The Human TREX2 3′ → 5′-Exonuclease Structure Suggests a Mechanism for Efficient Nonprocessive DNA Catalysis*

The 3′ → 5′-exonucleases process DNA ends in many DNA repair pathways of human cells. Determination of the human TREX2 structure is the first of a dimeric 3′-deoxyribonuclease and indicates how this highly efficient nonprocessive enzyme removes nucleotides at DNA 3′ termini. Symmetry in the TREX2 dimer positions the active sites at opposite outer edges providing open access for the DNA. Adjacent to each active site is a flexible region containing three arginines positioned appropriately to bind DNA and to control its entry into the active site. Mutation of these three arginines to alanines reduces the DNA binding capacity by ∼100-fold with no effect on catalysis. The human TREX2 catalytic residues overlay with the bacterial DnaQ family of 3′-exonucleases confirming the structural conservation of the catalytic sites despite limited sequence identity, and mutations of these residues decrease the still measurable activity by ∼105-fold, confirming their catalytic role.

The 3 3 5-exonucleases process DNA ends in many DNA repair pathways of human cells. Determination of the human TREX2 structure is the first of a dimeric 3-deoxyribonuclease and indicates how this highly efficient nonprocessive enzyme removes nucleotides at DNA 3 termini. Symmetry in the TREX2 dimer positions the active sites at opposite outer edges providing open access for the DNA. Adjacent to each active site is a flexible region containing three arginines positioned appropriately to bind DNA and to control its entry into the active site. Mutation of these three arginines to alanines reduces the DNA binding capacity by ϳ100-fold with no effect on catalysis. The human TREX2 catalytic residues overlay with the bacterial DnaQ family of 3exonucleases confirming the structural conservation of the catalytic sites despite limited sequence identity, and mutations of these residues decrease the still measurable activity by ϳ10 5 -fold, confirming their catalytic role.
During the multistep processes of DNA replication, repair, and recombination DNA ends are often remodeled by enzymes containing 3Ј 3 5Ј-exonuclease activities to remove mispaired, modified, fragmented, and normal nucleotides from DNA 3Ј termini. The major 3Ј 3 5Ј-exonuclease activity detected in mammalian cell extracts is catalyzed by the TREX1 1 enzyme (1)(2)(3). The gene encoding TREX1 and the closely related TREX2 gene were cloned (4,5), and the 3Ј 3 5Ј-exonuclease activities of the recombinant proteins confirmed the catalytically robust nature of these enzymes (6,7). In addition to the TREX enzymes, several other proteins containing 3Ј 3 5Ј-exonuclease activities have recently been identified in human cells (for recent reviews, see Refs. 8 and 9). The sequences and structures of these enzymes indicate a diverse collection that includes multidomain proteins such as the Werner syndrome (WRN) (10 -12) and p53 proteins (13) and the "proofreading" DNA polymerases ␦, ⑀, and ␥. The 3Ј 3 5Ј-exonuclease activities have been detected in the hMRE11 subunit of a doublestrand break DNA repair complex (14,15) and the base excision repair apurinic/apyrimidinic endonuclease, APE (16 -20).
The presence of 3Ј 3 5Ј-exonuclease activities in a multitude of proteins likely reflects the importance of this activity in the proper maintenance of DNA 3Ј-ends.
Defective 3Ј 3 5Ј-exonucleases impair critical DNA metabolic pathways and elicit devastating consequences in cells and animals. The Trex1Ϫ/Ϫ mice exhibit dramatically reduced survival resulting from inflammatory myocarditis leading to cardiomyopathy and circulatory failure (21). Deficiencies in the WRN protein result in the premature aging associated with Werner syndrome patients (22), and mice deficient in p53 or in the proofreading exonuclease of DNA polymerase ␦ show high incidences of cancers (23)(24)(25). Eliminating both APE alleles in mice results in early embryonic lethality (26), and defects in the hMRE11 complex have been associated with an ataxiatelangiectasia-like disorder and the Nijmegen breakage syndrome characterized by chromosome instability, increased cancer incidence, cell cycle checkpoint defects, and ionizing radiation sensitivity (27,28).
The nonprocessive, autonomous nature of the TREX exonucleases led to the suggestion that these enzymes might serve as proofreading exonucleases for the nuclease-deficient replicative DNA polymerase ␣ (4) and the repair DNA polymerase ␤ (5) or for the DNA lesion bypass polymerase (30). This 3Јediting activity has been demonstrated in reconstituted systems for exonuclease-deficient DNA polymerases (2,5) and for the exonuclease-proficient DNA polymerase ␦ under adverse conditions (8). Although an in vivo proofreading function for the TREX enzymes has not been demonstrated, such an editing function might also eliminate replication blocks incurred upon polymerization of the 3Ј-blocking nucleoside analogs commonly used as antiviral or anticancer compound (2,3). In an effort to understand better this concept of exonucleolytic proofreading by autonomous 3Ј-exonucleases to edit kinetically blocked DNA 3Ј termini we have determined the x-ray structure of the human TREX2 dimer with data extending to 2.47 Å resolution. The TREX2 protein is structurally very similar to the bacterial ⑀ subunit of DNA polymerase III (⑀ subunit) (31,32) and exonuclease I (exoI) 2 (33) proteins despite their low sequence homology (ϳ17% identity). The conservation of this protein fold from bacteria to mammals indicates that the TREX2 structure is an efficient scaffold to orient the catalytic residues for excision at DNA 3Ј termini. The position of the active sites in the TREX2 dimer and the adjacent flexible region provides an explanation for the tight DNA binding and robust catalytic activity of the TREX2 enzyme.

EXPERIMENTAL PROCEDURES
Enzyme Preparation-The plasmid construct used to express the human TREX2 enzyme and mutants in Escherichia coli and purification of the TREX2 used for structural studies has been described (7).
For the TREX2 and mutants used in exonuclease assays a 10/10 Mono Q column (Amersham Biosciences) was substituted for the phosphocellulose column in the final step. The TREX2 sample in 50 mM Tris-HCl, pH 7.5, 1 mM EDTA, 10% glycerol was loaded onto the Mono Q column, washed with the equilibration buffer, and developed with a 100-ml linear gradient of 0 -300 mM NaCl in the equilibration buffer. The peak fraction containing the TREX2 protein eluted at ϳ250 mM NaCl at a concentration of ϳ0.5 mg/ml and was stored in aliquots at Ϫ80°C. The TREX2 concentrations were determined by A 280 using the molar extinction coefficient ⑀ ϭ 16,860 M Ϫ1 cm Ϫ1 (52). All TREX2 mutant plasmid constructs were produced using a PCR site-directed mutagenesis strategy (53), and the constructs were confirmed by DNA sequencing. Selenomethionyl-derivatized protein was expressed utilizing methionine pathway inhibition (54) and behaved nearly identically to native protein with respect to solubility. Approximately 2 mg of purified protein/ liter of cell culture was obtained for both the native and selenomethionyl protein.
Crystallization and X-ray Data Collection-The purified TREX2 protein was dialyzed into 20 mM Tris, pH 7.6, 150 mM NaCl, 20 mM dithiothreitol. Crystallization was carried out by the hanging drop, vapor diffusion method. Protein solution (5 mg/ml) was mixed with equal volume well solution (0.4 M ammonium phosphate, 7% PEG 400). Crystals grow in about 2 weeks but can be grown overnight with microseeding from existing crystals. Prior to data collection crystals were soaked in 0.4 M ammonium phosphate, 25% PEG 400 for 2 min in preparation for freezing. Crystals were then mounted in a nylon loop and flash frozen in liquid nitrogen. The crystals belong to space group P2 1 2 1 2 1 with unit cell dimensions a ϭ 52.5, b ϭ 77.2, c ϭ 101.7 Å. Two TREX2 molecules occupy the asymmetric unit (V M ϭ 2.0, 38% solvent).
Phasing and Refinement-X-ray data were collected at three x-ray energies corresponding to the selenium peak (0.97832 Å), inflection (0.97876 Å), and remote (0.95000 Å) wavelengths at beamline X12-C of the National Synchrotron Light Source, Brookhaven National Laboratories (Upton, NY). Intensity data were processed using the programs DENZO/SCALEPACK (55). Five ordered selenium sites were located in the crystallographic asymmetric unit using the program SOLVE (56). Multiwavelength anomalous dispersion phases were calculated with the program MLPHARE (57), and density modification was performed with RESOLVE (34). Model building was performed with the program O (58). The model was refined with no noncrystallographic constraints against the peak wavelength data set by conjugate gradient minimization and torsion angle-restrained molecular dynamics using the program CNS (59). The success of the refinement was evaluated at each stage by the change in the free R factor (60) and by inspection of stereochemical parameters with the programs PROCHECK (61) and ERRAT (29). Model refinement converged with a final R factor of 21.6% FIG. 1. The TREX2 protein forms a dimer. A, the structure of TREX2 shows a five-stranded, twisted, antiparallel ␤-sheet surrounded by 9 helices. The protein dimerizes about a 2-fold axis perpendicular to the ␤-strands to form an extended ␤-sheet that extends across the dimer (monomers shown in red and blue). B, TREX2 dimerization positions the active sites (yellow circles) of each monomer on opposite edges of the same face of the dimer. The TREX2 protein possesses little positive charge except in the region adjacent to the active sites. The active sites themselves appear to have an overall negative charge because of the presence of the four conserved acidic residues required for magnesium ion coordination (metal ions not present in structure). All ribbon diagrams and electron density shown in Fig. 2 were made with the program SETOR (35) and protein electrostatic potential surfaces made with the program GRASP (39).
(R free ϭ 26.1%), using all observed x-ray data measurements in the resolution range 25-2.47 Å. A Ramachandran plot shows that greater than 91% of all residues in the model have and angles in the most preferred regions with no residues in the disallowed regions. The coordinates of the TREX2 dimer have been deposited in the Protein Data Bank (PDB code 1Y97).
Protein Sequence and Structure Alignments-The structural coordinates of the E. coli epsilon protein (1J53), exoI (1FXX), and Klenow fragment of DNA polymerase I (1KSP) were obtained from the Protein Data Bank (www.rcsb.org) and superimposed on TREX2 model coordinates by a least squares superpositioning of homologous residues in the exonuclease active site. The structure-based sequence alignment of each 3Ј-exonuclease with TREX2 was created based on structural alignments using the program DALI (www.ebi.ac.uk/dali/).
Exonuclease Assays-The exonuclease activities of TREX2 and the catalytic mutants were measured in 10-l reactions using 50 nM 32 Plabeled (dT) 20 oligonucleotide substrate as described (7). For kinetic analysis the (dT) 20 concentrations were varied for TREX2 (5-500 nM) and for TREX2 R163A/R165A/R167A (1,000 -100,000 nM), and kinetic constants were determined as described previously (7). All incubations were 20 min at 25°C. Enzyme concentrations are indicated in the figure and table legends.

RESULTS
Structure of TREX2-The structure of the human TREX2 3Ј 3 5Ј-exonuclease was determined by multiwavelength anomalous dispersion experiments at three x-ray energies ( Fig.  1 and Table I) using a crystal of the selenomethionyl protein.
After density modification by solvent flattening (34) the electron density was of high quality for most residues in the asymmetric unit (Fig. 2). Three segments spanning residues 159 -168, 185-187, and 229 -236 were disordered in both monomers and omitted from the crystallographic model. The crystallographic model of the TREX2 dimer has been refined to an R factor of 21.6% (R free ϭ 26.1%) using all x-ray data extending to a resolution limit of 2.47 Å (Table I).
TREX2 is a compact single domain protein consisting of a central, five-stranded, antiparallel, twisted ␤-sheet surrounded by nine ␣-helices (Fig. 1A). The TREX2 protein dimerizes about a noncrystallographic 2-fold axis perpendicular to the ␤-strands, connecting the individual ␤-sheets of each monomer at the interface to form an extended ␤-sheet core through the observed dimer. Formation of the crystallographic dimer is consistent with the observation of TREX2 dimerization in solution (6) and is the first structure determined of a dimeric 3Ј-deoxyribonuclease. The TREX2 dimer observed in the crys-tal is likely the same as the one present in solution for several reasons. The interface between subunits of the dimer, consisting mainly of strand ␤3 and helix ␣4 of each monomer, buries about 1,500 Å 2 of surface area from each monomer. This substantial interface, which consists of nearly 10% of the surface area of each monomer, is formed predominantly by 19 hydrogen bonds with additional contributions from hydrophobic interactions and three salt bridges. The hydrogen bonds are formed mostly by the backbone amide nitrogens and carbonyl oxygens between the ␤3-strands in the creation of the extended ␤-sheet. The interactions in this interface far exceed all other protein packing interactions present in the crystal. Overall, the individual protomers in the dimer adopt structurally similar conformations with a root mean square difference of only 0.42 Å between the backbone atoms of each monomer. The minor differences present are mostly the result of the N-terminal six residues in one monomer having a different packing arrangement in the crystal than in the other monomer.
One result of the observed interaction is the creation of two faces on the dimer. The "back" face is relatively flat with scattered positive charge, whereas the "front" face of the dimer contains both active sites (Fig. 1B) on opposite outer edges. Each active site is a spacious cleft framed by strand ␤1 and helices ␣5, ␣7, and ␣8. The five conserved residues (Asp 14 , Glu 16   R merge ϭ ⌺͉I Ϫ ͗I͉͘/⌺I, where I is the observed intensity and ͗I͘ is the average intensity. R factor ϭ ⌺ʈF o ͉ Ϫ ͉F c ʈ / ⌺͉F o ͉. R free is the same as R but calculated with 10% of reflections that were never used in crystallographic refinement. Data collection statistics for the highest resolution shell are shown in parentheses.
active sites appear to have an overall negative charge because of the conserved four acidic residues in each required for coordination of the magnesium ions that are necessary for catalysis.
Comparison of TREX2 with Other 3Ј-Exonucleases-The primary sequence of TREX2 most closely relates to the DnaQ family of 3Ј-exonucleases by virtue of three conserved exonuclease sequence motifs (exoI, exoII, and exoIII) that identify the divalent metal binding residues required for catalytic activity (36 -38). The presence of histidine rather than tyrosine in the exoIII motif of TREX2 places this enzyme in a subgroup of the exonuclease family of which TREX1 and TREX2 are the only known human members (4, 40 -42). The overall fold of the TREX2 protein monomer is very similar to the bacterial ⑀ subunit and the exoI members of this subgroup. A comparison of the structures of the bacterial exonucleases with TREX2 shows the conservation of the five central ␤-strands and seven ␣-helices (Fig. 3). The TREX2 enzyme contains an additional helix, ␣1, not present in the bacterial exonucleases. Residues in helix ␣1 participate in hydrogen bonding interactions across the TREX2 dimer interface with residues of helix ␣3 of the opposing monomer. The presence of helix ␣1 in TREX2 and not FIG. 3. Overall folds of bacterial 3exonucleases and TREX2. A, the TREX2 monomer shows strong structural conservation to the E. coli exoI and epsilon proteins. Unique to TREX2, however, is helix ␣1 and a ␤-hairpin formed by strands ␤2a-2b. Dimerization of TREX2 involves, in part, hydrogen bonding interactions between residues in helix ␣1 and helix ␣3 in the opposing monomer. B, a structural sequence alignment of the TREX2, epsilon, and exoI proteins indicates positional conservation of catalytic residues (red), as well as the insertion of helix ␣1 and a ␤-hairpin between strands ␤2 and ␤3. Also notable is the dissimilarity of the ␣6 -␣7 loop region (indicated in orange). Arginines 163, 165, and 167 within the loop are unique to TREX2 and strongly contribute to its high affinity for DNA (see Table III). The secondary structure elements from TREX2 are shown above the sequences.
FIG. 4. The TREX2 active site contains the conserved residues of the exoI, exoII, and exoIII motifs (shown in red). The acidic residues (Asp 14 , Glu 16 , Asp 123 , and Asp 193 ) in concert with His 188 are proposed to coordinate two magnesium ions that are involved in DNA binding and are necessary for catalysis. The exo motifs are also structurally conserved among all other members of the DnaQ family of 3Ј-exonucleases, suggesting that they share a common catalytic mechanism. Shown here is the superposition of the active site residues from the E. coli epsilon (cyan) and exoI (green) proteins superimposed on the TREX2 (blue) structure (stereo figure). in the bacterial exonucleases likely accounts, in part, for the unique dimeric structure of TREX2. The strands ␤2a and ␤2b of TREX2 constitute a unique ␤-hairpin positioned centrally on the top and bottom of the front face of the TREX2 dimer (Fig. 3,  A and B). There is also clear structural dissimilarity between TREX2 and the bacterial enzymes in the loop region positioned between helices ␣6 and ␣7 of TREX2. A superposition of the structures of ⑀ subunit and exoI onto TREX2 shows a strict structural conservation of the residues that comprise the active site; the Asp 14 , Glu 16 , Asp 123 , and Asp 193 are properly situated in TREX2 to coordinate two divalent magnesium ions (Fig. 4). By analogy to the ⑀ subunit (31, 32) and exoI structures, (33) His 188 in TREX2 is well positioned to deprotonate a water molecule positioned adjacent to the target phosphate. The hydroxyl nucleophile could then attack the phosphate to affect bond cleavage (43). To test the catalytic role of the conserved carboxylate residues and the His 188 a series of TREX2 mutants was prepared by changing these residues to Ala. The exonuclease activities of these TREX2 mutants were measured in the presence of increased amounts of enzyme using a (dT) 20 oligonucleotide and compared with the wild-type TREX2 (Table II). A small amount of activity was detected for all of the TREX2 mutants tested. Quantification of these activities indicated 10 4 -10 6 -fold reductions in activity relative to the wild-type TREX2 (Table II). These data confirm the catalytic role of these residues in TREX2 and suggest that the TREX2 exonuclease utilizes a catalytic mechanism similar to that of the bacterial members of this subgroup in the DnaQ family.
Model for DNA Binding-We have modeled the proposed interactions of TREX2 with DNA by superimposing the crystal structure of the exonuclease domain of the Klenow fragment of DNA polymerase I in complex with phosphorothioate containing DNA (44) on the crystal structure of TREX2 (Fig.  5). The superposition of active site exo motif residues provides very good docking of the DNA substrate onto TREX2, with favorable electrostatic interactions and no steric clashes. The model shows several protein-DNA interactions including those to the conserved exo motifs. Notably, the model revealed interactions with several other residues in and around the active site which are structurally conserved among DnaQ exonucleases. In addition to the four metalbinding residues of the exo motifs, Arg 152 in TREX2, which is conserved in the ⑀ subunit and exoI, is adjacent to the active site and positioned perfectly to interact with the phosphate backbone of the modeled substrate (Fig. 5). The Asn 118 is also conserved in the bacterial 3Ј-exonuclease active sites and is positioned to hydrogen bond with the deoxyribose oxygen of the penultimate nucleotide. The Arg 152 and Asn 118 along with the metal ion coordinating interactions of Asp 14 , Glu 16 , Asp 123 , and Asp 193 comprising the exo motifs appear to provide the major electrostatic interactions with oligonucleotide substrate within the 3Ј-exonuclease active site of the TREX2 protein.
The TREX2 structure provides a model to explain the nonprocessive, yet highly efficient, catalytic activity of this 3Ј 3 5Ј-exonuclease. The 15-amino acid loop between helices ␣6 and ␣7 in TREX2 is directly adjacent to the active site. This loop is largely disordered, indicating that it is somewhat mobile, and it contains the positively charged arginines at positions 163, 165, and 167. The equivalent ␣6 -␣7 loop in the ⑀ subunit structure (Fig. 3), although ordered, is only 6 amino acids long and is tightly packed against the protein surface. Moreover, it contains only a single arginine (Arg 142 ) that points away from the active site in the structure (31,32). Steady-state kinetic analyses indicate that ⑀ subunit binds DNA poorly (K m ϭ 16 M), requiring high concentrations to saturate the enzyme (45,46). In stark contrast, the TREX2 exonuclease binds DNA with an apparent high affinity (6). To test the contribution of the ␣6 -␣7 loop arginines to DNA binding in TREX2 these three residues were mutated to alanines, and a steady-state kinetic analysis of the mutant enzyme was performed (Fig. 6). The TREX2 R163A/R165A/R167A protein was expressed and purified, and the kinetic analysis revealed a dramatic reduction in the apparent binding capacity for DNA as reflected in the K m value of 9,200 nM for (dT) 20 relative to the K m value of 61 nM measured for the wild-type TREX2 enzyme ( Fig. 6 and Table III). Mutation of the three arginines to alanines had no effect on catalysis of 3Ј-nucleotides as reflected in the measured k cat value of 8.1 s Ϫ1 for the mutant enzyme relative to the k cat value of 4.8 s Ϫ1 for the wild-type TREX2 enzyme ( Fig. 6 and Table  III). These kinetic values indicate a 90-fold reduction in the apparent second order rate constant for DNA in the TREX2 R163A/R165A/R167A enzyme and provide direct evidence for the involvement of these residues in DNA binding in the TREX2 enzyme.  5. Model of nucleotide binding to TREX2 (stereo figure). The model was constructed by a superposition of the structure of the exonuclease domain of DNA polymerase I in complex with phosphorothioate containing DNA (PDB code 1KSP) onto the TREX2 structure (green). The model shows good positioning of the DNA (gold) relative to the conserved catalytic residues (red) with no major steric clashes. Arg 152 , which is structurally conserved in the exoI and ⑀ structures, is properly positioned adjacent to the active site to interact with the phosphodiester backbone of the substrate DNA.  20 . Reaction products were subjected to electrophoresis on 23% polyacrylamide gels (insets). The products were quantified, and the kinetic constants (Table III) were determined as described under "Experimental Procedures."

DISCUSSION
The three-dimensional structure of the human TREX2 enzyme is the first to be determined of a dimeric 3Ј-deoxyribonuclease from any species. The structure shows that the TREX2 protein monomer contains a fold very similar to that of the subgroup members in the DnaQ family of bacterial exonucleases characterized by the presence of a catalytic His rather than Tyr. A comparison of the TREX2 structure with these bacterial members reveals impressive structural conservation in the positions of the carboxylate residues required for metal binding and two additional conserved residues likely in contact with the DNA in the active site region. Conservation of this catalytic center in TREX2 provides, in part, an explanation for the potent 3Ј 3 5Ј-exonuclease activity of TREX2. The k cat values for 3Ј-nucleotide excision reported for ⑀ subunit (46) and exoI (47) are comparable with those of TREX2. The high catalytic efficiency of TREX2 allowed us to quantify contributions of each residue at the catalytic center in our mutagenesis analysis (Table II). This analysis showed similar contributions of the individual residues with losses in activity ranging from 3.4 ϫ 10 4 for His 188 to 1.7 ϫ 10 5 for Glu 16 . The two double mutants that were tested (D14A/E16A and H188A/D193A) exhibited further diminutions of activities, and it seems likely that these losses in activity reflect the inability of the mutant enzymes to position metals and waters precisely at the active site of TREX2.
The structure of the TREX2 enzyme is consistent with the nonprocessive action of this exonuclease. On the surface of TREX2 there is no apparent binding groove that might function to enclose a DNA polymer partially, and there is no apparent extended region on the enzyme surface that might interact with a DNA polymer as has been suggested for enzymes exhibiting processive properties (48). Comparisons between TREX2 and the bacterial exonucleases identified the unique ␣6 -␣7 loop present in TREX2. Mutagenesis of the arginine residues on this loop indicates that these positively charged residues are directly involved in DNA substrate binding and not in 3Јnucleotide excision (Table III). These data suggest a model for TREX2 catalysis in which the negatively charged phosphodiester backbone of DNA is first bound by the positively charged arginine residues present on this flexible loop, and the DNA is subsequently moved into the active site for catalysis. Additional contributions to movement of the DNA 3Ј-end into the active site might include the apparent negative charge of the catalytic center. Involvement of all three of the arginines positioned at 163, 165, and 167 on the flexible loop is supported by additional mutagenesis studies that demonstrate contributions to DNA binding by each of the individual arginines in TREX2. 3 We estimate that a contribution of ϳ100-fold is made to the overall catalytic efficiency of TREX2 by the ␣6 -␣7 flexible loop region (Table III). Together, the efficient excision of 3Ј-terminal nucleotides facilitated by a presumed two-metal ion mechanism of phosphodiester bond cleavage similar to that proposed for the exonuclease of E. coli DNA polymerase I (43) and the efficient binding and movement of the DNA substrate into the active site by the ␣6 -␣7 flexible loop can account for the robust nature of the TREX2 enzyme.
The view of 3Ј-exonucleolytic proofreading of DNA has been mostly limited to the removal of misinserted nucleotides by the DNA polymerase-associated exonucleases. However, we now know that there are at least 14 DNA polymerases encoded in the human genome with as many as 13 functioning in the nucleus (49). Only 3 of these DNA polymerases are known to contain intrinsic 3Ј 3 5Ј-exonuclease activities. There are at least 8 autonomously acting 3Ј 3 5Ј-exonucleases in human cells (8,9) which might provide a proofreading function to edit DNA 3Ј termini. Such an autonomous proofreading activity might require a 3Ј 3 5Ј-exonuclease consistent with the catalytic properties of TREX2. Upon replication arrest resulting from blocked 3Ј termini, access to DNA ends might be facilitated through protein-protein interactions to move TREX2 to DNA 3Ј termini to alleviate replication blocks. The tight binding of TREX2 to the DNA 3Ј terminus coupled with the catalytically robust excision potential could affect the necessary excision event. The nonprocessive nature of TREX2 would further support the movement of TREX2 into and out of the active sites of DNA maintenance.
TREX Dimerization-The dimeric structure of the TREX exonuclease is likely conserved among all species expressing this enzyme, suggesting that the dimeric architecture of this enzyme plays a critical role in its biological function. Three of the ionic interactions that stabilize the TREX dimer consist of salt bridges formed by Arg 55 to Glu 29 , Arg 107 to Asp 94 , and Lys 59 to Glu 191 . In each case the interactions occur between  residues in opposing monomers, across the dimer interface. These six amino acids are strictly conserved in TREX protein sequences from other species but are not conserved among any of the monomeric members of the DnaQ family of 3Јexonucleases (Fig. 7). The obvious question is what evolutionary advantage a dimeric state would provide for this enzyme. The phenomenon of biological molecules self-assembling into multimeric states is quite common (for a review, see Ref. 50) and is often integral to structure, control, or biological function. The formation of a dimer can be a mechanism for enzyme regulation controlled by protein concentration. The dimerization of TREX may be required for activity, and indeed in vitro assays of TREX2 demonstrate the potent catalytic activity of the enzyme as a dimer. However, TREX enzymes are expressed at extremely low levels in the cell (6), and at this low concentration it is unknown whether the protein is dimeric. One possibility is that under certain conditions in the cell the low concentration of TREX allows it to remain a monomer and inactive, and only in response to a stimulus is the protein expression increased, allowing TREX to dimerize and become active. Although it is unknown whether the TREX enzyme is inactive as a monomer, the TREX2 structure hints at the possibility. The residue Glu 191 , conserved in all TREX proteins, seems to anchor the catalytic residues His 188 and Asp 193 structurally at the N terminus of helix ␣8 in the proper position for catalysis. Glu 191 , which is also positioned near the N terminus of helix ␣8, makes contacts with Lys 59 of the opposing monomer, and without this interaction this helix might be free to move and compromise the catalytic integrity of the enzyme. We are currently working on ways to interrupt the extensive interface of the TREX dimer to test the catalytic and DNA binding properties of the monomer.
Another possible reason for dimerization may be that two active sites are required for proper TREX cellular function. The pathological changes in lymphoid organs observed in TREX1-null mice has suggested a role for the TREX1 enzyme in processing of DNA termini during a subset of V(D)J recombination events, receptor editing, or class switch recombination (21). Receptor editing by the processing of DNA termini during recombination activating gene-mediated V(D)J recombination allows B cells to maintain tolerance and increase their repertoire (51). The possible need for simultaneous processing of two DNA termini prior to end joining or other DNA repair/recombinational events may necessitate a dimeric nuclease. The 2-fold symmetry of the TREX2 dimer, which places the active sites on opposite sides of the molecule, along with our model of TREX-DNA interaction (Fig. 5) suggest that it is perfectly suited for the simultaneous processing of two 3Ј-ends from opposite directions.