Rad18-dependent SUMOylation of human specialized DNA polymerase eta is required to prevent under-replicated DNA

Translesion polymerase eta (polη) was characterized for its ability to replicate ultraviolet-induced DNA lesions that stall replicative polymerases, a process promoted by Rad18-dependent PCNA mono-ubiquitination. Recent findings have shown that polη also acts at intrinsically difficult to replicate sequences. However, the molecular mechanisms that regulate its access to these loci remain elusive. Here, we uncover that polη travels with replication forks during unchallenged S phase and this requires its SUMOylation on K163. Abrogation of polη SUMOylation results in replication defects in response to mild replication stress, leading to chromosome fragments in mitosis and damage transmission to daughter cells. Rad18 plays a pivotal role, independently of its ubiquitin ligase activity, acting as a molecular bridge between polη and the PIAS1 SUMO ligase to promote polη SUMOylation. Our results provide the first evidence that SUMOylation represents a new way to target polη to replication forks, independent of the Rad18-mediated PCNA ubiquitination, thereby preventing under-replicated DNA.

D NA polymerase eta (polZ) belongs to the Y family of specialized DNA polymerases, best characterized for their capacity to replicate DNA damages that block the progression of replicative DNA polymerases, a process called translesion synthesis (TLS) 1 . PolZ is particularly efficient and accurate on the most abundant damage induced by ultraviolet light, the cyclobutane thymine dimer (TT-CPD) 2,3 and hereditary mutations in the POLH gene are responsible for the skin cancerprone xeroderma pigmentosum variant (XPV) syndrome, highlighting the importance of TLS for genome stability. However, polZ, like other TLS polymerases, is highly errorprone on undamaged templates and its access to DNA is tightly regulated through several mechanisms. For instance, monoubiquitination of PCNA (Ub-PCNA) by the Rad18/Rad6 complex at stalled replication forks allows specific recruitment of polZ at damaged sites thanks to the cooperation of its PCNA-and ubiquitin-interacting motifs [4][5][6] . Direct interaction with Rad18 and phosphorylation also promote ultraviolet lesion bypass and cell survival 7-10 , whereas extraction from chromatin by the segregase valosin containing protein (VCP) and proteasomal degradation, presumably relying on ubiquitination of the TLS polymerase, were proposed to limit the extent of polZ-dependent synthesis after bypass and the subsequent mutagenesis [11][12][13] .
Recently, a new function of polZ at intrinsically difficult to replicate DNA loci was proposed in human cells 14,15 . Paragons of these loci are the common fragile sites (CFSs), which are DNA regions exquisitely prone to breakage upon mild replication stress, for instance when replicative polymerases are slowed down by a low dose of aphidicolin (APH). Incomplete replication of these loci generates DNA intermediates that can pass through mitosis, where they can be cleaved by endonucleases, generating gaps or breaks on metaphasic chromosomes 16,17 or form ultra-fine bridges resolved by the Bloom pathway 18,19 . Stigmata of incomplete DNA replication can also be observed in the G1 daughter cells by the formation of 53BP1 nuclear bodies (53BP1 NBs), which are proposed to shield the transmitted DNA damages until repair 20,21 . PolZ localizes at CFSs upon mild replication stress and is more efficient than the replicative pold to replicate CFS sequences able to adopt non-B conformations in vitro. Moreover, APH-challenged polZ-deficient cells show delayed completion of CFS replication, higher number of gaps and breaks in metaphase and accumulation of 53BP1 NBs compared with wild-type (WT) cells 14,15 . PolZ was therefore proposed to participate in the timely completion of CFS replication, thereby preventing the persistence of under-replicated DNA in mitosis and CFS instability. As most of the knowledge on polZ regulation comes from analysis of its canonical function at ultraviolet damage, it is not yet clear if this new lesion-independent function shares the same regulatory mechanisms.
Here, we show that, unexpectedly, polZ travels with replication forks during unperturbed S phase and that this relies on SUMOylation of the TLS polymerase on lysine K163. Abrogation of this post-translational modification (PTM) mimics the phenotype of polZ-deficient cells in response to low doses of APH, whereas it has a marginal impact after ultraviolet radiation. Rad18, independently of its ubiquitin ligase activity, promotes polZ SUMOylation by facilitating its interaction with its SUMO ligase PIAS1 and is required for polZ function at difficult to replicate loci. Permanently SUMOylated polZ overcomes the need for Rad18 and PIAS1 in this process. Altogether, these data unravel a new way to recruit polZ to replication forks, especially relevant during lesion-independent replication stress.

Results
polg and Rad18 travel with replication forks. The discovery of polZ involvement in the replication of difficult to replicate DNA loci suggests that the TLS polymerase can be recruited to replication forks in absence of DNA damage. It is known for long that overexpressed polZ forms nuclear foci that co-localize with replication foci (RF) in a subset of untreated S phase cells 22 but the localization of endogenous polZ remains elusive. We therefore performed iPOND experiment (isolation of proteins on nascent DNA) 23 in unchallenged MRC5-V1 fibroblasts. Nascent strands were pulse-labelled with the thymine analogue 5-ethynyl-2'-deoxyuridine (EdU) followed by conjugation of biotin on EdU and purification by streptavidin pull-down (Fig. 1a). Proteins associated to labelled DNA were analysed by western blot. PolZ was retrieved in the sample harvested immediately after the pulse but lost in the thymidine-chased sample (Fig. 1b). This behaviour is similar to the one of known replisome components, PCNA and the catalytic subunit of the replicative pold (p125), demonstrating that endogenous polZ travels with replication forks during unperturbed S phase. Interestingly, we found that Rad18, one of its regulators, also associated with nascent DNA.
SUMOylation on K163 drives Polg to nascent DNA. To better understand how polZ is recruited to replication forks, we made the assumption that it could rely on PTMs of the polymerase. We focused on the small ubiquitin-like modifier (SUMO) pathway, as it was shown that SUMOylated proteins are enriched at replication forks 24 and that SUMOylation was proposed to protect the C. elegans ortholog of polZ (polh-1) from degradation during DNA damage bypass 25 .
Therefore, to examine if human polZ is a SUMO target, 293FT cells were co-transfected with plasmids coding for WT polZ (polZ WT ) and His-tagged SUMO1 or SUMO3. SUMOylated proteins were purified on nickel (Ni) beads in denaturing conditions and analysed by western blot using three different anti-polZ antibodies (Fig. 2a). All the antibodies detected a slower migrating band in the pull-down, preferentially in the presence of His-SUMO3 (arrow). This band was no longer detected upon overexpression of the SUMO protease SENP1 but not of a catalytically dead SENP1 mutant (Fig. 2b), confirming that it is a SUMOylated species and suggesting that SENP1 is responsible for polZ deSUMOylation. SUMO-modified polZ was also detected  cells were pulse-labelled with EdU for 10 min. Cells were then crosslinked and harvested immediately or after a 1 h thymidine chase to allow replication forks moving away from the labelled DNA. Biotin was conjugated to EdU by click chemistry before cell lysis and chromatin fragmentation. EdU-containing DNA and associated proteins were purified on streptavidin beads. (b) Input and EdU-associated proteins (iPOND) were analysed by western blot using the indicated antibodies. No EdU: negative control processed as described in a but without EdU incorporation.
with Flag-polZ using an anti-Flag antibody ( Supplementary  Fig. 1a). The increase of the molecular weight of the polymerase (B40 kDa) suggests that SUMOylated polZ may contain more than one SUMO moiety. Mutation of K11 of SUMO3 to arginine (R), which prevents the formation of SUMO chains 26 , did not modify the apparent size of the modification ( Supplementary  Fig. 1b), showing that it is mono-SUMOylation(s). The two Ks SUMOylated in Caenorhabditis elegans polh-1 are conserved in human polZ; however, their mutations did not prevent its SUMOylation ( Supplementary Fig. 1c). To identify the SUMO acceptor site(s), we performed in silico analysis with three SUMOylation site-prediction software programs (SUMOplot http://www.abgent.com/sumoplot, seeSUMO 27 and SUMOsp 28 ) and tested K to R mutants of the common predicted sites. We identified K163 as the SUMO acceptor site using denaturing Ni pull-down ( Fig. 2c and Supplementary Fig. 1d). To confirm our findings, we co-expressed green fluorescent protein (GFP)-polZ WT or GFP-polZ K163R with HA-SUMO2 and purified GFP-polZ on GFP-trap beads followed by extensive washes in stringent denaturing conditions. A slower migrating band was detected by both anti-polZ and anti-HA antibodies only with GFP-polZ WT (Supplementary Fig. 1e). K163R mutation did not affect polZ ubiquitination ( Fig. 2c and Supplementary  Fig. 1d), in agreement with previous results mapping the ubiquitination sites in the C-terminus of the polymerase 29 and suggesting that SUMOylation is not a prerequisite for polZ mono-ubiquitination.
K163 lies in the catalytic domain of polZ, in the back of the palm domain, and the SUMOylation site is conserved in most vertebrates, at the exclusion of zebrafish (Fig. 2d,e and Supplementary Fig. 1f, ref. 30). To explore if SUMOylation can impact on the intrinsic activity of the polymerase, we generated, in addition to polZ K163R , a mimetic of constitutively SUMOylated polZ (polZ SUMO ) by inserting the sequence of SUMO2 in place of K163 ( Fig. 3a and Methods). PolZ K163R and polZ SUMO were fully competent for replication of undamaged DNA and for TT-CPD bypass in vitro (Fig. 3b-d). Hence, both non-SUMOylable and constitutively SUMOylated polZ retained full intrinsic polymerase activity and the introduced mutations do not alter the conformation of polZ catalytic site.   We next investigated the biological significance of polZ SUMOylation by establishing XPV cells stably expressing polZ K163R or polZ SUMO . Both mutants localized in the nucleus and we confirmed that polZ K163R is not SUMOylated in these conditions ( Supplementary Fig. 2). We first examined by immunofluorescence the capacity of these mutants to form spontaneous foci. Cells were pre-extracted with a detergent before fixation to unravel the fraction of polZ associated to nuclear structures and PCNA was used as a marker of RF 6,22 . Only 10% of polZ K163R S phase cells presented spontaneous polZ foci, compared with 40% for polZ WT (Fig. 4a,c). Moreover, polZ K163R foci were fainter although total polZ K163R amounts were similar to that of polZ WT ( Supplementary Figs 2 and 4a). In contrast, polZ SUMO was fully proficient in spontaneous focus formation (Fig. 4b,d).
To determine if the impairment of spontaneous focus formation reflects a defect of polZ K163R recruitment to replication forks, we performed iPOND in our stable cell lines. Whereas both polZ WT and polZ SUMO were found at replication forks, the K163R mutation abolished polZ recruitment to nascent DNA (Fig. 4e,f). In situ proximity ligation assay (PLA) between polZ and neo-synthesized DNA confirmed this finding (Supplementary Fig. 3). Importantly, although MRC5-V1 and polZ WT cells showed specific PLA signals compared with XPV cells, only background amplification was detected in polZ K163R cells, despite a 3-4-fold overexpression compared with endogenous polZ level ( Supplementary Fig. 2a). Altogether, these results strongly suggest that polZ association with the replication machinery in unchallenged conditions required its SUMOylation on K163.
SUMO-polg increases after replication stress. If SUMOylation on K163 constitutes a means to recruit polZ to replication forks, one obvious question is how this PTM impacts on the canonical and non-canonical functions of polZ during S phase. To answer this, we first determined the consequence of replication stress on polZ SUMOylation. PolZ WT cells were transfected with His-SUMO3 and exposed to ultraviolet-C or to low doses of replication inhibitors APH and hydroxyurea. Denaturing Ni pulldowns showed that SUMOylated polZ was readily observed in mock-treated cells and that its level increased after both DNA lesion-dependent or -independent replication stress ( Fig. 5a,b).
As previously observed 6,22 , ultraviolet-C exposure led to the accumulation of polZ WT in RF (Fig. 5c,d). PolZ K163R was also able to relocalize to RF after ultraviolet-C, although in only 45% of S-phase cells versus 70% for polZ WT and with a fainter staining (Fig. 5c,d and Supplementary Fig. 4a,b). In spite of this defect, polZ K163R was able to complement the ultraviolet sensitivity of XPV cells and polZ K163R cells were not further sensitized by addition of a low concentration of caffeine, a characteristic feature of XPV cells used for diagnostic 31 (Fig. 5e). Accordingly, polZ K163R prevented the accumulation of single-strand DNA during replication of ultraviolet-damaged DNA 32 as efficiently as polZ WT ( Supplementary Fig. 5). However, polZ K163R cells were significantly more sensitive than polZ WT cells at a higher ultraviolet-C dose ( Supplementary Fig. 4c), suggesting that polZ SUMOylation can contribute to its recruitment at ultraviolet-stalled forks. Interestingly, polZ SUMO cells displayed WT sensitivity to ultraviolet-C. However, we observed a slight but reproducible sensitization by caffeine at a high ultraviolet-C dose (Supplementary Fig. 4d-f), suggesting that deSUMOylation is required to ensure efficient polZ function at ultraviolet-damaged sites.
Abrogation of polg SUMOylation leads to under-replicated DNA.
In contrast to what was observed after ultraviolet, XPV and polZ K163R cells treated with a low dose of APH experienced (a) Schematic representations of the polZ mutants used in the study. In addition to WT and K163R polZ, a constitutively SUMOylated polZ (SUMO) was constructed by inserting the sequence of SUMO2 in place of K163. A track of 7 glycines (G) was added before SUMO2 to confer flexibility. The C-terminal di-G motif of SUMO2 was mutated to alanines to prevent cleavage by the SENPs. Two constructs were generated: SUMOa contains the atg of SUMO2, whereas it was mutated to G in SUMOb. A catalytically inactive polZ (polDEAD) was used as a negative control. similar replication problems, as evidenced by increased recruitment of RPA32 on chromatin compared with polZ WT cells ( Supplementary Fig. 5). Moreover, APH did not increase polZ K163R association to RF (Fig. 6a,b and Supplementary  Fig. 6a), indicating that mild replicative stress is not sufficient per se to restore polZ K163R focus formation. We then assumed that SUMOylation of polZ on K163 could be required to prevent the persistence of under-replicated DNA at difficult to replicate loci 14,15 . To test this hypothesis, we first analysed the transmission of DNA damage to the daughter cells in the next G1 phase following APH exposure. XPV cells displayed a higher number of 53BP1 NBs per G1 cell compared with polZ WT cells, as already shown 14 . This defect was not corrected by the stable expression of polZ K163R (Fig. 6c,d and Supplementary Fig. 6b). Interestingly, we found that polZ depletion in MRC5-V1 cells lead to segregation defects upon APH exposure with an increased number of anaphases presenting lagging chromosome fragments, in majority devoid of centromeric protein CENPA ( Fig. 6e-g). This phenotype, evocative of increased chromosomal breaks, was also observed in XPV cells compared with polZ WT cells and was not rescued by polZ K163R expression (Fig. 6h). Moreover, polZ K163R and XPV cells showed similar slightly higher sensitivity to a low dose of APH (Fig. 6i). In agreement with its efficient recruitment to replication forks, polZ SUMO complemented the APH-induced defects of XPV cells ( Supplementary Fig. 6c). However, this effect was only partial in the clone expressing the highest polZ SUMO level (#2), suggesting that overexpressed permanently SUMOylated polZ may interfere with the correct processing of some replication intermediates.
We then asked whether polZ SUMOylation impairment could affect genetic stability without impacting on cell survival after ultraviolet. We showed that polZ deficiency led to a dosedependent increase of anaphases with chromosome fragments after ultraviolet irradiation ( Supplementary Fig. 7). However, both polZ K163R and polZ SUMO were able to correct this phenotype, again arguing for a minor role of polZ SUMOylation at ultraviolet-induced DNA lesions.
Altogether, these results indicate that SUMOylation on K163 is required for polZ recruitment at replication forks and its subsequent involvement in preventing persistence of underreplicated DNA at difficult to replicate loci. Abrogation of this PTM mimics polZ deficiency in this specific function.
Polg recruitment on nascent DNA requires PIAS1 SUMO ligase.
To have a deeper insight into the regulation of polZ SUMOylation, we then aimed to identify the SUMO ligase responsible for this modification. In C. elegans, polh-1 is SUMOylated by GEI-17 (ref. 25), which belongs to the PIAS family of E3 SUMO ligases that counts four members in human cells (PIAS1-4). Although the SUMOylation sites are not conserved from worm to human, we asked whether the E3 SUMO ligase of human polZ could belong to this family. We showed that polZ co-immunoprecipitated with both PIAS1 and PIAS4 (Fig. 7a,b), two SUMO ligases already involved in the DNA damage response [33][34][35] . However, only PIAS1 depletion impaired polZ SUMOylation ( Fig. 7c and Supplementary  Fig. 8a). Conversely, PIAS1 overexpression enhanced polZ SUMOylation in a K163-dependent manner (Fig. 7d). These (e,f) iPOND experiments were performed as described in Fig. 1.
results indicate that PIAS1 is the E3 SUMO ligase of human polZ on K163. PLA between polZ and EdU showed that depletion of PIAS1 impaired the proximity of polZ with newly synthesized DNA in both polZ WT and MRC5-V1 cells but had no significant impact on the recruitment of polZ SUMO (Fig. 7e-g and Supplementary  Fig. 8b). Hence, recruitment of polZ to nascent strands requires PIAS1-mediated SUMOylation of the polymerase and all the above data strongly suggest that this modification occurred on K163.
Rad18 directly interacts with the last 158 aa of polZ via its polZ-binding domain (BD), which was mapped between amino acids (aa) 401 and 445 (ref. 8). To determine if this direct interaction is required for polZ SUMOylation, we first used a C-terminally truncated polZ (polZ  ) and showed that it was impaired in both SUMOylation and association to Rad18 ( Supplementary Fig. 9b). We next generated Rad18 truncation mutants lacking the polZ BD (Rad18 1-409 ) or the last 50 aa (Rad18 1-460 ), which contain a nuclear localization signal (NLS) between aa 488 and 494 (ref. 36). In addition, the NLS of SV40 T antigen was added to the N-terminus of the protein to restore nuclear localization of these mutants (Rad18 nls1-409 and Rad18 nls1-460 ). Disruption of the polZ BD abrogated polZ SUMOylation, independently of the presence of a NLS (Fig. 8e,f). Rad18 nls1-460 was able to promote polZ SUMOylation as efficiently as Rad18 WT , indicating that the last 50 aa of Rad18 were not required. Interestingly, Rad18 1-460 was able to interact with polZ ( Supplementary Fig. 9c)   APH Mock interaction between polZ and Rad18 is essential to promote polZ SUMOylation in the nucleus, in agreement with the known localization of PIAS1 (ref. 37).
As a matter of fact, we showed that Rad18 interacted with PIAS1 (Fig. 8g). This required a functional NLS but was independent of Rad18 association with polZ (Supplementary Fig. 9d). In contrast, depletion of Rad18 weakened the interaction between polZ and PIAS1 (Fig. 8h), indicating that Rad18 may target polZ to PIAS1 and/or bridge the two proteins together to allow efficient polZ SUMOylation. Interestingly, polZ SUMO overcame the need for Rad18 for its recruitment on nascent DNA (Supplementary Fig. 9e). Altogether, our data show that direct interaction between polZ and Rad18 promotes polZ SUMOylation and polZ recruitment to nascent DNA, independently of Rad18-mediated PCNA ubiquitination.
SUMO-polg and Rad18 act in the same pathway after APH. We next showed that depletion of Rad18 increased the number of anaphases with chromosome fragments (Fig. 9a and Supplementary Fig. 10a) and the number of 53BP1 NBs in the next G1 ( Supplementary Fig. 10b) after APH, in a similar manner than polZ depletion. Simultaneous depletion of the two proteins did not further aggravate these defects. We confirmed this in HCT116 cells, where depletion of polZ in WT cells increased the level of chromosomal fragmentation after APH to the one observed in mock-or polZ-depleted RAD18 À / À cells ( Fig. 9b and Supplementary Fig. 10c). Altogether, these data suggest an epistatic relationship between polZ and Rad18 in response to mild replication stress.
We then generated cell populations expressing WT or mutated Rad18 fused to GFP. Endogenous Rad18 was depleted using a siRNA directed against the 3 0 -untranslated region (3 0 -UTR) of the mRNA (siR18 3 0 -UTR) and cells were treated with a low dose of APH for 24 h before scoring anaphases with chromosome fragments in GFP-positive cells. Interestingly, both Rad18 WT and ubiquitin ligase deficient Rad18 C28F were able to rescue the segregation defects observed in endogenous Rad18-depleted cells ( Fig. 9c and Supplementary Fig. 10d). This suggests that ubiquitination of PCNA by Rad18 is not required in response to mild replicative stress, unlike what was previously observed after ultraviolet 38,39 . In agreement with that, depletion of polZ in cells expressing a non-ubiquitinable PCNA mutant (PCNA K164R ) led to increased chromosome fragmentation after APH ( Supplementary Fig. 10e). In contrast, analysis of cells expressing Rad18 C207F showed that integrity of the UBZ motif is critical for this pathway (Fig. 9d and Supplementary  Fig. 10d). These phenotypes correlate with the impact of the really interesting new gene (RING) and UBZ motifs on polZ SUMOylation.
Finally, we showed that depletion of Rad18, and to a lesser extent of PIAS1, increased the number of anaphases with chromosome fragments after APH in polZ WT but not in polZ SUMO expressing cells, which demonstrates that constitutively SUMOylated polZ overcomes the need for Rad18 and PIAS1 to act during mild replication stress (Fig. 9e,f). Interestingly, PIAS1 depletion significantly decreased the APH-induced mitotic defects in XPV cells, suggesting that PIAS1 may also be involved in the formation or processing of these fragments when the activity of polZ is compromised.
We propose that Rad18 promotes polZ SUMOylation by acting as a platform between the TLS polymerase and its SUMO ligase PIAS1, allowing polZ recruitment to replication forks and prevention of under-replicating DNA in response to mild replication stress.

Discussion
The regulation of polZ access to replicating damaged DNA has been under close scrutiny since its discovery, with two underlying issues: (i) how is polZ recruited to damaged sites, where its TLS activity is required, and (ii) how is TLS restricted to avoid mutagenesis on undamaged DNA? The recent discovery that polZ also acts at intrinsically difficult to replicate loci 14,15 modifies the way of apprehending TLS polymerase transactions on DNA. In this study, we uncovered a new mechanism, involving the SUMO pathway and Rad18, which regulates this non-canonical function of human polZ during S phase.
Our results showed that SUMOylation of polZ on K163 is required for its recruitment to RF during unperturbed S phase or under low replication stress and, to a lesser extent, after ultraviolet-C irradiation. This PTM is particularly important in response to APH, preventing accumulation of ssDNA during S phase, genetic instability and cellular sensitivity. In contrast, it is largely dispensable for the efficient bypass of ultraviolet-induced lesions. Furthermore, SUMOylation of polZ is promoted by direct interaction with Rad18 but independent of its ubiquitin ligase activity. We therefore propose that polZ is differentially regulated in response to DNA lesions and to intrinsic replication fork barriers (Fig. 10). During unperturbed S phase or under mild replication stress, when the amounts of Ub-PCNA are low, PIAS1-mediated SUMOylation on K163 targets or retains polZ in the vicinity of replication forks encountering difficult to replicate sequences, such as non-B DNA, promoting the timely completion of their replication. After ultraviolet exposure, polZ relocalizes to virtually all RF, where its accumulation mainly relies on PCNA ubiquitination, as already described 4,6 . Our results highlight a  central role for Rad18 in the regulation of polZ, as it is a key factor in both processes, which rely on distinct functional domains. Indeed Rad18, in complex with the E2 ubiquitin conjugating enzyme Rad6, is responsible for the ubiquitination of PCNA, a process requiring both its RING and SAP domains, and also directly targets polZ to damaged sites 8,40 . Here, we show that Rad18 promotes polZ SUMOylation in a UBZ-dependent manner by bridging polZ and its SUMO ligase PIAS1 and shares an epistatic relationship with the TLS polymerase in response to mild replication stress. Interestingly, these latter functions do not require a functional Rad18 RING domain and therefore the associated PCNA ubiquitination. However, as other ubiquitin ligases are able to ubiquitinate PCNA 41,42 , we cannot formally exclude that Rad18-independent ubiquitination of PCNA participates in polZ function at difficult to replicate DNA loci. In particular, it would be interesting to investigate the role of the E3 ubiquitin ligase CRL4 Cdt2 , as it is responsible for a fraction of PCNA ubiquitination in untreated cells 41 . Moreover, this E3 ligase targets some PCNA-interacting proteins to degradation after ultraviolet, a mechanism required for polZ focus formation 43,44 . As most CRL4 Cdt2 substrates are also degraded during the course of unperturbed S phase, it is tempting to speculate that this clearance pathway operates as well during the replication of difficult to replicate loci. We showed that the K163R mutation does not lead to a strong defect of ultraviolet-lesion bypass, as evidenced by cell survival experiments, lack of ssDNA accumulation in S phase and rescue of the mitotic defects observed in irradiated polZ-deficient cells. However, in agreement with the moderate impairment of focus formation after ultraviolet, polZ K163R cells display increased sensitivity to high ultraviolet doses than polZ WT cells, suggesting that SUMOylation on K163 indeed also participates to the accumulation of polZ at forks stalled by photoproducts. Recently, several studies have challenged the currently accepted model placing Ub-PCNA at the heart of TLS regulation, with data supporting Ub-PCNA independent pathway(s) for polZ activation [45][46][47] . We propose that SUMOylation on K163 provides an alternative way to recruit polZ at damaged sites when PCNA ubiquitination is compromised.
Hence, although canonical and non-canonical polZ functions during S phase could theoretically be reconciled in a unique tolerance mechanism requiring the same stalling/recruitment/ bypass steps irrespectively of the type of fork barrier, our data argue for a differential regulation of polZ at DNA damage and at   ARTICLE non-B DNA. We postulate that this may reflect requirement of different protein complexes or different impacts of these replication impediments on the structure of the replication intermediates, a subject that remains largely unexplored in human cells. Using iPOND to retrieve the proteins associated with nascent DNA, we showed that polZ and Rad18 travel with replication forks during unperturbed S phase. Noteworthy, during the preparation of this manuscript, two teams also identified Rad18 as a component of the replisome 48,49 . Our data on polZ are, to our knowledge, the first demonstration of a TLS polymerase association with protein complexes at nascent strands in unchallenged cells. This finding was rather unexpected, given the intrinsic low fidelity of the polymerase on undamaged templates. However, our observation fits well with the emerging concept of TLS polymerases involvement in the natural course of DNA replication 50 . Moreover, polZ presence in the replisome does not necessarily imply that it actively replicates DNA, a hypothesis supported by the limited number of interaction signals between polZ and nascent DNA observed by PLA. PolZ may be pre-recruited to rapidly cope with barriers impeding replication fork progression. Composition of the replisome varies in response to acute replication stress 23,48,51 . However, it is not yet precisely known if and how this complex is modulated in response to natural fork barriers and after mild replication stress. Therefore, it remains to be determined if polZ and Rad18 are constitutive component of the replisome or if they are specifically found in the vicinity of forks dealing with the replication of problematic DNA regions like CFSs.
The current model implies that Rad18 is recruited on chromatin through the ssDNA formed at stalled forks and therefore its DNA-binding domain SAP is required for ultraviolet-induced PCNA ubiquitination and polZ foci 38,52 . We found that polZ SUMOylation and prevention of segregation defects upon APH treatment rather rely on the UBZ domain of Rad18, a motif involved in Rad18 dimerization and subnuclear localization 36,38,53-55 but dispensable for PCNA ubiquitination, polZ foci formation and cell survival after ultraviolet 36,54 . Interestingly, the UBZ motif was shown to promote interaction of Rad18 with ubiquitinated chromatin components including histone H2A 38,56 . Hence, it may provide a way to recruit the Rad18/polZ complex to the replisome, independently of fork stalling, and/or may target them to specific DNA regions.
Our results showed that association of polZ with the replisome in unperturbed S phase required its PIAS1-mediated SUMOylation on K163. Interestingly, both PIAS1 and SUMOylated species are enriched on nascent DNA 24,48 . However, it is not yet clear if PIAS1 SUMOylates polZ in the vicinity of replication forks, despite the fact that we demonstrated that polZ is SUMOylated in the nucleus. As it has been reported for many other SUMOmodified proteins, the amount of SUMOylated polZ is very low compared with that of the unmodified protein and only unmodified polZ was detected at replication forks. SUMO conjugation can be a very transient event, yet having a prolonged impact on the target protein. The cycling model proposed to explain this apparent paradox stipulates that SUMOylation acts through cycles of conjugation/deconjugation, SUMOylation promoting an event, like the recruitment of the target to a protein complex, which can persist after SUMO clearance 57,58 . Based on this model, we hypothesize that highly dynamic SUMO attachment on polZ K163 allows polZ stable incorporation in the replisome. The factors chaperoning this process remain to be identified.
On the other hand, it is also tempting to speculate that polZ SUMOylated on K163 represent the DNA elongating form of the polymerase in vivo and that its small amount precludes any detection by the current methods. According to the crystal structure of human polZ, the K163 residue is located in the back of the palm domain, in the most flexible region of the catalytic domain 30 . Therefore, and in agreement with our in vitro data, it is unlikely that attachment of a SUMO moiety alters the conformation or polymerase activity of polZ. However, the fact that polZ K163R is proficient in ultraviolet lesion bypass in vivo suggests that SUMOylation is not a strict requirement for polZ activity. SUMOylation on K163 might then protect polZ from restrictive mechanisms during DNA synthesis, in reminiscence of what is observed in the nematode after damage 25 , the excessive turn-over of polZ K163R at DNA synthesis sites being partly compensated by its increased affinity for Ub-PCNA after ultraviolet.
Interestingly, polZ was found as a putative SUMO target by mass spectrometry analysis of cells treated with the proteasome inhibitor MG132 (ref. 59), suggesting that SUMOylation may be a prerequisite for polZ degradation. We confirmed this finding using denaturing Ni pull-down and showed that the SUMOylation events up regulated by inhibition of the proteasome are independent of K163. Therefore, SUMO pathway may fulfil two opposite roles: SUMOylation on K163 promotes polZ function at difficult to replicate loci, whereas multiple SUMOylations on other unidentified sites mark the polymerase for proteasomal degradation. Recently, the segregase p97/VCP, associated to its adaptator Spartan/DVC1, has been shown to extract polZ from the chromatin after lesion Whether polZ function at difficult to replicate DNA sequences also requires Rad18-independent PCNA ubiquitination remains to be established. After ultraviolet exposure, PCNA is ubiquitinated at forks stalled by photoproducts by the Rad18/Rad6 complex, which allows accumulation of polZ at damaged sites, as already described. However, SUMOylation of polZ on K163 may also contribute, to a minor extent, to the recruitment of the polymerase, constituting an alternative pathway in cells deficient in PCNA ubiquitination.
bypass 11,13 . VCP mostly acts on ubiquitinated proteins, but it is now demonstrated that it can also target SUMOylated proteins 60,61 . Interestingly, Spartan/DVC1 has been proposed as the functional homologue of the yeast metalloprotease wss1 (ref. 62), a partner of the yeast ortholog of VCP recently shown to bear a SUMO ligase activity 63 . One possibility is therefore that Spartan/DVC1 may be responsible for the SUMOylation events leading to polZ degradation. Further investigations are required to clarify this hypothesis. In summary, we identified a novel layer of regulation of polZ to prevent under-replicated DNA at difficult to replicate loci, which involves SUMO pathway and Rad18 but not Rad18-mediated ubiquitination of PCNA. We therefore propose that polZ is differentially regulated in response to DNA insults or to intrinsic replication fork barriers to maintain genome stability. The protein Rad18 serves as a common regulator for these distinct pathways.
Cell treatments. For ultraviolet-C irradiation (254 nm), cells were rinsed in pre-heated PBS and irradiated without any medium at a fluency of 0.65 J m À 2 s À 1 . APH and MG132 (Sigma) stock solutions were at 3 and 4 mM, respectively, in DMSO.
Transfections. Unless otherwise indicated, plasmids were transfected using jetPEI (Polyplus), according to the manufacturer's instructions. Cells were transfected with 30 nM of siRNAs using Interferin (Polyplus) according to the manufacturer's instructions and incubated for 48 h before treatment. In co-depletion experiments, 15 nM of each specific siRNA was used. siNT (15 nM) was added to ensure a final concentration of 30 nM when required. HCT116 cells were transfected with calcium phosphate. For analysis of SUMOylation by denaturing Ni pull-down, 293FT cells were seeded in 60 mm dishes 1 day before plasmid transfection (1 mg of pcDNA-POLH or GFP-POLH þ 2 mg of His or His-SUMO±1 mg of GFP-RAD18 or Flag-PIAS or Flag-SENP). Stable XP30RO-derived cell lines were seeded in 100 mm dishes and transfected with 7 mg of His or His-SUMO3 24 h before treatment. In depletion experiments, cells were transfected with siRNAs the day after seeding and further incubated 24 h before plasmid transfection. Plasmids were allowed to express for 24 h. Alternatively, co-transfection of siRNAs and plasmids were performed by calcium phosphate 48 h before harvesting. For denaturing GFP-trap, 293FT cells were transfected with 1 mg of plasmids expressing GFP, GFP-polZ WT or GFP-polZ K163R and 2 mg of HA-SUMO2. For immunoprecipitation experiments, 293FT cells were transfected with 2 mg of each of the indicated plasmids 24 h before harvesting.
Denaturing Ni pull-down. Cells were lysed in 500 ml of urea buffer (8 M urea and 20 mM imidazole in PBS) supplemented with 20 mM N-ethylmaleimide (NEM, Sigma) at room temperature and sonicated for 15 s with 30% amplitude (Vibracell, Bioblock Scientific). Extracts were centrifuged at 16,000g for 10 min at 15°C. 50 ml of supernatant was kept as input fraction and boiled for 10 min in 2 Â Laemmli buffer. Samples were incubated with nickel beads (His60 Ni Superflow resin, Clontech) for 45 min at room temperature on a wheel. Beads were washed four times for 5 min in 1 ml urea buffer. Proteins were eluted by boiling for 10 min in 2 Â Laemmli buffer with 30 mM EDTA and analysed by western blot.
Denaturing GFP-trap. Purification of GFP-polZ in stringent denaturing conditions was performed according to Chromotek's application note on ubiquitination of GFP-tagged proteins. Cells were lysed in GFP-trap lysis buffer (50 mM Tris-HCl pH 7.5, 150 mM NaCl, 1 mM EDTA, 0.5% Triton Â 100, 20 mM NEM, antiproteases Complete EDTA-free Roche) for 20 min on ice. Samples were sonicated twice for 10 s at 29% amplitude and cleared by centrifugation for 5 min at 9,500g at 4°C. Supernatants were incubated for 2 h 30 at room temperature on a wheel with 20 ml of GFP-trap agarose beads (Chromotek). Beads were washed once with GFP-trap dilution buffer (10 mM Tris-HCl pH 7.5, 150 mM NaCl, 0.5 mM EDTA, 20 mM NEM and antiproteases), three times with stringent washing buffer (8 M urea, 1% SDS in PBS) and once with 1% SDS in PBS. Bound proteins were eluted by boiling for 10 min in 2 Â Laemmli buffer.
Immunoprecipitation. Cells were lysed in NETN buffer (50 mM Tris-HCl pH 7.5, 150 mM NaCl, 1 mM EDTA, 0.5% NP40, antiproteases) for 30 min on ice and sonicated twice at 29% for 10 s. Samples were cleared by centrifugation at 9,300g for 5 min at 4°C. Immunoprecipitations were performed with 1 mg of antibodies (Bethyl rabbit anti-polZ #A301-230A, Sigma mouse anti-Flag-M2 #F4049 or Santa Cruz mouse anti-HA F-7 #sc-7392) for 3 h at 4°C on a wheel followed by 1 h 30 incubation in presence of sepharose-protein A beads (GE Healthcare). Beads were extensively washed in NETN, with 300 mM NaCl for the final wash, and denatured in 2 Â Laemmli.
In situ proximity ligation assay. Cells were pulse-labelled with 10 mM EdU for 5 min before pre-extraction and fixation as described above. PLA with nascent DNA was described elsewhere 69 . Briefly, cells were blocked with 3% BSA in PBS. Biotin-azide was conjugated to EdU by click chemistry and cells were incubated with primary antibodies against polZ and biotin (rabbit anti-polZ 1/300, Santa Cruz H300, and mouse anti-biotin 1/6,000, Jackson ImmunoResearch #200-002-211). PLA and EdU counterstaining were performed according to the manufacturer's instructions using the Duolink In Situ Red kit (Sigma) and goat anti-mouse Alexa Fluor 488 antibody.
In vitro transcription/translation of human polg and TLS assay. In vitro transcription/translation of full-length WT or mutant polZ was performed using a TNT-coupled reticulocyte lysate system (Promega) according to the manufacturer's instructions. The expression vector encoding polZ was added to the reaction mixture and incubated for 90 min at 30°C in the presence of [ 35 S] methionine. The catalytic activity of the DNA polymerase was analysed by primer extension on a circular single-stranded template (pUC118) and separation of the labelled products on a 20% polyacrylamide-7 M urea denaturing gel. Construction of single-stranded plasmids containing a single unique TT-CPD (pUC-CDP.ss) has been extensively described 70 . Primer extension analysis was performed as previously described 71 using a XP30RO cell extract supplemented with an equal amount of WT or mutated polZ. Briefly, the reaction mixture (6.25 ml) containing 10 fmoles of primed monomodified DNA and 20 mg of proteins was incubated 20 min at 37°C in 50 mM Hepes-KOH (pH 7.8), 7 mM MgCl2, 1 mM DTT, 4 mM ATP, 500 mM of dNTPs, 40 mM creatine phosphate, 100 mg per ml creatine kinase. The reaction was stopped by adding an equal volume of proteinase K-SDS (4 mg ml À 1 -2%) and incubated for 30 min at 37°C. Purified replication products were further digested with EcoRI and PvuII restriction enzymes and analysed by electrophoresis on a polyacrylamide-7 M urea denaturing gel. Radiolabelled products were visualized and quantified after phophorimaging (Typhoon FLA9500) using the ImageQuant TL software.
Data availability. The data that support the findings of this study are available from the corresponding authors upon request.