A gatekeeping function of the replicative polymerase controls pathway choice in the resolution of lesion-stalled replisomes

In an effort to determine how processive replication and TLS are coordinated, we reconstituted Pol IV-mediated TLS on a rolling-circle DNA template that contains a site-specific N²-furfuryl dG (N²-FFdG) lesion on the leading-strand template (Fig. 1A and SI Appendix, Fig. S1A). N²-FFdG is an attractive model lesion because Pol IV is proficient at replicating past N²-dG adducts and structurally related DNA lesions can be created in living cells by treatment with nitrofurazone (NFZ) (31). In rolling-circle replication, replication proceeds over the circular template multiple times, generating a long leading-strand tail that serves as a template for discontinuous lagging-strand synthesis (Fig. 1A) (2, 32, 33). Replication of the lesion-free control template by the reconstituted E. coli replisome resulted in the rapid formation of replication products that can be resolved by denaturing gel electrophoresis: A resolution-limited leading-strand product band and a distribution of smaller lagging-strand products along with a fraction of unreplicated templates (Fig. 1B and SI Appendix, Fig. S1B) (2). The resolution limit (RL) of our gel is ∼45 kilonucleotides, and therefore accumulation of leading-strand products at the RL indicates that at least 6 cycles of rolling-circle replication occurred on each template. Consistent with a prior observation that a single N²-FFdG blocks primer extension by Pol III (9), replication of the N²-FFdG–containing template in the absence of Pol IV was strongly attenuated by a single N²-FFdG on the leading-strand template (Fig. 1B). We also observed faint, discrete bands between resolution-limited replication products and unreplicated templates (Fig. 1 B and C). As these products were created only in the presence of N²-FFdG, they represent lesion-stalled replisomes that have undergone different multiples of replication around the template with TLS over the lesion inefficiently mediated by Pol III (34).

To determine whether Pol IV-mediated TLS might resolve this stalling, we examined the effect of Pol IV on replication of the N²-FFdG–containing template. Upon addition of increasing amounts of Pol IV, synthesis of both the leading and lagging strands was gradually restored and long leading-strand replication products accumulated at the RL before replication was fully inhibited by Pol IV at high concentrations (Fig. 1C and SI Appendix, Fig. S1E) (7). The robust formation of these resolution-limited products shows that TLS occurs efficiently over the lesion at the fork as repriming of leading-strand synthesis would result in a gap that terminates rolling circle replication. This termination results from the displacement of the circular leading-strand template that occurs when the helicase runs into the strand discontinuity resulting from repriming (SI Appendix, Fig. S1F).

Consistent with the requirements of Pol IV-mediated TLS in cells, the ability of Pol IV to promote replication of the N²-FFdG–containing template required both its catalytic and clamp-binding activities (SI Appendix, Fig. S1 D, E, and G) (17). Pol IV-mediated TLS occurred less efficiently over an N²-(1-carboxyethyl)-2′-deoxyguanosine (N²-CEdG) lesion and was strongly blocked by tetrahydrofuran (THF) (SI Appendix, Fig. S1H), indicating that our rolling-circle assay is sensitive to the efficiency by which Pol IV mediates TLS over various lesions (35, 36).

Moreover, at optimal Pol IV concentrations, replication of the N²-FFdG–containing template, as measured by quantifying leading-strand replication products (SI Appendix, Fig. S1C), was slowed only by ∼40% compared to replication of the control template (SI Appendix, Fig. S1I). Given that the polymerization rate of the Pol IV-based E. coli replisome is ∼10 nt/s (7, 9), which is over 50 times lower than that of the Pol III-based replisome, these results indicate that Pol IV only briefly switches with Pol III (9, 37). Collectively, these results demonstrate that within our reconstitution, Pol IV can efficiently mediate TLS at the replication fork by switching with Pol III, synthesizing a small patch of DNA over the lesion.

As Pol III core (αεθ) occupies both clefts of the β₂ clamp during processive replication (26) and Pol IV must bind to a cleft for its action, we sought to determine how Pol III disengages from the clamp to allow for Pol IV to mediate TLS at the fork. To address this question, we varied the strength of the interaction between the Pol III core and the β₂ clamp and examined the effect on TLS. The α and ε subunits of the Pol III core (αεθ) each occupy a cleft via independent CBMs, albeit with vastly different affinities; the ε–cleft contact is ∼250-fold weaker than the α–cleft interaction (Fig. 2A) (24, 26). Replacing the wild-type CBM of the ε subunit with a mutant CBM (ε_L), which binds the cleft ∼500 times tighter than the wild-type CBM (26), suppressed Pol IV-mediated TLS compared with the wild-type Pol III core (αεθ) and required higher Pol IV concentrations for optimal TLS (Fig. 2B). Normalizing replication of the lesion containing template by replication of the lesion-free control template (SI Appendix, Supplementary Methods) showed that the ε_L mutation suppressed Pol IV-mediated TLS to ∼40% (Fig. 2 B, Right Inset). Strengthening the interaction also modestly enhanced replicative activity on a lesion-free template (Fig. 2 B, Left Inset) (26). These results suggest that the disengagement of the ε subunit from the β₂ cleft at a lesion is necessary for Pol IV to bind to the β₂ clamp and perform TLS.

As the α subunit also contains a CBM, we next considered if disengagement of α was additionally required for Pol IV-mediated TLS (Fig. 2A). To address this possibility, we examined the effect of strengthening the α–cleft interaction within our biochemical reconstitution. Unlike the effects of strengthening the ε–cleft interaction, replacing the wild-type CBM of the α subunit with a mutant CBM (α^M3), which binds the cleft ∼100 times tighter than the wild-type CBM (24)—even ∼50 times tighter than the ε_L–cleft interaction—had little effect on replication of the lesion-free or lesion-containing template (Fig. 2C). Collectively, these results demonstrate that disengagement of only the ε subunit, not the α subunit, within lesion-stalled replisomes promotes Pol IV to bind to a cleft and perform TLS.

To validate these in vitro observations in cells, we introduced the dnaQ(ε_L) and dnaE(α^M3) mutations, individually or in combination, into their respective genomic loci and examined their effects on cellular sensitivity to DNA damaging agents NFZ and methyl methanesulfonate (MMS) (Fig. 2D). Cells lacking Pol IV (ΔdinB) were strongly sensitized to NFZ and MMS, indicating that Pol IV-mediated TLS contributes to cellular tolerance to these agents. Consistent with our in vitro observations, the dnaQ(ε_L) strain was sensitized to both NFZ and MMS, whereas the dnaE(α^M3) strain retained wild-type tolerance. Furthermore, the strain containing both the dnaQ(ε_L) and dnaE(α^M3) mutations [dnaQ(ε_L) dnaE(α^M3)] resembled the sensitivity of the dnaQ(ε_L) strain, indicating that the strengthened ε–cleft interaction is responsible for the increased sensitivity. This increased sensitivity of the dnaQ(ε_L) strain is due to defective Pol IV-mediated TLS because dnaQ(ε_L) was epistatic to ΔdinB (SI Appendix, Fig. S2A). The dnaQ(ε_L) and dnaE(α^M3) strains grew normally and retained nearly wild-type DNA content (SI Appendix, Fig. S2 B and C). Furthermore, the dnaQ(ε_L) strain retained similar numbers of replisome foci to the dnaQ⁺ strain both in untreated and NFZ-treated cells (SI Appendix, Fig. S2 D and E). The dnaQ(ε_L) and dnaE(α^M3) strains also retained nearly wild-type SOS-responses to both NFZ and MMS (SI Appendix, Fig. S2F). Collectively these results rule out the possibility that the increased sensitivity of the dnaQ(ε_L) strain results from a general defect in DNA replication or the DNA damage response.

Because the dnaQ(ε_L) mutation stimulates proofreading of Pol III core (αεθ) (27, 38), we next addressed whether the ε_L-containing replisome suppressed TLS through futile cycles of TLS and proofreading, impeding the transition from TLS to processive replication (SI Appendix, Fig. S3A) (39, 40). If the proofreading function of the ε subunit counteracts Pol IV-mediated TLS, abrogating the exonuclease activity of the ε subunit should promote Pol IV-mediated TLS. Indeed, when the catalytic residues of the ε subunit were mutated (αε^D12A,E14Aθ) (41), Pol IV-mediated TLS was promoted (Fig. 3A), indicating that the proofreading function of the ε subunit antagonizes Pol IV-mediated TLS over N²-FFdG, likely through futile cycles. In contrast to the effect on TLS, abrogating the exonuclease activity of the ε subunit reduced replication on the control template (Fig. 3A), indicating that exonuclease activity facilitates processive replication. Importantly, we found that the dnaQ(ε_L) mutation still strongly suppressed Pol IV-mediated TLS within the exonuclease-dead dnaQ-containing replisome (αε^D12A,E14Aθ), which was comparable to the suppression within the wild-type replisome (αεθ) (Fig. 3 A, Inset and SI Appendix, Fig. S3B). This result indicates that strengthening the ε–cleft interaction suppresses Pol IV-mediated TLS by inhibiting the interaction of Pol IV with the β₂ clamp rather than promoting futile cycles of TLS and proofreading.

Given that these results suggest an exonuclease activity-independent gatekeeping role of the ε subunit, we hypothesized that weakening the ε–cleft interaction or removing ε would make replication more potently inhibited by Pol IV and increase Pol IV-mediated TLS. To weaken the ε–cleft interaction, we introduced the dnaQ(ε_Q) mutation, which substantially decreases affinity for the clamp (K_D > 2 mM) (26). Indeed, the ε_Q-containing replisome (αε_Qθ) exhibited reduced replicative activity on the control template and replication was more potently inhibited by Pol IV as compared with the wild-type replisome (αεθ) (Fig. 3 B, Left Inset). These results indicate that access of Pol IV to the primer/template (P/T) junction during processive replication is inhibited by the ε subunit. Consistent with this gatekeeping role, spontaneous mutagenesis was elevated when the ε–cleft interaction was weakened (SI Appendix, Table S1). However, identifying the origin of this increase is difficult because weakening the ε–cleft interaction impacts not only the access of TLS polymerases but also the proofreading activity of Pol III core (27, 38). On the N²-FFdG–containing template, Pol IV-mediated TLS occurred more efficiently in the ε_Q-containing replisome (αε_Qθ) compared with the wild-type replisome (αεθ) when normalized for replication on the control template (Fig. 3 B, Right Inset). Removal of the ε subunit further increased the efficiency and potency of Pol IV-mediated TLS (Fig. 3 B, Right Inset), suggesting that in the absence of the ε subunit, Pol IV may frequently bind to the β₂ clamp even during processive replication. Similar observations were also made with the N²-CEdG–containing template (SI Appendix, Fig. S3C). These results demonstrate that it is a gatekeeping activity of the ε subunit rather than its exonuclease activity that limits the access of Pol IV and potentially other clamp-binding proteins to lesion-stalled replisomes.

Given prior observations that the reconstituted E. coli replisome can reprime leading-strand synthesis (42, 43), we next asked if suppression of TLS at the fork by strengthening the ε–cleft interaction increased repriming. To determine whether the N²-FFdG–stalled replisome reprimed downstream of the lesion, we employed Southern blotting to detect the expected repriming products (Fig. 4A and SI Appendix, Fig. S1F). In the absence of Pol IV, replication of the N²-FFdG–containing template created replication products that were detected with leading-strand specific Southern blot probes (Fig. 4B and SI Appendix, Fig. S4 A–C) (2). Replication products that ran as discrete bands above the template were leading-strand replication products that resulted from processive replication because these were detected only with leading-strand probes and were longer than the template (Fig. 4B and SI Appendix, Fig. S4C). Additionally, we detected diffuse replication products that ran below the template. These replication products were not created in the absence of DnaG, but unlike Okazaki fragments these were detected with leading-strand probes (Fig. 4B and SI Appendix, Fig. S4 C and D). These short leading-strand products were better detected with a distal leading-strand probe (1,901-nt probe) than a proximal leading-strand probe (40-nt probe) (Fig. 4A and SI Appendix, Fig. S4E), consistent with prior reports that repriming occurs a few hundred nucleotides downstream of the lesion (42). Collectively, these results indicate that replisomes stalled at N²-FFdG reprime DNA synthesis downstream of N²-FFdG, leaving an ssDNA gap between the lesion and a newly synthesized leading-strand RNA primer (Fig. 4A).

The addition of increasing concentrations of Pol IV to replication reactions of the N²-FFdG–containng template led to a gradual increase in TLS at the fork and a concomitant decrease in repriming, suggesting that Pol IV-mediated TLS competes with repriming (Fig. 4B and SI Appendix, Fig. S4 F–H). Indeed, when wild-type Pol IV (100 nM) was added with varying time delays after the initiation of replication, stalled replisomes were rapidly released into the TLS pathway and no further increase in repriming was observed (Fig. 4C).

Notably, for the ε_L-containing replisome, while TLS was suppressed, repriming persisted in the presence of higher concentrations of Pol IV as compared with the wild-type replisome (Fig. 4B). To examine whether strengthening the ε–cleft interaction increases the intrinsic propensity of the replisome to reprime, we compared the time course of repriming in the absence of Pol IV between the wild-type and ε_L-containing replisomes (SI Appendix, Fig. S4I). Both replisomes had a similar partitioning between Pol III-mediated TLS over the lesion and repriming with repriming occurring at similar rates (t_1/2,apparent ∼4 min). Therefore, the persistent repriming of the ε_L-containing replisome at higher concentrations of Pol IV most likely results from suppression of Pol IV-mediated TLS.

Our in vitro observations demonstrate that the ε–cleft interaction plays a decisive role in determining the fate of lesion-stalled replisomes. To test whether this also happened in living cells, we employed an in vivo assay that quantitatively measures the fraction of stalled replisomes that are resolved either by TLS or via the DA pathway at a site-specific lesion located in the chromosome. The experimental system (SI Appendix, Fig. S5 A and B), which is based on phage-λ site-specific recombination, consists of 2 major components: A recipient E. coli strain with a single attR site, and a nonreplicating plasmid construct containing the single lesion of interest, an attL site, and an ampicillin resistance gene. A site-specific recombination reaction between attL and attR is controlled by ectopic expression of phage-λ integrase and excisionase proteins (44) and leads to the integration of the lesion-containing plasmid into the chromosome. Integrants are selected based on their resistance to ampicillin. The chromosomal region where integration takes place carries the 3′-end of the lacZ gene fused to attR (minute 17 in the E. coli chromosome), while the remaining 5′-end is located on the incoming plasmid fragment in fusion with attL. Precise integration restores a functional β-galactosidase gene (lacZ⁺). A short sequence heterology inactivates the lacZ gene in the “opposite” strand across from the single lesion and serves as a genetic marker to allow strand discrimination during replication (45). The recipient strain carries both uvrA and mutS deletions to prevent the single lesion and the sequence heterology from being repaired by nucleotide excision repair and mismatch repair, respectively. Functional LacZ is expressed only when the lesion-containing leading-strand template is bypassed by TLS events with the exception of events in which a frameshift mutation is introduced (SI Appendix, Fig. S5 B and C). Therefore, colonies resulting from TLS events appear as blue/white-sectored colonies on X-Gal–containing plates (46). On the other hand, white colonies result from RecA-mediated HDGR (47). For most blue/white-sectored colonies observed, the blue and white sectors were roughly equal, indicating that in the majority of cases TLS occurred in the first round of replication.

To investigate Pol IV-mediated TLS, we introduced a single (−)-trans-anti-benzopyrene-N²-dG adduct [dG-BaP(−)] into the E. coli genome, known to be bypassed only by Pol IV (48, 49). In the dnaQ⁺ background, the majority of G-BaP(-)-stalled replisomes were resolved by the DA pathway, while only 20% by TLS (Fig. 5A). However, in the dnaQ(ε_Q*) background, which weakens the ε–cleft interaction, 55% of the stalled replisomes were resolved by TLS with a corresponding drop in the use of the DA pathway. As the basal SOS response in the dnaQ(ε_Q) strain was comparable to that in the dnaQ⁺ strain (SI Appendix, Fig. S5D), we ruled out the possibility that the increase in Pol IV-mediated TLS in the dnaQ(ε_Q*) strain resulted from elevation of Pol IV expression levels.

As the dnaQ(ε_Q*) mutation also reduces the processivity of proofreading (27, 38), we examined if the increase in TLS in dnaQ(ε_Q*) background was primarily due to a decrease in proofreading activity. To test this possibility, we abolished the exonuclease activity of the ε subunit with the mutD5 mutation (SI Appendix, Table S1A) (50, 51) and assessed the fraction of DA and TLS events over the G-BP(-) lesion. In the mutD5 background, participation of TLS increased from ∼20 to ∼34% but still remained significantly below the 55% value seen in the dnaQ(ε_Q*) strain (Fig. 5A). Therefore, the additional increase in TLS seen upon weakening the ε–cleft contact supports the model that the ε subunit plays a gatekeeping role in regulating access of Pol IV during TLS in living cells.

Next, we asked whether the ε–cleft interaction plays a similar gatekeeping role in Pol V-mediated TLS across TT-cyclobutane pyrimidine dimer (CPD) and TT(6–4) lesions. In the dnaQ⁺ background, replisomes stalled at TT-CPD and TT(6–4) were fully processed by the DA pathway as no Pol V-mediated TLS was measured. This lack of TLS is likely due to the absence of the active form of Pol V (UmuD′₂C) as UmuD is not processed into UmuD′ under non-SOS–induced conditions (Fig. 5B) (52). To observe Pol V-mediated TLS, we engineered a strain that carries umuD′ instead of umuD at the native umuDC locus (umuD′C strain); this strain constitutively expresses the active form of Pol V (SI Appendix, Table S1B) (53). In this background, Pol V-mediated TLS corresponded to ∼20% and ∼5% for TT-CPD and TT(6–4) lesions, respectively (Fig. 5B). Introduction of the dnaQ(ε_Q*) allele in the umuD′C background further stimulated TLS to ∼70 and ∼50% for TT-CPD and TT(6–4), respectively (Fig. 5B). In contrast, introducing the proofreading-deficient mutD5 allele in the umuD′C background, had no effect on the level of TLS, suggesting that the robust increase in TLS in the dnaQ(ε_Q*) strain resulted from an increased accessibility of Pol V to the clamp rather than from a defect in proofreading (Fig. 5B).

Finally, to investigate Pol II-mediated TLS, we introduced a N²-acetylaminofluorene dG (G-AAF) lesion in the NarI sequence context (SI Appendix, Supplemental Methods); this lesion can be bypassed by either Pol II- or Pol V-mediated TLS leading to −2 or error-free bypass, respectively (Fig. 5C) (5, 54). In contrast to what was observed for Pol IV and Pol V cognate lesions, Pol II-mediated TLS over the G-AAF lesion was comparably stimulated in the dnaQ(ε_Q*) and mutD5 strains (Fig. 5D). The similar levels of TLS in the dnaQ(ε_Q*) and mutD5 backgrounds are most likely related to the peculiar mechanism of Pol II-mediated −2 frameshift mutagenesis, which involves elongation of a “slipped lesion terminus” (Fig. 5C) (54, 55). As revealed in the crystal structure, Pol II is uniquely suited to elongate this frameshift intermediate (56). Upon stalling at the G-AAF lesion, Pol III dissociates from the P/T junction, allowing for the formation of a −2 slippage intermediate; this intermediate is thermodynamically favored by the stacking energy provided by insertion of the aromatic fluorene residue between the neighboring base pairs (57, 58). We speculate that Pol II-mediated TLS was promoted in both backgrounds because in the mutD5 strain the formation of the −2 slippage intermediate is facilitated due to the lack of Pol III-dependent proofreading, while in the dnaQ(ε_Q*) strain, Pol II has increased access to the clamp, resulting in promoted extension of this slippage intermediate.

With respect to error-free bypass of G-AAF by Pol V, TLS is stimulated (∼7-fold) in the dnaQ(ε_Q*) strain compared with the mutD5 strain (Fig. 5D), consistent with the effect of the dnaQ(ε_Q*) on Pol V-mediated TLS across TT-CPD and TT(6–4) lesions. As weakening the ε–cleft interaction in cells significantly increased the utilization of TLS over cognate lesions for all 3 TLS polymerases, we conclude that the ε–cleft contact acts as a gatekeeper to regulate access to the P/T junction.

During replication, replisomes encounter a diverse set of challenges that can stall replicative DNA polymerases, including DNA lesions and protein–DNA complexes (59, 60). It remains poorly understood how replisomes select among distinct rescue mechanisms to cope with these different situations. In this study, we demonstrated that for replisomes stalled at leading-strand lesions, the dynamic interaction between the exonuclease subunit (ε) of Pol III and the β₂ clamp plays a crucial role in pathway choice between TLS at the fork and repriming (Fig. 6A). We showed that the ε–cleft interaction acts as a steric gate to limit access of TLS polymerases to the P/T junction in a manner that is largely independent of its exonuclease activity. This model is consistent with in vivo imaging demonstrating that Pol IV is only weakly colocalized with the replisome in undamaged cells (14).

Upon polymerase stalling at a lesion, Pol IV gains transient access to the P/T junction by first binding to the cleft on the β₂ clamp that is occupied by the ε subunit during processive replication (b through d in Fig. 6A). Weakening the ε–cleft interaction (“opening the gate,” transition from b to c in Fig. 6A) markedly increased TLS at the fork while only modestly decreasing processive replication. Strengthening the ε–cleft interaction (“closing the gate,” transition from c to b in Fig. 6A) exerted the opposite effects. Therefore, the strength of this interaction is finely tuned to maximize processive replication on undamaged DNA while still enabling replisome remodeling by TLS polymerases upon the encounter of DNA damage.

Within stalled replisomes, recurring binding and unbinding between the ε subunit and the ε–cleft form a kinetic barrier to Pol IV-clamp binding. Unbinding may be stimulated by a bulky lesion in the template strand; the inability of the narrow catalytic cleft of the replicative polymerase to accommodate the lesion-containing template could promote opening of the Pol III–clamp complex. Alternatively, replicating Pol III core complexes may sample the open state (reversible transition between b and c in Fig. 6A) regardless of the presence of blocking lesions (61). In either case, strengthening the ε–cleft interaction suppresses these conformational transitions and therefore TLS, while weakening this interaction has the opposite effect.

Recent observations demonstrate that the E. coli replisome readily exchanges Pol III holoenzyme from cytosol during processive replication (62–64), suggesting that Pol III core dissociation from the clamp may be required for Pol IV binding (61). In contrast, our results suggest that the Pol III core remains bound to the β₂ clamp during Pol IV-mediated TLS as strengthening the α–cleft interaction had little effect on TLS in vitro and in cells. This model is consistent with prior observations that both Pol III core and Pol IV can bind the clamp simultaneously, the toolbelt model (9, 65, 66). Single-molecule imaging of a minimal reconstitution of β₂ clamp, Pol III core, and Pol IV demonstrated that higher concentrations of the Pol III core and Pol IV promoted simultaneous binding of the clamp (61); this simultaneous binding could be facilitated by interactions between polymerases and the replisome that locally concentrate Pol III core and Pol IV. We propose that during exchange of the Pol III holoenzyme, another Pol III core rapidly binds the β₂ clamp due to the preassociation of the Pol III holoenzyme with the replisome through its interaction with the DnaB helicase (67, 68) (SI Appendix, Fig. S6A). In essence this is a “dynamic toolbelt” model, in which rapid exchange of the Pol III core happens through the α–cleft while Pol IV carries out synthesis. Within this model the α–cleft interaction is maintained, which is essential to prevent other factors from accessing the P/T junction and compromising processive replication (7, 24).

It is also likely that the ε–cleft interaction limits the size of the TLS patch. Pol IV is expected to synthesize only a short patch before it falls off the template and loses its CBM–cleft interaction (transitions from c through f in Fig. 6A) (9, 37). Following the dissociation of Pol IV from the ε–cleft, the Pol III core must reestablish its ε–cleft interaction before it can resume processive replication (transition from f to a in Fig. 6A). This step would be much faster than expected for diffusion-controlled binding of the Pol III core if Pol III remained bound to the β₂ clamp while Pol IV performs TLS. Conversely, weakening the ε–cleft interaction could result in longer TLS patches and slower replication due to rebinding of Pol IV (Fig. 6A, transition from f to d rather than to a).

We demonstrated that TLS and repriming compete kinetically to resolve stalled replisomes. Failure to carry out TLS at the fork in a timely manner results in repriming, which acts as a failsafe to ensure replisome progression. A consequence of repriming is the formation of an ssDNA gap behind the fork (SI Appendix, Fig. S6B). This gap is initially coated by an ssDNA binding protein and subsequently converted to an RecA-ssDNA filament, which induces the SOS DNA damage response. Alternatively, TLS at the fork yields a continuous DNA product and therefore does not lead to induction of the SOS response.

Within our reconstitution, repriming downstream of an N²-FFdG adduct on the leading-strand template occurred quite slowly compared with Pol IV-mediated TLS at the fork. Similar to prior observations of repriming downstream of CPD lesions, we observed repriming occurring on the timescale of ∼4 min, implying that the rate of repriming is likely independent of lesion identity (42, 43). While direct measurements of the rate of repriming in cells have not been made, estimates (∼10 s) suggest that it could be much faster than in vitro observations (69, 70). Therefore, it is possible that other factors present in the cellular environment, but not in current biochemical reconstitutions, may facilitate repriming.

On the other hand, the rate of TLS is controlled by at least 2 factors: 1) the rate at which a TLS polymerase can bind the P/T junction and 2) the rate at which the TLS polymerase can insert and extend past the lesion. As we have observed here, access of a TLS polymerase to the P/T junction at the replication fork is regulated, in part, by the dynamics of the ε-gate. Increasing concentrations of Pol IV led to more frequent TLS at the fork within our biochemical reconstitution. Similarly, when the level of Pol IV was constitutively elevated as a result of eliminating LexA repressor binding sites in the promoter of the dinB gene (dinBp-lexA-mut), the MMS-induced SOS response was decreased relative to the wild-type strain, indicating that creation of ssDNA gaps is suppressed (Fig. 6B and SI Appendix, Fig. S6 B and C). These results show that higher levels of Pol IV help to overcome the gatekeeping barrier of the ε subunit by enabling Pol IV to more readily capture the ε open state.

As cellular concentrations of all 3 TLS DNA polymerases are increased upon induction of the DNA damage-dependent SOS response, it is likely that TLS at the fork plays a more prominent role in rescuing lesion stalled replisomes after SOS induction. During the early stages of the SOS response, when levels of TLS polymerases are relatively low, the majority of lesion-stalled replisomes are resolved by lesion skipping and most of the resulting ssDNA gaps are repaired by DA (Fig. 6B). This is supported by our in vivo observations that only small fractions of lesion-stalled replisomes were resolved by TLS, which is likely through a gap-filling process (Fig. 5 A, B, and D). A potential exception is Pol IV, which is present at relatively high levels even in the SOS uninduced state and can efficiently bypass cognate lesions (SI Appendix, Fig. S1I). In a ΔdinB strain, replisomes stalled at Pol IV cognate DNA lesions would be predominantly resolved by lesion skipping, leaving more ssDNA gaps than in a dinB⁺ strain. Indeed, when Pol IV-mediated TLS was completely abolished by the deletion of dinB, the MMS-induced SOS response was highly elevated compared with the wild-type strain (SI Appendix, Fig. S6C), consistent with an increase in repriming. Collectively, pathway choice is a net result of 2 counteracting molecular events—1) gatekeeping by the ε subunit and 2) mass action of TLS polymerases (71)—with the latter changing throughout the SOS-response. Failure of the TLS polymerase to gain access to the P/T junction or the inability to quickly bypass the lesion upon binding results in repriming and subsequent gap repair.

TLS at the fork prevents the creation of ssDNA gaps by outcompeting repriming. This gap-suppressing effect of TLS at the fork may play a pivotal role in determining the extent of damage-induced mutagenesis. A recent study has demonstrated that damage-induced mutagenesis is not limited to the lesion site but instead can spread a few hundred nucleotides from the lesion (72). This extended patch of low-fidelity synthesis likely results from gap filling synthesis where bypass of the lesion by the cognate TLS polymerase is followed by the sequential action of multiple polymerases, including the highly mutagenic Pol V, to fill in the remaining ssDNA gap (Fig. 6B) (73, 74). However, if TLS occurs at the fork, error-prone TLS may be limited to around the blocking lesion, as Pol III can readily regain control of the P/T junction (Fig. 6B).

Within cells the extent to which TLS occurs at the fork versus in an ssDNA gap created by repriming remains unclear. Lesion identity, which plays a role in determining the rate of TLS, almost certainly affects the partitioning between these pathways. Lesions that are strongly blocking, such as UV-induced lesions, are likely to be predominantly resolved through repriming due to the absence of Pol V in the early stage of the UV-induced SOS response (75). Even in the presence of highly elevated levels of Pol V, gap filling may still be the predominant mechanism as Pol V activation requires a RecA-ssDNA filament downstream of the stalled replisome (76–78). Weakening the ε-gate still dramatically increased Pol V-mediated TLS in cells (Fig. 5B), suggesting that the interaction between Pol III and the β₂ clamp likely influences polymerase switching independent of whether TLS occurs at the fork or in a gap (Fig. 6B). Biochemical studies presented here and in prior work, demonstrate that Pol IV can efficiently carry out TLS at the fork for at least a subset of lesions (79, 80). Cellular evidence for Pol IV-mediated TLS at the fork remains limited but the role of dinB in MMS-induced mutagenesis provides circumstantial evidence. 3meA is the major replication-blocking lesion created in MMS-treated cells and is bypassed primarily by Pol IV in a largely error-free manner (81–84). Intriguingly, deletion of dinB increases MMS-induced mutagenesis, which may result from more mutagenic TLS over 3meA by other polymerases at the fork (84, 85). However, we speculate that this increase is likely related to the creation of ssDNA gaps due to frequent repriming. In the absence of Pol IV, other polymerases mediate TLS over methyl lesions primarily in a postreplicative manner because inefficient synthesis past the lesion is likely outcompeted by repriming (Figs. 4B and 6B). As is the case for untargeted mutagenesis by UV lesions (72), the remainder of the gap can be filled in by mutagenic gap-filling synthesis with Pol V being the major mutator. In contrast, MMS-induced mutagenesis is reduced by constitutive expression of Pol IV (85). These observations are consistent with Pol IV-mediated TLS at the fork and suppression of ssDNA gaps (Fig. 6B).

Our result that the ε-gate antagonizes TLS raises the question if other factors may stimulate TLS in cells and therefore contribute to mutagenesis. Intriguingly, we previously showed that Pol IV is highly enriched near replisomes upon DNA damage in a manner that was only partially dependent on interactions with the β₂ clamp (14), suggesting that additional interactions, possibly with replisome components, are required for localization of Pol IV to stalled replisomes. Identifying these putative Pol IV–replisome interactions, along with factors that may promote repriming, will be important areas of future investigation.

Components of the E. coli replisome and other proteins were expressed and purified as described in SI Appendix, Supplemental Materials. E. coli strains were constructed by λ recombinase-mediated allelic exchange and P1 transduction as described in SI Appendix, Supplemental Methods. Rolling-circle templates were constructed by inserting a lesion-containing or control oligomer at the EcoRI site into M13mp7(L2) phage genome as described in SI Appendix, Supplemental Methods. Lesion-containing plasmids for in vivo tolerance assay were constructed as previously described (46).

The E. coli replisome was assembled and stalled on the template in the presence of ATPγS (50 nM), dCTP, and dGTP (60 μM each). Replication reactions were initiated by adding dATP and dTTP (60 μM each) together with [α−³²P]dATP. Quenched reactions were run on a 0.6% alkaline agarose gel, and replication products were visualized by autoradiography. Details can be found in SI Appendix, Supplemental Methods.

The GFP-based SOS-response reporter strains were treated with DNA damaging agents for a period and subsequently fixed with PBS containing 4% formaldehyde. Fixed cells were washed with and resuspended in PBS. GFP fluorescence from individual cells was measured by flow cytometry

Rolling-circle replication reactions were performed in the absence of [α−³²P]dATP. Replication products were separated in a 0.6% alkaline gel and transferred to a nylon membrane. Transferred DNA was fixed to the membrane by UV illumination. To probe replication products, fixed membranes were incubated with ³²P end-labeled oligomers, complementary to either a leading- or lagging-strand replication products and unbound oligomers were removed by washing. Replication products were visualized by autoradiography. Details can be found in SI Appendix, Supplemental Methods.

To assess partitioning of stalled replisomes into either DA or TLS, a single replication blocking lesion was incorporated into the E. coli genome, as previously described (46). Subsequently lesion-incorporated cells were grown on X-gal–containing LB agar plates, and the next day white and blue/white-sectored colonies, representing DA and TLS events, respectively, were counted. These colony counts were used to calculate fractional DA and TLS as described in SI Appendix, Supplemental Methods. To examine sensitivity of strains to DNA-damaging agents, overnight cultures were serially diluted and spotted onto NFZ- or MMS-containing LB-agar plates and incubated overnight. Dilution spots were pictured and analyzed.

All data are included in the main text and the SI Appendix. More details on materials and methods can be found in the SI Appendix.