An additional proofreader contributes to DNA replication fidelity in mycobacteria

Significance

Proofreading ensures the fidelity of DNA replication by removing the mis-incorporated nucleotide. In bacteria, proofreading relies on either the ε-exonuclease DnaQ or the polymerase and histidinol phosphatase (PHP) domain of the replicative polymerase. Here, we show that a noncanonical DnaQ acts as a second proofreader that cooperates with the PHP domain proofreader to correct replicative errors in mycobacteria. Furthermore, our results show that the second proofreader preferentially prevents AT-biased mutations and insertions/deletions in the homopolymer tract, possibly through targeting mismatches with distorted DNA strand geometry. We provided real-world evidence that dnaQ is subject to positive selection in some Mycobacterium tuberculosis (Mtb) sublineages and is associated with hypermutability. These results provide a coproofreading model and reveal a mutator-driven evolutionary pathway in Mtb.

Abstract

The removal of mis-incorporated nucleotides by proofreading activity ensures DNA replication fidelity. Whereas the ε-exonuclease DnaQ is a well-established proofreader in the model organism Escherichia coli, it has been shown that proofreading in a majority of bacteria relies on the polymerase and histidinol phosphatase (PHP) domain of replicative polymerase, despite the presence of a DnaQ homolog that is structurally and functionally distinct from E. coli DnaQ. However, the biological functions of this type of noncanonical DnaQ remain unclear. Here, we provide independent evidence that noncanonical DnaQ functions as an additional proofreader for mycobacteria. Using the mutation accumulation assay in combination with whole-genome sequencing, we showed that depletion of DnaQ in Mycolicibacterium smegmatis leads to an increased mutation rate, resulting in AT-biased mutagenesis and increased insertions/deletions in the homopolymer tract. Our results showed that mycobacterial DnaQ binds to the β clamp and functions synergistically with the PHP domain proofreader to correct replication errors. Furthermore, the loss of dnaQ results in replication fork dysfunction, leading to attenuated growth and increased mutagenesis on subinhibitory fluoroquinolones potentially due to increased vulnerability to fork collapse. By analyzing the sequence polymorphism of dnaQ in clinical isolates of Mycobacterium tuberculosis (Mtb), we demonstrated that a naturally evolved DnaQ variant prevalent in Mtb lineage 4.3 may enable hypermutability and is associated with drug resistance. These results establish a coproofreading model and suggest a division of labor between DnaQ and PHP domain proofreader. This study also provides real-world evidence that a mutator-driven evolutionary pathway may exist during the adaptation of Mtb.

The emergence of drug-resistant (DR) Mycobacterium tuberculosis (Mtb) poses a major challenge to the global control of tuberculosis, which kills 1.5 million people annually (1). In contrast to many other bacteria, where the development of drug resistance is usually driven by horizontal gene transfer, genome rearrangement, and mutations, all drug resistance characterized in Mtb is the result of chromosomal mutations, suggesting that the mutational capacity of Mtb is one of the key mechanisms supporting the de novo generation of drug resistance (2, 3).

Nucleotide mispairing during DNA replication is a major cause of mutagenesis. In the model organism Escherichia coli, fidelity of DNA replication is ensured by base selection via the DNA polymerase α subunit (DnaE), the proofreading by the ε subunit 3′-5′ exonuclease (DnaQ), and postreplicative mismatch repair by a separate enzyme system (MutHLS) (3–5). Proofreading acts as a double check, where the mis-incorporated nucleotide is removed by 3′-5′ exonuclease activity (6). In E. coli, naturally occurring mutants with a defect in proofreading activity can lead to a 100- to 1,000-fold increase in the mutation rate and accelerate the development of drug resistance (7, 8). In addition, the deletion of E. coli dnaQ leads to growth defects, insufficient polymerization capacity, and constitutive SOS (cellular responses induced by DNA damage or replication fork blockages) phenotype (6, 9, 10).

Mycobacteria encode an annotated dnaQ and a second potential dnaQ homolog (11). Despite the de facto exonuclease activity of Mtb DnaQ in vitro, previous fluctuation analysis results have shown that deletion of the annotated dnaQ (Rv3711c) in Mtb or deletion of dnaQ (Ms6275) in combination with the second potential dnaQ homolog gene (Ms4259) in Mycolicibacterium smegmatis (Msm) did not result in the “expected” mutator phenotype (11). This functional divergence from E. coli DnaQ led to the identification of a novel proofreading activity mediated by the polymerase and histidinol phosphatase (PHP) domain of dnaE1-encoded replicative polymerase (11). These results demonstrated that mycobacterial DnaQ is a noncanonical exonuclease that is functionally distinct from E. coli DnaQ.

Phylogenetic and structural analyses suggest that the PHP domain may be the most abundant replicative exonuclease in the bacteria (4, 11–13). Interestingly, although the presence of an active PHP domain appears to be mutually exclusive with the presence of an E. coli-like DnaQ, it extensively coexists with the noncanonical DnaQ in a large proportion of bacteria (11). In addition, a recent study showed that Mtb DnaQ and DnaE1 can coexist in the DNA core replicase by simultaneously binding to the β clamp (DnaN), suggesting the possibility of coproofreading in bacterial replicases (14). However, the biological function of this type of noncanonical exonuclease remains unclear (15).

In this study, we provide independent evidence that the mycobacterial DnaQ functions in proofreading. We showed that the second proofreader exhibits a preference for correcting AT-biased mutations and maintaining the stability of homopolymer tracts (HTs), potentially through targeting mismatches with distorted DNA strand geometry. By analyzing the sequence polymorphisms of 51,229 clinical Mtb isolates, we demonstrated that dnaQ is under positive selection in some sublineages and has a functional impact on the mutation rate. These results demonstrate that mycobacterial DnaQ is an additional proofreader and provides a model for coproofreading involving two proofreaders (PHP domain and DnaQ), which may have broader relevance to DNA metabolism. This study also demonstrates that a mutator-driven evolutionary pathway may exit during the adaptation of Mtb.

Results

Deletion of dnaQ Results in Increased Mutagenesis in Msm.

Sequence analyses showed that the two DnaQ homologs encoded by mycobacteria showed considerable similarity in their N-terminal exonuclease domain to the DnaQ of E. coli but generally differ in overall size and domain organization (Fig. 1A and SI Appendix, Fig. S1A). The annotated DnaQ contains an additional C-terminal domain of 86 residues (Fig. 1 A and B), which is homologous to the human breast cancer suppressor protein C-terminal domain (BRCT). This domain functions as a protein–protein interaction module in a variety of proteins involved in DNA replication, repair, and recombination (16). The second potential DnaQ homolog (annotated as a hypothetical protein) has a larger C-terminal region of ~400 residues, which is highly similar to the endonuclease domain of the nucleotide excision repair protein UvrC (SI Appendix, Fig. S1B).

Fig. 1.

Deletion of dnaQ in Msm results in a mutational bias for AT and increased indels in the HT. (A) Domain architectures of DnaQ homologs from E. coli and Mtb. The Mtb Rv2191 encodes a second potential DnaQ homolog. The domains were predicted by InterPro and the blue lines and sequences indicate the clamp-binding motif (CMB). (B) Structural models of DnaQ from E. coli and Mtb. (C) Fluctuation analysis of Msm strains. dnaQ Exo⁻, DnaQ[D28A/E30A/D112A] lacking exonuclease activity; H37Rv_dnaQ, dnaQ allele from the Mtb strain H37Rv. Circles represent the mutant frequency (number of rifampicin-resistant mutants per cell plated in a single culture). Significance was determined using one-way ANOVA for log-transformed values. (D–F) Mutation spectra identified using the MA assay. Significance was determined using the Mann–Whitney U test compared with the WT counterpart. (D) Mutation rates in the BPS mutation spectra. (E) Mutation rates of AT/>G/C and G/C>A/T events normalized to the genomic AT and GC content, respectively. (F) Mutation rates of indel events in the HT. ND, not detected. All data are shown as the mean ± 95% CI. *P < 0.05, **P < 0. 01, ***P < 0.001, ****P < 0.0001.

To investigate whether mycobacterial dnaQ contributes to DNA replication fidelity, we constructed the null mutant strains ΔdnaQ, ΔMs4259 (encoding the second potential dnaQ homolog), and ΔdnaQΔMs4259 in Msm and estimated the mutation rate by fluctuation analysis (Fig.1C). The deletion of these genes had no effect on bacterial growth in vitro (SI Appendix, Fig. S2 A and B). Consistent with a previous study (11), our results showed that these mutant strains did not exhibit a significant increase in spontaneous mutation rate (defined by nonoverlapping 95% CI) (Fig. 1C and SI Appendix, Table S1).

Because the fluctuation assay relies on the selection of mutants (e.g., rifampicin resistance [Rif^R]) that exhibited mutations in a specific locus (e.g., rpoB), the mutations detected using this method may have significant biases and may not be representative of the whole-genome (17, 18). To obtain an unbiased view of mutation events across the genome, we used a mutation accumulation (MA) experiment in combination with whole-genome sequencing (WGS). For each strain, 10 to 12 MA lines were generated from a single colony (Table 1). Every 2 d, a single colony from each line was streaked onto a fresh 7H10 plate. In this process, strong bottlenecks minimize selection pressure, allowing mutations to accumulate in a neutral, unbiased manner (17, 18). The application of WGS to MA lines also ensures the detection of a complete and almost unbiased profile of mutation events (5, 17, 18). The MA experiment was carried out for 100 d, which meant >8,150 generations for each strain (Table 1). In total, 44 MA lines were successfully sequenced, covering an average of 99.96% of the reference genome with >10× reads (Dataset S1). By identifying base pair substitutions (BPSs) and indels (defined as insertions or deletions of ≤30 nucleotides), we found that the basal mutation rate for Msm wild-type (WT) was 4.88 × 10⁻¹⁰ mutations per base pair per generation (95% CI: [2.97 to 6.78] × 10⁻¹⁰), which is highly consistent with estimates from previous studies (95% CI: [4.52 to 5.27] × 10⁻¹⁰) (11, 19). Among the mutant strains, the mutation rate of the ΔdnaQ strain was 11.68 × 10⁻¹⁰ mutations per base pair per generation (95% CI: [7.05 to 16.31] × 10⁻¹⁰), a 2.4-fold increase over the WT (Table 1). Deletion of Ms4259 in either the dnaQ⁺ (ΔMs4259) or in the ΔdnaQ strain (ΔdnaQΔMs4259) resulted in a moderate increase (~1.4-fold) in the mutation rate (Table 1).

Table 1.

Deletion of dnaQ results in a mutator phenotype in MA assay

Strain	No. of BPSs	No. of indels	No. of lines	Generations per line	Total No. of generations	Mutation rate per base pair ± 95% CI (×10–10)	Mutation rate per genome ± 95% CI (×10–3)
WT	23	10	12	815	9,780	4.88 ± 1.90	3.39 ± 1.29
ΔdnaQ	50	16	10	815	8,150	11.68 ± 4.63	8.08 ± 3.14
ΔMs4259	32	7	10	815	8,150	6.83 ± 1.95	4.78 ± 1.40
ΔMs4259ΔdnaQ	44	11	12	930	11,160	7.09 ± 2.36	4.94 ± 1.62

Mutation rates were estimated using a MA assay. Mutations were classified according to BPSs or indels (defined here as insertions or deletions of <30 nucleotides). CI, confidence interval.

Depletion of DnaQ Leads to a Mutation Bias for AT and Increased Indels in HT.

In the WT, we identified 23 BPSs in the 12 lines, showing a transition/transversion ratio of 1.30 (13/10) (SI Appendix, Table S2), which is very similar to the previous observation (1.48) (19). However, in the ΔdnaQ mutant, this ratio increased to 2.13 (34/16), resulting in a 3.1-fold (P < 0.001) increase in the transition mutation rate over the WT (Fig. 1D and SI Appendix, Table S3). Among the transitions, the G:C>A:T mutation rate was significantly increased by 4.2-fold (P < 0.01) in the ΔdnaQ mutant compared to that of the WT (Fig. 1D and SI Appendix, Table S3). Given that the G:C>A:T mutation has a major impact on genomic GC content during long-term evolution (20), we analyzed the direction of the mutation bias (Fig. 1E and SI Appendix, Table S4). In WT, the mutation rate of A/T>G/C (i.e., A to G or C, T to G or C) and G/C>A/T normalized to the genomic GC and AT content, respectively, was 3.40 × 10⁻¹⁰ and 2.69 × 10⁻¹⁰, showing a mutation bias for GC (the expected GC content was 55.8% and the actual genomic GC content of Msm was 65.6%), which is consistent with a previous MA study (the expected GC content was 58.2%) (19). While the mutation rate of A/T>G/C remained unchanged in the ΔdnaQ mutant, the G/C>A/T mutation rate (95% CI: [4.75 to 13.98] × 10⁻¹⁰) increased significantly by 3.5-fold (P < 0.01), resulting in a striking propensity for AT mutation (expected GC content of 37.4%) (Fig. 1E). Deletion of the dnaQ homolog (Ms4259) in the ΔdnaQ background (ΔdnaQΔMs4259) resulted in mutational spectra and bias for AT similar to that of the ΔdnaQ mutant, but these effects were not observed in the ΔMs4259 mutant (Fig. 1 D and E and SI Appendix, Tables S3 and S4). Overall, these results suggest that mycobacterial dnaQ plays a role in maintaining the genomic GC content. Notably, the mutational spectrum of the mycobacterial ΔdnaQ mutant is markedly different from that of the dnaQ-deficient E. coli, which shows a mutation bias for transversion and no effect on the GC content (21).

Our MA analysis identified 44 indels, 1 to 30 bp in length (Table 1 and Dataset S1), and all strains exhibited a mutational propensity for insertion events (SI Appendix, Table S2). The indel rate of WT (1.48 × 10⁻¹⁰) was very similar to the previous estimate (1.27 × 10⁻¹⁰) (19). While the ΔdnaQ mutant showed a 1.9-fold increase in indel mutation rate (SI Appendix, Table S5), the rate was slightly reduced in the ΔMs4259 strain, indicating a functional divergence between DnaQ and Ms4259. In the WT and the ΔMs4259 strains, approximately 40% of indels occurred in the HT; however, in the ΔdnaQ mutant and the ΔdnaQΔMs4259 mutant, the proportion increased to 63% and 82%, respectively (SI Appendix, Tables S5 and S6). Accordingly, the mutation rate of indels occurring in the HT (almost G/C tract) was significantly increased threefold in the ΔdnaQ mutant compared to that in the WT (P < 0.05) (Fig. 1F and SI Appendix, Table S5). Considering that indel events in the HT are mainly caused by strand slippage during replication (22), these results suggest a role for mycobacterial dnaQ in resolution of replication conflicts induced by distorted DNA conformation. Overall, the results of the MA experiment showed that mycobacterial dnaQ contributes to the conservation of genetic information. In addition, our data indicate that the second potential DnaQ homolog encoded by Ms4259 has a different mutational phenotype than the annotated Msm DnaQ. Our results also showed that the ΔMs4259 mutant had a mutational bias for the noncoding region (31% of the BPSs) (SI Appendix, Table S2), which was significantly different from that of the ΔdnaQ mutant (6%) and from the expected value of 10% (χ²= 16.6, P < 0.01). Therefore, the second potential DnaQ homolog encoded in mycobacteria is functionally different from annotated DnaQ.

Mycobacterial DnaQ Specifically Corrects DNA Replication Errors.

Spontaneous mutations can arise due to replication errors or as a result of intrinsic DNA damage (18). To determine whether the antimutational role of mycobacterial dnaQ is due to the correction of replication errors, we performed a fluctuation analysis and compared the mutation rate of WT and the ΔdnaQ mutant expressing the DnaE1[D228N] variant deficient in PHP domain-mediated proofreading (11). DnaE1 [D228N] expression had no effect on bacterial growth (SI Appendix, Fig. S2C). While expression of WT DnaE1 had no effect on the mutation rate of the dnaQ⁺ prototype strain (95% CI: [2.21 to 4.07] × 10⁻⁹) and the ΔdnaQ strain (95% CI: [2.11 to 4.01] × 10⁻⁹), expression of DnaE1[D228N] in the ΔdnaQ strain resulted in a mutation rate of 5.03 × 10⁻⁸ (95% CI: [4.18 to 5.95] × 10⁻⁸), nearly two times higher than that of the WT strain expressing DnaE1[D228N] (95% CI: [2.21 to 3.41] × 10⁻⁸) (Fig. 2A). Thus, the combination of dnaQ depletion and lack of PHP domain-mediated proofreading leads to a synergistic effect on the mutation rate, demonstrating that mycobacterial DnaQ can efficiently correct replication errors when the PHP-mediated proofreading activity is lacking. The synergistic effect also suggests that the mutagenesis phenotypes of the ΔdnaQ strain are not due to defects in the PHP domain proofreader. To investigate the role of mycobacterial DnaQ in DNA damage repair, we measured the stress-induced Rif^R frequency in strains exposed to hydrogen peroxide (H₂O₂) or ultraviolet (UV) radiation, which are the most common sources of endogenous and exogenous DNA damage, respectively (23). Exposure of the WT to H₂O₂ and UV resulted in 4- and 90-fold increases in Rif^R frequency, respectively; however, these mutational effects were independent of dnaQ, as the ΔdnaQ mutant had a Rif^R frequency similar to that of WT (Fig. 2B). These results indicate that mycobacterial DnaQ specifically corrects DNA replication errors but not DNA damage.

Fig. 2.

Mycobacterial DnaQ specifically corrects the errors produced during DNA replication. (A) Fluctuation analysis in the Msm WT and the ΔdnaQ strains expressing Msm dnaE1 or dnaE1[D228N]. Both strains contain a WT endogenous dnaE1 allele. The second allele (dnaE1 or dnaE1[D228N]) was integrated at the L5 attB site of the genome. Circles represent the mutant frequency. Estimated mutation rates (mutations conferring Rif^R per generation) are expressed as mean ±95% CI. Significance was determined using an unpaired t test for log-transformed values. (B) Rif^R mutation frequencies in the indicated strains exposed to 5 mM H₂O₂ or 25 mJ/cm² UV. (C) Eluates of immunoprecipitation (with high washing stringency) analyzed by silver staining. Arrows indicate DnaQ and DnaN identified by LC–MS/MS. (D) Peptide coverage of precipitated proteins identified by LC–MS/MS. HS, high washing stringency, LS, low washing stringency. ND, not detected. (E) Sequences of CBMs identified in mycobacterial DnaQ, DnaE1, NucS, and E. coli (Ec) DnaQ. The numbers indicate the peptide positions. (F) Pull-down eluates analyzed by Coomassie blue staining. DnaQ_Trx_His6, DnaQ fused to an N-terminal Trx domain containing His6 tag; DnaQ CMB⁻_Trx_His6, CBM-truncated DnaQ_Trx_His6. DnaN, native DnaN without tag. (G) The exonuclease activity of Msm DnaQ on single-stranded DNA (ssDNA). Reactions were performed with 24 nt 5′ fluorescein (FAM)-labeled ssDNA for 5 min. (H) Rif^R mutation frequencies in the indicated strains. All data are shown as the mean ±95% CI. Significance was determined using one-way ANOVA. *P < 0.05, ***P < 0.001.

The Clamp-Binding Motif (CBM) Is Essential for Binding of DnaQ to β Clamp but Is Dispensable for Its Antimutational Function.

Previous studies have suggested that the mycobacterial core replicase consists of αβ₂ε (encoded by dnaE1, dnaN, and dnaQ, respectively), with the β clamp acting as a bridge between α and ε (11, 14); however, in contrast to E. coli (24), mycobacterial ε does not form a stable complex with α. Our results showed that Coimmunoprecipitation (Co-IP) of whole-cell lysates with anti-FLAG (DYKDDDDK motif) antibodies coprecipitated the β subunit and a protein of unknown function (Ms4171) from the strain expressing the DnaQ fusion protein with a FLAG tag at the N terminus (DnaQ_N-FLAG), but not from the lysate containing DnaQ_C-FLAG or untagged DnaQ (Fig. 2 C and D and SI Appendix, Fig. S3 A–C). The failure of Co-IP of α in our assay could be due to the low affinity of β for α (14, 24). As expected, a reduction in the washing stringency of the Co-IP assay led to additional precipitation of α (DnaE1) and the clamp loader components δ’ (HolB) and τ (DnaX) (Fig. 2D). We found that mycobacterial DnaQ contains a highly conserved CBM Q[Y/L]ALF in its C terminus, and pull-down results showed that CBM is essential for the direct interaction between DnaQ and DnaN in vitro (25) (Fig. 2 E and F). Therefore, the failure of DnaQ_C-FLAG to coprecipitate DnaN is likely due to the inaccessibility of CBM to DnaN because of the fused tag. Consistent with a previous study (14), our results showed that the presence of DnaN did not inhibit the 3′-5′ exonuclease activity of DnaQ in vitro (Fig. 2G and SI Appendix, Fig. S3D). Surprisingly, however, the expression of the CBM-truncated DnaQ variant (DnaQ CBM⁻) in the ΔdnaQ mutant completely restored Rif^R frequency to WT level (Fig. 2H), suggesting that CBM was dispensable for the antimutational role of DnaQ. To determine whether the effect of DnaQ is solely dependent on the exonuclease domain, we removed the BRCT domain and CBM from DnaQ (DnaQ BRCT⁻). This truncation had little effect on DnaQ expression or stability in vivo (SI Appendix, Fig. S3E). However, expression of DnaQ BRCT^- could not complement the Rif^R frequency in the ΔdnaQ strain (Fig. 2H). Therefore, the BRCT domain is essential for the antimutational role of DnaQ, possibly through interaction with the other subunits of the core replicase (see also Discussion below).

Depletion of DnaQ Results in Replication Fork Perturbation.

These results indicate that mycobacterial replicase utilizes two proofreaders (the PHP domain and DnaQ). To investigate the physiological role of the additional proofreader DnaQ, we performed transcriptional profiling and analyzed the DNA damage response (DDR) signature during normal growth. According to previously published criteria (fold change of ≥1.5, FDR < 0.001) (26), 46 DDR genes were differential regulated (mostly up-regulated compared to the dnaQ⁺ prototype strain) in the ΔdnaQ mutant, accounting for 40% of the differentially regulated genes (46/114) (Fig. 3A and Dataset S2). In contrast, only three DDR genes were up-regulated in the ΔMs4259 strain. Interestingly, a dominant proportion (33/46) of the differentially expressed DDR genes in the ΔdnaQ mutant belonged to the PafBC regulon (26), resulting in over 29-fold enrichment (P < 10⁻¹⁵, Fisher’s exact test) (Fig. 3B and SI Appendix, Fig. S4A). These transcriptional signatures were further validated using qRT-PCR (SI Appendix, Fig. S4B). Because the PafBC regulator responds specifically to quinolones and replication fork perturbation (26), as shown by the increased expression of well-characterized fork repair genes including adnAB (a functional homolog of E. coli recBCD) and sbcD (Fig. 3C) (27–29), these results show that the loss of DnaQ leads to increased replication conflicts during normal growth. In contrast to the E. coli ΔdnaQ strain, which exhibited a constitutive SOS phenotype and growth defect (9), we did not observe apparent SOS signature in the ΔdnaQ strain (Dataset S2). In addition, our qRT-PCR results showed that the deletion of dnaQ had no effect on the induction of DNA repair genes under oxidative stress (SI Appendix, Fig. S4C).

Fig. 3.

DnaQ deficiency results in replication fork perturbation and affects DNA replication rate and fidelity upon exposure to sub-MIC FQs. (A) DnaQ-depleted cells induced a DDR. DDR genes (fold change ≥ 1.5 and FDR < 0.001) were categorized according to the established mycobacterial DDR regulons (PafBC/SOS regulon, PafBC regulon, and unknown). (B) Enrichment analysis of the differentially expressed genes in the DnaQ-depleted cells. (C) Differentially expressed genes in the PafBC regulon (fold change ≥1.5 and FDR < 0.001) involved in DNA replication and repair. End, endonuclease. (D) Growth of the indicated strains assessed by 10-fold serial dilutions on 7H10 plates with or without sub-MIC Ofx or Cip. (E) Growth rate of the indicated strains in the presence of sub-MIC Ofx. *P < 0.05, **P < 0.01 with an unpaired t test compared to WT. (F) Rif^R mutation frequencies in the indicated strains after exposure to sub-MIC Ofx. *P < 0.05, **P < 0.01 using an unpaired t test on log-transformed values. Data are presented as mean ±95% CI.

The mild upregulation of imuB (an accessory factor in DnaE2-dependent mutagenesis) and dinB homolog (Fig. 3C), both of which are crucial mediators of the translesion DNA synthesis (30, 31), raises the possibility that the increased mutation rate in the ΔdnaQ mutant could be due to error-prone mutagenesis (30). However, a recent study showed that deletion of dnaE2 or dinB123 in Msm does not alter the Rif^R frequency and the mutation spectra, and overexpression of dinB1 leads to GC-biased mutations (31). Therefore, the influence of dnaE2 and dinB homologs on the mutation spectrum was clearly different from that of the ΔdnaQ mutant (AT-biased mutations), suggesting that the mutagenesis phenotypes of the ΔdnaQ mutant were not due to error-prone mutagenesis.

Loss of DnaQ Reduces DNA Replication Fidelity upon Exposure to Subinhibitory Fluoroquinolones (FQ) Possibly due to Increased Vulnerability to Fork Collapse.

The loss of dnaQ had no effect on bacterial growth (SI Appendix, Fig. S2B), indicating that depletion of DnaQ probably does not affect replisome assembly (9, 10). Therefore, the increased expression of PafBC regulon, especially the fork repair genes, in the ΔdnaQ mutant could reflect a replication conflict under certain circumstances (23, 32). To identify this condition, we examined the susceptibility of the dnaQ⁺ prototype and the ΔdnaQ mutant to a panel of replication inhibitors and genotoxic agents. We observed no difference in growth or survival between WT and the ΔdnaQ mutant upon exposure to the DNA-damaging agents 4-nitroquinoline-1-oxide, UV (causing cyclobutene pyrimidine dimers, cross-links, and strand breaks), reactive oxygen species (menadione and tert-butyl hydroperoxide), the cross-linking and alkylating agents mitomycin C, and the topoisomerase IV inhibitor etoposide (SI Appendix, Fig. S5 A–D). However, the ΔdnaQ strain showed attenuated growth compared to the dnaQ⁺ prototype at the subminimal inhibitory concentration (MIC) ofloxacin (Ofx, MIC = 0.3 μg/mL) and ciprofloxacin (Cip, MIC = 0.12 μg/mL), both of which belong to the FQ that inhibits DNA gyrase A and topoisomerase IV (Fig. 3D and SI Appendix, Fig. S5 E and F). Interestingly, this effect was not observed with novobiocin, an inhibitor of DNA gyrase B and topoisomerase IV (SI Appendix, Fig. S5D). Consistent with the Rif^R results (Figs. 1C and 2H), this growth phenotype was complemented by WT DnaQ and DnaQ CBM⁻, but not by Exo⁻ or BRCT^- variants (Fig. 3D). Unlike the E. coli dnaQ mutant strain (33, 34), deletion of dnaQ in Msm had no effect on Ofx/Cip MICs or survival at bactericidal Ofx concentration, suggesting that mycobacterial DnaQ is not involved in FQ-mediated killing (SI Appendix, Fig. S5 F and G).

Both FQs and novobiocin induce the accumulation of supercoils ahead of replication forks. Therefore, the different growth behaviors of the ΔdnaQ strain toward these antibiotics indicate that topological stress alone is not sufficient to cause the growth phenotype. In addition to inducing topological stress, FQs can trigger DSBs and SOS responses (23). However, induction of DSB/SOS in the ΔdnaQ strain by genotoxic agents (e.g., UV and mitomycin C) did not phenocopy the effects of sub-MIC FQs (SI Appendix, Fig. S5 A–D), suggesting that FQ-induced growth failure may not be due to DSB or SOS.

A recent study on Msm showed that nalidixic acid (a quinolone) at concentrations above the MIC can induce rapid replication fork collapse, whereas novobiocin causes replication fork slowdown (35). These results suggest that the growth attenuation of the ΔdnaQ strain on sub-MIC FQs was likely a result of reduced DNA replication due to increased fork collapse. To test this, we measured the exponential growth rate, which correlates well with the DNA replication rate of cells on sub-MIC FQs (36). We found that the ΔdnaQ mutant exhibited a growth rate of ~90% of the WT in the presence of 0.33 × or 0.5 × MIC Ofx (0.1 and 0.2 μg/mL) (P < 0.05). The expression of the dnaQ in the mutant strain completely restored the growth defects (Fig. 3E). Furthermore, when cultured with sub-MIC Ofx, the ΔdnaQ mutant exhibited increased Rif^R frequency compared with the WT (P < 0.05) (Fig. 3F). Overall, these results indicate that the loss of DnaQ reduces DNA replication and replication fidelity upon exposure to sub-MIC FQs, possibly due to increased vulnerability to fork collapse.

dnaQ Is Subject to Positive Selection in Mtb L4.3/LAM Sublineage.

Early studies have shown that dnaQ exhibits high polymorphism in clinical Mtb isolates, suggesting a possible role for DnaQ in Mtb evolution (37, 38). We analyzed the dnaQ sequences of 51,229 clinical Mtb isolates collected worldwide (39). Compared with the most recent common ancestor DnaQ of the Mtb complex, we identified two lineage-defining mutations of DnaQ, with DnaQ A164V affecting L7 and L2 to L4 and DnaQ D76G affecting L4 (Fig. 4A). While dnaE1 was under strong purifying selection (dN/dS = 0.44), we found that the selection pressure on dnaQ was almost neutral (dN/dS = 0.97). This could be attributed to either relaxed purifying selection or a mixture of positive and negative selections on dnaQ. We examined the selection on dnaQ in different sublineages and found that dnaQ was subject to positive selection in some sublineages while under negative selection in others (SI Appendix, Table S7). The strongest positive selection was found in sublineage L4.3 that was formerly known as LAM sublineage and is considered to be a generalist with worldwide distribution (40).

Fig. 4.

A DnaQ V88A variant prevalent in the Mtb L4.3 sublineage confers hypermutability and is associated with drug resistance. (A) A representative phylogenetic tree highlighting the stepwise mutations of dnaQ in Mtb lineages. MTBC, Mtb complex; MRCA, the most recent common ancestor. (B) Phylogenetic tree of L4.3, highlighting the mutation events of dnaQ. Asterisk refers to a pre-stop mutation. (C) Comparison of SNP counts between Mtb strains within L4.3. WT neighbor refers to the closest phylogenetic clades to the dnaQ V88A or dnaQ G151R clades. (D and E) Fluctuation analyses in the Msm ΔdnaQ strain (D) and Mtb ΔdnaQ strain (E) expressing the dnaQ alleles identified in clinical Mtb. Circles represent the mutant frequency. Estimated mutation rates are expressed as mean ±95% CI. ****P < 0.0001, using a one-way ANOVA for log-transformed values compared to ancestral-type (D) or H37Rv WT (E). (F) Exonuclease activity of Mtb DnaQ on ssDNA. Exo⁻, DnaQ[D28A/E30A/D112A] lacking exonuclease activity. (G) Number of DR mutation events identified in each Mtb isolate with DnaQ A88 or V88 (A88 neighbor). n indicates the number of DR Mtb isolates. Significance was analyzed using the Mann–Whitney U test. (H) Proportion of resistance to individual antituberculosis drugs. Significance was analyzed using the Chi-square test. (I) Resistance profiles of Mtb isolates with DnaQ A88 or V88 (A88 neighbor) in Peru, the UK, and all 58 countries (ALL). The numbers denote Mtb isolates. DS, drug sensitive. DR#, number of resistant drugs.

In total, we observed 31 dnaQ mutation events in L4.3, with 26 of them being nonsynonymous (dN/dS = 1.96) (SI Appendix, Table S7). Among these mutations, dnaQ V88A arose at an early stage of L4.3 diversification and affected 16.5% (1,203/7,284) of the L4.3 strains. The G151R mutation also arose at mid-root position and formed a clade (204/7,284) (Fig. 4B). To test whether the positive selection on dnaQ has functional effects on the mutation rate of L4.3 strains, we compared the number of single-nucleotide polymorphisms (SNPs) accumulated in the dnaQ-WT and dnaQ-mutants. Overall, the dnaQ mutants accumulated an average of 14 more SNPs than the dnaQ-WT strains (P < 0.0001) (Fig. 4C). We also found that the dnaQ V88A and G151R clades had longer tip-to-root lengths than their closest neighbors without dnaQ mutation (Fig. 4B), providing further evidence that the naturally selected dnaQ mutants have functional effects on Mtb mutation rate.

The global distribution of L4.3 indicates that the selection on dnaQ was not driven by a specific host population (Dataset S3) (40). Alternatively, this unique evolutionary path could also be shaped by the genetic background of the sublineage. We therefore analyzed the genome sequence and found 62 nonsynonymous mutations occurred specifically in L4.3 (SI Appendix, Table S8). Two genes belonging to the DNA replication and repair pathways, DinB1 and RecC, were affected by these mutations (DinB1 T306P and RecC R535M). Interestingly, both dinB and recC were up-regulated in the dnaQ-null Msm (Fig. 3C), suggesting a functional relation between DnaQ and DinB1/RecC.

Naturally Selected DnaQ V88A Mutant Is Associated with Hypermutability and Extensive Drug Resistance.

To experimentally investigate the mutational effect of naturally selected DnaQ variants, we introduced dnaQ mutants into the dnaQ-null Msm or Mtb strain H37Rv via an integrative plasmid, from which the dnaQ allele is expressed via a constitutive promoter. The expression of these DnaQ variants in dnaQ-null Msm had no effect on bacterial growth (SI Appendix, Fig. S2D). By fluctuation analyses, we found that the DnaQ V88A mutant caused a sixfold (P < 0.0001) increase in the mutation rate compared with WT Msm (Fig. 4D), whereas the expression of other natural variants in the parallel experiments did not significantly alter the mutation rate. Consistent with the results of Msm, the expression of DnaQ V88A in the dnaQ-null Mtb strain H37Rv also increased the mutation rate by sixfold (P < 0.0001), suggesting the mutagenesis effect of DnaQ V88A was not derived from a strain-specific effect (Fig. 4E). Notably, the increase in the mutation rate was an effect of more Rif^R colonies, not driven by lower culture CFU (Dataset S4). These results indicated that the naturally selected dnaQ V88A is a potential mutator gene that leads to an intermediate level of hypermutability (41).

Interestingly, the expression of DnaQ Exo⁻ resulted in only a 2.4-fold increase in the mutation rate (Fig. 1C and SI Appendix, Table S1), suggesting that V88A does not simply impair exonuclease activity. The enzyme assay showed that the V88A mutation did not affect the exonuclease activity (Fig. 4F). These data suggest a role for the V88 region in regulating the overall fidelity of replicase. Indeed, most of the naturally selected dnaQ mutations in L4.3 were located outside the catalytic core of the exonuclease (SI Appendix, Fig. S1A), but not loss-of-function mutations (e.g., frameshift and nonsense mutations) (Fig. 4B).

A mathematical modeling study has shown that Mtb strains with eightfold increased mutation rate could result in a significantly increased risk of multidrug resistance (MDR) (42). To test whether the DnaQ[A88]-containing isolates were associated with MDR, we compared the number of DR mutations identified in each isolate between the DnaQ[A88]-containing population and the closest phylogenic neighboring isolates containing DnaQ[V88] (Dataset S3) (39, 43). These isolates were distributed across 58 countries. Among the isolates that harbored at least one DR mutation, those containing DnaQ[A88] accumulated more DR mutations than the DnaQ[V88] isolates (mean, 6.6 vs. 3.8; median, 6 vs. 4) (Fig. 4G). Further analysis revealed that the DnaQ[A88]-harboring isolates were more frequently associated with resistance to the second-line drugs, including levofloxacin, moxifloxacin, ethionamide, and the injectable drug kanamycin (Fig. 4H). Accordingly, the proportion of isolates resistant to 5 to 11 drugs was 2.2-fold higher in DnaQ[A88]-containing isolates than in DnaQ[V88]-containing isolates (36.7% vs. 16.9%) (Fig. 4I). To determine whether there was a country-dependent effect, we also examined Mtb isolates from Peru and the UK, where each had >30 sequenced isolates containing DnaQ[V88] or DnaQ[A88]. Again, in both countries, the proportion of Mtb isolates resistant to 5 to 11 drugs was at least twofold higher for the DnaQ[A88] isolates than for the DnaQ[V88] isolates (Fig. 4I). Overall, these data provide evidence that DnaQ[A88] can facilitate the de novo generation of DR mutations.

Discussion

In this study, we showed that the noncanonical DnaQ of mycobacteria acts as an additional proofreader and thus establishes a coproofreading activity involving both the PHP domain proofreader and noncanonical DnaQ (4). Although the contribution of DnaQ to overall fidelity is small compared to that of PHP domain proofreader, our results have shown that this additional proofreader plays a critical role in preventing AT-biased mutations and indels in HT, suggesting a division of labor between DnaQ- and PHP domain-mediated proofreading (Fig. 5).

Fig. 5.

Proposed model of action of DnaQ in Mycobacteria. (A) Schematic view of the architecture of core replicase between Mycobacteria and E. coli. The subunits shown are α (Mtb DnaE1 or E. coli DnaE), β (DnaN), and ε (DnaQ). (B) Structural model of the Mtb αε complex (predicted by AlphaFold Multimer) aligned with the E. coli cryo-EM structure (PDB 5M1S) (44). The V88 residue, located on a surface-exposed α-helix, is marked. The boxed region indicates the predicted interaction between α and ε. (C) Proposed model of action of DnaQ. See also Discussion. (D) Possible mechanisms underlying DnaQ V88A-induced hypermutability.

The independent evidence presented in this study suggests that mycobacterial 3′-5′ exonuclease DnaQ plays a role in maintaining replication fidelity. First, the results of the MA experiment showed that the loss of dnaQ leads to a significantly increased spontaneous mutation rate, a phenotype thought to be largely determined by the replication fidelity of DNA polymerases (5, 17, 19). Importantly, the antimutational role of DnaQ was strikingly dependent on exonuclease activity (Fig. 1C and SI Appendix, Table S1). Second, the synergistic effect of DnaQ depletion and lack of PHP domain-mediated proofreading on the mutation rate means that DnaQ can efficiently correct replicative errors when the PHP domain proofreader is defective (11). Furthermore, this synergistic effect also showed that the increased mutagenesis in the ΔdnaQ strain was not due to a secondary effect on the polymerase caused by the loss of DnaQ. Third, biochemical experiments suggest that mycobacterial DnaQ binds to the replication sliding clamp via its C-terminal CBM, linking the antimutational effect of DnaQ to replicase (see also below). Finally, deletion of dnaQ had no effect on stress-induced mutagenesis or bacterial tolerance to DNA damage caused by a panel of genotoxic agents, suggesting that mycobacterial DnaQ does not function in DNA damage repair. From these data, we concluded that mycobacterial DnaQ is a second proofreader that cooperates with the PHP domain proofreader to maintain replication fidelity (Fig. 5A).

A previous study using fluctuation analysis showed that deletion of dnaQ in Msm or Mtb does not significantly affect the mutation rate (11). These results were reproducible in our study by the same method (Figs. 1C and 4E). The discrepant results between the fluctuation assay and the MA in the ΔdnaQ strain could be largely due to the difference in the methodology of detecting mutations (17–19). For example, since the Rif^R phenotype is mainly (over 95%) caused by nonsynonymous BPSs in the 81 bp cluster I region of rpoB (45, 46), the increased indel events in the ΔdnaQ strain (detected by the MA assay) would not be adequately detected in the fluctuation assay by the selection of Rif^R. Moreover, the striking mutation bias in the ΔdnaQ strain (dominated by G:C>A:T) could also influence the mutation rate estimated from the fluctuation analysis (17). Therefore, although the fluctuation assays are more rigorous and give us a thorough idea of mutational frequency, the mutation rate extrapolated from these locus-specific measurements do not incorporate the contributions of all types of mutations. The MA assay is designed such that mutations occur in a neutral manner, devoid of selection pressure and, in combined with WGS, allows complete and unbiased detection of mutation events on the entire chromosome (17, 47).

Consistent with previous biochemical results (11, 14), our Co-IP results showed that mycobacterial DnaQ has a high binding affinity for β clamp (the only replicase subunit that is coprecipitated by high washing stringency). Through identification of the CMB and the subsequent domain deletion and complementation analyses, we found the BRCT domain, rather than the CMB, plays an essential role in the functions of DnaQ, such as antimutation and growth on sub-MIC FQs. Because the functions of DnaQ are also highly dependent on the 3′-5′ exonuclease, the most likely scenario for these results is that the BRCT domain can bind DnaQ to the replicase even in the absence of CMB. This prediction is supported by a previous study in which an unstable interaction between Mtb DnaQ and DnaE1 was observed in vitro (11), as well as by the results of structural modeling, which showed that the BRCT domain is in contact with the PHP domain of DnaE1 (Fig. 5B and SI Appendix, Fig. S6 A and B).

Unlike the E. coli DnaQ, mycobacterial DnaQ shows a preference for correcting G/C>A/T mutations. Previous studies have shown that AT-biased mutations are associated with mismatches caused by oxidized and deaminated DNA bases (48–50). However, our data suggest that the AT-biased mutations induced by the deletion of dnaQ may not be due to deficient repair of oxidative DNA damage, as the loss of dnaQ had no effect on H₂O₂-induced mutagenesis. Considering that mycobacterial DnaQ specifically corrected replicative errors and had no effect on the A/T>G/C mutation rate (GC-biased mutations), we propose that mycobacterial DnaQ preferentially corrects the A/T-G/C mismatches (primer-template, i.e., A mismatches with G or C, and T mismatches with G or C). A previous structural study showed that DNA polymerase I fragment (containing intramolecular exonuclease activity) tend to extend G-T, C-T, and G-G mismatches (leading to T>C, T>G, and G>C mutations, respectively), but is unable to extend A/T-G/C mismatches (leading to a mutation bias for AT) due to distortions in both the catalytic site of the polymerase and the geometry of DNA strand (51). This notion was corroborated by a recent cryo-EM study of the E. coli replicase, which showed the T-C mismatch enhances the fraying of the 3′ terminus of the primer, resulting in a ~55 Å translocation from the polymerase to the DnaQ (44). The preference of mycobacterial DnaQ for A/T-G/C mismatches is probably due to drastic distortions in DNA strand geometry that can make the DNA terminus inaccessible to the intramolecular PHP exonuclease (52). Failure of correcting A/T-G/C mismatches may result in AT-biased mutation or collapse of replication forks (Fig. 5C). Further biochemical and structural studies are required to fully elucidate the mechanisms underlying DnaQ substrate preference.

The preference of mycobacterial DnaQ to mismatches with distorted DNA strand geometry can be also inferred from the greater tendency of the ΔdnaQ strain to accrue indels at HTs (Fig. 5C). Indel mutations in HT are perceived as a result of DNA strand slippage, in which the HT misaligns to complementary sequences due to the formation of secondary structure in DNA strands (22). Such structures were found to be a causative reason for replication fork collapse (53). Of note, recent studies showed that the indels occurring in HTs have a strong impact on Mtb’s pathogenesis and drug tolerance (54–56).

These mutagenesis features suggest that the loss of DnaQ in mycobacteria results in inefficient repair of replication errors with drastically distorted geometry of DNA strand and thus leads to increased collapse of replication forks (Fig. 5C). This notion is supported by the observation that depletion of DnaQ causes upregulation of the PafBC regulon, in particular, the well-characterized fork repair genes adnAB and sbcD (27–29). In addition, through chemical genetic analyses, we showed that the reduced replication fidelity of the ΔdnaQ stain on sub-MIC FQs is likely a result of increased vulnerability of the ΔdnaQ stain to fork collapses.

The positive selection of dnaQ in Mtb L4.3/LAM suggests a role for DnaQ in the evolution of Mtb. Of the two clade-forming DnaQ mutations (V88A and G151R), the DnaQ V88A variant resulted in an intermediate level of hypermutability in both Mtb and Msm (41). In line with earlier findings on naturally evolved mutators of gram-negative pathogens and the mathematical modeling study in Mtb (41, 42, 57, 58), we showed that the DnaQ[A88]-carrying DR Mtb isolates accumulated more DR mutations and were more frequently associated with resistance to second-line drugs. Therefore, we propose that patients infected with the Mtb DnaQ V88A strain may be at an increased risk of treatment failures because this strain is more capable of acquiring DR mutations. The V88A mutation occurred at the mid-root position of L4.3 sublineage, with an average tip-to-root length of approximately 199.6 mutations. Considering that the long-term molecular clock rate of Mtb has been estimated to be around 4.6 × 10⁻⁸ substitutions per site per year (or ~0.2 mutations per genome per year) (59), this number of mutations dates the emergence of DnaQ V88A to approximately 998 y B.P., which is long before the use of antibiotics (1940 s). Thus, the selection pressure on dnaQ was not originally caused by the use of antibiotics, but its functional effects favor the evolution of drug resistance in Mtb. While we found that the DnaQ V88A mutation can lead to a sixfold increase in the mutation rate in vitro, the long-term substitution rate is also affected by natural selection, which purges deleterious mutations (60). As a result, the long-term substitution rate is often much slower than the mutation rate (60). However, even if we use a sixfold higher substitution rate, the estimated age (~166 y B.P.) of the DnaQ V88A mutation still predates the usage of antibiotics.

Interestingly, our enzyme assay showed that the hypermutable phenotype of DnaQ V88A was not due to loss of exonuclease activity. Considering that the mutagenesis effect of DnaQ V88A is more potent than that of DnaQ Exo⁻, a possible explanation for the hypermutability of DnaQ V88A is that the mutation impairs PHP domain-mediated proofreading. By structural modeling of the Mtb DnaE1-DnaQ (αε) complex, we found that the V88 residue is located on a surface-exposed α-helix and not in contact with DnaE1 (Fig. 5B and SI Appendix, Fig. S6 A and B). By aligning the model to the cryo-EM structure of E. coli core replicase (44), we found the V88 region is likely in a position close to the β clamp (Fig. 5B and SI Appendix, Fig. S6C). Based on these experimental and structural modeling data, we speculated that the V88A mutation may interfere with the interaction of DnaQ with other subunits (e.g., β), resulting in improper positioning of DnaQ in the replicase complex or impaired switching (possibly requiring subunit movement) between PHP domain- and ε-mediated proofreading (Fig. 5D). Further structural studies are needed to decipher the mechanism of DnaQ V88A-mediated mutagenesis, which may further our understanding of the dynamic regulation of mycobacterial proofreading activity.

Materials and Methods

More detailed information can be found in SI Appendix, Materials and Methods.

Strains and Mutants.

Msm mc²155 (The American Type Culture Collection, 706) and Mtb H37Rv (The American Type Culture Collection, 27294) were used in this study. Knockout mutant strains were generated by allelic exchange using a specialized phage transduction method (61). For the expression of dnaQ and its variants, the dnaQ alleles were PCR-amplified and cloned into an integrative single-copy plasmid pMV361 or the multicopy plasmid pMV261 using a Tet-on expression system (62). The strains and primers used in this study are listed in SI Appendix, Tables S9 and S10.

Culture Conditions.

Mycobacterial strains were grown in Difco Middlebrook 7H9 broth or on 7H10 agar supplemented with 0.5% glycerol, 0.05% Tween80, and 10% OADC (Mtb). Experimental cultures were started by inoculating the overnight culture into fresh medium to achieve an OD₆₀₀ of 0.01 to 0.02, and then incubated at 37 °C with shaking at 100 rpm.

MA assay.

The MA assay was performed as previously described (11, 19). The number of generations (n) was calculated as n = log₂N, where N is the number of cells per colony. See SI Appendix, Materials and Methods for a detailed description of MA assay.

WGS analysis.

Sequencing was performed on the Illumina X-10 instrument. The variant calling was performed with SAMtools and Genome Analysis Toolkit (63). A SNP and indels (≤30 bp) were called and filtered if i) we have at least 10 reads covering the site, ii) it was found at a frequency of >0.8. SNP that was observed in all lines of a strain was excluded (17).

Estimate of Mutation Rate from MA Assay.

The mutation rate estimate for the MA line was generated using the equation (11, 17, 50): μ = m/(N × g). m is defined by the number of observed variants (SNPs and indels), N was determined based on >99.9% coverage of Msm genome, and g is an estimate of the number of generations that occurred during the passage.

Fluctuation Analysis.

Fluctuation analysis was performed as described previously (42). The mutation rate was calculated using the FALCOR web tool (https://lianglab.brocku.ca/FALCOR/) using the Ma–Sandri–Sarkar maximum likelihood method (18, 64, 65).

Mutagenesis Assay.

Mutagenesis assays were performed to measure the frequency of Rif^R in the population. See SI Appendix, Materials and Methods for a detailed experimental procedure.

Immunoprecipitation.

FLAG-tagged dnaQ under the control of the native promoter was integrated into the attB site of the ΔdnaQ mutant strain. Immunoprecipitation was performed by using the magnetic beads and Monoclonal ANTI-FLAG^® M2 monoclonal antibody (Sigma Aldrich, F1804).

Protein Expression and Purification.

The pET32a and pET22b plasmids were used to express DnaQ and DnaN. Proteins were purified using Ni-Sepharose. See SI Appendix, Materials and Methods for a detailed experimental procedure.

Pull-Down Assay.

The N-terminal Trx_His_FLAG tag was removed by recombinant enterokinase. To perform the pull-down assay, Ni-Sepharose prebound with His-tagged DnaQ or DnaQ CMB⁻ was mixed with DnaN proteins and incubated for 5 min at 4 °C. The Ni-Sepharose was washed two times and eluted with sodium dodecyl-sulfate polyacrylamide gel electrophoresis sample buffer.

Exonuclease Assay.

The exonuclease assays were performed in a 30 μL reaction system with 20 mM Tris-HCl (pH 8.0), 100 mM NaCl, 10% glycerol, 2 mM MgCl₂, 2 mM dithiothreitol, and 50 μM bovine serum albumin. Reactions were performed at 25 °C with purified protein and DNA substrate 5′-fluorescein-GTTCACGAGACCTACTGACACTGA-3′. See SI Appendix, Materials and Methods for a detailed experimental procedure.

WGS Data Analysis.

WGS data for 51,229 Mtb isolates from 203 Sequence Read Archive projects were described previously (39). To test whether dnaQ was under positive selection in Mtb population, we used a previously described method (pNpS) to generate mutations in silico through a random substitution process (66). A more detailed description of WGS data analysis can be found in SI Appendix, Materials and Methods.

Protein Structure Modeling.

The Mtb αε complex model was predicted using AlphaFold Multimer (67), which had an AlphaFold Model Confidence (pLDDT) value of 88 for almost all residues. Structure alignment was performed by using MM-align (68).

Statistical Analysis.

Significance tests were performed using GraphPad Prism version 9.4.1. Normality and lognormality tests were performed for each dataset. All statistical tests were two-sided. For unpaired nonparametric tests, the Mann–Whitney U test was performed. For unpaired parametric tests, t tests, or one-way ANOVA followed by Dunnett’s test for multiple comparisons were performed.

Supplementary Material

Appendix 01 (PDF)

Dataset S01 (XLSX)

Dataset S02 (XLSX)

Dataset S03 (XLSX)

Dataset S04 (XLSX)

Data, Materials, and Software Availability

WGS and RNAseq data were deposited and are available as FASTQ files in the National Omics Data Encyclopedia (https://www.biosino.org/node) under Project ID OEP005532 (69). All other data are included in the article and/or supporting information.