Structural basis for recognition of distinct deaminated DNA lesions by endonuclease Q

Significance

Deamination of cytosine and adenine bases in DNA is one of the most common sources of genetic mutation in biology and a major driver of virus and tumor evolution. Recent studies have identified a novel class of DNA-repair enzyme from archaea and bacteria, designated endonuclease Q (EndoQ), which has unique activity to initiate the repair of both these deaminated lesions. Our X-ray crystallographic and biochemical analyses presented here reveal how EndoQ achieves high selectivity toward these chemically divergent substrates without cleaving undamaged DNA. We also demonstrate that EndoQ, as the only known nuclease that cleaves DNA at deoxyuridine nucleotide, is useful for studying DNA deamination and repair in mammalian systems.

Abstract

Spontaneous deamination of DNA cytosine and adenine into uracil and hypoxanthine, respectively, causes C to T and A to G transition mutations if left unrepaired. Endonuclease Q (EndoQ) initiates the repair of these premutagenic DNA lesions in prokaryotes by cleaving the phosphodiester backbone 5′ of either uracil or hypoxanthine bases or an apurinic/apyrimidinic (AP) lesion generated by the excision of these damaged bases. To understand how EndoQ achieves selectivity toward these structurally diverse substrates without cleaving undamaged DNA, we determined the crystal structures of Pyrococcus furiosus EndoQ bound to DNA substrates containing uracil, hypoxanthine, or an AP lesion. The structures show that substrate engagement by EndoQ depends both on a highly distorted conformation of the DNA backbone, in which the target nucleotide is extruded out of the helix, and direct hydrogen bonds with the deaminated bases. A concerted swing motion of the zinc-binding and C-terminal helical domains of EndoQ toward its catalytic domain allows the enzyme to clamp down on a sharply bent DNA substrate, shaping a deep active-site pocket that accommodates the extruded deaminated base. Within this pocket, uracil and hypoxanthine bases interact with distinct sets of amino acid residues, with positioning mediated by an essential magnesium ion. The EndoQ–DNA complex structures reveal a unique mode of damaged DNA recognition and provide mechanistic insights into the initial step of DNA damage repair by the alternative excision repair pathway. Furthermore, we demonstrate that the unique activity of EndoQ is useful for studying DNA deamination and repair in mammalian systems.

Deamination of the nucleobases is one of the most common types of damage in DNA, which can result from spontaneous hydrolysis, nitrosative stress, or activities of cellular deaminase enzymes (1–3). The loss of the exocyclic amino group from cytosine (C) and adenine (A) bases in DNA generates uracil (U) and hypoxanthine (Hx), respectively, which mimics thymine (T) and guanine (G) in base-pairing capacity and if left unrepaired, leads to point mutations upon DNA replication (4, 5). Spontaneous deamination of cytosine is estimated to occur over 100 times per mammalian cell per day and is a significant source of mutation that accounts for approximately half of all known pathogenic single nucleotide polymorphisms in humans (6–9). Deamination of adenine base in DNA is slower but still occurring at 2 to 3% the rate of cytosine deamination (10, 11). Because of its sensitivity to temperature, the hydrolytic cytosine deamination is also thought to have contributed to the rapid evolution of organisms on a warm earth (12). Enzymatic cytosine deamination also contributes to human diseases; in multiple human cancers, genomic mutations introduced by the APOBEC family of single-stranded DNA cytosine deaminases play major roles in tumor evolution, promoting the development of metastases and chemotherapeutic resistance (13–15).

As premutagenic lesions, deaminated DNA bases are repaired by multiple cellular pathways. A highly conserved base excision repair pathway entails removal of U or Hx base from DNA by a lesion-specific N-glycosylase, followed by repair of the resulting apurinic/apyrimidinic (AP) lesion by the concerted action of DNA endonuclease, polymerase, and ligase enzymes (16–18). The Uracil-DNA glycosylase (UDG/UNG) and alkyladenine DNA glycosylase excises U and Hx from DNA, respectively (19, 20). In eukaryotes, U:G mismatches generated as a result of cytosine deamination are also repaired by the mismatch repair pathway (21). In prokaryotes, endonuclease V (EndoV) cleaves the second phosphodiester bond on the 3′ side of deoxyinosine (dI: 2′-deoxynucleotide form of Hx) in double-stranded DNA, which has been proposed to initiate an alternative excision repair pathway for deaminated purine bases (22–24).

Recent studies by Ishino and colleagues have identified a novel enzyme from archaea and some bacteria, designated endonuclease Q (EndoQ), which exhibits a unique dual specificity for U and Hx in DNA (25–27). EndoQ is structurally and mechanistically distinct from EndoV and cleaves the phosphodiester bond immediately 5′ of 2′-deoxyuridine (dU) or dI in either single-stranded or double-stranded DNA to generate 5′-phosphate and 3′-hydroxyl termini (27). In addition to DNA strands containing deaminated bases, EndoQ also cleaves the phosphodiester bond 5′ of AP lesions in DNA. However, EndoQ shows no activity toward normal (undamaged) DNA and, unlike EndoV (28), it does not recognize mismatched bases as substrates. The precise mechanisms underlying the selective recognition of deaminated pyrimidine and purine bases has remained unknown. In this study, we determined the crystal structures of EndoQ from a hyperthermophilic archaeon Pyrococcus furiosus (pfuEndoQ) in complex with dU, dI, and AP site-containing DNA substrates. These structures and supporting functional analyses combined reveal a unique mode of DNA lesion recognition by EndoQ and its utility in the studies of DNA deamination and repair in mammalian systems.

Results

Structure Determination of EndoQ–DNA Complexes.

To understand the structural basis for the unique substrate selectivity of EndoQ, we crystallized pfuEndoQ(1–400), which lacks the C-terminal proliferating cell nuclear antigen-interacting motif present in the full-length 424-residue pfuEndoQ (29), in complex with 27-bp double-stranded DNA substrates. The DNA substrates contained a central dU:G or dI:T mispair representing deaminated C:G or A:T base pairs, respectively, or an AP-site analog (1′,2′-dideoxyribonucleotide) opposite either G or T. To prevent DNA strand cleavages and enable the capture of stable enzyme–DNA substrate complexes, we used a catalytic mutant variant of EndoQ, D193N. The crystal structures were determined by molecular replacement phasing using the DNA-free EndoQ structure (30) and refined to resolution ranging from 2.3 to 3.1 Å (Fig. 1). The overall structures of pfuEndoQ bound to different DNA lesions are very similar to each other, with an rmsd of ∼0.3 Å for all protein main-chain and DNA backbone atoms. As observed in the DNA substrate-free structure (30), pfuEndoQ consists of three structural domains: The N-terminal polymerase and histidinol phosphatase (PHP) domain containing the Zn-bound endonuclease active site, the Zn-binding domain featuring a noncatalytic 4xCys-type zinc finger, and the C-terminal helical domain consisting of a bundle of α-helices with a protruding hairpin insertion (antiparallel β15/β16 sheet with a β-hairpin at the tip) (Figs. 1A and 2B).

Fig. 1.

Overall structure of the EndoQ–DNA complex. (A) Ribbon schematic with the three structural domains of pfuEndoQ colored differently. Zinc and magnesium ions are shown as gray and light-green spheres, respectively. The deaminated nucleotide (dU) in an extrahelical position and engaged by the enzyme is colored in red. (B) Protein surface colored based on the electrostatic potential (blue: positive; red: negative). Some of the key residues involved in DNA binding are indicated.

Fig. 2.

EndoQ–DNA interactions stabilizing the distorted DNA conformation. Zoomed views (A, C, D) are shown for three regions indicated by the correspondingly colored and labeled boxes in B. (A) C-terminal helical domain (pink) interacting with a distal region of the nondamaged DNA strand. Hydrogen bonds or salt bridges are indicated by yellow dashed lines. (B) Overall structure of the EndoQ–DNA complex with the three structural domains of pfuEndoQ colored differently. (C) A view along the DNA helical axis from the 5′ side of the scissile (damaged) strand, highlighting the insertion of Arg82 and Trp144 side chains into the DNA minor groove. (D) Minor groove engagement by the catalytic PHP domain (green). Note insertion of a loop centered on Lys15 and Ala16, which are stacked against an unstacked G-C base pair at the kink. (E) Further zoomed view of DNA around the flipped-out dU nucleotide from a different angle. Van der Waals radii for the DNA atoms are shown as transparent spheres whereas those for Lys15 and Ala16 are shown as dots.

Overall Structure.

The three structural domains of EndoQ together form a positively charged cleft that accommodates lesion-containing DNA substrates (Fig. 1B). The catalytic and Zn-binding domains interact extensively with the scissile (damage-containing) strand and engage the DNA minor groove adjacent to the lesion, whereas the C-terminal helical domain is positioned over the major groove and interacts primarily with a distal region of the nonscissile strand (Fig. 2). The EndoQ-bound DNA is sharply bent by ∼40° away from the catalytic domain. The damaged nucleotide is at the apex of the kink, rotated out of the double helix and buried in a deep pocket formed between the catalytic and Zn-binding domains (SI Appendix, Fig. S1). DNA double helices up- and downstream of the lesion show a regular base stacking, but there is a significant shift and roll (up to –4.5 Å and 66°, respectively) along the base pair step across the scissile phosphodiester bond, which results in complete unstacking of the bases at the kink. A catalytic domain loop centered on Lys15 and Ala16 inserted into the DNA minor groove makes hydrophobic or van der Waals contact with the face of an unstacked G-C base pair adjacent to the site of damage, which appears to stabilize the sharply bent DNA conformation with unstacked bases (Fig. 2 D and E).

A comparison between the DNA-bound and free EndoQ structures shows that the Zn-binding and C-terminal helical domains move in a concerted fashion with respect to the catalytic domain (displacement of the Zn position = 7.9 Å), swinging in toward DNA (Fig. 3A and Movies S1 and S2). The elongated β15/β16 hairpin protruding from the C-terminal helical domain contains a unique and highly conserved sequence motif consisting of four consecutive glycine residues (³⁷²GGGG³⁷⁵), which forms a type-I′ β-hairpin turn (31) and is sandwiched between an aromatic side chain Tyr253 and the Lys243-Tyr244 stretch from the Zn-binding domain (Fig. 3B). This interaction serves as a tether to couple the Zn-binding and C-terminal helical domains, likely facilitating concerted movement of these two domains in clamping down on DNA substrates.

Fig. 3.

A concerted domain movement of EndoQ in clamping down on DNA substrate. (A) Comparison between the DNA-free (light gray) (30) and DNA-bound (colored) EndoQ structures, superimposed with the catalytic domain as a reference. (B) A sharp β-hairpin turn at the ³⁷²GGGG³⁷⁵ sequence motif coupling the C-terminal helical domain (pink) to the Zn-binding domain (teal). Movies S1 and S2 show a morphed conformational transition between the DNA-bound (colored) and unbound (light gray) (30) states.

EndoQ–DNA Interactions.

EndoQ engages the damage-containing DNA strand both up and downstream of the lesion. Several nucleotides on the 3′ side of the lesion fit in a narrow groove formed on the catalytic domain surface, which leads to the deep active-site pocket (Fig. 1B). The backbone interaction in this region includes Ser18, Asn198, Arg201, and Arg204 from the catalytic domain. In addition, Tyr377, which follows the ³⁷²GGGG³⁷⁵ motif within the β15/β16 hairpin insertion from the C-terminal helical domain, is hydrogen-bonded to the phosphate 3′ of the damaged nucleotide. Notably, Lys243 from the Zn-binding domain also interacts with this phosphate and simultaneously with the 5′ phosphate 1 nucleotide upstream of the lesion, positioned close to each other across the rotated-out damaged nucleotide (Fig. 4). The neighboring residue Tyr244 from the Zn-binding domain and Thr145 from the catalytic domain are hydrogen-bonded to the backbone phosphate of the nucleotide 5′ to the lesion as well. These interactions with the flanking nucleotides stabilize a scrunched DNA backbone conformation with the damaged nucleotide extruded out of the helix. A similar conformation was also observed for the AP lesion-containing DNA, with its sugar-phosphate backbone flipped (Fig. 5). While EndoQ primarily interacts with the scissile DNA strand, which explains the efficient cleavage of both single- and double-stranded DNA substrates (SI Appendix, Fig. S1), a helix–hairpin–helix motif that includes Lys315, Gly316, Lys320, and Ala321 from the C-terminal helical domain forms a highly basic patch and binds to a region of the noncleaved strand distal to the site of damage (Fig. 2A). This is consistent with the greatly reduced DNA-binding affinity of the K320A mutant of pfuEndoQ reported earlier, although this interaction is not essential for the catalysis (30).

Fig. 4.

DNA backbone contacts surrounding the extruded deaminated nucleotide. EndoQ–DNA contacts supporting the scrunched DNA backbone conformation across the lesion are indicated by yellow dashed lines.

Fig. 5.

A comparison between the dU and AP substrate-bound structures. The dU-containing structure is shown in yellow, and the AP site-containing structure in cyan. A Zn ion in the active site and the magnesium ion mediating U/Hx recognition are shown as spheres with respective van der Waals radii.

Several structural elements from the catalytic domain of EndoQ are inserted into the DNA minor groove, which widens significantly (∼13 Å) compared to canonical B-form DNA (5∼6 Å). As mentioned above, the loop centered on Lys15 and Ala16 contacts the unstacked bases at the kink (Fig. 2 D and E). In addition, Arg82 makes hydrogen-bonds with a cytosine base and a sugar moiety from the third base pair from the site of damage, which is the only observed direct base contact made by EndoQ (Fig. 2C). Trp144 stacks against the deoxyribose moiety of the scissile DNA strand on the 5′ side of the lesion, which appears to help shape the DNA backbone around the active site (Fig. 4). The importance of this interaction is supported by a W144A amino acid substitution severely compromising the activity of EndoQ on both the dI and AP site-containing DNA substrates (30). Notably, while K15A amino acid substitution modestly affected the activity of pfuEndoQ, R82A substitution caused varying defects in its activity on dU, dI, and AP site-containing DNA substrates (Fig. 6), highlighting importance of the minor groove contacts on different types of DNA lesions.

Fig. 6.

DNA cleavage activities of pfuEndoQ mutant derivatives. Cleavage of 27-bp double-stranded DNA substrates containing either a deaminated mis-pair (dU:G or dI:T) or an abasic lesion (abasic nucleotide opposite T) by full-length wild-type pfuEndoQ or its point mutant derivatives. The DNA substrates at 0.2 μM were incubated at 60 °C for 10 min with 0.1 μM of the indicated proteins.

DNA Damage Recognition.

The U and Hx bases of the dU- and dI-containing DNA substrate, respectively, are flipped out of the double helix and fit in a deep pocket formed between the Zn-binding and catalytic domains of EndoQ (Figs. 1, 2 B and E, 4, and 5). The U base is recognized through a set of three hydrogen bonds: The O2 atom of the pyrimidine ring is bonded to the main-chain amide group of Lys243, N3 to the side chain hydroxyl group of Ser171, and O4 to the main chain amide group of Ser171 (Fig. 7A). The Hx base is also recognized through three hydrogen bonds, but not with the same set of amino acid residues; the purine ring O6 and N1 atoms are bonded to Ser171 main-chain amide and side-chain hydroxyl groups, respectively, whereas the N7 atom accepts a hydrogen bond from the main-chain amide group of Leu170 (Fig. 7B). These interactions by EndoQ with the deaminated bases are distinct from the recognition of U by UDG through hydrogen bonds with an Asn side chain in the active-site pocket (18, 32) or the recognition of Hx specifically in its lactim (enol) tautomeric form by EndoV (28, 33).

Fig. 7.

Specific recognition of deaminated DNA bases by EndoQ. (A) A set of hydrogen bonds with U base in the dU-bound structure. The magnesium and zinc ions are depicted as light green and gray spheres, respectively, showing their van der Waals radii. (B) A set of hydrogen bonds with Hx base in the dI-bound structure, in the same view as in A. (C) Superposition of the two (dU vs. dI) structures, with a transparent protein surface shown for EndoQ to highlight the pocket.

The U and Hx bases bound to EndoQ are oriented slightly differently inside the pocket, with an ∼30° difference in tilt of the base plane, to accommodate the unique hydrogen bonds (Figs. 7C and 8 and SI Appendix, Fig. S2). In addition to the polar contacts described above, the U and Hx bases are surrounded by nonpolar portions of His139, Gly169, Leu170, Lys243, and Tyr244, which line the pocket and make van der Waals or hydrophobic interaction with the deaminated nucleobases. Leu170 side-chain stacks on one face of the U and Hx base, whereas residues on the other side of the base plane (Thr142/Tyr244) make less intimate contact with the deaminated bases. As mentioned above, Lys243 and Tyr244 from the Zn-binding domain swing toward the catalytic domain and make key contacts with the DNA phosphate backbone. Concomitantly, these residues close down on each flipped-out nucleobase and shape the U/Hx-binding pocket, playing a key role in the recognition of deaminated bases (Figs. 3B, 4, and 7 A and B).

Fig. 8.

pfuEndoQ active site. (A) dU-bound structure with single Zn ion in the active site. (B) dI-bound structure with two Zn ions. (C) Proposed catalytic mechanism of EndoQ-mediated DNA strand cleavage. The pfuEndoQ residues essential for its endonuclease activity (30) are shown in this schematic. R, purine or pyrimidine nucleotide.

Consistent with the observed roles of Lys243 in both DNA backbone and base contacts, as well as structurally linking the Zn-binding and C-terminal helical domains, the K243A amino acid substitution was reported to abolish the endonuclease activity of pfuEndoQ on a dI-containing substrate (30). On the other hand, while pfuEndoQ Y244A was shown to be less active than the wild-type enzyme, Y244F mutant is fully active (30), suggesting that the DNA base or the protein β-hairpin interaction made by the aromatic side chain of Tyr244 is more important than the hydrogen-bond it makes with a DNA backbone phosphate. Among the other residues lining the U/Hx-binding pocket, Ser171 is absolutely conserved and plays a central role in the deaminated base recognition by making direct hydrogen bonds with the carbonyl groups of U and Hx bases introduced by the hydrolytic deamination reaction. As expected from these structural observations, our mutation analysis shows that S171A significantly affects the activities of pfuEndoQ on dU- and dI-containing substrates and, less severely, the cleavage of AP site-containing DNA (Fig. 6).

It was shown previously that DNA-binding of EndoQ is dependent on Mg²⁺ (27), but the role of the divalent cation besides Zn²⁺ for EndoQ activities was unknown. In the electron density map for all our EndoQ–DNA complexes, we observed strong density (2mFo-DFc peak > 8.0σ) for a putative metal ion, octahedrally coordinated by the main-chain carbonyl oxygen atoms of Gly169, Leu170, Ala172, and Leu237, and the side chains of Glu205 and Tyr299 (Fig. 7). Because we included 1 mM MgCl₂ in the EndoQ–DNA complex samples for all crystallization experiments, we interpreted this to be a bound magnesium ion. The magnesium-binding site is adjacent to the U/Hx-binding pocket and involves a stretch of amino acid residues that constitutes the lining of the U/Hx-binding pocket. Importantly, the octahedral Mg²⁺ coordination by the main-chain atoms of Gly169-Ala172 restricts the conformation of Leu170 and Ser171, which are key residues in the recognition of U and Hx bases through both main-chain and side-chain interactions. Thus, the magnesium ion is likely to play an important structural role by helping position key residues for recognition of deaminated bases in DNA. Because the Mg²⁺ site is only present in the closed DNA-bound conformation of EndoQ, the magnesium ion binding may also help coordinate the domain movements described above (Fig. 3 and Movies S1 and S2).

Enzyme Active Site.

The scissile phosphodiester bond 5′ of the extruded damaged nucleotide is engaged in the Zn-bound active site. Unexpectedly, we observed two distinct configurations of the EndoQ active site. The highest-resolution (2.3 Å) structure containing the dU:G mispair shows an active site with single Zn ion, tetrahedrally coordinated by Glu76, His84, His139, and the pro-Rp nonbridging oxygen atom of the scissile phosphodiester group. The other (pro-Sp) nonbridging oxygen atom is surrounded by His10, Arg114, His195, and Asp193 (mutated to Asn in our structures) (Fig. 8A and SI Appendix, Fig. S2). Electron density maps for the crystals containing an AP-site analog, albeit at lower resolution, also show the active site with single Zn ion. This Zn ion is observed in all our EndoQ–DNA complex structures, and the E76A, H84A, and H139A single-amino acid substitutions all completely abolish the DNA-cleavage activity of EndoQ (30). Thus, it is likely to be essential for stabilizing a highly positively charged transition state of the phosphodiester bond cleavage.

Interestingly, the structure with a dI:T mispair refined to 2.9 Å resolution uniquely shows a second Zn ion in the active site with less idealized coordination geometry, surrounded by His8, His10, Glu76, Asp193 (Asn), and the pro-Sp nonbridging oxygen atom of the scissile phosphodiester (Fig. 8B and SI Appendix, Fig. S2). The positions of the two Zn ions in this structure match those observed in the previously reported DNA-free structure of pfuEndoQ (30). A superposition of all EndoQ–DNA structures obtained in this study shows that despite an overall high similarity, the positioning of the dU, dI, and AP nucleotides have a considerable variation (pairwise rmsd of 0.75 [dU vs. dI], 1.0 [dU vs. AP:G], and 1.5 Å [dU vs. AP:T] for the sugar-phosphate backbone atoms of the damaged nucleotide, compared to 0.37 [dU vs. dI], 0.47 [dU vs. AP:G], and 0.55 Å [dU vs. AP:T] for those from all nucleotides), reflecting their different chemical structures and protein interactions (Figs. 5 and 7C). In particular, the variation in the positioning and orientation of the scissile phosphodiester group is likely to account for the differences observed in metal coordination.

Applications of the EndoQ Activity.

The unique biochemical activity of EndoQ may be useful in probing DNA base deamination in vitro or in mammalian cells. For example, in the widely used DNA oligonucleotide cleavage assay to examine the activity of DNA cytosine deamination in vitro (14), EndoQ could be substituted for the UDG treatment and subsequence heating steps to cleave DNA backbone at the site of deamination. As a proof of concept, we analyzed the DNA cytosine deamination activity and chemical inhibition of human APOBEC3B (A3B) using pfuEndoQ to selectively cleave dU-containing deamination product DNA. A fluorescently labeled single-stranded DNA containing the 5′-TC dinucleotide target motif was incubated with A3B in the absence or presence of a 2′-deoxyzebularine (dZ)-containing mechanism-based inhibitor of A3B (34), which was followed by treatment with wild-type pfuEndoQ. As expected, pfuEndoQ readily cleaved the deamination product of A3B, and the dZ-oligo inhibited the generation of this smaller DNA fragment in a dose-dependent fashion (Fig. 9). We confirmed that this was not due to the inhibition of the EndoQ nuclease activity by the dZ-based A3B inhibitor, as a control (lanes 7∼9). These results show that EndoQ serves as a useful reagent in biochemical assays to examine DNA deaminase activities.

Fig. 9.

Human APOBEC3B-mediated DNA deamination and its chemical inhibition probed using EndoQ. A fluorescently labeled 15-nucleotide single-stranded DNA substrate (dC-oligo) was subjected to deamination on its single cytosine base by APOBEC3B (A3B), which was subsequently cleaved by pfuEndoQ in this coupled assay. The dZ-containing oligo (TTTTdZAT) inhibited A3B activity in a dose-dependent fashion (lanes 3 to 6). The cleavage of dU-oligo, which mimics the product of deamination on dC-oligo by A3B, is not affected by TTTTdZAT (lanes 7 to 9).

To determine if EndoQ nuclease activity could be utilized in mammalian cells, we overexpressed and purified a Flag-tagged Thermococcus kodakarensis EndoQ (TkoEndoQ, 73% sequence identity to pfuEndoQ) (27) from 293T cells (SI Appendix, Fig. S3A). As a control, Flag-tagged enhanced green fluorescent protein (eGFP) was purified in parallel. TkoEndoQ purified from human cells showed robust nuclease activity on a ssDNA dU-containing substrate at both 37 °C and 60 °C (73% and 94% cleavage at 100 nM enzyme, respectively) (SI Appendix, Fig. S3C). Alternatively, activity on a double-stranded DNA dU-containing substrate was strongly improved at 60 °C compared to 37 °C (13 to 50% at 100 nM) (SI Appendix, Fig. S3D). To address whether EndoQ activity can be detected in cellular lysate of the human cells overexpressing EndoQ, we performed the nuclease activity assays using whole-cell lysate (SI Appendix, Fig. S3B). Although EndoQ activity was detected at both 37 °C and 60 °C on both double-stranded and single-stranded dU-containing DNA substrates, activity was improved at 60 °C (SI Appendix, Fig. S3 E and F). Together, these data demonstrate that the unique biochemical activity of EndoQ may be useful in mammalian cells, for example, with a chromatin immunoprecipitation-sequencing approach to identify the sites of base deaminations in the genomic DNA.

Discussion

Our structural results suggest that DNA binding and cleavage by EndoQ depends on a highly distorted conformation of the substrate DNA strand, in which the sugar-phosphate backbone of the target nucleotide is rotated out to present the scissile phosphodiester group to the Zn-bound active site. This would be accommodated readily by an abasic (AP) lesion, which lacks the base moiety to clash into the protein. In addition, the unique U/Hx-recognition pocket adjacent to the active site allows EndoQ to engage dU and dI in DNA as well (Fig. 7). However, the observed mode of DNA substrate engagement, in which the base of the target nucleotide is flipped toward the protein, precludes other nucleotides from binding into the active site (Fig. 10 and Movies S3 and S4). T and G bases have similar hydrogen-bonding potentials to U and Hx, respectively, but neither would fit in the pocket for steric reasons: The methyl group of T would clash into His139 side chain and Gly169 C_α atom, and similarly the amino group of G would have a clash with Lys243. On the other hand, if adenine (A) and cytosine (C) bases were to be positioned similarly to the near isosteric Hx and U, respectively, their amino groups would clash with the main-chain amide group of Ser171. Thus, none of the normal (undamaged) nucleotides in DNA can be recognized as substrates for EndoQ (SI Appendix, Fig. S4). We reason that the endonuclease activity of EndoQ is restricted to dU, dI, and AP sites or their close chemical analogs with relatively small substitutions (e.g., 5,6-dihydrouracil, 5-hydroxyuracil) (35), based on a negative selection by steric exclusion and a positive selection by specific hydrogen bonding within the pocket.

Fig. 10.

Modeling normal (undamaged) bases in the U/Hx-binding pocket of EndoQ. Pyrimidine (T and C) bases in B and C were modeled based on superposition to U in the dU-bound structure (A), whereas purine (G and A) bases in E and F were modeled based on superposition to Hx in the dI-bound structure (D). Movies S3 and S4 show these models from various orientations. Clashing atoms are highlighted by transparent spheres showing their van der Waals radii.

Various DNA repair or modification enzymes have been shown to bind double-stranded DNA with a nucleobase flipped into an extrahelical position, where the resulting increased accessibility of the base facilitates molecular recognition and catalysis (36–38). A hallmark feature of these enzymes is the insertion of an amino acid “wedge” into the base stack, filling the gap generated by base-flipping. For example, UDG inserts a leucine side chain into the DNA base stack to replace the flipped-out U, and bacterial EndoV inserts a tyrosine side chain to replace the flipped-out Hx base (33, 39, 40). Similarly, two different classes of AP-endonucleases promote nucleotide-flipping through intercalation of an arginine side chain into the DNA double helix (41, 42). In contrast, base-flipping by EndoQ does not involve insertion of an amino acid residue into the base stack, but is rather driven by extensive backbone and minor groove contacts to unstack base pairs and force the extrusion of the damaged nucleotide (Fig. 4). Base-flipping facilitated by DNA backbone distortion but without protein side chain intercalation has also been observed for a restriction endonuclease (43). The highly distorted structure of DNA bound to EndoQ suggests that substrate recognition by EndoQ would benefit from instability of the DNA double helix at the lesion, consistent with recent biochemical studies showing a negative correlation between base pair stability and EndoQ cleavage rates as well as rapid cleavage of DNA with AP-sites (35). PfuEndoQ was reported to bind dU-containing double-stranded DNA more tightly than dI-containing double-stranded DNA (30), which may reflect difference in the thermodynamics or kinetics of the flipping of U versus Hx bases. We observed a significant enhancement of TkoEndoQ cleavage activity for double-stranded DNA substrates by raising the reaction temperature (60 °C vs. 37 °C) (SI Appendix, Fig. S3 B and C), consistent with a requirement for thermal energy in distorting the double-helical DNA helix.

The active site of EndoQ contains two Zn ions in its DNA-free state (30), one fewer compared to the typical trinuclear active site containing three Zn or Fe ions for other PHP domain family enzymes (44–46). Unexpectedly, we observed only one of these Zn ions in most of our DNA-bound EndoQ structures, coordinated by Glu76, His84, His139 (Fig. 8A). Because all of these amino acids were shown to be essential for the endonuclease activity of pfuEndoQ (30), this Zn ion is very likely to be catalytically required. On the other hand, the role for the second Zn ion, which is coordinated by His8, His10, Glu76, and Asp193 (mutated to Asn in our structures) and was only observed in the dI-containing structure (Fig. 8B), is less clear. It is possible that this second Zn ion is also important for catalysis, perhaps particularly on dI-containing DNA substrates. However, H8A and H10A single-amino acid substitutions only modestly affect the endonuclease activity of pfuEndoQ toward a dI-containing substrate, only ∼50% and ∼30% reduction respectively, suggesting that the second Zn ion is not essential (30). We therefore hypothesize the following catalytic mechanism (Fig. 8C). Asp193 plays a role as the general base in deprotonating a water molecule for in-line S_N2 nucleophilic attack of the scissile phosphodiester bond. The resulting highly negatively charged pentavalent intermediate would be stabilized by the directly coordinated Zn ion and positive charges of the surrounding residues, including Arg114 and His195. These two residues were shown to be catalytically essential (30). His10 may also play a role in stabilizing the transition state. His8 may interact with Asp193 to facilitate its positioning or the catalytic function. This mechanism is distinct from the two Mg²⁺/Mn²⁺ ion-dependent catalytic mechanism employed by EndoV, whose active site consists of a conserved set of carboxylate (Asp and Glu) residues (33, 47). However, further studies are needed to fully understand the catalytic mechanism of EndoQ, which could also involve transiently bound metal ion cofactors as observed for mouse EndoV (47).

In summary, our work reveals a structural basis for the novel mode of recognition of diverse deaminated lesions by EndoQ, and further suggests that this unique enzymatic activity could be leveraged in biochemical studies or potentially for probing DNA deamination in human cells.

Materials and Methods

Protein Purification.

Full-length PfuEndoQ, PfuEndoQ(1–400), and their mutant derivatives were expressed in Escherichia coli strain BL21(DE3) with a C-terminal noncleavable His-tag (LEHHHHHH), using the pET-24a expression vector and codon-optimized coding sequences. The proteins were expressed in LB medium supplemented with 50 μM ZnCl₂ by induction with 0.5 mM isopropyl β-d-1-thiogalactopyranoside for ∼16 h at 18 °C. All proteins were purified from the soluble fraction of E. coli lysate using nickel-affinity and Superdex 75 size-exclusion chromatography. The purified proteins in 20 mM Tris⋅HCl (pH 7.4), 0.2 M NaCl, 5 mM β-mercaptoethanol were concentrated by ultrafiltration, flash-frozen in liquid nitrogen, and stored at −80 °C. Protein concentrations were determined based on UV absorption and theoretical extinction coefficient calculated from the amino acid sequences.

Crystallization.

PfuEndoQ(1–400) D193N was mixed with an equimolar amount of blunt-ended 27-bp substrate (top strand: 5′- GCAGACCGACGACXTGTAGCGAACGAC-3′ annealed with bottom strand: 5′- GTC​GTT​CGC​TAC​AYGTC​GTC​GGT​CTG​C-3′, boldface letters denote the damaged base-pair) at an approximate protein concentration of 20 mg mL⁻¹ in the final condition of 15 mM Tris⋅HCl pH 7.4, 0.15 M NaCl, 1 mM MgCl₂, and 4 mM β-mercaptoethanol. Crystals with a substrate representing the deaminated C:G pair (boldface X = dU, Y = G) was obtained by the hanging-drop vapor diffusion method using a reservoir solution containing 0.2 M tripotassium citrate and 15% (wt/vol) polyethylene glycol (PEG) 3,350. In addition to native crystals, we obtained crystals of the dU:G complex with the bottom strand labeled with bromine (5′-GTCGTTCGCTACAGG/5-Br-dU/CGTCGGTCTGC-3′), which helped to confirm the register of DNA bases in the complex as well as improve resolution (SI Appendix, Fig. S5). Crystals with a substrate representing the deaminated A:T pair (X = dI, Y = T) were obtained using the reservoir solution containing 0.125 M sodium fluoride, 0.125 M trisodium citrate, and 12.5% (wt/vol) PEG3,350. Crystals with a substrate mimicking the abasic lesion (X = 1′,2′-dideoxyribonucleotide, Y = G or T) were grown using reservoir solutions containing 20% (wt/vol) PEG3,350 and either 0.1 M tripotassium citrate or 75 mM each of sodium fluoride and trisodium citrate.

Structure Determination and Refinement.

The crystals were cryoprotected in the reservoir solutions supplemented by 25% of ethylene glycol and flash‐cooled in liquid nitrogen. X‐ray diffraction data were collected at the NE‐CAT beamline 24-ID-C and 24‐ID‐E of the Advanced Photon Source (Argonne National Laboratory, Lemont, IL). The X-ray wavelength used was 0.9191 Å for the dU:G complex containing 5-Br-dU (SI Appendix, Fig. S5) and 0.9791 Å for all other crystals. All datasets were processed using XDS (48). Because our EndoQ–DNA complex crystals had severely anisotropic diffraction, the data were subjected to ellipsoidal truncation and anisotropic scaling (49) prior to structure calculations. Molecular replacement was performed with PHASER (50) using the crystal structure of DNA-free EndoQ (PDB ID code 5ZB8) (30) and a 27-mer B-DNA. Model building was performed using COOT (51) and the structure refinement using PHENIX (52). The data collection and model refinement statistics are summarized in Table 1. DNA helical parameters were calculated using CURVES (53).

Table 1.

Data collection and refinement statistics

	dU:G (7K30)	dI:T (7K31)	abasic:G (7K32)	abasic:T (7K33)
Data collection
Wavelength (Å)	0.9191	0.9791	0.9791	0.9791
Space group	H3	H3	H3	H3
Unit cell (a, b, c in Å)	152.70, 152.70, 118.17	151.12, 151.12, 117.58	154.03, 154.03, 118.51	151.50, 151.50, 119.38
Total reflections	509,733 (47,179)	51,018 (5,259)	41,306 (4,168)	46,128 (4,621)
Resolution range (Å)	88.1–2.32 (2.40–2.32)	87.5–2.9 (3.0–2.9)	77.0–3.1 (3.2–3.1)	88.3–3.1 (3.2–3.1)
Unique reflections	44,238 (4,311)	21,492 (2,238)	17,827 (1,343)	17,765 (1,840)
Completeness (%)	99.50 (98.90)	94.20 (96.10)	94.70 (67.20)	95.11 (95.47)
Multiplicity	11.5 (10.8)	2.4 (2.4)	2.3 (2.3)	2.6 (2.6)
I/σ(I)	11.92 (0.50)	10.64 (1.54)	7.62 (0.95)	7.83 (0.80)
Rmerge (%)	14.39 (444)	5.63 (78.2)	11.4 (106)	7.70 (130)
Rmeas (%)	15.02 (466)	7.28 (101)	15.3 (140)	9.80 (164)
Rpim (%)	4.31 (140)	4.56 (63.45)	10.0 (92.4)	5.90 (99)
CC1/2	0.999 (0.644)	0.992 (0.876)	0.980 (0.738)	0.998 (0.705)
After anisotropic correction
High-resolution limit along a*, b*, c* (Å)	2.9, 2.9, 2.34 (2.42–2.34)	3.4, 3.4, 2.9 (3.0–2.9)	3.7, 3.7, 3.1 (3.2–3.1)	3.8, 3.7, 3.1 (3.2–3.1)
Unique reflections	28,597 (143)	15,268 (127)	12,475 (89)	12,345 (97)
Completeness (%)	65.99 (3.29)	67.51 (5.57)	65.55 (4.70)	67.40 (5.30)
Refinement
Reflections	28,591	15,265	12,328	12,336
No. for Rfree	1,411	790	562	618
Rwork/Rfree	20.55/22.84	18.2/23.7	19.9/24.7	19.1/23.0
No. of non-H atoms	4,301	4,253	4,218	4,228
Macromolecules	4,208	4,214	4,215	4,225
Ligands	38	29	3	3
Solvent	55	10
Protein residues	395	395	395	395
RMSD
Bond lengths (Å)	0.004	0.007	0.003	0.003
Bond angles (Å)	0.67	0.85	0.52	0.49
Ramachandran plot
Favored (%)	95.93	97.71	93.64	91.60
Allowed (%)	4.07	2.29	5.85	7.64
Outliers (%)	0.00	0.00	0.51	0.76
Average B-factor	70.52	58.63	131.86	145.29
Protein	53.99	42.88	114.13	125.64
DNA	120.13	105.25	182.65	201.80
Ligands	53.78	41.26	145.57	129.77
Solvent	45.97	12.73

Statistics for the highest-resolution shell are shown in parentheses.

DNA Cleavage Assay.

Fluorescein-labeled 27-base DNA oligonucleotides containing dU, dI, or the abasic site analog in the middle were like those used in the crystallization experiments described above. The “undamaged” substrates (SI Appendix, Fig. S4) were unmodified oligonucleotides. These top strands were annealed with unlabeled bottom strands to form double-stranded DNA substrates. Various DNA substrates at 0.2 μM were mixed with 0.1 μM of full-length wild-type pfuEndoQ or its mutant derivatives in 50 mM Tris–HCl (pH 8.0), 1 mM MgCl₂ and 0.01% Tween-20 as reported previously (30). After incubation for 10 min at 60 °C, the reactions were stopped by the addition of formamide to 80% and heating to 100 °C. The reaction products were separated on a 15% TBE-Urea gel, which was scanned on a Typhoon FLA 9500 imager to visualize the DNAs.

DNA Deamination Assay.

A 5′-fluorescein-labeled 15-base DNA oligonucleotide with single TC target sequence for APOBEC3B-mediated deamination (5′-TAG​GTCATT​ATT​GTG-3′, the deaminated base is underlined) at 0.2 μM was incubated with 1.0 μM of a double-mutant (L230K/F308K) of the human APOBEC3B catalytic domain (A3B-DM) (54) in the presence of 0 to 10 μM the dZ-containing mechanism-based inhibitor of A3B (34), 5′-TTTTdZAT-3′. After incubation at 37 °C for 70 min, 1.0 μM wild-type pfuEndoQ was added and the mixture was further incubated at 60 °C for 10 min. The control substrate (5′-TAGGTdUATTATTGTG-3′) represented the product of APOBEC3B-mediated deamination. The reactions were stopped and the products analyzed as described above for the DNA cleavage assay.

Human Cell Experiments.

HEK 293T cells were obtained from ATCC (#CRL-3216) and were maintained in RPMI (HyClone) supplemented with 10% FBS (Gibco) and 0.5% pen/strep (50 units). C-terminal 2xStrep and 3xFlag-tagged eGFP construct was previously described (55). T. kodakarensis EndoQ gene was codon-optimized for mammalian expression and cloned into pcDNA4/TO-3xFlag using KpnI and NotI restriction sites. All constructs were validated using Sanger sequencing. For protein expressions, 293T cells were transfected with pcDNA4/TO-GFP-3xStrep3xFlag or EndoQ-3xFlag using Transit LT1 (Mirus). Cells were harvested in 1× PBS 48 h posttransfection. Cells were washed two times with 1xPBS followed by lysis (150 nM NaCl, 50 mM Tris⋅HCl pH 7.4, 0.5% Tergitol, 1× Protease Inhibitor Mixture [Roche], RNase and DNase). Lysates were cleared by centrifugation and then added to equilibrated Flag M2 resin (Sigma) followed by end-over-end rotations overnight at 4 °C. Following immunoprecipitation, the anti-Flag resin was washed five times in wash buffer (50 mM Tris⋅HCl pH 8.0, 1 mM MgCl_2, 200 mM NaCl, 0.01% Tween 20, and 1× Protease Inhibitor Mixture). Protein was eluted from the resin in elution buffer (wash buffer + Flag peptide [Sigma]). Purified protein was validated using Coomassie staining and quantitated against a BSA gradient.

DNA Cleavage Assay on Proteins from Human Cells.

Fluorescein-labeled 27-base DNA oligonucleotides containing dU in the middle were like those used in the crystallization experiments. These top strands were annealed with unlabeled bottom strands to form double-stranded DNA substrates. For whole-cell lysate assays 293T cells were transfected with pcDNA4/TO-GFP-3xStrep3xFlag or EndoQ-3xFlag using Transit LT1 (Mirus). Cells were harvested in 1× PBS 48 h posttransfection. DNA cleavage reactions were performed at either 37 °C or 60 °C for 120 min for whole-cell lysate (0.025 U of UGI [New England Biolabs] was added to lysates) or 60 min for purified protein, 0.2 μM of DNA substrate in 50 mM Tris–HCl pH 8.0, 1 mM MgCl₂ and 0.01% Tween-20, as reported previously (30). The reactions were stopped by the addition of formamide and heating to 100 °C. The reaction products were separated on a 20% TBE-Urea gel, which was scanned on a Fujifilm FLA 7000 imager to visualize the DNAs and quantified using ImageJ.

Data Availability

The atomic coordinates and structure factors have been deposited in the RCSB Protein Data Bank, https://www.rcsb.org/ (PDB ID codes 7K30, 7K31, 7K32, and 7K33, for the dU:G, dI:T, AP:G, and AP:T complex structures, respectively).