Base excision repair AP endonucleases and mismatch repair act together to induce checkpoint-mediated autophagy

Cellular responses to DNA damage involve distinct DNA repair pathways, such as mismatch repair (MMR) and base excision repair (BER). Using Caenorhabditis elegans as a model system, we present genetic and molecular evidence of a mechanistic link between processing of DNA damage and activation of autophagy. Here we show that the BER AP endonucleases APN-1 and EXO-3 function in the same pathway as MMR, to elicit DNA-directed toxicity in response to 5-fluorouracil, a mainstay of systemic adjuvant treatment of solid cancers. Immunohistochemical analyses suggest that EXO-3 generates the DNA nicks required for MMR activation. Processing of DNA damage via this pathway, in which both BER and MMR enzymes are required, leads to induction of autophagy in C. elegans and human cells. Hence, our data show that MMR- and AP endonuclease-dependent processing of 5-fluorouracil-induced DNA damage leads to checkpoint activation and induction of autophagy, whose hyperactivation contributes to cell death.


5
-Fluorouracil (5-FU) remains a central component of systemic treatment of a wide range of solid cancers in the adjuvant setting 1 . The active metabolite, 5-fluoro-2 0 -deoxyuridine monophosphate, inhibits thymidylate synthase, which leads to imbalanced nucleotide pools with subsequent incorporation of dUTP and 5-fluoro-2 0 -dUTP into DNA, and the corresponding ribonucleotides into RNA 2 . As the resulting uracil and fluorouracil (FUra) bases in DNA do not lead to formation of strand breaks directly, it is thought that repair intermediates generated through incomplete or aberrant processing of the original lesions by DNA repair enzymes is the basis for DNA-directed toxicity 1 . Uracil and FUra in DNA are primarily repaired via the base excision repair (BER) pathway 3 . BER is initiated by a uracil-DNA glycosylase (UDG), which excises the damage as a free base. The resulting abasic (apurinic/apyrimidinic (AP)) site is incised by an AP endonuclease to generate a singlestrand break, and further processing leads to replacement of one or two nucleotides. All five mammalian UDGs may process uracil or FUra (for review see ref. 4), but conflicting reports exists as to whether they mediate DNA-directed toxicity 2,[5][6][7] . Furthermore, no good correlation between BER deficiency and therapeutic response has been observed in clinical material 4 .
In contrast, the DNA mismatch repair (MMR) pathway is an important determinant for 5-FU toxicity, and MMR deficiency is associated with resistance to 5-FU (ref. 1). In the MMR pathway, the MutS complex (MSH-2/MSH-6) binds DNA damage in a mismatch context 8 and recruits the MutL complex (MLH-1/PMS-2). The MutS/MutL complex then travels away from the mismatch to search for a nick, which is required for loading of an exonuclease (EXO-1) that removes the lesion together with an extended stretch of the surrounding DNA. Thus, processing of DNA damage through the MMR pathway leads to the generation of long stretches of singlestranded DNA, which will be coated by replication protein A (RPA) before replicative polymerases are recruited to fill in the gap 9 .
The mechanistic basis for involvement of MMR in response to 5-FU is puzzling, as 5-FU primarily leads to incorporation of uracil and FUra opposite adenine to generate BER substrates. The high degree of redundancy among UDGs effectively prevents further clarification of the division of labour between BER and MMR in activation of DNA-directed toxicity in human cells. Thus, we used Caenorhabditis elegans as a model to investigate the function of the two DNA repair pathways in eliciting 5-FU toxicity. C. elegans has only one characterized UDG, the enzyme UNG-1 (ref. 10), and has significantly contributed to our understanding of the role of DNA repair pathways in initiating DNA damage response (DDR) signalling in response to misincorporated uracil 11 . Furthermore, as C. elegans also has only one MutS complex, it allows genetic interrogation of the separate role of BER and MMR in eliciting DNA-mediated toxicity in response to 5-FU.
Here we show that the DNA damage recognition complex of MMR, but not BER, acts as a sensor of DNA damage induced by 5-FU. Furthermore, epistasis analyses show that the BER AP endonucleases APN-1 and EXO-3 function in the same pathway as MMR to induce toxicity in response to 5-FU. Immunohistochemical analyses suggest that EXO-3 generates DNA nicks required for MMR activation, whereas APN-1 is required for checkpoint activation. Processing of DNA damage via this pathway, in which both BER and MMR enzymes are required, leads to induction of autophagy in C. elegans and in human U2OS cells. This suggests that failure to process 5-FU-induced DNA damage via this pathway is a basis for resistance.

Results
C. elegans MMR and BER mutants are resistant to 5-FU. Removal of uracil or FUra bases from DNA is generally believed to be required for generating the repair intermediates leading to DNA-directed toxicity induced by 5-FU. Thus, we asked whether mutants of the lesion recognition enzymes in the BER or MMR (Fig. 1a) pathways led to 5-FU resistance in C. elegans.
A null mutant of the UNG-1 enzyme, ung-1(qa7600; ref. 11, also see Supplementary Fig. S1 for description of the mutant alleles) was as sensitive to 5-FU as the wild type (N2; Fig. 1b). In contrast, pronounced 5-FU resistance was observed after RNA interference (RNAi)-mediated depletion of the AP endonucleases EXO-3 and APN-1 (Fig. 1c). As the AP endonucleases act downstream of UNG-1 in the BER pathway (Fig. 1a), this surprising result indicates that DNA-mediated toxicity proceeds via an UNG-1-independent alternative pathway that also requires AP endonucleases.
To test whether the AP endonucleases act in the same pathway as MMR to mediate 5-FU-directed toxicity, we depleted APN-1 and EXO-3 in msh-6 and msh-2 mutant animals ( Fig. 1g,h, respectively). In both cases epistasis was observed, as no statistically significant difference in median survival was found between the groups. Thus, the BER AP endonucleases act in the same pathway as MMR to mediate 5-FU toxicity. Further epistasis analyses in exo-3(tm4374) mutants suggested that MSH-6 operates upstream of EXO-3, whereas MSH-2, MLH-1 and APN-1 probably act downstream of EXO-3, as their depletion resulted in 5-FU sensitivity indistinguishable from the exo-3(tm4374) mutant-fed control RNAi (Fig. 1i). Analyses in exo-3;msh-6 and exo-3;mlh-1 double mutants (Fig. 1j,k, respectively) confirmed the epistasis between EXO-3, and both MSH-6 and MLH-1. Finally, the msh-6 single and exo-3;msh-6 double mutants showed most pronounced tolerance to 5-FU, supporting an upstream role of MSH-6 in eliciting 5-FU toxicity. Hence, the epistasis experiments support a model where the BER AP endonucleases function downstream of MutS to elicit DNAdirected toxicity in response to 5-FU.
DNA repair-dependent checkpoint activation. Next, we asked whether the processing of 5-FU-induced DNA damage led to apoptosis. Despite the appearance of cytological markers for DDR activation ( Supplementary Fig. S3a,b), apoptotic cell death was not observed upon 5-FU treatment neither in the germline ( Supplementary Fig. S3c) 13 nor in embryos ( Supplementary  Fig. S3d).
The epistasis of MSH-6 between EXO-3 and APN-1 (Fig. 1g,j) could be explained if these endonucleases provide the DNA nicks required for the loading of EXO-1. If so, we would expect that depletion of the BER AP endonucleases prevented RPA-1 focus formation. This was indeed the case for EXO-3 (Fig. 2b and  Supplementary Table S1). In contrast, depletion of APN-1, which, according to our epistasis analysis, acts downstream of EXO-3, did not significantly suppress RPA-1 focus formation in the wildtype background (Fig. 2a,b). However, depletion of APN-1 in exo-1(tm1842) mutants completely abrogated RPA-1 focus formation ( Fig. 2c and Supplementary Table S2). The different outcomes with respect to RPA-1 focus formation confirmed that APN-1 and EXO-3 have non-redundant functions. The finding that EXO-3 was required for RPA-1 filament formation suggested that it might contribute to generate nicks required for the activation of MMR.
RPA-1 filament formation in response to DNA damage is an indication of DNA damage checkpoint activation that, in 5-FUtreated human cells, involves CHK-1 phosphorylation 14 . 5-FU induced-phosphorylation of CHK-1 in wild-type C. elegans but not in msh-6(pk2504) mutant embryos (Fig. 2d,e). Interestingly, CHK-1 phosphorylation was not induced in APN-1-depleted embryos (Fig. 2d,e) despite the presence of RPA-1 foci. The immunohistochemical analysis therefore supports that checkpoint activation depends on DNA repair. Although MSH-6 function is required both for RPA-1 focus formation and CHK-1 phosphorylation, APN-1 is dispensable for RPA-1 filament formation.    The absence of RAD-51-positive foci (Supplementary Table S3) showed that checkpoint activation was not likely to be a result of DNA double-stranded break (DSB) formation. However, compared with untreated nuclei, 5-FU-treated nuclei were larger with faint DAPI (4 0 ,6-diamidino-2-phenylindole) staining already at the four-to six-cell stage embryo (Fig. 2a). Taken together, this suggested that 5-FU primarily leads to chromatin decompaction rather than DSB formation. Consistently, the maximum nuclear diameter, defined by the perimeter of lamin-1 (LMN-1) staining in the four-cell stage embryo, increased from 8.2 to 9.0 mm (Student's t-test, Po0.05) by 5-FU treatment concomitant with reduced DAPI staining intensity in an MSH-6-dependent manner ( Supplementary Fig. S4). Chromatin decompaction was confirmed by the loss of linker histone H1X in 5-FU-treated embryos (Fig. 2f,g). H1X levels were not reduced upon 5-FU treatment in the msh-6(pk2504) mutant. This gives independent confirmation that chromatin decompaction and checkpoint activation depended on DNA-repair-mediated processing of 5-FU-induced DNA damage.
5-FU induces autophagy in C. elegans. As we recently showed that uracil incorporation induces cell death with both apoptotic and autophagic features 15 , we tested whether 5-FU may induce autophagy as an alternative cell death response 16 . During the process of macroautophagy (hereafter, it is referred to as autophagy), parts of the cytoplasm are sequestered by a double lipid layer that grows to form a small vesicle (autophagosome). The autophagosome then fuses with a lysosome to generate an autolysosome, in which the cargo is degraded by acidic hydrolases. To monitor the activation of autophagy, we used the transgenic reporter strain expressing LGG-1 (C. elegans Atg8/ LC3) in fusion with green fluorescent protein (GFP). LGG-1 remains associated with autophagic membranes from the early nucleation step until fusion with lysosomes. Increased LGG-1 expression and appearance of GFP::LGG-1-positive puncta indicated that autophagy was activated in the embryos of 5-FU-treated hermaphrodites (Fig. 3a).
LGG-1 accumulation was dependent on the core autophagic machinery, as it was suppressed after depletion of UNC-51 (C. elegans ULK1/Atg1 homologue), a serine/threonine kinase that regulates autophagy induction; the class III phosphatidylinositol 3-kinase (PI3K) complex subunit BEC-1 (C. elegans beclin1/Atg6), which is required for vesicle nucleation; and ATG-7, an E1-like enzyme involved in conjugation of the ubiquitin-like proteins LGG-1 and ATG-12 to autophagic membranes (Fig. 3a). Depletion of LGG-1 itself suppressed puncta formation in a BEC-1::GFP reporter strain ( Supplementary Fig. S5a-c). To exclude the possibility that 5-FU-induced autophagy could be an artefact of the transgenic reporter system used, we monitored the expression Actin N2 msh-6 (pk2504) Embryos with RPA-1 foci (%) of the endogenous VPS-34 protein (a homologue of human class III PI3K) that interacts with BEC-1 (ref. 17). Confirming autophagy activation in response to 5-FU treatment, an overall increase in VPS-34 expression was observed together with the appearance of VPS-34-positive puncta around the nuclear periphery and in the cytoplasm (Fig. 3b). Distinct VPS-34 foci were also observed in the nucleoli, indicating that VPS-34 shuttles between the nucleus and cytoplasm, similar to what was previously described for its interaction partner Beclin 1 (ref. 18). Furthermore, as LGG-1 itself is degraded as part of the autophagic process, the appearance of a 25-kDa band corresponding to GFP in 5-FU-treated embryos, which disappeared upon addition of the PI3K inhibitor 3-methyladenine (3-MA), confirmed increased autophagic flux upon 5-FU treatment (Fig. 3c). Although autophagy primarily acts as a prosurvival mechanism following various forms of stress, including starvation, excessive activation of autophagy can promote cell death 15,16 . The significant increase in progeny survival in animals depleted for bec-1 and atg-7 before 5-FU treatment showed that autophagy, in this case, contributes to toxicity (Fig. 3d).
A connection was recently described between the ATR (Ataxia telangiectasia and Rad3 related) DNA damage checkpoint and activation of autophagy in yeast 19 . Depletion of ATL-1 (the C. elegans ATR orthologue 20 ) reduced autophagy induction by about 50% in C. elegans ( Fig. 3e and Supplementary Table S4), whereas no effect was seen upon loss of CEP-1 (the C. elegans p53 orthologue; Fig. 3e and Supplementary Table S4). In atl-1(tm853) mutants, there was weaker cytoplasmic anti-VPS-34 staining and the fraction of embryos with VPS-34-positive puncta was halved compared with the N2 control ( Fig. 3f,g), supporting a partial requirement of ATL-1 for 5-FU-induced autophagy. In contrast, the fraction of embryos with GFP::LGG-1-positive puncta was indistinguishable in the atm-1(gk186) mutant background and the wild-type background. (Fig. 3h).
DNA repair-dependent induction of autophagy. Next, we asked whether autophagy induction, similar to chromatin decompaction, RPA-1 focus formation, and CHK-1 phosphorylation also depended on DNA repair. This was addressed by counting embryos with GFP::LGG-1-positive foci in animals after RNAi-  Table S4). Similar results were obtained using accumulation of VPS-34 as a read-out for autophagy activation ( Supplementary Fig. S5d,e). Thus, 5-FUinduced autophagy depends on DNA repair. As predicted, as depletion of core autophagy genes attenuated 5-FU toxicity (Fig. 3d), suppression of autophagy further improved survival in the already-tolerant msh-6(pk2504) mutant (Fig. 4c).
We also monitored the expression of the ATF-2 transcription factor, which negatively regulates the expression of autophagy genes such as lgg-1 and bec-1 in C. elegans 15 . In line with an upstream function of the MutS complex, we found that depletion of msh-2 and msh-6 resulted in 41% and 36% (93 out of 225, or 54 out of 150 embryos scored) of embryos showing excessive ATF-2 expression, which was not seen in the control (1,825 embryos scored; Fig. 5a,b). In contrast, no excessive ATF-2::GFP expression was observed in exo-3 or apn-1-depleted worms (150 embryos scored). Hence, the MutS complex functions to instigate the cellular response cascade leading to DNA-mediated toxicity in response to 5-FU.
DNA repair-dependent ribosome biogenesis defect. The DNA repair-dependent accumulation of VPS-34 in the nucleoli indicated that processing of 5-FU-induced DNA damage led to specific problems in this compartment. It is known that 5-FU inhibits ribosome biogenesis at the level of late ribosomal RNA processing 21 . These phenotypes are commonly interpreted to be resulting from incorporation of 5-fluorouridine triphosphate into RNA 2 . As regulation of ribosome biogenesis is tightly linked to regulation of chromatin condensation state, we suspected that MMR-dependent chromatin decompaction in response to 5-FU might contribute to these phenotypes. In C. elegans, ribosome biogenesis defects can be monitored as the appearance of visible (enlarged) nucleoli during early embryonic development. The appearance of enlarged nucleoli in B70% of the four-cell stage embryos treated with 5-FU (Fig. 5c) is therefore a phenotype consistent with an RNA biogenesis defect. The enlarged nucleoli phenotype is in this case, however, unlikely to be a consequence of incorporation of 5-fluorouridine triphosphate into RNA, as it was absent in the msh-6 mutant (Fig. 5c,d). Moreover, the enlarged nucleoli phenotype was not observed after depletion of MSH-2, MLH-1 and EXO-3, and was partially suppressed by APN-1 depletion. In contrast, enlarged nucleoli were seen at wildtype levels after depletion of UNG-1 (Fig. 5d). Hence, the appearance of enlarged nucleoli depended on DNA repair, and its suppression coincided with the suppression of nucleolar VPS-34 staining and failure to induce autophagy. This suggests that RNAand DNA-mediated 5-FU toxicity have a common mechanistic component.

5-FU induces DNA repair-dependent autophagy in human cells.
To confirm the relevance of our finding for human cells, we monitored induction of autophagy in human U2OS cells treated with 10 mM 5-FU. Upon induction of autophagy, cytosolic LC3 (orthologous to C. elegans LGG-1; LC3-I, 18 kDa band) is cleaved and bound to phosphatidylethanolamine in the autophagic membranes (LC3-II, 16 kDa band) and, as LC3-II remains bound to autophagic membranes throughout the pathway, it can be used as a marker for autophagy. A weak but consistent induction of autophagy was seen following 18 h of exposure to 5-FU, with maximum LC3-II expression between 24 and 48 h (Fig. 6a,e). Autophagy was not activated secondarily to apoptosis as neither poly (ADP-ribose) polymerase 1 nor caspase-3 cleavage was observed. Induction of autophagy was suppressed by inhibiting the major mammalian AP endonuclease APE1 by methoxyamine (Mx; Fig. 6a,b). Similarly, a reduction of 5-FU-induced LC3positive puncta was seen in the presence of Mx in U2OS cells (Fig. 6c,d). Induction of autophagy by 5-FU also depended on MMR in human cells, as depletion of MSH-2 effectively suppressed LC3-II induction (Fig. 6e,f) and formation of LC3-positive puncta (Fig. 6g,h). Similar results were obtained after depletion of MSH-6 by short interfering RNA (siRNA) as expected, as the stability of MSH-2 protein depends on MSH-6 in mammalian cells 12 (Supplementary Fig. S6a).
Finally, we asked whether nucleolar proteins or nucleic acids were degraded by autophagy as suggested by the nucleolar VPS-34-positive foci seen in 5-FU-treated C. elegans. Expression of the RPS3 subunit of the 40S small ribosomal particle, which shuttles between the nucleolus and cytoplasm 22 , was gradually lost in 5-FU-treated U2OS cells (Fig. 6i and Supplementary Fig. S6b), although RPS3 mRNA levels were unaffected (Fig. 6i). Indeed, depletion of Ulk1, a kinase required for induction of autophagy, prevented the 5-FU-induced degradation of RPS3 ( Supplementary Fig. S6c,d) confirming that autophagic activity contributed to RPS3 degradation.
In summary, we have identified a novel conserved mechanism for 5-FU-induced toxicity. MutS and the BER AP endonucleases are required to elicit DNA-directed toxicity, leading to checkpoint activation and autophagy induction in response to 5-FU in C. elegans embryos and in human cell lines. Epistasis analyses in C. elegans are consistent with MutS and the BER AP endonucleases acting in the same pathway with EXO-3, providing a nick for activation of the MMR pathway in vivo.

Discussion
Preclinical and clinical evidence show that MMR deficiency is associated with poor therapeutic response to a wide range of cytotoxic drugs 23 . It is not clear whether this effect is due to a classical DNA repair function of MMR or a role for MMR in signalling DNA damage 4 . The DNA damages induced by 5-FU can be repaired by BER and MMR, but little is known about the nature of the crosstalk between these DNA repair pathways beyond the fact that they share substrates. Here we present genetic and molecular evidence to show that the MMR MutS complex is the upstream requirement for DNA-mediated toxicity. Importantly, we show that the BER AP endonucleases EXO-3 and APN-1 act in the same pathway as classical MMR in eliciting DNA-directed toxicity in response to 5-FU in C. elegans and in a human cell line. We present evidence that EXO-3, the C. elegans orthologue of human APN-1, generates the nicks required for MMR activation in vivo. Processing of 5-FU-induced DNA damage via this pathway led to autophagy activation, which coincided CHK-1 phosphorylation, showing that autophagy is activated as a consequence of DNA damage processing and checkpoint activation. Failure to induce autophagy correlated with resistance to 5-FU. Autophagy was, as such, a determinant for sensitivity. We identified a specific requirement for MMR and BER to elicit 5-FU toxicity. Depletion of genes participating in other DNA repair pathways did not result in similar resistance ( Supplementary Fig. S7). Genetic studies in Saccharomyces cerevisiae suggest that the nucleotide excision repair (NER) pathway may contribute to repair of AP sites in the absence of BER 24 , which might explain the somewhat increased 5-FU sensitivity of a mutant in XPA-1, a DNA damage binding protein required for NER ( Supplementary Fig. S7a). In contrast, depletion of ERCC-1 or XPF-1, two proteins that function together as a structure-specific endonuclease in the NER pathway, resulted in a mild tolerance ( Supplementary Fig. S7b), but this effect might be due to the role of ERCC-1/XPF-1 in repair of DSB 25 rather than in NER. Similar tolerance was also observed after depleting proteins acting in the two main DSB repair pathways, homologous recombination and non-homologous end joining ( Supplementary Fig. S7c,d, respectively). These results suggest that both NER and DSB repair pathways may process DNA-repair intermediates generated by MMR and BER in C. elegans, but that they have modest effects on overall toxicity. Moreover, our results revealed that the MutS complex, in particular the MSH-6 protein, acts as the initial sensor of 5-FUinduced DNA damage, as none of the cellular effects induced by 5-FU in the wild-type background was seen in MSH-6-deficient animals; msh-6 mutants showed no chromatin decompaction, failed to activate the DNA damage checkpoint as measured by RPA-1 focus formation and CHK-1 phosphorylation, and failed to induce autophagy. The nature of the in vivo substrates for cytotoxic MMR processing remains to be demonstrated. However, as only BER is expected to be able to process 5-FUra/Ura base paired with adenine, these lesions probably do not contribute much to DNA-mediated toxicity, as lack of UNG-1 does not affect overall survival. Nevertheless, UNG-1-initiated BER probably repairs 5-FU-induced DNA damage, as depletion of  exo-3 and apn-1 gave less-effective suppression of toxicity in ung-1 mutants than in the wild type ( Supplementary Fig. S2b), and depletion of UNG-1 improved survival in the exo-3 mutant (Supplementary Fig. S2d). Interestingly, the sharp reduction in F1 survival in exo-3-depleted animals treated with higher concentrations of 5-FU (Fig. 1c) was lost in ung-1 mutants ( Supplementary  Fig. S2b) and, conversely, upon depletion of UNG-1 in exo-3 mutants (Supplementary Fig. S2d). This suggests that BER becomes more important at higher 5-FU concentrations as was previously observed in human cell lines 26 . BER and MMR both remove 5-FUra/Ura in a mismatched context. Some competition for substrates is supported experimentally with depletion of MSH-2 and MSH-6 being epistatic in ung-1 mutants (Supplementary Fig. S2c). However, MMR is the only pathway expected to be able to detect the mismatched normal bases expected to arise from the nucleotide pool imbalance induced by 5-FU. Hence, MMR might respond to many different substrates in 5-FU-treated cells and it is likely to be that all of these substrates contribute to DNA-mediated toxicity. Importantly, only MMR-initiated processing was associated with checkpoint activation and downstream DNA damage signalling. This is consistent with resistance to 5-FU-based chemotherapy in MMRdefective human cells and tumours 4 , whereas loss of UNG-1 and SMUG1 DNA glycosylases have little effect 2,5 . The requirement of mammalian TDG (thymine-DNA glycosylase) for DNAmediated toxicity 7 might reflect that TDG might compete with MMR for mismatched substrates 27 .
It was recently demonstrated that MMR may act outside of S-phase in response to cytotoxic drugs 28 . This process, called non-canonical MMR, is uncoupled from replication with the consequence that the DNA ends associated with the replication fork, which direct and activate classical MMR, are not automatically available. It was therefore proposed that noncanonical MMR may hijack nicks generated by BER 28,29 . Noncanonical MMR is associated with extensive gap formation 28 and the prominent RPA foci observed here would therefore be compatible with non-canonical MMR. If the activating nicks for non-canonical MMR were intermediates generated by the BER pathway, we would also expect to see a dependency of UNG-1 for RPA filament generation and toxicity. However, no such dependency was observed. Hence, MMR does not hijack nicks generated through ordinary UNG-1-initiated BER. Although, we cannot exclude the possibility that MMR may hijack BER intermediates arising through processing by a cryptic, uncharacterized UDG activity 10 , the epistasis analyses suggested that the BER AP endonucleases EXO-3 and APN-1 function downstream of MutS. Non-redundant functions for APN-1 and EXO-3, as previously postulated 30 , were confirmed here, where only depletion of EXO-3 suppressed RPA-1 focus formation. These observations offer in vivo indications that EXO-3 provides nicked intermediates required for recruitment of EXO-1 and thereby activation of the MMR pathway. The in vivo substrates of EXO-3 and APN-1 remain unknown. As neither enzyme is known to incise undamaged DNA, it is likely to be that base damage is involved. As UNG-1-deficiency does not confer resistance, however, this base damage is likely not restricted to UNG-1 substrates. Approximately 77% of APN-1-depleted embryos still displayed prominent RPA-1-positive foci, indicating that the recently described nucleotide incision repair activity of APN-1 (ref. 31) contributes little to this function. RPA-1 focus formation in response to 5-FU depended on MutL, which distinguishes the pathway described here from the alternative mode for repair of oxidative DNA damage in which recruitment of low-fidelity DNA polymerase Z depended on MutSa but not on MutLa (ref. 32). Consistently, EXO-1 was the main exonuclease required for RPA-1 focus formation, although the obliteration of RPA-1 focus formation in exo-1 mutants upon APN-1 depletion suggested that the APN-1 3 0 -5 0 exonuclease activity might contribute to DNA resection 31 . RPA-1 focus formation is used as an early marker of DDR activation. We were therefore interested to note that although RPA-1 foci formed in response to 5-FU in APN-1-depleted cells, they failed to mount a DNAdamage checkpoint response as measured by induction of CHK-1 phosphorylation. Hence, APN-1 functions downstream of RPA-1 filament formation to allow recruitment of the downstream factors required for checkpoint activation. Further experiments are required to pinpoint which activity of the multifaceted APN-1 enzyme is required for checkpoint activation, but the observation that some APN-1-depleted cells had extended RPA-1-positive tracts could point to a possible function of APN-1 in limiting MMR-mediated resection.
The observed APN-1-dependent CHK-1 phosphorylation further argues that autophagy was primarily induced as a consequence of DNA damage checkpoint activation resulting from processing the 5-FU-induced DNA damage via the pathway described. As autophagy is a cytoplasmic degradation pathway, it is not obvious how it is activated by DNA damage, but two observations allow us to present a model for autophagy activation in response to 5-FU. First, the appearance of VPS-34 foci in the nucleolus suggested that cargo in this organelle might be collected and targeted for degradation in 5-FU-treated cells. The degradation of RPS3 by autophagy in human cells supported this interpretation. Second, we showed that 5-FU primarily leads to global chromatin decompaction rather than DNA strand breaks. Our data therefore allow us to speculate that a relaxed chromatin landscape induced by MSH-6-initiated processing of 5-FUinduced DNA damage interferes with rDNA organization in the nucleoli leading to autophagy induction. Consistent with this model, the specific degradation of RPS3 by autophagy in human cells indicated that autophagy contributes to remove faulty or excess nucleolar proteins.
In summary, the data presented here allow us to propose a model for crosstalk between the BER and MMR pathways in eliciting DDR activation in response to 5-FU where the BER enzyme EXO-3 generates a nick required for MMR activation. Our data suggest that MSH-6-dependent processing of 5-FUinduced DNA damage either within or in the vicinity of the nucleoli is an event leading to toxicity and induction of autophagy. Together, our data strongly suggest that the role of autophagy in mediating DNA damage-induced cell death is more significant than previously anticipated.
C. elegans autophagy. A transgenic reporter strain expressing functional LGG-1 in fusion with GFP (GFP::LGG-1) was used to monitor autophagy induction. Young adults grown on RNAi-expressing food were exposed to 100 mM 5-FU along with RNAi food for 24 h. Worms were anaesthetized with 10 mM levamisol (Sigma) and mounted on 4% agarose pads. Induction of autophagy was monitored under a Zeiss LSM-510 META Confocal microscope with Â 63 Plan-Apochromat 1.4 numerical aperture (NA) objective. The number of worms having GFP-positive foci in embryos was scored in 10-60 worms per condition in four independent experiments. To monitor the induction of autophagy in seam cells, L1-to L2-stage larvae were exposed to 100 mM 5-FU for 24 h before scoring the average number of GFP-positive puncta in the seam cells at the L3 stage under a Zeiss Axiovert 200M inverted microscope with Â 100 Plan-Apochromat 1.45 NA objective and standard epifluorescence filters. At least four to eight seam cells were observed in each worm in three independent experiments. Expression of the ATF-2 transcription factor in fusion with GFP (ATF-2::GFP) was measured using a transgenic reporter strain in 5-FU-treated hermaphrodites fed control, msh-2, msh-6, or exo-3 RNAi. ATF-2::GFP expression was scored in at least 150 animals per condition.
C. elegans immunofluorescence. For immunohistochemistry, embryos or dissected germlines were collected 24 h after treating synchronized L4 hermaphrodites with 125 Gy ionizing radiation or 100 mM 5-FU.
C. elegans nucleolar stress. Induction of nucleolar stress was monitored as previously described 33 . Briefly, embryos were dissected out of 5-FU-treated hermaphrodites and observed continuously from the three-cell through the fourcell stage under differential interference contrast in a Zeiss Axiovert 200M inverted microscope with Â 100 Plan-Apochromat 1.45 NA objective. A fraction of four-cell stage embryos with visible nucleoli was scored in three independent experiments comprising a minimum of eight embryos per experiment.
C. elegans chromatin decompaction. To examine the chromatin decompaction state after 5-FU treatment, dissected embryos were immunostained with a-LMN-1 as described above to visualize the nuclear membrane. The maximum nuclear diameter was measured in 4-cell embryos in 15-32 embryos per condition after identification of the appropriate focal plane from a z-stack under a Zeiss LSM-510 META MK14 Confocal microscope with Â 63 Plan-Apochromat 1.4 NA objective.
Human cell immunofluorescence. U2OS cells were seeded in Lab-Tek II Chambered Coverglass (Nunc) 1 day before the experiments were performed and treated with 5-FU (10 mM) or Mx (1 mM) for the indicated times. The cells were then fixed in 100% methanol for 10 min and blocked in 5% FCS before incubation with anti-LC3 antibody (Clone 5F10, Nanotools), followed by incubation with Alexa488-labelled donkey anti-mouse antibody (Molecular Probes). The nuclei were stained with DAPI (1 mg ml À 1 ) in PBS. Pictures were taken using a Cell Observer microscope (Zeiss) equipped with a Â 40 objective. The number of LC3 spots per cell was quantified from 200 to 1,600 cells per condition in each experiment. Four independent experiments were performed and the mean number of LC3 spots per cell was normalized to that of untreated control cells.
For preparation of human whole-cell extracts, U2OS cells were washed in cold PBS, lysed in RIPA buffer supplemented with a mixture of protease inhibitors (Roche) and scraped. After a brief sonication, the lysates were cleared and the protein concentrations determined by the BCA Assay (Pierce). Fifteen micrograms of protein per sample was loaded and resolved on 4-20% gradient gels (Bio-Rad) followed by electroblotting to Immobilon-P membranes (Millipore). The blots were probed with specific antibodies, which were detected using standard ECL reagents. The band signal intensities were quantified by the Quantity One software (Bio-Rad) and normalized to those of actin. Full blots can be found in Supplementary Fig. S8.