Identification of novel DNA-damage tolerance genes reveals regulation of translesion DNA synthesis by nucleophosmin

Cells cope with replication-blocking lesions via translesion DNA synthesis (TLS). TLS is carried out by low-fidelity DNA polymerases that replicate across lesions, thereby preventing genome instability at the cost of increased point mutations. Here we perform a two-stage siRNA-based functional screen for mammalian TLS genes and identify 17 validated TLS genes. One of the genes, NPM1, is frequently mutated in acute myeloid leukaemia (AML). We show that NPM1 (nucleophosmin) regulates TLS via interaction with the catalytic core of DNA polymerase-η (polη), and that NPM1 deficiency causes a TLS defect due to proteasomal degradation of polη. Moreover, the prevalent NPM1c+ mutation that causes NPM1 mislocalization in ~30% of AML patients results in excessive degradation of polη. These results establish the role of NPM1 as a key TLS regulator, and suggest a mechanism for the better prognosis of AML patients carrying mutations in NPM1.

T he high abundance of DNA damage has led to the evolution of a variety of DNA-damage tolerance and repair pathways that minimize potential deleterious effects on DNA replication and gene expression, thereby preventing genome instability and a plethora of pathological conditions 1 . Most of the damage is repaired by accurate DNA repair mechanisms, primarily during the G1 phase of the cell cycle. However, during S phase, DNA replication encounters DNA lesions that have escaped repair, or that were newly formed, leading to the arrest of replication forks and/or the formation of single-stranded gaps, which may further lead to the formation of double-stranded breaks (DSB) and genome instability 2 . These replication obstacles are dealt with by DNAdamage tolerance 3 , of which two main strategies are known: (1) translesion DNA synthesis (TLS), whereby specialized low-fidelity DNA polymerases replicate across the damaged DNA region in a process that is inherently error prone [4][5][6][7] ; and (2) homologydependent repair in which the gap opposite the DNA lesion is filled-in by either physical transfer of the complementary strand from the sister chromatid or by using the latter as a template for copying the missing strand (also termed error-free post-replication repair or template switch repair) [8][9][10][11][12][13] . The importance of TLS is highlighted by the hereditary disease xeroderma pigmentosum variant (XPV), which is characterized by sunlight sensitivity and very high predisposition to skin cancer, caused by germline mutations that inactivate DNA polymerase-Z (polZ), a major TLS DNA polymerase 14,15 . The realization that TLS maintains a low mutagenic burden despite its inherent error-prone nature, and protects cells against genome instability and cancer, raised great interest in this process 5,6 . TLS usually involves two DNA polymerases: an inserter, which incorporates a nucleotide opposite the damaged template base, and an extender, which continues DNA synthesis beyond the damaged base 16,17 . Several layers of TLS regulation are known, including damaged-induced monoubiquitination of proliferating cell nuclear antigen (PCNA), the sliding DNA clamp, which serves to recruit TLS DNA polymerases to damaged sites in the DNA [18][19][20][21][22][23][24] and clearance of TLS polymerases from the DNA by the activity of DVC1 and p97 (refs 21,22). In terms of cell physiology, TLS largely operates uncoupled from DNA replication, during late S and early G2 phases of the cell cycle [25][26][27] , and is also regulated by the DNAdamage response via the ataxia telangiectasia and Rad3-related (ATR) protein 28,29 and via p53/p21 (refs 30-32). The high complexity of TLS in mammalian cells and its involvement in the development of cancer drug resistance [33][34][35][36][37] highlights the importance of understanding how this process is regulated. While previous studies aimed to systematically identify TLS genes in Saccharomyces cerevisiae proved to be highly useful in the field, to the best of our knowledge, screening for mammalian TLS genes has not been yet reported, and high-throughput assays for mammalian TLS are currently not available.
Here we present the development of a high-throughput assay for TLS in mammalian cells, and its implementation in screening 1,000 candidate genes. We further describe the validation of 17 novel TLS players, and the mechanistic and clinical insights revealed by investigating one of them, nucleophosmin, encoded by the NPM1 gene. We show that NPM1 regulates TLS by protecting polZ from proteasomal degradation, and that a deficiency in NPM1 as well as expression of the acute myeloid leukaemia (AML)-related NPM1c þ mutation results in decreased polZ levels and defective TLS. Our results uncover multiple novel TLS regulators in mammalian cells and implicate NPM1 in the proteolytic regulation of TLS polymerases.

Results
Two-stage functional siRNA screen for mammalian TLS genes. We performed a two-stage functional short interfering RNA (siRNA) screen designed to identify new mammalian TLS genes. In the first stage, we assayed ultraviolet sensitivity using an XPA cell line that is deficient in nucleotide excision repair (NER), and therefore defective in the repair of ultraviolet-induced DNA damage. Consequently, ultraviolet survival of the XPA cells exhibits a greater dependence on DNA-damage tolerance compared with NER-proficient cells 38 , making the screen more selective to DNA-damage tolerance genes. siRNAs that were identified in this stage as significantly affecting ultraviolet survival were re-screened with a second more stringent assay, which measured TLS. This strategy was used to screen 1,000 siRNAs directed to genes involved in DNA repair, ubiquitination and deubiquitination, cell cycle regulation and cancer. The ultravioletsensitivity screen (Fig. 1a) was performed in three biological replicas, exhibiting good reproducibility (Fig. 1b). Of the 1,000 genes assayed, we found 192 genes for which knockdown resulted in elevated ultraviolet sensitivity, and 45 genes that reduced ultraviolet sensitivity (false discovery rate (FDR) o8%; Fig. 1c, Supplementary Data 1, Supplementary Table 1 and Supplementary Fig. 1). Known TLS genes, as well as genes related to the ATR DNA-damage response pathway, but not other DNA repair pathways, were highly represented among the hits (Fig. 1d), suggesting that the XPA deficiency indeed enriched for genes involved in TLS.
The second screening stage was performed using a newly developed high-throughput TLS assay. It is a modification of an assay based on plasmids containing a site-specific DNA lesion opposite a gap, previously successfully used to study TLS 17,32,38 . In the new assay, the gap lesion is positioned between a cytomegalovirus (CMV) promoter and a firefly luciferase (Fluc) reporter gene (Fig. 2a). The presence of a gap in the transcribed strand does not allow expression of the Fluc gene, unless the missing segment of the coding strand is synthesized by TLS. A lesion-free gapped plasmid expressing Renilla luciferase (Rluc) served to normalize for transfection and gap-filling efficiencies (Fig. 2b). Of note, the product of a successful TLS event in the Fluc system still carries the lesion on the non-transcribed strand, which might interfere with Fluc expression. To test this possibility, we constructed the expected TLS product, namely a fully double-stranded plasmid with the lesion on the nontranscribed strand, and found that its Fluc expression in XPA cells was essentially identical to that obtained with a control plasmid without the lesion ( Supplementary Fig. 2a).
The high-throughput TLS system was utilized to screen 237 siRNA hits from the primary screen for involvement in TLS across each of two major ultraviolet irradiation-induced DNA lesions, namely thymine-thymine 6-4 photoproduct (TT 6-4 PP) and thymine-thymine cyclobutane pyrimidine dimer (TT CPD; Fig. 2c-e). Screening was performed in four independent biological replicas performed on different days and exhibited high reproducibility (Fig. 2f,g). To normalize for differential damage-independent knockdown effects on Fluc and Rluc expression, each siRNA was also tested using Fluc and Rluc plasmids both carrying gaps but no lesions (NLs), and the results served to correct the data ( Supplementary Fig. 2b,c). TLS elevation or inhibition scores were calculated with respect to control siRNAs, and were subjected to statistical analysis (Supplementary Fig. 2d-f). This screen resulted in 47 novel hits and 12 known TLS genes (FDR o10%; Supplementary Data 1 and Supplementary Table 1). Importantly, known TLS genes, but not other DNA repair pathways, were further enriched among the hits (Fig. 2h and Supplementary Table 2), indicating high specificity of the screening method towards the TLS pathway. The ATR signalling pathway was also enriched to some extent, in agreement with recent reports 28,29 . Repetitive examination of each hit using four different siRNA oligos, to minimize off-target effects, excluded 30 of the genes, yielding 17 validated novel TLS genes (Table 1).
Functional validation of candidate TLS genes. To further validate the hits, a second version of the TLS assay was developed based on quantitative PCR (qPCR) readouts rather than gene expression (Fig. 3a). In brief, following TLS in mammalian cells, the plasmids were extracted under alkaline conditions, which denatured gapped plasmids but not the covalently closed products of TLS or the control gap-filling reactions. Remnants of gapped plasmids were digested with S1 single-stranded endonuclease. TLS efficiency was calculated as the ratio between the products of qPCRs that targeted the filled-in gap-lesion plasmid and the filled-in control gapped plasmid, and was normalized to control siRNA-treated samples (Fig. 3a). The qPCR-based TLS assay was utilized to examine the involvement of 10 selected hits in TLS across the TT 6-4 PP ultraviolet lesion in mouse embryonic fibroblasts (MEFs). In parallel, TLS was measured across the major tobacco smoke-induced DNA adduct benzo[a]pyrene-guanine (BP-G), a non-ultraviolet lesion. Six hits, namely Papd7, Ruvbl2, Trip11, Npm1, Abh2 and Ube2e1, significantly affected TLS across both DNA lesions, although the effects obtained for Abh2 and Ube2e1 were rather small (Fig. 3b, t-test, P values o0.05; knockdown efficiencies are shown in Fig. 3c). Cyld, Mcm3, Dclre1a and Ercc4 affected TLS across TT 6-4 PP but not across BP-G (Fig. 3b, Cyld knockdown reduced TLS across BP-G, but gave a marginal P value of 0.06). Of notice, all the 10 tested human genes were validated in the qPCR-based TLS assay with at least one DNA lesion in MEFs. Four genes were further tested for their impact on TLS extent and mutagenicity in a gapped plasmid assay in which the readout is based on transformation of Escherichia coli cells, and therefore enables determining the sequence signature of TLS events 17,32,38 . Knocking down the expression of each of the tested genes, namely Cyld, Npm1, Papd7 and Ruvbl2, significantly reduced TLS efficiency across the BP-G adduct ( Fig. 3d; two-tailed t-test, P values o0.02). DNA sequence analysis showed that in control cells TLS was 47% accurate (insertion of a dCMP opposite BP-G), and 53% were point mutations, most of which (87%) were caused by insertion of dAMP opposite the lesion (Fig. 3e), consistent with previous results 17,39 . Interestingly, knocking down the expression of each of the four tested genes also caused, in addition to the decrease in the extent of TLS, a lower error rate among the TLS events (Fig. 3e).
NPM1 regulates DNA polg promoted TLS. Among the novel TLS genes, we concentrated for further analysis on NPM1, a gene frequently mutated in AML 40 . NPM1 (also termed nucleophosmin or B23) is a multi-functional protein involved in diverse cellular processes such as ribosome biogenesis, histone assembly, protein chaperoning and cell proliferation 41 . To explore the possibility that NPM1 affects polZ-promoted TLS, we tested TLS across three lesions that are bypassed by polZ: the sunlight-induced TT CPD and two adducts formed by the chemotherapy drug cisplatin, namely cisPt-GG and cisPt-GTG. As can be seen in Fig. 4a, in MEF cells in which Npm1 was knocked down, TLS across each of the three lesions was significantly reduced (two-tailed t-test, P values o0.02), with the TT CPD being most affected. In contrast, knocking down Npm1 in polZ-deficient PolH À / À MEF cells did not significantly affect TLS across the TT CPD lesion ( Supplementary Fig. 3a,b), further suggesting that the effect of NPM1 on TLS across TT CPD is mediated via polZ. This effect was not indirectly mediated via APE1, the base excision repair endonuclease that is regulated by NPM1, since knocking down the expression of Ape1 in the MEF cells had no effect on TLS across TT CPD ( Supplementary  Fig. 3c,d). Since NPM1 was reported to control the levels of PCNA 42 , we examined whether the effect on TLS is via reducing PCNA monoubiquitination, which is needed for effective recruitment of polZ to damaged sites in DNA 19,20 and for   effective TLS 23 . Interestingly knocking down the expression of NPM1 increased PCNA monoubiquitination rather than decreasing it ( Supplementary Fig. 4). This rules out the possibility that the TLS deficiency is owing to a PCNAubiquitination defect. A possible explanation for this observation is that the TLS defect caused by knockdown of NPM1 increases the number of stalled replication forks that leads to increased PCNA monoubiquitination.
NPM1 interacts with DNA polg in the nucleus. We next examined the interaction between polZ and NPM1 by coimmunoprecipitation (co-IP) of the endogenous NPM1 with endogenous polZ. As can be seen in Fig. 4b, immunoprecipitation (IP) of polZ with a specific antibody caused co-precipitation of NPM1. No NPM1 was observed when a control IgG was used (Fig. 4b, lane 2). To probe the subcellular compartment in which NPM1 and polZ interact, we resorted to the proximity ligation assay (PLA). As can be seen in Fig. 4c, interaction foci between polZ and NPM1 were clearly observed in human MRC5sv cells and localized mainly in the nucleus (the green spots in Fig. 4c upper panel, and quantified in Fig. 4d). These interaction foci were not observed in control XPV cells that lack polZ ( Fig. 4c middle panel, and Fig. 4d), but appeared in XPV cells complemented with the POLH gene, encoding polZ ( Fig. 4c lower panel, and Fig. 4d). The interaction foci appear to be distributed throughout the nucleus but were excluded from the nucleoli (Supplementary Fig. 5a). Similarly, in MRC5sv cells co-expressing green fluorescent protein (GFP)-tagged polZ and mCherry-tagged NPM1, GFP-polZ was distributed throughout the nucleus and in reduced amounts in the nucleoli ( Supplementary Fig. 5b). Taken together, these results indicate that NPM1 interact with polZ in the nucleoplasm. We next examined whether DNA-damage induction affects the interaction between NPM1 and polZ. Exposure of the MRC5sv cells to ultraviolet irradiation resulted in a decrease in the amount of polZ while the total levels of NPM1 were not significantly altered (Fig. 4e, input blot). Interestingly, following ultraviolet irradiation the interaction between polZ and NPM1 was transiently lost, and reappeared at later time points, as detected by co-IP (Fig. 4e). Similarly, analysis by the PLA showed a transient loss of interaction between polZ and NPM1 following ultraviolet irradiation (Fig. 4f, quantified in Fig. 4g). Ultraviolet irradiation of MRC5sv cells co-expressing GFP-tagged polZ and mCherry-tagged NPM1 resulted in recruitment of polZ but not of NPM1 into replication foci ( Supplementary Fig. 5b), further supporting the observation that NPM1 releases polZ following ultraviolet irradiation. To identify the region of polZ that interacts with NPM1, a series of FLAG-polZ deletion mutants was constructed and expressed in HEK293 cells. While IP of Repair pathway NPM1 protects DNA polg against proteasomal degradation.
To examine whether NPM1 affects polZ stability, we measured by flow cytometry the level of ectopically expressed GFP-POLH in MRC5sv cells that stably express this construct. Knocking down NPM1 in these cells resulted in a threefold decrease in the expression of the GFP-polZ construct (Fig. 5a, median values). When cells expressing GFP-RAD18, a gene fusion of the main E3 ligase responsible for monoubiquitination of PCNA were analysed, no reduction in RAD18 expression was obtained following knockdown of NPM1 ( Supplementary Fig. 6), which is in agreement with the functional PCNA ubiquitination in these cells. Next, we tested whether NPM1 regulates the protein level of endogenous polZ. Knocking down the expression of NPM1 in XPA cells caused a significant fourfold decrease in the amount of the NPM1 protein (Fig. 5b, lanes 1 and 2). A similar result was obtained with the MRC5sv cells (3.4-fold decrease; Fig. 5c, lanes 1 and 2). Under these conditions, the amount of polZ was significantly reduced both in the XPA cells (3.8-fold decrease; Fig. 5b, lanes 1 and 2) and the repair-proficient MRC5sv cells (2.9-fold decrease, Fig. 5c, lanes 1 and 2), with no significant change in POLH mRNA level, as measured by qPCR (Fig. 5d). A similar effect was observed in ultraviolet-irradiated cells (Fig. 5b lanes 3 and 4, and Fig. 5c lanes 3 and 4, corresponding to XPA and MRC5sv cells, respectively). Importantly, ectopic expression of an siRNA-resistant NPM1 construct rescued the reduction in polZ caused by the silencing of endogenous NPM1 (Fig. 5e). Thus, a reduction in the amount of NPM1 causes a decrease in the amount of endogenous polZ at the post-transcriptional level.
A similar reduction was also observed with polk, which is another TLS DNA polymerase, but not with poli, the closest homologue of polZ in mammalian cells (Fig. 5f). Inhibiting the proteasome using MG132 caused an increase in polZ levels in cells pre-treated with siRNA against NPM1, but not in those treated with control siRNA (Fig. 5g). This indicates that NPM1 has an important function in maintaining the stability of polZ, and protecting it from proteasomal degradation. The E3 ubiquitin ligases PIRH2 and MDM2 were implicated in the proteasomal degradation of polZ. Nevertheless, knocking down each of these enzymes or both, in addition to knocking down NPM1, did not rescue the polZ level ( Supplementary Fig. 7a,b), suggesting that the degradation of polZ upon NPM1 deficiency is not mediated solely by PIRH2 or by MDM2.
The C terminus of NPM1 is mutated in B30% of all AML patients, resulting in loss of the nucleolar localization signal and the generation of a de novo nuclear export signal. This leads to mislocalization of mutant NPM1 (designated NPM1c þ ), and partially of wild-type (wt) NPM1 interacting with it, in the cytoplasm 40,43 . These mutations strongly correlate with better response of patients to chemotherapy and better clinical outcome 44,45 . Because of the role of NPM1 in stabilizing polZ in the nucleus, we hypothesized that mislocalization of NPM1 to the cytoplasm might cause destabilization of polZ, similar to the effect that we have observed in cells in which NPM1 was knocked down. To this end, we examined polZ levels in OCI/AML2 and OCI/AML3 cell lines derived from AML patients, either without or with the NPM1c þ mutation, respectively. As can be seen in Fig. 6a, the amount of polZ was much lower in the NPM1c þ AML cell line compared with the NPM1wt AML cell line, whereas the POLH mRNA levels were similar (Fig. 6b). Consistently with the lower polZ amount, TLS across a TT CPD was lower (twotailed t-test, P value o0.003), and more mutagenic (w 2 -test, P value o0.01) in the NPM1c þ AML cells compared with the NPM1wt AML cells (Fig. 6c). Proteasome inhibition using The ARTICLE MG132 rescued the level of polZ in the NPM1c þ AML cell line (Fig. 6d), indicating that polZ is subjected to excessive proteasomal degradation, as was the case in cells in which NPM1 was knocked down. To further support the notion that it is the mislocalization of NPM1 in the AML cells that causes the polZ deficiency, we treated the cells with Leptomycin B, an inhibitor of nuclear export via exportin 1, previously shown to prevent the exclusion of NPM1 from the nucleus 46 . As can be seen in Supplementary Fig. 8, NPM1 was present mainly in the cytoplasm in NPM1c þ AML cells, but not in the NPM1wt AML cells, as expected. Upon addition of Leptomycin B, The NPM1c þ AML cells exhibited nuclear localization of NPM1, similarly to the NPM1wt cells. Examining the level of polZ under these conditions revealed that Leptomycin B treatment did not change the amount of polZ in NPM1wt AML cells, but caused a significant increase in polZ in the NPM1c þ AML cells (Fig. 6e), consistent with a role of NPM1 in stabilizing polZ in the nucleus. Furthermore, ectopic expression of mCherry-NPM1c þ in HEK293 cells that stably express GFP-polZ caused a significant twofold decrease in the amount of GFP-polZ as indicated by flow cytometry analysis (Fig. 6f,g). This effect was not observed when mCherry-NPM1wt or mCherry alone were expressed (Fig. 6h-k).  Taken together, these results establish a role for NPM1 in the regulation of TLS, and suggest that the prevalent NPM1c þ mutation results in reduced levels of polZ, thereby leading to defective TLS.

Discussion
The multiplicity of TLS DNA polymerases and their low fidelity suggest the existence of a complex regulation to ensure their action at the right place and the right time. The molecular basis of the regulation that enables a low and tolerable mutation burden is only partially understood. We reasoned that given its complexity, TLS in mammalian cells is regulated by additional as yet unidentified genes, and searched for those using a two-stage functional siRNA screen. Having two functional assays, one of which was specific for TLS, and a subsequent stringent validation protocol that included analysis by additional TLS assays, was anticipated to reduce the number of false hits. The high selectivity of the screens for TLS genes is demonstrated by the high enrichment of known TLS genes within the hits, in contrast to other DNA repair pathways. The switch to mouse cells for the subsequent qPCR validation enabled us to gain more validation power by using different siRNA oligos to target the mouse genes, and by demonstrating that the hits are not cell type-or humanspecific. It should be noted that the secondary gapped plasmidbased screen used in this study measures TLS outside the chromosomal context, and hence TLS genes that are associated with chromatin structure or replication fork progression are unlikely to be scored. Such TLS genes might have been detected in the primary ultraviolet-sensitivity screen. The E3 ligase RAD18 that mediates PCNA monoubiquitination was not scored in our screen, although we have previously shown that it is required for effective TLS in the gap-lesion plasmid assay 23 . This might have been caused by inefficient knockdown of the gene. Our screening strategy had led to the identification of 17 new TLS genes, six of which, namely NPM1, SMURF2, UBE2E1, CYLD, OTUB2 and VCPIP1, do not have obvious homologues in E. coli or S. cerevisiae. Of the 17 TLS genes, 10 were further examined by a transcription-/translation-independent qPCR version of the TLS assay in mouse cells, and at least four, namely CYLD, NPM1, PAPD7 and RUVBL2, also affected mutation rates across the BP-G lesion, further supporting their role in TLS. Most of our hits were detected when screening for TLS across the TT 6-4 PP lesion, with only few exceptions detected in the TT CPD screen, perhaps owing to the robustness of CPD bypass by the highly specialized polZ. Virtually, all of our hits seem to have a positive role in TLS, with the exception of USP1, which is a known negative TLS regulator responsible for de-ubiquitination of monoubiquitinated PCNA 47 .
Most of the new TLS hits can be classified into four functional categories, including DNA replication and repair genes, genes affecting proteasomal degradation, chaperons and Golgi-related genes ( Table 1). The replication/repair category contains five genes, including the helicase subunit MCM3, which is involved in the formation of replication forks; ALKBH2, which is involved in a direct reversal repair pathway; DCLRE1A, which functions in the repair of DNA crosslinks; ERCC4, which function in HRR as well as in NER; and RUVBL2. Scoring those genes as regulators of TLS suggests cross talk between DNA replication, repair and damage tolerance. Interestingly, the recruitment of DCLRE1A to sites of damage requires the ubiquitination of PCNA by RAD18 in a similar manner to TLS polymerases 48 . Since TLS polymerases are implicated in crosslink repair, it is tempting to speculate that DCLRE1A participates in coordinating the TLS and crosslink repair pathways, a task in which FAAP20 was recently implicated 49 . RUVBL2 plays an essential part in several chromatin-remodeling complexes including INO80 that regulates transcription by sliding nucleosomes on DNA. Of note, the INO80 complex is also required for removal of histones at sites of damage and was recently suggested to regulate DNAdamage tolerance 50,51 .
The next two categories include four genes involved in protein degradation by the proteasome and two genes encoding chaperons, consistent with the importance of regulating the stability of TLS components. UBE2E1 and UBE2G2 are E2 ubiquitin ligases; SMURF2 is an E3 ubiquitin ligase; and RPN1 forms part of the regulatory subunit of the 26S proteasome. The chaperons group includes NPM1, which is discussed below, and DNAJC6, which belongs to the DNAJ/HSP40 family of chaperons, and promotes uncoating of clathrin-coated vesicles. (c) TLS in OCI/AML2 and OCI/AML3 cells. TLS across a site-specific TT CPD using the gapped plasmid-based, colony-forming TLS assay is shown. Mean values±s.e.m. of three replicas are presented, two-tailed t-test, P value o0.003). Mutagenicity was addressed by sequencing 96 TLS events per cell line, w 2 -test, P value o0.01. (d) Proteasomal inhibition rescued PolZ deficiency in NPM1c þ AML cells. PolZ levels in OCI/ AML2 and OCI/AML3 cells pretreated with MG132 for the indicated time to inhibit proteasomal activity. Tubulin served as a loading control. (e) NPM1 nuclear retention rescued PolZ deficiency in NPM1c þ AML cells. PolZ levels in OCI/AML2 and OCI/AML3 cells pretreated with Leptomycin B (LMB) for the indicated time or its solvent as a control. Tubulin served as a loading control. (f-k) Density plots and histograms representing fluorescence-activated cell sorting analysis of HEK293 cells stably expressing GFP-polZ, and partially transfected with mCherry-tagged NPMc þ (f,g), wt NPM1 (h,i) and control mCherry (j,k). Data from mCherry-negative (untransfected) cells in (g,i,k) are in red and correspond to the left y axis, while that from mCherry-positive (transfected) cells are in blue and correspond to the right y axis. All blots are representative of three independent experiments. siCont, control siRNA. NATURE COMMUNICATIONS | DOI: 10.1038/ncomms6437 ARTICLE Of note, the Caenorhabditis elegans homologue of DNAJC6 (w07a8.3) was previously scored as a hit in a mutagenesis screen, which further supports its involvement in TLS 52 . The category of Golgi-related genes includes TRIP11 and VCPIP1. VCPIP1 is a deubiquitinating enzyme that mediates the reassembly of Golgi stacks after mitosis. Importantly, it also interacts with VCP/p97, which was recently found to regulate TLS 21,22,53 . There are four additional TLS hits including two de-ubiquitination enzymes, namely OTUB2 and CYLD, the latter being a negative regulator of the NF-kB pathway; SENP2 that participates in SUMO signalling; and PAPD7 that is an RNA-specific ribonucleotidyl transferase that plays a role in the turnover of aberrant mRNAs 54 . The new TLS genes, which belong to diverse pathways and functions, are likely to be useful for further analysis of TLS in mammalian cells and its interaction with other cellular pathways.
The finding that NPM1 is a regulator of TLS adds an unexpected dimension to the role of this multi-functional protein in preventing genome instability [55][56][57] , and uncovers a new key player of TLS regulation via proteasomal degradation. By protecting polZ from degradation, NPM1 enables TLS to fulfil its function of overcoming replication obstacles, thereby preventing DSB and genome instability 58 . It was previously shown that two E3 ubiquitin ligases promote polyubiquitination of polZ and target it for proteasomal degradation: MDM2 (ref. 59) and CRL4(CDT2) (ref. 60). In addition, PIRH2 targets polZ to proteasomal degradation via a ubiquitinationindependent pathway 61 . Thus, the cellular levels of polZ are tightly regulated. Human HSP90 was reported to facilitate the corrected folding of polZ into its active form 62 , and GEI-17 was shown to prevent the polyubiquitination and subsequent degradation of polZ by CRL4(CDT2) following genotoxic stress in C. elegans 60 . While GEI-17 is dispensable for polZ stability at steady state, NPM1 is critical for maintaining a functional pool of polZ in the absence of external DNA damaging agents. Thus, even if GEI-17 activity is conserved in mammals, it does not seem to overlap with the role of NPM1 in protecting polZ stability, but rather work subsequently. The transient release of polZ from NPM1 following ultraviolet irradiation might enable polZ to reach damaged sites in the DNA through its affinity to monoubiquitinated PCNA and preform TLS, however, this dynamics of polZ needs further investigation.
It was recently reported that polZ is involved in the resistance to araC, a drug routinely used to treat AML, which exerts its therapeutic activity via incorporation into DNA and inhibiting replication 63 . Thus, our results may explain, at least in part, the better prognosis and response to chemotherapy treatment of AML patients carrying the NPM1c þ mutation, which causes exclusion of NPM1 from the nucleus, leading to a polZ deficiency, and therefore more effective killing of the leukaemic cells by drugs such as araC. PolZ was recently implicated in the cellular response to doxorubicin treatment 64 . Although we demonstrated that the NPM1c þ OCI/AML3 AML cells carry low levels of polZ, they were reported to be more resistant to doxorubicin than OCI/AML2 cells 65 . The sensitivity of the AML cells to doxorubicin is most likely governed by processes other than TLS such as DSB repair and apoptosis, which might also be regulated by NPM1. Such processes can mask the effect of polZ, which might be secondary for this drug. The results presented here underscore the critical role of NPM1 in maintaining genome stability, and highlight polZ and NPM1, and in particular their interaction, as potential targets for development of drugs for AML therapy.

Methods
Plasmids construction. pGL4.13 and pGL4.73 reporter plasmids (Promega) served as backbones for the reporter gapped plasmids used in the screen. Restriction-free cloning was performed using PfuUltra II Fusion High-Fidelity DNA Polymerase (Stratagene). The following cloning steps were performed: (i) Bsa1 and BstX1 restriction sites were mutated; (ii) the SV40 promoter was replaced with a CMV promoter, taken from pRL-CMV (Promega); (iii) the ColE1 origin of replication was relocated to the region in between the CMV promoter and the reporter gene; and (iv) the AMP resistance cassette was replaced with a chloramphenicol resistance one, and was relocated in the firefly construct to the region in between the CMV promoter and the reporter gene. The mCherry-tagged NPM1 and NPM1c þ constructs were produced by fusing the mCherry coding region (Clontech) to the carboxyl terminus of the human NPM1 or NPM1c þ coding region. FLAG-tagged polZ truncations were produced by introducing the FLAG epitope in the amino terminus of the human POLH coding region, and deleting the indicated fragments by restriction-free cloning. siRNA-resistant human NPM1 construct was design by introducing silence mutations in the sequence corresponding to the siRNA oligo: 5 0 -GAGCACCAGUUAUCUUUAA-3 0 .
Ultraviolet-sensitivity screen. Human XPA cells were reverse-transfected in 96well plates with 25 nM siGenome SMARTpool siRNA libraries (Dharmacon), using Hiperfect transfection reagent (Qiagen). Following incubation of 2 days, cells were washed twice with Hanks' buffer (Sigma) and irradiated in Hanks' buffer at a dose of 1 J m À 2 ultraviolet C using a low-pressure mercury lamp (TUV 15W G15T8, Philips). Following irradiation, the cells were maintained under growth conditions for 48 h. Cell viability was determined using CellTiter-Glo (Promega). In order to reduce misinterpretation of toxic effects of certain siRNAs, each sample was normalized using its corresponding non-irradiated sample transfected with the same siRNA. Ultraviolet sensitivity was calculated with respect to control wells.
TLS screen. Human XPA cells were transfected with the siRNA hits from the ultraviolet-sensitivity screen at a concentration of 10 nM. The cells were incubated for 2 days, after which they were co-transfected with a mixture of 2.5 ng Fluc gaplesion plasmid and 2.5 ng Rluc control gapped plasmid. Following incubation of 24 h, Fluc and Rluc signals were sequentially measured using the dual-luciferase reporter assay (Promega). TLS extent values were calculated as the ratio between Fluc and Rluc signals in respect to samples transfected with control siRNA, and were corrected for having two different reporters as described in the statistical analysis section. Exclusion of false-positive hits due to off-target effects of the siRNA was achieved by de-convolution of selected siRNA pools and testing each of the four siRNA oligos separately. Gap-lesion plasmids were constructed as described 66 .
Statistical analysis of the primary TLS-specific screen. Viability was measured using luminescence readouts for a set of 1,100 siRNA pools with or without ultraviolet irradiation in three independent biological replicas. This can be represented by: I UV,ij ; I no UV,ij where 'i' stands for the gene index (siRNA pool) and 'j' for the repeat index. All intensity values are given after log 2 transformation. The assay has a 96-well format such that each plate contains nine negative controls of several kinds (non-targeting siRNA, RNA-induced silencing complex (RISC)-free siRNA and siRNA targeting XPA, which is already defective in the cells). Normalization was done by setting the mean intensity of the negative controls in each plate to be equal to the mean intensity of the negative control wells over all the plates. This was done separately for unirradiated plates and ultraviolet-irradiated plates, and the data were corrected accordingly. For each plate, the corresponding ultraviolet effect of the negative control wells is given by: 8 j ¼ 1, 2, 3, where average is over i A negative control wells. Estimation of the noise level was done using intensity-dependent noise model 67 . This estimation assigns a s.d. value to each siRNA according to the mean intensity value. The approach was used since it provides better statistical power and is justified by the log Gaussian nature of the light intensity measurements. The statistic used for the analysis is: This represents the ultraviolet effect associated with a specific siRNA in a specific repeat with respect to the control ultraviolet effect. According to the null hypothesis, this statistic is distributed with zero mean and an intensity-dependent variance, A P value was calculated based on this null hypothesis for each siRNA in each repeat. The three P values calculated for each siRNA were then combined using a Fisher meta P value to get a single P value per siRNA pool. We call a gene 'ultraviolet sensitive' if two conditions are satisfied: 8% FDR, and a median fold change (FC) of either FC 41.5 or FC o0.8; these values, which are represented by the red dashed vertical lines on Fig. 1b, were set according to the distribution of the negative control siRNAs.
Statistical analysis of the secondary TLS-specific screen. The TT 6-4 PP screen, TT CPD screen and the NL correction assay where measured each in four independent biological repeats. Thus for each lesion, each siRNA (i) in each replica (j), we have the measured values: F i j ; R i j ; F i NL;j ; R i NL;j ; j ¼ 1; 2; 3; 4, where 'F' stands for Fluc intensity and 'R' for Rluc intensity. Each measured value is compared with a similar experiment with the negative control siRNAs discussed above.
A natural statistic then is (the bar represents average over the four replicas): A difficulty arises when analysing this data, namely, the two numbers F i j and R i j , measured from the same well, are strongly dependent in terms of transfection efficiency, which is not uniform between repeats and between wells in the same plate. The way we overcome this problem is by a normalization that keeps the difference in each well as was measured originally. This was done by setting the Rluc values to be the same for all replicates and the Fluc values to fit the differences.
Under this normalization: and the intensity distribution does not change. All three assays (TT 6-4 PP screen, TT CPD screen and NL correction) were normalized by this method. When using the new values, the variance of the statistic T i defined above becomes: where the values Var F i h i are taken from intensity-dependent noise estimation, as was done for the ultraviolet-sensitivity screen. Thus, formulating the null hypothesis as: We chose a threshold that combines 10% FDR and a median FC of either FC 41.5 or FC o0.7 (represented by the red dashed vertical lines on Fig. 1f,g), which was set according to the distribution of the negative control siRNAs.
qPCR-based and colony-based TLS assays. MEFs of wt origin were reversetransfected with 10 nM siRNA. Following incubation for 2 days, the cells were cotransfected with the gap-lesion plasmid and the control gapped plasmid that had been used in the TLS screen. Cells were maintained for 12 h, after which their plasmid content was extracted under alkaline conditions, in which only covalently closed plasmids remained non-denatured. In the qPCR-based TLS assay, remnants of nonfilled gapped plasmids were digested by the single-stranded DNA-specific S1-endonuclease (1.6 units ml À 1 , 30 min incubation). Reactions were terminated by adding 33 mM EDTA, followed by heat inactivation and purification by a Wizard SV PCR clean-up kit (Promega). TLS analysis was performed by qPCRs using a SYBR Green reagent (KAPA Biosystems). Discrimination between the filled-in gap-lesion plasmid and the control gapped plasmid was achieved based on two primer sets; one targeted the Fluc gene and the other the Rluc gene. The ratio between the two qPCR products, normalized to samples treated with control siRNA, served to assess the extent of TLS. In the colony-based TLS assay, the extracted plasmids were transformed into a TLSdefective E. coli recA strain, which was plated in parallel on Luria broth-kanamycin (for repaired gap-lesion plasmids) and Luria broth-chloramphenicol (for repaired control plasmids). The ratio of kanamycin/chloramphenicol resistant E. coli colonies represents the efficiency of TLS in the mammalian cells, and sequence analysis of plasmids extracted from individual colonies provides DNA sequence alterations during TLS. The siRNA sequences are listed in Supplementary Table 3.
qPCR for mRNA evaluation. Total RNA was extracted from the cells 48-72 h after transfection with siRNA, using the Perfect-Pure RNA cultured cells kit (5-PRIME). cDNA was produced using High capacity cDNA reverse transcription kit (Applied Biosystems). A quantity of 5 ng cDNA was taken for qPCR analysis using SYBR Green reagent (KAPA Biosystems). The data was normalized according to measurements of the mouse HPRT and RLP19 genes, or the human GAPDH and HPRT genes. The primers used for the qPCRs are listed in Supplementary Table 4.
Co-immunoprecipitation. Cells were rinsed twice in cold PBS and lysed in Triton lysis buffer (20 mM Tris-Cl, pH 7.5, 150 mM NaCl, 0.5 mM EDTA, 10% glycerol and 1% Triton X-100) supplemented with protease and phosphatase inhibitors (Sigma) for 1 h at 4°C. Samples were pre-treated with protein A/G PLUS agarose beads (Santa Cruz Biotechnology), and then incubated with 2 mg rabbit anti polZ (H300, Santa Cruz Biotechnology), 2 mg rabbit IgG (Santa Cruz Biotechnology) or 2 mg mouse anti FLAG (M2, Sigma) for 2 h, followed by adding protein A/G PLUS agarose beads or Protein A/G MagBeads (Genscripts) for additional 1 h. Beads were washed three times with lysis buffer, and two additional times with lysis buffer that contained also 250 mM NaCl. Beads were boiled in SDS buffer, and were taken for immunoblot analysis.