Caught in motion: human NTHL1 undergoes interdomain rearrangement necessary for catalysis.

Base excision repair (BER) is the main pathway protecting cells from the continuous damage to DNA inflicted by reactive oxygen species. BER is initiated by DNA glycosylases, each of which repairs a particular class of base damage. NTHL1, a bifunctional DNA glycosylase, possesses both glycolytic and β-lytic activities with a preference for oxidized pyrimidine substrates. Defects in human NTHL1 drive a class of polyposis colorectal cancer. We report the first X-ray crystal structure of hNTHL1, revealing an open conformation not previously observed in the bacterial orthologs. In this conformation, the six-helical barrel domain comprising the helix-hairpin-helix (HhH) DNA binding motif is tipped away from the iron sulphur cluster-containing domain, requiring a conformational change to assemble a catalytic site upon DNA binding. We found that the flexibility of hNTHL1 and its ability to adopt an open configuration can be attributed to an interdomain linker. Swapping the human linker sequence for that of Escherichia coli yielded a protein chimera that crystallized in a closed conformation and had a reduced activity on lesion-containing DNA. This large scale interdomain rearrangement during catalysis is unprecedented for a HhH superfamily DNA glycosylase and provides important insight into the molecular mechanism of hNTHL1.

INTRODUCTION

The base excision repair (BER) pathway recognizes and repairs oxidized or alkylated DNA lesions, as well as mispaired uracil or thymine bases (1–3). This repair pathway entails a ‘hand off’ of the DNA substrate in sequential steps (4,5). DNA glycosylases initiate the BER pathway by probing, recognizing, and excising DNA lesions. These enzymes are classified into two groups, monofunctional and bifunctional. The monofunctional enzymes possess only glycosylase activity, leaving an abasic site upon removal of the DNA lesion, whereas the bifunctional glycosylases have an additional lyase activity, which nicks the DNA backbone either once (β-elimination, leaving a 3′-aldehyde; as is the case for endonuclease III (Endo III, Nth), or twice (β,δ-elimination, leaving a 3′-phosphate), thus creating a single-strand break (SSB) (6,7). Apurinic endonuclease 1 (APE1) or polynucleotide kinase (PNK) process the respective products, leaving a free 3′ hydroxyl for DNA polymerase β (Pol β) to insert the correct nucleotides using the undamaged strand as template. The nick is subsequently sealed by the DNA ligase IIIα (LIGIII)-X-ray repair cross complimenting 1 (XRCC1) complex (8). DNA glycosylases have a high affinity for their product and remain bound to the highly reactive abasic (AP) site until the hand-off to APE1 (9–13). The BER machinery likely does not scan the DNA as a complex but rather the transient interactions aid in recruitment of the downstream enzymes protecting the cells from potentially lethal intermediates (14,15).

Nth is a bifunctional DNA glycosylase with a marked preference for oxidized pyrimidine substrates (16). While both Escherichia coli Nth (EcoNth) and human NTHL1 (hNTHL1) glycosylases display a preference for excising lesions opposite G, the human enzyme is slower than the bacterial enzyme on most substrates (17–19). The first bacterial Nth crystal structure, EcoNth, revealed that the enzyme consists of two globular α-helical domains, a six-helical bundle domain and a [4Fe4S] cluster domain. This crystal structure identified the first HhH motif-containing glycosylase and the first [4Fe4S] cluster in a DNA-binding protein (20). Later, the Geobacillus stearothermophilus Nth was crystallized in the presence of DNA in two forms, bound to the non-hydrolysable AP site analogue, tetrahydrofuran (THF), and as a covalently trapped intermediate. Nth glycosylases harbour two residues important for catalysis, an aspartate and a lysine: The aspartate acts as the carboxylate anion during catalysis while the lysine side chain forms a transient covalent Schiff-base intermediate with the abasite site. The crystal structures showed that DNA binds along the cleft between the two domains, with the lysine located in the six-helical barrel domain and the aspartate in the [4Fe4S] cluster domain. The Nth DNA-bound models also revealed extensive contacts of the [4Fe4S] domain to the DNA backbone (21).

The BER pathway has been implicated in the progression of cancers because mutations in BER enzymes can reduce the effectiveness of the variants to repair damaged DNA, resulting in mutations (22). Loss of NTHL1 is a driver of adenomas and other tumour types (see review (23)). Whole exome sequencing has directly linked NTHL1 to familial inherited colorectal cancer (CRC), and adenomatous polyposis (24,25). Multiple studies have found that NTHL1-associated polyposis (NAP) is associated with biallelic germline nonsense mutations that renders the DNA glycosylase inactive. Biallelic mutations have been found in 14 different tumour types, notably colorectal, breast and endometrial cancers (24,26–31). Weren et al. estimate the prevalence of NAP at 1 in 114 770 in individuals of European decent (32). More recently, a paper reported a 1.9% prevalence of NTHL1 biallelic mutations in polyposis patients, which typically present as adenomas potentially with CRC, serrated polyps, and multi-tumour phenotypes (33). The NAP tumours revealed a strong C > T transition pattern (24,34). NTHL1 deficiency was also identified as the root of COSMIC mutation signature 30, using human intestinal organoids (35). Signature 30 was also identified in some breast cancers and, retrospectively, the breast tumour in which signature 30 was identified was determined to be NTHL1-deficient (36). Recently, four more breast tumours, where signature 30 accounts for 80% of the mutations, had an NTHL1 deficiency, suggesting that lack of NTHL1 had driven the formation of these tumours (28). Transcriptome sequencing of a pancreatic neuroendocrine tumour revealed signature 30 and NTHL1 loss, implicating NTHL1 deficiency as a driver of another tumour type (37). It was also reported that overexpression of NTHL1 causes genomic instability (38).

The bacterial and eukaryotic homologs share limited sequence homology, around 30%, and notable differences in activity. Mammalian NTHL1 DNA glycosylases, including hNTHL1, harbour a disordered N-terminal extension, which encompasses nuclear and mitochondrial localization sequences; it is also a site of post-translational modifications and has been posited to play a part in protein-protein interactions (39–41). Additionally, hNTHL1 exhibits apparent positive cooperativity that has not been observed in the bacterial homologs (39,41). These differences, and the need for structural mapping of cancer variants, highlight the need for a crystal structure of the human enzyme. Here, we report the first crystal structure of hNTHL1, which was captured in a novel open conformation. A large-scale conformational change would be necessary to accomplish catalysis. We show that a protein segment linking the two domains is necessary for the open conformation and that the freedom of movement between the two domains facilitates the excision of thymine glycol (Tg) from DNA oligomers.

MATERIALS AND METHODS

Expression of hNTHL1

The protein constructs, full-length hNTHL1, hNTHL1Δ63 (lacking the first 63 residues) and the hNTHL1Δ63 chimera (lacking the first 63 residues and harbouring a shorter linker similar to the bacterial sequence), were overexpressed from the pET30a vector in E. coli Rosetta2 DE3 pLysS cells (Novagen), using autoinduction as previously described (42). Briefly, cells were grown in Terrific Broth media supplemented with 5052 sugar mix, kanamycin, and chloramphenicol at 20°C for 60 hours. The cells were lysed by sonication at 4°C in 500 mM NaCl, 20 mM Tris pH 8, 20 mM imidazole, 10% (v/v) glycerol, 3 mM β-mercaptoethanol, 1 mM PMSF. The cell lysate was cleared at 23 000 × g for 1 h and then passed over a Ni-NTA resin using gravity flow. The protein was eluted using 5 column volumes (CV) of elution buffer, 100 mM NaCl, 20 mM Tris–HCl pH 8, 250 mM imidazole, 10% glycerol, and 3 mM β-mercaptoethanol. The protein was further purified over a heparin column (Cytiva), in 100 mM NaCl, 20 mM Tris–HCl pH 8, 20 mM imidazole, 10% glycerol, 1 mM TCEP with a 20 CV salt gradient (0.1–1M NaCl). The protein typically elutes around 300 mM NaCl. A Superdex 75 gel filtration column (Cytiva) was used for the final purification step with a buffer composed of 100 mM NaCl, 20 mM HEPES pH 8, 10% glycerol and 1 mM TCEP. Proteins were concentrated to ∼8 mg/ml using a 30 000 Dalton cut-off centrifugal filter unit (Amicon), flash frozen in LN2, and stored at −80°C.

Selenomethionyl-protein purification

The pET30a hNTHL1Δ63 construct was overexpressed in E. coli Rosetta2 DE3 pLysS cells (Novagen), in minimal medium containing selenomethionine at 125 μg/mL, as described in (43). Briefly, the cells were grown to an optical density of 0.6 at 37°C, and then induced with 500 mM IPTG at 25°C for 4 hours. The protein was purified using the procedure described above.

Purification of DNA substrates

The following DNA substrates were chemically synthesized (IDT): 35mer oligonucleotides were used: damaged strand 5′ TGTCAATAGCAAG(X)GGAGAAGTCAATCGTGAGTCT 3′ and complementary strand 5′ AGACTCACGATTGACTTCTCC(G/A)CTTGCTATTGACA 3′, where X is the DNA lesion, either tetrahydrofuran (THF) or thymine glycol (Tg). (G/A) represents the opposite base. Oligonucleotides were resuspended in 200 μl TE buffer and 800 μl formamide and purified by electrophoresis on a 30% polyacrylamide-urea gel (National Diagnostics), run at 55 W for 6 h. The bands were cut out of the gel, crushed, and soaked overnight in 50 mM NaCl and 50 mM Tris–HCl pH 8. The gel pieces were filtered out and the solution was run over a SepPak C18 column (Waters). The oligonucleotides were eluted in 75% acetonitrile. The acetonitrile was evaporated using a Speedvac, and the oligonucleotides were resuspended in annealing buffer (50 mM NaCl, 50 mM Tris–HCl pH 8). The oligonucleotides were annealed at an equimolar ratio using a hot water bath at 95°C and allowed to cool to room temperature slowly in the water.

Crystallization of hNTHL1

Full-length NTHL1 failed to crystallize, presumably due to the presence of a flexible N-terminal extension. We therefore attempted to crystallize an N-terminal deletion construct missing the first 63 residues, hNTHL1Δ63. We chose this construct because it expresses well, is soluble, and is as active as full-length hNTHL1 (Supplementary Figures S1 and S2). Initial hits for hNTHL1Δ63 were obtained in the Index screen HT (Hampton Research) condition F12 (0.2 M NaCl, 0.1 M HEPES pH 7.5, 25% PEG 3350) set in a 96-well sitting drop plate by the NT8 drop setter (Formulatrix). Both hNTHL1Δ63 and hNTHL1Δ63 SeMet crystals were grown in 24-well hanging drop trays with a final protein concentration of 3.5 mg/ml, incubated at 18°C, and streak seeded 18 hours later. The reservoir solution ranged from 0.5% to 2% Polyethylene glycol 5000 monomethyl ether (PEG 5K MME, Hampton Research), 75–100 mM NaCl and 100 mM tricine pH 8.5. Long needle-like crystals were cut to ∼600 μM and soaked for 30 minutes in a 1:1 solution of mother liquor and a cryoprotecting solution of 10% (w/v) PEG 5K MME, 10 mM NaCl, 50% (w/v) glycerol, 50 mM Tricine pH 8.5. The hNTHL1Δ63 chimera crystals were grown using the hanging-drop method at a final concentration of 3 mg/ml, incubated at 18°C, and streak seeded 3 hours later. The reservoir solution ranged from 0.5% to 2% Polyethylene glycol 6000 (PEG 6K, Hampton Research), 55 mM NaCl, and 100 mM tricine pH 8.3. Needle-like crystals grew to approximately 400 × 70 × 70 μm3 over 5 days. Crystals were soaked for 30 min in a 1:1 solution of mother liquor and a cryoprotecting solution composed of 10% (w/v) PEG 6K, 10 mM NaCl, 50% (w/v) glycerol, 50 mM tricine pH 8.3.

Data collection and processing of crystals

hNTHL1Δ63 crystals diffracted to 2.5 Å at the APS synchrotron (beamline 23-ID-B). Data were collected using the 20-micron beam tuned to 12 kEV. A complete dataset was obtained by using the vector function along the length of the crystal. The hexagonal (P63) crystals were integrated, scaled, and truncated using iMOSFLM, AIMLESS/POINTLESS and CTRUNCATE (44–48). Data were collected on hNTHL1Δ63 SeMet crystals at the APS synchrotron (beamline 23-ID-B) at a peak wavelength of 0.9795Å. The crystals diffracted past 3.2 Å. Data were collected using a 20-micron beam and the vector function described above. The SeMet crystals were hexagonal (P63) and were processed using HKL2000. hNTHL1Δ63 chimera crystals diffracted to 2.1 Å at the APS synchrotron (beamline 23-ID-D) at 12 kEV. The vector function was again employed in order to obtain a complete dataset. The hNTHL1Δ63 chimera crystallized in the orthorhombic space group, P212121. Data were integrated, scaled, and truncated using iMOSFLM, AIMLESS/POINTLESS and CTRUNCATE (44–48).

Structure solution and refinement

hNTHL1Δ63 crystals were experimentally phased by Single-wavelength Anomalous Dispersion (SAD) methods. The selenium sites were identified by PHENIX autoSOL (49,50). The hNTHL1Δ63 model was refined using PHENIX and Translation/Libration/Screw (TLS) model (49,51). There is one molecule per asymmetric unit (ASU) with a Matthew's coefficient of 3.32 and a solvent content of 63%. The hNTHL1Δ63 model was refined using PHENIX (49,51–54) to a Rwork/Rfree of 19.96%/24.45% at 2.5 Å with 96.96% preferred and 3.04% allowed residues in the Ramachandran plot. The average B-factor is 88.6 Å2; the high B-factor is most likely due to the high solvent content and flexibility of the two domains. The following residues were built in the electron density map: 86–317. The hNTHL1Δ63 chimera dataset was solved by molecular replacement using hNTHL1Δ63 as a search model. The two domains were separated and searched for individually using PHENIX autoMR (49). There is one molecule per ASU, with a calculated Matthew's coefficient of 2.36% and 48% solvent content. The hNTHL1Δ63 chimera model was refined using PHENIX (49,51–54) to a Rwork/Rfree of 18.12%/23.06% at 2.1 Å with 98.1% preferred, and 1.9% allowed residues in the Ramachandran plot. The average B-factor is 29.5 Å2. The following residues were built in the electron density map: 86–106, 111–304.

Radiolabelling DNA

DNA oligonucleotides were radio-labelled with 32P at a ratio of 10% hot DNA to 90% cold DNA for the single-turnover experiments. The damage-containing oligonucleotide was incubated with 32P γ-ATP and polynucleotide kinase for 30 minutes. The reaction was quenched with 25 mM EDTA, and heat inactivated for 1 min at 95°C. The DNA was cleaned up by either ethanol precipitation for the 10% hot oligonucleotide. The damaged oligo and complementary strand were brought up to a final concentration of 250 nM in a DNA annealing buffer (10 mM Tris–HCl pH 8 and 50 mM NaCl) and annealed in 1:1 ratio in a 95°C water bath and allowed to cool slowly to room temperature.

Single-turnover kinetics experiments

To perform kinetics experiments under single-turnover conditions 100 nM of hNTHL1, hNTHL1Δ63 or hNTHL1Δ63 chimera was added to a solution of 20 nM of radiolabelled DNA containing Tg:A, 2 mg/ml BSA (New England Biolabs), 10 mM Tris–HCl pH 8, 75 mM NaCl, and 1 mM DTT at 37°C. Time points were taken at 0, 0.25, 0.5, 1, 2, 3, 4, 5, 8, 10, 20, 30, 45 min and quenched in either equal volume of formamide stopping dye (95% formamide, bromophenol blue, xylene) or 0.1 M NaOH, and boiled for 5 min before adding equal volume formamide stopping dye. Samples were loaded onto a 12% sequencing gel and run at 55 watts for 1 h. The gels were dried and exposed on phosphorescence screens (Kodak) and scanned using a STORM imager (GMI). Phosphorescence was quantified using Quantity One. The data were fit to the equation: Y = Y0 + (plateau − Y0)*(1 − exp(−K*x)) using Graph Pad Prism (GraphPad Prism version 8.1.2 for Windows, GraphPad Software, La Jolla, CA, USA).

Stopped-flow tryptophan fluorescence

Tryptophan fluorescence experiments were conducted on a stopped-flow SX-20 instrument (Applied Photophysics) with samples excited at 280 nm and emission filtered with a 350 nm long-pass filter at 37°C. Data were collected using the pretrigger setting for 60 s. Artifacts from the initial flow and mixing of the solutions and instrument dead time were accounted for. Each trace reported is an average of multiple traces. The glycosylase reaction buffer was used: 10 mM Tris–HCl pH 8, 75 mM NaCl, 1 mM DTT. The final mixture contained 2 μM hNTHL1Δ63 or hNTHL1Δ63 chimera and 2 μM DNA substrate (same sequence as above).

Selection of single nucleotide variants

We used the Genomic Data Commons (GDC) Application Programming Interface (API) to select subjects in The Cancer Genome Atlas (TCGA) that had at least one normal and one cancer sample. For the normal samples we used blood-derived normal, buccal cell normal, and solid tissue normal samples. Tumor samples were solid tumor samples. Of the 10 999 subjects in the TCGA, 9962 met these requirements. After selecting the subjects, we used the GDC API to perform BAM slicing for NTHL1. Using VarScan 2.4.4 and the GRCh38 as a reference, we tested for germline mutation using every possible combination of tumor and normal samples for each subject and thresholds of minimum variant frequency > 0, P-value < 0.1 and a minimum depth of eight sequence reads covering each variant. Once the variants were called for each subject, we selected the exon chromosome positions that had mutations in at least 10 subjects. For those exons with at least 10 subjects, we determined if the mutation was a missense, nonsense, or silent and recorded the number of cancers associated with each mutation. The potential pathogenicity of the mutations was assessed using REVEL, where higher scores are correlated with a higher likelihood of being disease-causing (55).

RESULTS

Human hNTHL1 structure adopts an open conformation

We present the first crystal structure of hNTHL1, which was solved using single-wavelength anomalous dispersion (SAD) to 2.5 Å using the peak anomalous signal of selenium with a resulting Rwork/Rfree of 19.96%/24.45% (Table 1, Figure 1). For crystallization, we selected hNTHL1Δ63, an active deletion construct lacking the first 63 residues (previously described as hNTHL1Δ55 (39,40) due to a discrepancy in the identity of the initiation methionine) because the flexible N-terminal region appeared to hinder crystal formation of full-length hNTHL1. The crystal structure contains 227 residues (aa 86–312). Like its bacterial homologs, hNTHL1 comprises two globular helical domains: a six-helical bundle domain, which contains a helix-hairpin-helix (HhH) DNA-binding motif and a helical domain containing a [4Fe- 4S] cluster domain (Figure 1) (20,21,56) (57). Both the N- and C- termini of hNTHL1 reside in the [4Fe−4S] cluster domain; the two domains are therefore connected by two linkers: linker 1 (aa 104–125) joins the [4Fe−4S] cluster domain to the six-helical bundle domain and linker 2 (aa 230–240) connects the six-helical bundle domain to the [4Fe4S] cluster domain (Figure 1). Surprisingly, our hNTHL1Δ63 model reveals a novel open conformation (Figure 2A), which is not observed in any of the bacterial Nth glycosylases previously crystallized without DNA: Escherichia coli Nth (EcoNth) (PDB ID 2ABK (58), RMSD 7.4 Å, calculated with PyMOL (The Pymol Molecular Graphics System, Version 1.2r3pre, Schrodinger, LLC.) and Deinococcus radiodurans Nth (DraNth) (PDB IDs 4UNF & 4UOB (57), RMSD 8.4 and 10.3 Å, respectively). The unliganded bacterial Nth structures have a domain orientation similar to that of the DNA-bound Geobacillus stearothermophilus Nth (GstNth) (PDB ID 1ORN (21); RMSD between GstNth and EcoNth is 1.9 Å).

Table 1.

hNTHL1 Data Processing and Refinement Statistics. Data collection and refinement statistics for the hNTHL1Δ63, hNTHL1Δ63 chimera, and hNTHL1Δ63 SeMet crystals. The highest resolution shell is shown in parentheses

	hNTHL1Δ63	hNTHL1Δ63 chimera	hNTHL1Δ63 SeMet
PDB ID	7RDS	7RDT
Data Collection
Beamline	APS 23-ID-B	APS 23-ID-D	APS 23-ID-B
Space group	P 63	P 21 21 21	P 63
Cell dimensions
a, b, c (Å)	124.8 124.8 42.25	42.37 71.68 86.52	125.0, 125.0, 42.4
α, β, γ (o)	90 90 120	90 90 90	90 90 120
Resolution range (Å)	40.85- 2.5	38.05 - 2.1	40 - 3.2
R-pim (%)	6.8 (62)	5.3 (21)	10.6 (55.6)
CC1/2	99.5 (57.3)	99.5 (84.3)	97.6 (16.7)
I/sigma	8.3(1.8)	8.9 (3.4)	12.1 (2.87)
Completeness (%)	100 (100)	99.7 (99.3)	100 (100)
Multiplicity	22.2 (22.3)	4.1 (4.0)	8.4 (8.5)
Wavelength (Å)	1.033	1.033	0.9795
Wilson-B factor (Å2)	50.72	22.48
Refinement
Resolution range (Å)	40.85−2.5	38.05−2.1
R-work/R-free (%)	19.96/24.45	18.12/23.06
No. reflections	25 423	28 900
No. atoms
Protein	1756	1725
Ligands	8	8
Water	115	200
RMS deviations
Bond lengths (Å)	0.002	0.002
Bond angles (o)	0.431	0.471
B-factor (Å2)	88.64	29.49
Protein	89.34	28.83
Ligands	73.19	28.03
Water	78.77	35.11
Ramachandran
Preferred (%)	96.96	98.10
Allowed (%)	3.04	1.90
Outlier (%)	0	0
Rotamer outlier	1.06	0.56
Clashscore	6.64	2.05
Number of TLS groups	3

Figure 1.

The active site of human NTHL1 is unassembled in the absence of DNA. (A) An overview of hNTHL1Δ63 coloured by domain and motifs. The [4Fe4S] domain is shown in blue, the six-helical bundle domain in red. Linker 1 is coloured orange, the HhH DNA binding motif in yellow, linker 2 in cyan, and the [4Fe4S] cluster in green. (B) The peaks (green mesh) from an anomalous difference Fourier map calculated with the peak selenium data set overlay nicely on the sulphurs (spheres) from the five methionine residues in the hNTHL1Δ63 model. There is a sixth anomalous peak, which corresponds to the [4Fe4S] cluster. (C) A cartoon representation of the hNTHL1 domain organization, from N- to C-terminus, with the same colour code as in panel A: missing residues (light grey not included in the protein construct, dark grey included in the protein construct but not built in the model), six-helical bundle (red), linker 1 (orange), HhH (yellow), linker 2 (cyan), [4Fe4S] domain (blue); the catalytic residues, Lys 220 and Asp 239 (magenta), and the [4Fe4S] cluster (green). (D) A detailed view of hNTHL1 Δ63 highlighting the unassembled active site illustrates the distance between the catalytic Lys and Asp (23.5 Å). Same colour code as in (C).

Figure 2.

An extended linker 1 in hNTHL1 creates a flexible hinge region. (A) The unexpected domain orientation of hNTHL1Δ63 (purple) relative to G. stearothermophilus (PDB ID code 1ORN, transparent blue (21)) became evident from a superposition of the two orthologs. Linker 1 is shown in orange and linker 2 is shown in cyan. (B) A multiple sequence alignment (63,80) between human, C. elegans, E. coli, and G. stearothermophilus confirms the variability of linker 1 (orange line), where eukaryotic NTHL1 has a seven amino acid insertion in addition to four additional residues extending the helix immediately after linker 1. (C) The multiple sequence alignment shows linker 2 (cyan line) is more conserved, with a single residue insertion in the eukaryotes. The CONSURF (64) representation of sequence homology (red, conserved and blue, variable) also shows a higher degree of variability in (D) linker 1 than in (E) linker 2.

The flexible linker is necessary for the open state

In the open state, the strictly conserved catalytic residues, Lys 220 and Asp 239, which are located ∼5 Å apart in the bacterial enzymes, are found ∼23 Å apart (Figure 1). Therefore, hNTHL1 must undergo a conformational change in order for catalysis to occur (17,19,59–61). This finding was unexpected as the bacterial homologs were captured in a similar closed conformation, in the presence and absence of DNA (Figure 1). We noted that E. coli Endonuclease 8 (EcoNei), a DNA glycosylase from the Fpg/Nei family, undergoes a conformational change between the unliganded and DNA-bound forms (Supplementary Figure S3). EcoNei, like hNTHL1, also harbours a flexible hinge region (62). Based upon this precedent, we investigated the role of linkers in hNTHL1. A multiple sequence alignment with mammalian, lower eukaryotes, and prokaryotes created using MUSCLE showed that linker 1 is not highly conserved (63). To visualize the conservation of Nth residues, we used CONSURF (64) to generate a multiple sequence alignment of 150 orthologs, and colour the residues by conservation level from red to blue (conserved to variable). As mentioned above, linker 1 exhibits a degree of divergence, with a single conserved turn, whereas linker 2 is more conserved across species (Figure 2). A simplified sequence alignment shows an eleven-amino acid insertion compared to bacterial Nth sequences in the linker 1 region: seven residues extend the loop to 21 residues and four residues extend helix B in hNTHL1. In contrast, 14 residues compose the linker in E. coli (Figure 2, orange). Linker 2 is more conserved, with a single residue insertion in the eukaryotic NTHL1 (Figure 2). To test whether the open conformation in hNTHL1 was due to the increased flexibility imparted by the extended linker 1, we engineered a chimera of hNTHL1Δ63 and EcoNth by replacing 16 residues (residues 110–125) of the human linker with the shorter EcoNth linker (residues 21–28) (hNTHL1Δ63 chimera, Table 2). Since the length of linker 2 was quite similar across bacterial and mammalian sequences, we hypothesized that the increased flexibility in the human enzyme was due to linker 1, and linker 2 was therefore not altered in this report.

Table 2.

Nth linker 1 sequences. The linker sequences for hNTHL1, EcoNth, and hNTHL1Δ63 chimera are listed. The human linker (linker 1) was substituted with the corresponding EcoNth linker. The last four residues (KVRR) are an extension of helix B and were therefore left in the hNTHL1Δ63 chimera to not disrupt local secondary structure elements.

Construct	Linker sequence
hNTHL1Δ63	VDHLGTEHCYDSSAPKVRR
hNTHL1Δ63 chimera	TTELNFSSPKVRR
EcoNth	TTELNFSSP

We solved the crystal structure of hNTHL1Δ63 chimera to 2.1 Å (Rwork/Rfree of 18.12%/23.06%) by molecular replacement using hNTHL1Δ63 as the search model (49,65). To determine the phases via molecular replacement, the six-helical bundle domain was placed prior to the [4Fe4S] domain. By searching with each domain individually, the domains were allowed the freedom necessary to rearrange relative to each other. The hNTHL1Δ63 chimera crystallized in a closed conformation, similar to unliganded EcoNth (RMSD 2.21 Å), DraNth (RMSD: 2.77 and 1.84 Å), and GstNth bound to DNA (RMSD 1.40 Å). To examine the degree of rotation, we superimposed the open and closed models in COOT (66) using the six-helical bundle domain as a reference, and then measured the angle of rotation between three residues, Asp 239 hNTHL1Δ63, Lys 220 hNTHL1Δ63/hNTHL1Δ63 chimera, and Asp 239 NTHL1Δ63 chimera, calculating a ∼85° rotation (Figure 3A, Supplementary Figure S4). When comparing the WT to the chimera construct, the analogous catalytic aspartates were 23.6 Å apart, and the [4Fe4S] clusters were 38.7Å apart (Supplementary Figure S4). We were unable to completely trace the hNTHL1Δ63 chimera linker itself because of disorder in the electron density map. Nevertheless, the shortened linker decreased the interdomain flexibility within the enzyme. DNA interacting regions are highly conserved while the outer edges of the enzyme display more variability, as visualized with CONSURF (64) (Supplementary Figure S5). The closed conformation may be stabilized by interactions between the globular domains and linker 2: Arg 272 of the [4Fe4S] domain interacts with the backbone carbonyl of Val 233 of linker 2; a second hydrogen bond is formed between His 223 (six-helical bundle domain) and Ser 234 (linker 2); and the third interaction occurs between Glu 267 ([4Fe4S] domain) and the backbone amide of Leucine 236 (linker 2) (Figure 3). The refined B-factors, or temperature factors, provide a measure of the relative movement of atoms, where higher numbers (shown in warmer colours in the protein cartoon) signify increased movement (Figure 3). The reduced movement of the chimeric enzyme is reflected in the B-factors: not only is the average B-factor lower overall, but the [4Fe−4S] cluster domain has reduced B-factors relative to the six helical bundle domain, as indicated by the average B-factors ratios for the six helical bundle domain:[4Fe−4S] cluster domain: 1:1.5 for hNTHL1Δ63 chimera versus 1:1.8 for hNTHL1Δ63.

Figure 3.

Shortening linker 1 in the hNTHL1 chimera assembles the active site. (A) Shortening linker 1 in hNTHL1Δ63 chimera (green) decreased the interdomain flexibility and yielded a closed domain orientation compared to hNTHL1Δ63 (purple). Linker 1 is shown in orange, and light orange for hNTHL1Δ63 and hNTHL1Δ63 chimera, respectively. Linker 2 is shown as cyan and light blue for hNTHL1Δ63 and hNTHL1Δ63 chimera, respectively. The catalytic residues (Lys 220 and Asp 239) are shown as sticks. (B) Comparison of hNTHL1Δ63 chimera (green) to DNA-bound G. stearothermophilus Nth (transparent blue, PDB 1ORN (21)) reveals a similar domain orientation. (C) Zooming in on the active site (blue dashed box) reveals a complete active site with the catalytic lysine and aspartate 5.3 Å apart (2Fo− Fc map is shown as a grey mesh). This corresponds to the distances observed in the bacterial homologs. (D) The closed conformation may be stabilized by interactions between the globular domains and linker 2. hNTHL1Δ63 chimera is shown in cartoon format with the highlighted residues shown as sticks; the [4Fe4S] domain is shown in blue, linker 2 in cyan, and the six-helical bundle domain in red. Arg 272 interacts with the backbone carbonyl of Val 233, a second hydrogen bond is formed between His 223 and Ser 234, and the third interaction occurs between Glu 267 and the backbone amide of Leu 236. B-factor analysis suggests that the shortened linker 1 reduces movement of hNTHL1. (E) A putty representation of hNTHL1Δ63 coloured by B-factor shows that the [4Fe4S] domain has more movement than the six-helical bundle domain (A high B-factor signifies a higher degree of movement, indicated by warmer colours and a thicker putty). (F) The same B-factor representation of hNTHL1Δ63 chimera shows decreased movement within the entire protein but especially in the [4Fe4S] cluster relative to the six-helical bundle domain.

The closed conformation of hNTHL1Δ63 chimera puts the two catalytic residues in close proximity

In the hNTHL1Δ63 chimera model, the catalytic Asp and Lys are 5.3 Å apart. These residues occupy an identical orientation in the closed structures of homologs. Closure of the enzyme therefore restores the glycosylase active site to an apparent active configuration (Figure 3). Comparing these new structures of hNTHL1 to prior models of homologs suggests that the conformation change is essential for catalysis. These conformational changes observed in the hNTHL1Δ63 crystal structures are reminiscent of the conformational changes reported for EcoNei, which also closes upon binding DNA (62,67). In the apo-EcoNei structure, the C-terminal domain requires a reported ∼50° rotation to form the closed ligand complex (62) (Supplementary Figure S4). Another glycosylase of the Fpg/Nei family, Nei-like 2 (NEIL2), is also predicted to undergo a similar conformational change between the unliganded and bound forms (68).

The hNTHL1Δ63 chimera has reduced glycosylase activity

To determine if the hNTHL1Δ63 chimera retains activity we performed glycosylase activity assays under single-turnover conditions. In the glycosylase assays, the 5′ end of the DNA strand containing the lesions was radiolabelled with 32P. hNTHL1 nicks the DNA backbone as the damaged base is removed, resulting in a single product band with increased electrophoretic mobility on a PAGE gel. When the assays are quenched with NaOH the glycosylase activity is measured, as NaOH will resolve the Schiff base and reduce the AP site. Alternately, quenching with a formamide dye yields a single product band representing both glycosylase and lyase activity, because hNTHL1 must catalyse the resolution of the Schiff base and β-elimination. Based upon the published catalytic scheme for hNTHL1, k2 is the rate of base excision, and k3 is the rate of β-lyase activity (Supplementary Figure S6) (59). Full-length hNTHL1 is active on Tg:A, AP:G and DHU:G, with k2 = 4.53, 5.53, and 3.79 min–1, respectively (Table 3, Supplementary Figure S1A). The hNTHL1Δ63 construct is as active as the wild-type enzyme on Tg:A, with k2 = 2.95 min–1 (Table 3; Figure 4A, Supplementary Figure S2A). We found that the hNTHL1Δ63 chimera retains some glycosylase activity but is impaired compared to the hNTHL1Δ63 and full-length hNTHL1 (Table 3, Figure 4A and Supplementary Figure S2). The hNTHL1Δ63 chimera is highly deficient when it must provide the lyase reaction (Table 3, Figure 4B & Supplementary Figure S2B). It was not possible to calculate the rate constant k3 for the Tg:A substrate for the chimera construct because the data did not fit a one-phase exponential association model. These results indicate that when hNTHL1Δ63 is trapped in the closed conformation, it is deficient in glycosylase and lyase activity. We speculate that the reduction in catalytic activity may be due to impaired DNA binding, because when the chimeric DNA glycosylase is provided with additional time, it can cleave all the DNA substrate in the assay (Figure 4).

Table 3.

Single-turnover experiments with hNTHL1Δ63, hNTHL1Δ63 chimera, and full-length hNTHL1. The rate constants are reported in this table. The rate constant k2 is the observed rate of glycosylase only activity, quenched with NaOH. The rate constant k3 is the observed rate of β-elimination activity, quenched with formamide stopping dye. Experiments were repeated three times. The rate constants were calculated by fitting the data to a one-phase association exponential, Y = Y0 + (plateau-Y0) * (1 − exp(−K*x)).

		Tg:A
	k 2 (min–1)	99% CI	k 3 (min–1)
hNTHL1Δ63	2.95	1.53–5.65	0.45
hNTHL1Δ63 Chimera	0.070	0.021–0.15	n/a
hNTHL1 FL	4.53	3.40–6.35	–

Figure 4.

Reduced interdomain flexibility of the hNTHL1Δ63 chimera decreases its activity. Single-turn over experiments measuring the activity of hNTHL1Δ63 and hNTHL1Δ63 chimera on Tg:A substrate: (A) glycosylase reaction and (B) combined glycosylase and lyase reaction. Stopped flow experiment measuring tryptophan fluorescence of hNTHL1Δ63 (C) and hNTHL1Δ63 chimera (D) incubated with T:A, THF:A and Tg:A.

The hNTHL1Δ63 chimera has reduced interdomain flexibility

To determine if a conformational change could be observed during DNA binding, we performed stopped-flow kinetics experiments and measured tryptophan emission at 490 Abs with hNTHL1Δ63, hNTHL1Δ63 chimera, and hNTHL1 FL in the presence of Tg:A, THF:A, or undamaged DNA. Supplementary Figure S7 shows the location of the seven tryptophan residues present in hNTHL1Δ63. We observed a conformational change for hNTHL1Δ63 and hNTHL1 FL when processing Tg:A, but not THF:A or undamaged DNA (Figure 4C and Supplementary Figure S8). The tryptophan fluorescence curves for full-length hNTHL1 are similar to those of hNTHL1Δ63, supporting the idea that the observed change in fluorescence is not affected by the N-terminal extension. There was no conformational change detected with the hNTHL1Δ63 chimera on any of the DNA oligonucleotides (Figure 4D). The fact that we observed a change in conformation in the hNTHL1 constructs during lesion processing, but not in the hNTHL1Δ63 chimera, suggests that a conformational change occurs during lesion removal. Moreover, the conformational change that was observed in hNTHL1Δ63 was seen only in the context of a lesion, Tg, suggesting that the structural changes resulting in a difference in tryptophan environment happen during or after base cleavage.

DISCUSSION

The first crystal structure of human NTHL1 reveals an open conformation that had not been observed previously in any of the bacterial Nth homologs, or other members of the HhH family such as OGG1 or MutY (69). In the open conformation the DNA glycosylase is catalytically inactive as the two active site residues (K220 and D239) are much too far apart, 23.5 Å, to perform a nucleophilic attack. Thus, a conformational change must occur upon DNA binding to assemble the active site. We showed that the interdomain rearrangement requires an extended linker in the hNTHL1 enzyme. The hNTHL1Δ63 chimera with the substitution of the E. coli linker into hNTHL1Δ63 crystallized in the closed form, like that of bacterial Nth (16,20,21,56–58). Even with the active site residues K220 and D239 in close proximity, the hNTHL1Δ63 chimera exhibits a decreased glycosylase activity in vitro. This finding suggests that the movement between the two domains is critical for lesion search and processing in vitro.

Mapping residues onto a three-dimensional structure is crucial when designing and interpreting mutagenesis studies, and for understanding the potential outcome of cancer variants. A series of hNTHL1 point mutations can now be interpreted considering the new crystal structures. Robey-Bond et al. investigated several residues which were deemed important for catalysis in E. coli and yet had a different amino acid residue at the analogous position in mammalian NTHL1 (19), including Asn279, Gly280, and Gln287 (19). Our crystal structure of hNTHL1 shows that Asn 279 and Gly 280 are located near linker 2, with the former contacting the backbone carbonyl of Ala 137 in that linker. hNTHL1 Gln 287 is predicted to contact the DNA backbone. The analogous residue in GstNth, Arg186, contacts the DNA backbone at the site of the lesion, and hydrogen bonds Glu 24 to linker 1. Additionally, we were able to map germline hNTHL1 single-nucleotide variants (SNVs) from the TCGA dataset (Supplementary Table S1), onto the open and closed hNTHL1 models. The curated SNVs were predicted to be deleterious using REVEL (55). Interestingly, these mutations appear to cluster near the two linker regions, and therefore could potentially affect the predicted conformational change (Supplementary Figure S9). While none of the SNVs appear close enough to contact the DNA, Ile 176 is part of the HhH motif and Thr 289 is close to the [4Fe4S] cluster.

DNA glycosylases were generally thought to be relatively rigid enzymes, with modest interdomain changes, as the apo- and DNA-bound forms are nearly identical in numerous X-ray crystal structures where both DNA-bound and unbound models are available (Supplementary Figure S3). There are now several examples of glycosylases that show interdomain rearrangement when comparing the unliganded to the DNA-bound form. Human uracil DNA glycosylase (UDG) undergoes a 10° conformational change, in which the two globular domains move together and ‘pinch’ the DNA, causing a 45o kink in the DNA backbone (70). EcoNei has a much more dramatic 50o interdomain rotation upon binding DNA (62). In the same H2TH family as EcoNei Neisseria meningitidis formamidopyrimidine DNA glycosylase (Fpg) was shown to undergo a 22° conformational change upon binding DNA (69). Additionally, the unliganded mammalian NEIL2 crystalized in an open conformation with a large conformational change upon binding DNA shown using small angle X-ray scattering(68). NTHL1, is the first reported member of the HhH family of DNA glycosylases to show structural evidence of a large conformational change and joins the UDG and H2TH glycosylase families in this respect (62,70). A structure based alignment using Expresso showed EcoNei has a 7 amino acid insertion in the linker region compared to human NEIL1 and mimivirus Nei1 (71). There were no notable differences in linker size between the FPG enzymes, although this conformational change is much more modest compared to EcoNei and hNTHL1. With the accumulation of structural evidence, interdomain movements of the DNA glycosylases may in fact be a more common mechanism than previously thought, albeit with differing degrees of movement.

Investigations of protein movement in solution with tryptophan fluorescence have shown that EcoNei, EcoFpg, human NEIL1 and human OGG1 undergo protein isomerization events prior to catalysis, but the degree of movement is unknown because tryptophan fluorescence simply indicates that the environment of the tryptophan has changed (67,72–74). Changes in fluorescence were not detected for EcoNth, suggesting that either EcoNth does not undergo a conformational change during catalysis, or that it was simply not observed under the studied conditions because of the non-optimal location of tryptophan residues within the protein (75). Our tryptophan fluorescence data yielded a large difference in signal between hNTHL1Δ63 and hNTHL1Δ63 chimera in the presence of Tg:A - containing DNA, but not with THF:A or undamaged DNA. This finding suggests that there is a large conformational change at some point after lesion recognition, as the undamaged and THF curves look similar for NTHL1Δ63.

In addition to its interdomain flexibility, hNTHL1 also differs from its prokaryotic homologs because it possesses a flexible N-terminal extension (residues 1 to 86) (Figure 1C). This N-terminal extension was posited to inhibit product release when hNTHL1 is at low concentration (40). The emergence of a sigmoidal curve when DNA glycosylase activity is plotted vs. enzyme concentration hinted at the possibility of positive cooperativity, leading to the hypothesis that hNTHL1 forms dimers (39,41). A crosslinking experiment with BS3, an amine-to-amine crosslinker, showed that serial truncations of the N-terminal extension reduced the amount of crosslinked hNTHL1 and putative dimers (39). However, we propose that the decreased propensity to form dimers as the N-terminus was progressively trimmed is likely to be due to the serial elimination of crosslinkable lysines. We sought to identify the predicted hNTHL1 dimer, but the protein consistently eluted off the sizing column (Superdex 75; Cytiva) as a monomer (Supplementary Figure S10). Additionally, analysis of the crystal lattice of hNTHL1Δ63, which still maintains 27 residues of the N-terminal extension, did not reveal any putative dimerization interface.

How, then, could NTHL1 display positive cooperativity in the absence of dimer formation? Another means to achieve apparent enzymatic cooperativity independent of multimerization is kinetic cooperativity (76). A well-documented instance of kinetic cooperativity (or monomeric enzyme cooperativity) is glucokinase, which exemplifies this particular behaviour in both kinetic and structural data (76). Glucokinase performs the first step in glycolysis by phosphorylation of glucose to glucose-6-phosphate. Glucokinase comprises two domains, and the active site is formed in the cleft between those domains. Much like hNTHL1, the glucokinase structures revealed a hinge region, with a 99° rotation of the two domains between the open and closed conformations (77). When quantifying the biochemical reaction trajectories, glucokinase exhibits positive cooperativity, evident as a sigmoidal curve relating substrate concentration to rate, in the presence of increasing glucose concentration (78). Kamata et al. interpreted the crystal structures, postulating the existence of a ‘super-open’ ground state conformation that is driven to an open glucose-bound state. This equilibrium depends on the concentration of glucose and is rate-limiting in the greater reaction scheme. Sigmoidal reaction curves therefore emerge due to bypass of this rate-limiting step at high concentrations of glucose, rather than by the conventional cooperative model in which the affinity of the binding sites for a ligand is increased.

In the context of our current understanding of hNTHL1, we speculate that the DNA glycosylase forms an open scanning configuration that can slide along the DNA searching for a lesion. Upon encountering a lesion, hNTHL1 would close on the DNA to process the lesion. After processing, the enzyme would release back into the scanning complex to continue its search for lesions (Figure 5). This model is supported by the two conformations of hNTHL1, in open and closed states. Furthermore, the tryptophan fluorescence stopped-flow data are consistent with this hypothesis. hNTHL1 exhibits a low affinity for undamaged DNA (30 μM (79) vs. nM range for lesion-containing DNA (19)), therefore we can assume that there is little appreciable DNA binding signal in the undamaged DNA curve. hNTHL1 binds to DNA containing THF but, as this lesion is non-cleavable, the enzyme cannot cycle through catalysis, thus keeping the enzyme in a lesion-bound state. In contrast, hNTHL1 can bind, cleave, and release Tg. This finding suggests that a conformational change occurs either during or after the cleavage event, perhaps to protect undamaged bases from being erroneously cleaved. It is important to note that tryptophan fluorescence relies on a change in environment around a particular tryptophan (Supplementary Figure S7), and therefore we cannot rule out the possibility of additional conformational changes during the catalytic cycle.

Figure 5.

Proposed model of hNTHL1 lesion searching, recognition and removal, which is consistent with kinetic cooperativity. The [4Fe4S] domain is shown in blue, with the [4Fe4S] cluster represented in yellow; the six-helical bundle domain is red and the interdomain flexible linker region is orange. In Step 1, the apo enzyme exists in an open conformation; when it associates with undamaged DNA, the DNA glycosylase adopts a semi-closed state, or ‘scanning conformation’, depicted in Step 2. As hNTHL1 scans the DNA, if it encounters a lesion (cyan pentagon) the closed or ‘active’ complex is achieved, Step 3. After lesion removal the enzyme may relax into the scanning conformation to continue a search for DNA damage (Step 4).

Additional studies are needed to characterize the conformational change of hNTHL1. We developed the hNTHL1Δ63 chimera to gain understanding about the hNTHL1 mechanism. Identification of point variants in NTHL1 that disrupt or enhance the closed conformation, but are less extensive than the chimera linker, will aid in the characterization of the hNTHL1 conformational change. Additional fluorescence studies would be informative, such as FRET, but provide the additional challenge of not disrupting the cysteines that coordinate the [4Fe–4S] cluster. These characterization studies could examine a broader range of DNA substrates to determine if different conformations are observed with different base lesions.

CONCLUSION

The open conformation observed in hNTHL1 suggests a novel molecular mechanism for this DNA glycosylase. The bacterial homologs that laid the foundation for the current kinetics models do not appear to undergo a large conformational change, based on the available X-ray crystal structures and solution fluorescence studies (20,21,56,75). We have established that a truncation of linker 1 in hNTHL1 shifts the equilibrium towards the closed conformation. The reduced interdomain flexibility of the hNTHL1Δ63 chimera decreased its glycosylase activity. Interdomain movements have been observed in the H2TH and UDG families of glycosylases, and now the HhH family, indicating that this mechanism may be more common than previously thought for DNA glycosylases. Analysis of previously reported variant data and germline SNVs suggests that the conformational change of hNTHL1 is crucial for proper function.

ACCESSION NUMBERS

Atomic coordinates and structure factors for the reported crystal structures have been deposited with the Protein Data Bank (www.rcsb.org) under accession numbers 7RDS (hNTHL1Δ63) and 7RDT (hNTHL1Δ63 chimera).

Click here for additional data file.