conformation of the full-length protein. The only known biochemical effect of deletion of the PHP domain is to reduce the structure modulated nuclease activity of PolX Dr . Because the PHP seems to participate in stabilizing the stretched conformation of PolX Dr , it is tempting to speculate that the stretched conformation is involved in the stem-loop-specific nuclease activity of PolX Dr .
Mutations of two conserved glycines 104 and 106 does not affect the 3Ј35Ј exonuclease activity on homopolymeric DNA but leads to a complete loss of the pausing site on stem-loop oligonucleotides (6). These glycines are part of the conserved GXG motif shown to interact with the DNA phosphate backbone in Pol␤ (27). These glycines are situated at the entrance of the fingers and the interaction with DNA requires the cusp configuration (where they are located at ϳ15-Å distance from the catalytic carboxylates). In the stretched PolX Dr configuration, this glycine motif is far removed (31 Å) from the palm catalytic carboxylates, and it is hard to imagine how it could interact with stem loop DNA without a drastic rearrangement of the domains.
We have tested whether the conserved glutamates of the palm domain of Polymerase active site might be important for exonuclease activity. The E199A, E234A, and E253A mutants exhibited the same structure modulated exonuclease activity as the wild-type enzyme, indicating that the polymerase active site is not involved (results not shown). It was recently shown that PolX Bs contains a 3Ј35Ј exonuclease activity that resides in its PHP domain (34). Mutation of two evolutionary conserved histidines (His 339 and His 441 ) belonging to a motif predicted to bind putative catalytic zinc ions abolished the 3Ј35Ј exonuclease activity. The equivalent histidines of PolX Dr (His 353 and His 355 ) are direct ligands of the zinc cluster, and mutagenesis results of PolX Bs seem therefore compatible with the PolX Dr structure. As mentioned the zinc cluster is totally inaccessible in our present PolX Dr structure, which probably represents an inactive conformation as far as the exonuclease activity concerns. The PHP domain in PolX Dr makes two direct salt bridges with the ␤5-␤6 loop of the palm (Glu 193 -Arg 500 and Arg 190 -Glu 305 , Fig. 2). The homologous position to Arg 190 in Pol␤ binds the incoming nucleotide. This offers a possibility to couple polymerase and exonuclease activities in PolX Dr . The PHP domain may change its position upon encounter of mismatches in the active site, thereby liberating its active site zinc-cluster.
Conformation of PolX Dr in Solution-To study the extent of the flexibility and to establish whether or not the crystal structure corresponds to the conformation in solution, we have investigated its structure by SAXS measurements. The radius of gyration value derived from the pair distribution function is 27.9 Ϯ 0.1 Å, close to the value 27.8 Ϯ 0.1 Å obtained from the Guinier law. The maximal extension of the protein is estimated from the pair distribution function: D max ϭ 88 Ϯ 3 Å. These values of R g and D max are very close to the values corresponding to the crystal structure (27.5 Å and 91 Å, respectively). Fig. 3B shows that the calculated scattering curve from the crystal structure does not fit the experimental curve (the best adjustment gives ϭ 5.2). Thus the protein in solution adopts a conformation different from the crystal structure. We have also tried to adjust the experimental curve to the curve calculated from a model of PolX Dr in the closed conformation (see "Experimental Procedures" for modeling details), but this conformation does account neither for the data in solution ( ϭ 11.3). The orientations of the different PolX Dr domains were therefore adjusted by rigid-body refinement (see "Experimental Procedures" for details). Optimizing the orientation of the 8-kDa domain could not account for the experimental data (not shown). However, optimized models where both the fingers and 8-kDa domains were both reoriented (as rigid bodies around the link between Arg 158 and Gln 159 (Fig. 1)) gave good fits to the experimental data with typical value of ϭ 1.3 (Fig.  3). Several independent runs starting from either the crystal structure or the closed conformation converged to very similar models where the first domain has turned of ϳ20°from the twisted conformation (Figs. 1D and 3). The results demonstrate that the conformation of PolX Dr is different from the crystal structure but is close to the twisted conformation. One can expect that the structure is dynamic, and the model shown in the Fig. 3 must be regarded as an "average" rather than as a unique conformation adopted by the protein.

CONCLUSION
The extended conformation observed in our crystal structure raises interesting questions as to the function of PolXDr. At first sight, our observations suggest that the stretched conformation does not represent a (near) catalytically competent state of the enzyme and that, for polymerase activity, the enzyme must undergo drastic structural rearrangements to adopt the canonical Pol␤ cusp-like arrangement. The present PolX Dr configuration for which the fingers and 8-kDa domains are far removed from the palm catalytic domain could be compared with the polymerase of the African swine fever virus (PolX ASFV ), an X family polymerase, which consists uniquely of a thumb and palm domain. PolX ASFV binds gapped DNA as efficiently as does Pol␤ (35). The lack of the fingers and 8-kDa domain in this polymerase is thought to contribute to the low fidelity of nucleotide insertion. The crystal structure and conformation in solution of PolX Dr enrich the structural diversity of family X polymerases with a new twisted conformation and suggest that polymerases can explore a much greater conformational space than was previously thought. The twisted conformation of PolX Dr might present specific advantages in the repair of some specific subsets of DNA lesions. Alternatively, PolX Dr might adopt different conformations, for example in the presence of specific structures in damaged DNA, which might expose either polymerase or exonuclease activities. The presence of non-standard residues in the active site might indicate that proteins with polymerase activities might have been overlooked in sequence data base based on search against the canonical motifs.