APRIN is a cell cycle specific BRCA2-interacting protein required for genome integrity and a predictor of outcome after chemotherapy in breast cancer

The EMBO Journal (2012) 31, 1160 1176 & 2012 European Molecular Biology Organization All Rights Reserved 0261-4189/12 www.embojournal.org is a cell cycle specific -interacting protein required for genome integrity and a predictor of outcome after chemotherapy in breast cancer THE EMBO JOURNAL Rachel Brough 1,2, Ilirjana Bajrami 1,2, Radost Vatcheva 2, Rachael Natrajan 2, Jorge S Reis-Filho 2, Christopher J Lord 2, * and Alan Ashworth 1,2, * 1 Cancer Research UK Gene Function and Regulation Group, London, UK and 2 The Breakthrough Breast Cancer Research Centre, The Institute of Cancer Research, London, UK Mutations in confer an increased risk of cancer development, at least in part because the protein is required for the maintenance of genomic integrity. Here, we use proteomic profiling to identify (PDS5B), a cohesion-associated protein, as a -associated protein. After exposure of cells to hydroxyurea or aphidicolin, and other cohesin components associate with in early S-phase. We demonstrate that expression is required for the normal response to DNA-damaging agents, the nuclear localisation of RAD51 and and efficient homologous recombination. The clinical significance of these findings is indicated by the observation that the / interaction is compromised by missense variants of previously unknown significance and that expression levels are associated with histological grade in breast cancer and the outcome of breast cancer patients treated with DNA-damaging chemotherapy. The EMBO Journal (2012) 31, 1160 1176. doi:10.1038/ emboj.2011.490; Published online 31 January 2012 Subject Categories: genome stability & dynamics; molecular biology of disease Keywords: ; breast cancer; DNA repair Introduction Inheritance of a single mutated allele of strongly predisposes to breast cancer and a number of other malignancies (Wooster and Weber, 2003). encodes a very large protein (3418 amino acids (aa)) that is likely to have multiple cellular functions but is particularly important for the repair of DNA double-strand breaks (DSBs) by the process of homologous recombination (HR). HR is a sequence-conservative process that involves the identification of a homologous DNA sequence to that present at the DSB, and the utilisation of this sequence as a template for repair. As part of this process, *Corresponding authors. CJ Lord and A Ashworth, The Institute of Cancer Research, Fulham Road, London SW3 6JB, UK. Tel.: þ 44 20 7153 5333; Fax: þ 44 20 7153 5340; E-mail: chris.lord@icr.ac.uk or alan.ashworth@icr.ac.uk Received: 12 April 2011; accepted: 15 December 2011; published online: 31 January 2012 sequesters the DNA recombinase RAD51, mobilises it to the site of damage and then facilitates the formation of helical RAD51-single-stranded (ss) DNA nucleoprotein filaments either side of the DSB. These nucleoprotein filaments invade double-stranded (ds) DNA, usually the sister chromatid, that has homology to the site of DNA damage. Following strand invasion, DNA synthesis is instigated using the homologous sequence as a template. This ultimately leads to the restoration of the original sequence at the damaged site. However, in the absence of functional, cells use alternative, more error-prone forms of DNA repair, with the inevitable consequence that the genome becomes peppered with chromosomal rearrangements and breaks. This genetic instability is thought to foster the development of malignancy (Gudmundsdottir and Ashworth, 2006). In addition to RAD51, has also been shown to interact with a number of other proteins that control HR including PALB2 (Xia et al, 2006), FANCG (Hussain et al, 2003), FANCD2 (Hussain et al, 2004), BRCA1 (Chen et al, 1998) and DSS1 (Marston et al, 1999). In a similar fashion to deficiency, mutations in -binding proteins can also result in compromised HR efficiency and sensitisation to DNA damage. Notably, biallelic mutations in the - interacting proteins PALB2, FANCD2 and FANCG (and also biallelic mutations in itself) cause Fanconi anaemia (FA), a disease characterised by cellular sensitivity to DNA cross-linking agents (Moldovan and D Andrea, 2009). The interplay between other FA susceptibility genes and is currently unclear, although it has been demonstrated that FANCD2 and associate in response to damage and co-localise at stalled replication forks (Hussain et al, 2004). Orthologues of have also been identified in lower organisms. Bioinformatic analyses identified a candidate orthologue in Drosophila melanogaster, CG30169/ dmbrca2, (Lo et al, 2003) and subsequent analysis suggests that dmbrca2 shares many of the characteristics of its human orthologue despite being a much smaller and simpler protein in structure (Brough et al, 2008). The mechanism of DSB repair and the essential components of this process are also well conserved between mammals and Drosophila (Rong and Golic, 2003). Here, we exploited dm to identify additional -interacting proteins and in doing so identify as a novel determinant of RAD51 localisation, HR and the clinical response to chemotherapy. Results Identification and validation of as a -interacting protein We reasoned that a rapid approach to identifying novel -interacting proteins was to exploit the ease by which the relatively small dmbrca2 protein could be 1160

interacts with and is required for genomic integrity A 50 kda MW kda 250 160 105 75 50 35 pha Empty pha Empty pdmbrca2 pha dmbrca2 HA pha dmbrca2 * * * * Tubulin dmbrca2 Pds5 B 38 kda D HA IP pha Empty pha dmbrca2 MYC IP pmyc CON pmyc MYC pha Empty pha dmbrca2 IP PDS5A RAD51 FANCD2 HA E C Co-transfection FLAG + pmyc CON FLAG + pmyc MYC FLAG FLAG IP Control IP IP IR +IR IR +IR IR +IR Figure 1 interacts with. (A) (Top panel) Western blot analysis of whole cell lysates () from Drosophila Kc cells transfected with pha Empty, pdmbrca2 or pha dmbrca2 expression constructs. Immunoblots were probed with an HA-epitope tag-specific antibody or a b- tubulin antibody as shown, indicating expression of the HA dmbrca2 fusion protein. (Bottom panel) Coomassie-stained polyacrylamide gel containing anti-ha immunoprecipitated material from Drosophila Kc cells transfected with either pha Empty or pha dmbrca2 expression constructs. Position of dmbrca2 and Pds5-containing bands removed for Q-TOF mass spectrometry analysis are shown. Asterisks indicate additional bands excised for mass spectrometry analysis. MW, molecular weight. (B) Western blot analysis of anti-ha immunoprecipitates or from human 293T cells transiently expressing either pha Empty or pha dmbrca2 constructs. Blots were probed with anti-human, PDS5A, RAD51, FANCD2 and HA antibodies as shown. (C) Western blot analysis of anti-flag immunoprecipitates from human 293T cells transiently expressing FLAG-epitope-tagged human as well as MYC-epitope-tagged human or a control (CON) construct. Blots were probed with anti-flag or anti-myc antibodies as shown, suggesting a human / interaction. (D) Western blot analysis of anti- MYC immunoprecipitates or from human 293T cells transiently expressing MYC-epitope-tagged or a control (CON) construct. Blots were probed with anti-human, anti-human or anti-myc antibodies as shown, suggesting a human / interaction. (E) Western blot analysis of anti-human immunoprecipitates or from untransfected human 293T cells exposed to 10 Gy of IR. Lysates were collected 2 h following treatment and blots were probed with anti-human or anti-human antibodies as shown, suggesting an endogenous human / interaction following damage. Figure source data can be found in Supplementary data. manipulated. To identify novel interactions, we expressed haemaglutanin (HA)-epitope-tagged dmbrca2 in Drosophila embryonic Kc cells and identified dmbrca2-interacting proteins by using anti-ha immunoprecipitation (IP) from total cell lysates, followed by gel electrophoresis and mass spectrometric (MS) analysis (Figure 1A). As expected, this approach identified three peptides with 100% identity to regions of the dmbrca2 protein (Supplementary Figure S1A). In addition, we also identified 11 peptide sequences with 100% identity to fragments of the Drosophila protein Pds5 (CG17509) (Supplementary Figure S1B and C), the likely orthologue of the yeast Pds5 protein (Celniker et al, 2002) and the human proteins PDS5A and /PDS5B (Hartman et al, 2000; Losada et al, 2005). Sequence alignment analysis (using clustalw software, http://www.ebi.ac.uk/tools/msa/clustalw2/) indicated that PDS5A and PDS5B were not only very similar to each other (65% amino acid sequence identity) but also similar to the Drosophila Pds5 orthologue (35% amino acid sequence identity between PDS5A and Pds5; Supplementary Figure S1D) (Chenna et al, 2003), suggesting that there may be functional conservation between the different PDS5 species. Yeast Pds5 is a cohesin-associated protein that is involved in the maintenance of sister chromatid cohesion (Peters et al, 2008), but the role of the mammalian orthologue in this process is less clear (Losada et al, 2005; Zhang et al, 2007). To extend the observations made using MS, we used IP and immunoblot analysis. In human 293T cells, we observed interactions between HA-epitope-tagged dm and endogenous human, PDS5A, RAD51 and FANCD2 (Figure 1B). The interaction between dmbrca2 and human was far stronger than the dm human PDS5A interaction. In addition, dysfunction has also been linked to cancer (Chen et al, 2007) and therefore our subsequent studies focused upon the / interaction. We went on to observe interactions between epitope-tagged human and epitope-tagged human (Figure 1C) and between epitope-tagged human and endogenous & 2012 European Molecular Biology Organization The EMBO Journal VOL 31 NO 5 2012 1161

interacts with and is required for genomic integrity human (Figure 1D). Demonstrating an interaction between endogenous human and endogenous human initially proved problematic but we noted that this interaction was enhanced could be observed following exposure of cells to ionising radiation (IR) (Figure 1E), suggesting that this interaction could play a role in the response to such a cellular insult. binds the BRC1 repeat region of To gain insight into the nature of the interaction, we expressed fragments of the protein and defined the domains of that interact with (Figure 2A). Using 293T cells transiently expressing FLAGtagged fragments of, we demonstrated that a fragment encompassing aa 786 1909 (fragment X in Figure 2A) co-immunoprecipitated with, whereas fragments representing N-terminal (aa 1 627) or C-terminal (aa 2126 3418) fragments did not (Figure 2B). The -binding fragment of (aa 786 1909) contains BRC repeats 1 6 (Figure 2A). Using additional fragments encompassing repeats BRC1 or BRC3/4 (Figure 2A), we demonstrated that a 166-aa fragment (998 1164) including BRC1 bound, whereas the BRC3/4 fragment did not (Figure 2C). In addition to these co-ip experiments, we validated the interaction of 998 1164 with using a mammalian two-hybrid system. This analysis also confirmed a lack of association between and the N- and C-portions of (Supplementary Figure S2A). To assess the potential significance of these data, we examined the Breast Cancer Information Core (BIC) database to identify sequence polymorphisms that mapped to the 166 aa -binding fragment. The BIC database (http://research.nhgri.nih.gov/bic/) encompasses germline sequence information obtained principally from breast cancer patients but also from women potentially at an increased risk of the disease. Within the aa 998 1164 region, 37 missense variants with unknown functional significance have been identified. We assessed the effect of 15 variants on the / interaction. To do this, we used site-directed mutagenesis to introduce each mutation into constructs expressing the 998 1164 aa fragment and used these in a mammalian two-hybrid analysis, where each variant fragment represented the bait and a full-length (FL) cdna expression construct encoded the prey. This analysis demonstrated that a number of the variants exhibited a compromised interaction with (Supplementary Figure S2B). To validate these observations, we selected the six variants that demonstrated the most reduced association with for further validation (T1011R, Q1073R, N1101Y, A1011V. M1149V and S1069F; Figure 2A). We performed immunoblot analysis on material that co-immunoprecipitated with each of the six FLAG-epitope-tagged variant fragments in human cells. This validation experiment (Figure 2D) indicated that while all six variants were expressed, two variants, T1011R and S1069F, almost completely ablated the / interaction, while Q1073R and M1149V had less significant effects. Since the DNA recombinase, RAD51, also binds to the BRC repeat sequences, we assessed whether these same variants modulated the interaction between and RAD51. As expected, we observed decreased RAD51-binding capacity in the T1011R variant located within the BRC1 repeat but did not observe a similar reduction in RAD51 binding for any of the other variants that we had shown to modulate binding (Figure 2D). Taken together, these observations suggested that the S1069F, Q1073, A1109V and M1149V variants altered the interaction independently of the BRC1 RAD51 interaction. To examine these effects further, we introduced five of the variants into a cdna sequence (PIR1) that encodes a minimal protein isoform that is capable of mediating HR and which serves as a useful tool for manipulating for functional studies (Edwards et al, 2008) (Supplementary Figure S2C). We introduced each of the variants into PIR1 cdna expression constructs and transfected these into -deficient DLD1 cells (Hucl et al, 2008) harbouring a synthetic HR reporter substrate DR-GFP (Green Fluorescent Protein) that contains an inducible DSB (Pierce et al, 1999) (Supplementary Figure S2D). We then estimated the ability of variants to restore HR in -deficient cells by measuring GFP fluorescence generated by the HR-repaired DR-GFP construct. As expected, the non-manipulated PIR1 cdna construct restored HR in null DLD1 cells, matching previous observations (Edwards et al, 2008). Contrastingly, cdnas with either Q1073R, A1109V, M1149V or S1069F variants did not rescue HR in DR-GFP (Figure 2E). We did note, however, that there was a lack of a direct correlation between the level of interaction impairment (Figure 2D) and the impairment of HR (Figure 2E). For example, the variants S1069F, A1109V and M1149V impaired the interaction Figure 2 interacts with the BRC1 repeat. (A) Schematic diagram of FLAG-tagged fragments of human used in the experiments described in (B E). FL ¼ aa 1 3418; fragment N ¼ aa 1 627; fragment X ¼ aa 786 1909; fragment C ¼ aa 2126 3418; fragment BRC1 ¼ aa 998 1164; fragment BRC3/4 ¼ aa 1386 1560. In addition, the position of six missense mutations within the BRC1 fragment (aa 998 1164) that were investigated in (D, E) are shown. Not to scale. (B) Western blot analysis of whole cell lysates () and anti- FLAG immunoprecipitates from 293T cells transiently expressing fragments from pflag EMPTY, pflag (FL), pflag N, pflag X and pflag C constructs. Blots were probed with anti- and the control antibodies anti-palb2, anti-rad51 and anti-flag, as shown. This demonstrates an interaction between and fragment X of (aa 786 1909), as well as FL. Confirmation of this association was also seen using a mammalian two-hybrid method (Supplementary Figure S2A). Validation of appropriate fragment behaviour was controlled by detection of PALB2 binding to the N fragment of only and RAD51 binding to both the X- and C-terminus. (C) Western blot analysis of anti-flag immunoprecipitates from 293T cells transiently expressing pflag EMPTY, pflag X, pflag BRC1 or pflag BRC3/4. Blots were probed with anti- or anti-flag antibodies as shown, demonstrating an interaction between and the BRC1 fragment of (aa 998 1164). (D) Western blot analysis of anti-flag immunoprecipitates and from 293T cells transiently expressing six of the FLAG-tagged mutated fragments of (aa 998 1164) assessed in Supplementary Figure S2B and illustrated in (A). Blots were probed with anti-, anti-flag or anti-rad51 antibodies as shown. A number of variants abrogate the / interaction. (E) species containing the indicated missense variants were expressed in / DLD1 cells along with DR-GFP and ISCE1 reporter constructs. After 48 h, GFP-positive cells were quantified by flow cytometry. Error bars represent standard errors of the mean. *Po0.05 (Student s t test) compared to wild type. 1162

interacts with and is required for genomic integrity to varying levels but had similar effects on HR efficiency. However, it is possible that a biological threshold effect occurs in this case, such that both complete ablation or a modest impairment of the interaction have similar effects on HR efficiency. As an additional control, the variant N1101Y, that did not abrogate the interaction (Figure 2D), did restore HR in null cells (Figure 2E). Taken together, these results suggested that missense variants present in the BIC database that modulate the interaction also impair HR. is required for efficient DNA repair Given the DNA damage-dependent interaction of with (Figure 1E), we assessed whether played a non-redundant role in the cellular sensitivity to agents that cause DNA damage. silencing by RNA interference A FL N X C B pflag N pflag EMPTY pflag pflag X pflag C BRC1 BRC3/4 PALB2 IP BRC repeat 1 T1011R Q1073R N1101Y M1149V S1069F A1109V RAD51 38 kda FLAG 75kDa C FLAG IP E In null DLD1 cells: pflag EMPTY pflag BRC3/4 pflag X pflag BRC1 GFP* PURO GFP I-SceI + I-SceI + mutant 75 kda 24 kda FLAG GF P* PURO GFP Homologous recombination efficiency of mutant is measured by GFP fluorescence D FLAG IP GFP PURO GFP 24 kda Empty BRC1 aa 998 1164 mutants T1011R Q1073R N1101Y A1109V M1149V S1069F FLAG IP Fold-GFP 1.4 1.2 1.0 0.8 0.6 * * * * 38 kda RAD51 0.4 24 kda 38 kda FLAG RAD51 0.2 0.0 Wild-type Q1073R N1101Y A1109V M1149V S1069F 6174delT Empty variant & 2012 European Molecular Biology Organization The EMBO Journal VOL 31 NO 5 2012 1163

interacts with and is required for genomic integrity A Survival fraction 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 0 0.001 Aphidicholin 0.01 0.1 1 10 [Aphidicholin] μm B Survival fraction 1.2 1.0 0.8 0.6 0.4 0.2 0 0.001 0.01 0.0 Hydroxyurea (HU) 0.1 [HU] μm 1 10 C Survival fraction 1.0 0.9 0.8 0.7 0.6 0.0 2.5 5.0 7.5 0.5 Mitomycin C (MMC) [MMC] nm 10.0 12.5 D Survival fraction 1.2 1.0 0.8 0.6 0.4 0.2 Irradiation (IR) 0.0 0 1 2 3 4 E Survival fraction 1.2 1.0 0.8 0.6 0.4 0.2 0.0 0 10 9 10 8 PARP inhibitor IR (Gray) [Olaparib] M F sicon si si 10 7 10 6 10 5 sicon si si-3 si-4 10 μm 10 μm 10 μm Figure 3 silencing increases genomic instability and heightens sensitivity to DNA-damaging agents. (A C) Survival curves of 293Tcells transfected with sirna targeting and then treated with either (A) aphidicholin or (B) HUor(C) MMC for 5 days. (D, E) Survival curves generated from colony formation assays on adherent Hela cells transfected with sirna targeting then exposed to either (D) IR or(e) PARP inhibitor. For (A E), two different sirna species (si-3 and si-4) were used in separate experiments. Non-targeting control sirna (sicon) and sirna targeting were used as negative and positive controls, respectively. Error bars represent standard errors of the mean from three replicate experiments. (F) Metaphase chromosome images generated from 293T cells treated with control, or sirna after MMC exposure. The prevalence of specific chromosomal aberrations between the different sirna treatments was compared using a two-tailed heteroscedastic t-test (n ¼ 50) and are summarised in Table I. Scale bar represents 10 mm. (Supplementary Figure S3A) sensitised human kidney 293T cells to the polymerase a inhibitor, aphidicholin (Figure 3A), the DNA synthesis inhibitor, hydroxyurea (HU; Figure 3B), and the interstrand cross-linking reagent, mitomycin C (MMC; Figure 3C) as measured by short-term viability assays. Using Hela cells, which are more adherent than 293T cells, we also used colony formation assays, to demonstrate that silencing sensitised cells to IR (Figure 3D) as well as the PARP inhibitor, olaparib (Figure 3E). Similar results for IR and MMC sensitivity were also seen in breast epithelial cancer T47D cells and non-tumour-derived breast epithelial MCF10A cells (Supplementary Figure S3B, C and Supplementary Table SI). In addition, silencing caused genomic instability in response to MMC treatment, as demonstrated by an increased prevalence of chromosomal aberrations (Figure 3F and summarised in Table I). These DNA-damaging reagents cause the formation of DNA lesions such as interstrand cross-links that inhibit replication forks and cause fork collapse and DNA DSBs. Previously, it has been demonstrated that forms S-phase nuclear foci that co-localise with other repair molecules following DNA damage (Wong et al, 1997). Using a MYC-tagged expression construct, we found that also forms nuclear foci that occasionally co-localise with RAD51 (Supplementary Figure S4A) and (Supplementary Figure S4B). Notably the RAD51 1164

interacts with and is required for genomic integrity Table I Unrepaired chromosomal aberrations in cells transfected with sirna against and and treated with MMC sicon a sirna a sirna a P (CON versus ) b P (CON versus ) b P ( versus ) b Chromatid gaps and breaks 1 (0 12) 4 (0 33) 3 (0 17) 0.000174 0.000759 0.414088 Chromosomal gaps and breaks; acentric fragments 0 (0 3) 0 (0 7) 0 (0 6) 0.19571 0.272194 0.815242 Chromatid or chromosome gaps and breaks; 1.5 (0 12) 4.5 (0 33) 4 (0 17) 0.000108 0.000383 0.4027 acentric fragments Triradial 0 (0 2) 0 (0 3) 0 (0 2) 0.13393 0.244443 0.593049 Quadriradial 0 (0 1) 0 (0 2) 0 (0 6) 0.00274 0.094461 0.897594 Di or tricentric 0 (0 0) 0 (0 1) 0 (0 1) 0.322223 0.159386 0.562501 Complex chromosomal aberrations 0 (0 2) 0 (0 6) 0 (0 6) 0.292147 0.269018 1 Total 2 (0 14) 5.5 (0 39) 4 (0 18) 0.000116 0.000267 0.501974 a Median number of unrepaired aberrations (range). b Heteroscedastic two-tailed t-test. and co-localisation events were less frequent than RAD51 co-localised foci and this could be due to being localised elsewhere in the cell. Nevertheless, we were confident that the immunostaining was specific as the frequency of foci was reduced following transfection of either or sirna (data not shown). The formation of nuclear foci involving RAD51 and most likely represents one of the critical events in the faithful repair of DNA damage by HR (Ammazzalorso et al, 2010), namely the -coordinated loading of RAD51 onto DNA (Forget and Kowalczykowski, 2010). Silencing of caused a significant reduction in IR-induced RAD51 (Figure 4A) and (Supplementary Figure S4C and D) nuclear foci, supporting the hypothesis that modulates HR. Consistent with these observations, we also noticed that silencing led to reduced nuclear RAD51 and localisation after DNA damage, as estimated by western (Figure 4B; Supplementary Figure S4E). As RAD51 and foci primarily occur in S phase, we also assessed whether the reduction in RAD51/ foci caused by silencing could be secondary to a change in the cell cycle. silencing did not obviously alter the cell cycle profile of cells (Supplementary Figure S4F), suggesting that limiting the extent of S phase was an unlikely explanation for the reduction in nuclear foci. To directly measure the effect of deficiency upon HR, we assayed HR-mediated DNA repair using Hela cells harbouring a stable single-copy integration of the DR-GFP reporter system (Pierce et al, 1999; Slabicki et al, 2010). Silencing of caused a reduction in HR repair similar in scale to that caused by silencing of (Figure 4C). Since appeared to modulate HR capacity, we investigated whether was localised at the site of a DSB. To assess this, we designed a chromatin IP (ChIP) assay based upon the HR reporter described in Figure 4C and similar to that previously described by Potts and colleagues (Potts et al, 2006). Cells harbouring a genomic DR-GFP reporter were transfected with an I-SceI expression construct that caused a DSB within the reporter construct. Twenty-four hours after transfection, chromatin-bound proteins were cross-linked to DNA and then the cells were lysed. After sonication of the genomic DNA, we performed IP using antibodies specific for either or a phosphorylated form of the histone H2AX (gh2ax) as a control. H2AX around the site of DSBs is phosphorylated as part of the normal DNA damage response (Bonner et al, 2008). To assess whether either of these two proteins were associated with the reporter DNA bearing a DSB, we purified DNA from immunoprecipitates and performed PCR using primers specific for regions flanking the DSB caused by I-SceI expression. After transfection of the I-SceI expression construct, the amount of DR-GFP DNA that co-immunoprecipitated with gh2ax was elevated when compared to levels in cells not expressing I-SceI, thus validating the assay (Figure 4D). We also observed that the amount of DR-GFP DNA that co-immunoprecipitated with when cells were transfected with the I-SceI expression construct was also elevated (Figure 4D), suggesting that is present at or near the site of DSBs. Taken together with the effects of silencing upon RAD51 foci formation, DNA damage sensitivity and HR capacity, the localisation of at the site of a DSB implicated as a critical HR/DSB repair factor. Further dissection of the / interaction To further characterise the nature of the / interaction, we assessed its temporal modulation in response to agents that stall replication forks, cause fork collapse/ DSB formation and induce HR. In the first instance we used aphidicolin, a replicative DNA polymerase inhibitor (Rothkamm et al, 2003; Van et al, 2010). We exposed 293T cells to aphidicolin for 18 h and then analysed the / interaction following release from aphidicolin blockade. Immunoprecipitation of endogenous, followed by western blot analysis of G 1 /S synchronised cells indicated that the / interaction was greatest at 4 h after release from aphidicolin-induced arrest (Figure 5A), a time point at which the majority (68%) of cells were in the first half of S-phase (Figure 5B). The / interaction was not readily observed in asynchronous (AS) cells, in cells predominantly at G 1 /S (0 h), or when the majority of cells were in late S/G 2 (6 h) (Figure 5A and B). A similar pattern was also observed in the breast cell lines, T47D and MCF10A (Supplementary Figure S5A D) and was observed whether anti- or anti- IPs were used (Figure 5C and D). We also observed a very similar pattern of / interaction after release from HU exposure (Supplementary Figure S5E and F). Recent models suggest that chronic HU exposure causes replication fork collapse, followed by new replication origin firing (Petermann et al, 2010). The firing of new replication origins eventually generates a DNA template that is used in long-tract HR to repair the collapsed replication fork (Petermann et al, 2010). It is possible that the delay in / interaction after release from & 2012 European Molecular Biology Organization The EMBO Journal VOL 31 NO 5 2012 1165

interacts with and is required for genomic integrity HU block could represent the time taken for the firing of replication origins and the generation of new DNA prior to HR repair. In the light of these observations, we assessed whether other interactions mirrored the temporal nature of the / interaction. We assessed interactions with the partner protein PALB2 (Xia et al, 2006) and the DNA recombinase RAD51 (Wong et al, 1997). Using IP and western blot analysis of aphidicolin-treated cells, we found that the /PALB2 and /RAD51 interactions were relatively consistent following release from aphidicolin blockade (Figure 5A). We also assessed whether coimmunoprecipitated with these other -binding proteins. RAD51 and PALB2 both co-immunoprecipitated with (Figure 5C); notably, the /, / RAD51 and /PALB2 interactions were only observable 4 h after release into S-phase (Figure 5C and D). When taken together with the observations that silencing modulates the response to various DNA-damaging agents (Figure 3), RAD51 foci formation (Figure 4A) and the repair of a synthetic HR substrate (Figure 4C), the increase in the / interaction following chronic aphidicolin or HU exposure is consistent with the hypothesis that the / interaction is associated with the repair of DSBs by HR following replication fork collapse. The HR process is, in part, orchestrated by the protein kinase ATM (Zhou et al, 2000; Bakkenist and Kastan, 2004), which phosphorylates a series of protein substrates in response to damage and cellular stress. In fact, has already been identified as a potential ATM/ATR target (Matsuoka et al, A RAD51 DAPI Merge sicon KD KD 10 μm 10 μm 10 μm 10 μm 10 μm 10 μm % of cells with >5 FOCI 60 40 20 0 sirna: sicon si si-3 si-4 B GFP* PURO GFPΔ Nuclear I-SceI + I-SceI C IR: sicon si si + + + GFP* PURO GFPΔ 38 kda 10 kda 50 kda 25 kda RAD51 γ H2AX Tubulin PCNA Fold-GFP 1.2 *P < 0.0001 1.0 0.8 0.6 0.4 * * 0.2 0.0 sicon si si-3 si-4 D 1. Express I-SceI 2. Cross-link proteins to DNA 3. Immunoprecipitate 4. Amplify DSB region GFP* I-SceI 3 2 GF P* PURO 1 + I-SceI PURO GFPΔ GFPΔ 4 Inputs I-SceI +I-SceI PCR VE CON H2AX-P ChIP I-SceI +I-SceI CON H2AX-P GFP IGFR Quantification: 1.0 1.3 1.4 1.0 3.1 3.1 1166

interacts with and is required for genomic integrity 2007). To investigate whether ATM signalling might be required for the / association, we treated cells with the small molecule ATM inhibitor (ATMi), KU0055933 (Hickson et al, 2004). Inhibition of ATM abrogated the / association in early S-phase (Figure 5E and F), suggesting that this interaction was governed by ATM catalytic activity, and implicates as part of the cellular response to DNA damage. and interact with replication and cohesion factors is thought to be associated with DSB repair at stalled replication forks (Hussain et al, 2004; Thompson and Hinz, 2009). Furthermore, the parasite Trypanosoma brucei orthologue of interacts with the pre-replication complex protein CDC45 (Oyola et al, 2009). We investigated whether the early S-phase / association coincided with binding of to components of the replication fork machinery. Using western blot analysis of immunoprecipitated proteins from human 293T cells, we first confirmed the interaction between human and human CDC45 (Figure 6A). As for the / interaction, the /CDC45 interaction was restricted to early S-phase. Similarly, an interaction between and the DNA replication/fork progression protein PCNA was also restricted to early S-phase (Figure 6A). We then assessed whether also interacted with these same replication proteins; / CDC45 and /PCNA interactions were also identified from immunoprecipitates, predominantly at early S-phase (Figure 6B)., and its lower organism orthologues, have previously been shown to interact with elements of the cohesin complex (Losada et al, 2005; Lengronne et al, 2006). This complex includes SMC3 and RAD21 and is involved in the tethering of sister chromatids after DNA synthesis (Lengronne et al, 2006). Having confirmed that associates with both SMC3 and RAD21 in 293Tcells (Figure 6B), we investigated whether also interacted with these cohesin-related factors. We found that interacted with RAD21 in early S-phase. We also identified a weaker but reproducible association between and SMC3 (Figure 6A). We next assessed the possibility that acts as a determinant of the ability of to bind other interacting partners in early S-phase. Using RNA interference to silence the expression of, we demonstrated that was required for s interaction with replication and cohesin components (e.g., CDC45, PCNA and RAD21) (Figure 6C and D). Furthermore, although the effects were more modest, silencing of also reduced the binding of RAD51 and PALB2 to (Figure 6C; Supplementary Table SII), in part corroborating the reduction in nuclear RAD51 localisation and foci formation observed using sirna (Figure 4A and B). Collectively, these data suggest that sits at a nodal point in terms of the interaction between and elements of the cohesin, replication and repair machinery. Clinical relevance of in breast cancer To date, a number of preliminary observations have suggested associations between expression levels and cancer (Beckmann et al, 1996; Chen et al, 1998; Edwards et al, 1998; Harada et al, 2001; Zhang et al, 2008). To investigate this in more detail, we examined the expression of in a panel of 160 invasive breast tumours using an anti- antibody optimised for immunohistochemistry (IHC). Tumours were derived from patients with no obvious family history of cancer. After IHC staining, each section was scored using a semi-quantitative scoring system based upon a Figure 4 is involved in DNA repair. (A) (Left panel) Representative confocal microscopy images from 293T cells transfected with nontargeting control sirna or sirna targeting. Forty-eight hours after transfection, cells were exposed to 10 Gy IR and then fixed and stained using a nuclear dye (DAPI, blue in image) and an anti-rad51 antibody. Nuclear IR-induced RAD51 foci are shown in red. The scale bar represents 10 mm. (Right panel) Graphical representation of this data is shown. Cells were treated with control (sicon), or sirna (si and si, respectively), as shown. The proportion of cells containing five or more nuclear RAD51 foci after IR was estimated from three independent experiments (n4100 for each experiment). Error bars for each individual experiment represent standard errors of the mean. (B) Western blot analysis of nuclear proteins from 293T cells treated as in (A). Cells were transfected and irradiated as in (A). Following cytosolic/nuclear extraction, the nuclear fraction was analysed by western blot using antibodies detecting RAD51,,, gh2ax, b-tubulin and PCNA. PCNA (nuclear specific) and b-tubulin (cytosolic) expression confirmed successful fractionation. See Supplementary Figure S3D for analysis of the cytosolic fraction. (C) (Top panel) Schematic of the pdr-gfp HR assay. The DR-GFP recombination substrate encompasses (from left to right); an hcmv enhancer/chicken b-actin promoter (black box); a modified GFP* (a GFP gene containing an integrated I-SceI endonuclease restriction site that leads to a premature stop codon); a puromycin drug selectable marker (PURO); and a second modified GFP-coding sequence (GFPD), which harbours 5 0 and 3 0 truncations. Neither GFP* nor GFPD are functional GFP ORFs. Expression of I-SceI in cells carrying a DR-GFP reporter induces a DSB in GFP*. Repair of this DSB by HR/gene conversion uses the GFPD gene as a template, and results in the removal of the termination codons from GFP*, reconstitution of a functional GFP ORF and GFP-mediated fluorescence. Non-conservative forms of repair do not reconstitute the GFP ORF and do not lead to GFP expression. (Bottom panel) Twenty-four hours following transfection with the indicated sirna, Hela cells harbouring a single-copy genomic integration of the DR- GFP reporter, with or without I-SceI. GFP expression was estimated by FACS analysis and is represented in the graph. Error bars represent standard errors of the mean from three separate experiments. (D) (Left panel) Schematic of the ChIP PCR assay used, which is based upon the HR reporter construct described in (C). 293T cells harbouring a genomic HR reporter were transfected with an I-SceI expression construct that caused a DSB within the GFP* gene. Twenty-four hours after transfection, chromatin-bound proteins were cross-linked to DNA and cells lysed. After sonication of the genomic DNA, IP was performed using an anti- antibody. To ascertain the presence of DNA flanking, the DSB site in immunoprecipitates, PCR amplifying a GFP sequence was used. Positive control (anti-gh2ax) and negative control (anti-flag) IPs were used to validate the assay. A region of the IGFR gene was also amplified to demonstrate specificity of gh2ax and to DSB regions rather than genomic DNA as a whole. (Right panel) 293T cells harbouring the pdr-gfp reporter construct were transfected with the I-SceI expression construct and 24 h later, ChIP PCR was performed as described above. In the PCR-negative control, DNA was substituted with water in the PCR reaction. This analysis indicated that after transfection of the I-SceI expression construct, the amount of DSB-flanking DNA that co-immunoprecipitated with gh2ax was elevated, compared with levels in cells not expressing I-SceI, thus validating the assay. In addition, the amount of DSB-flanking DNA that co-immunoprecipitated with when cells were transfected with the I-SceI expression construct was also elevated, suggesting that is present at or near the site of DSBs. Quantification of the PCR bands are shown, normalised to the control ChIP without I-SceI treatment. & 2012 European Molecular Biology Organization The EMBO Journal VOL 31 NO 5 2012 1167

interacts with and is required for genomic integrity A IP CON B AS 0h IP AS 0 h 4 h 6 h AS AS 0 h 4 h 6 h 38 kda RAD51 4h 6h PALB2 No. cells PI C D IP AS 0h CON IP AS 0h 4 h AS AS 0h 4 h 38 kda RAD51 4h PALB2 No. Cells PI E No. cells PI Asynchronous Early S-phase ATMi +ATMi F ATMi + + AS IP Early S Figure 5 The / interaction is cell-cycle dependent. (A) Western blot analysis of immunoprecipitated material or whole cell lysates () from AS or synchronous 293T cultures. For synchronous cultures, cells were arrested at the G 1 /S checkpoint by aphidicolin treatment (18 h) and then released into the cell cycle by the removal of aphidicolin (t ¼ 0 h). Samples were taken for IP at 0, 4 and 6 h after release as shown. Western blots were probed with antibodies detecting human endogenous,, RAD51 or PALB2 as shown. (B) PI cell-cycle FACS profiles of cells treated in (A). (C) Western blot analysis of immunoprecipitated material or from AS 293T cultures or 293T cells treated with aphidicolin as in (A). Samples were taken for IP at 0 and 4 h after release as shown. Western blots were probed with antibodies detecting human endogenous,, RAD51 or PALB2 as shown. (D) PI cell-cycle FACS profiles of cells treated in (C). (E) PI cell-cycle FACS profiles of 293T cells treated with the ATM catalytic inhibitor KU0055933. Samples were collected for analysis in early S-phase. Cell-cycle arrest and synchronisation was achieved using the aphidicolin method (see (A)). (F) Western blot analysis of immunoprecipitates and from cells in (E). Blots were probed with anti- or antibodies as shown, suggesting that the / interaction is abrogated by inhibition of ATM. IP combination of staining intensity and the proportion of IHCpositive cells in each tumour (see Materials and methods). The median expression of in the whole cohort was defined as a score equal to 8 and we considered tumours with a score of o8 having reduced expression (Figure 7A). The frequency of IHC scores for the breast tumour panel is summarised in Supplementary Figure S6A. From this analysis, we found that 46.9% (75/160) of breast tumours had reduced expression and six of these tumour samples appeared to express no detectable. We then crosscompared expression with histological grade in the 160 breast cancer biopsies. The frequency of low expression correlated with histological grade (P ¼ 0.005). In general, the expression of tended to be lower in grade III tumours, where 56.4% (57/101) of samples exhibited lowlevel expression. In grade I tumours, the frequency of lowlevel expression was 25% (3/12) and 31% (14/45) in grade II tumours (Figure 7B). We also compared levels 1168

interacts with and is required for genomic integrity A IP CON IP AS 0 h 4 h 6 h AS AS 0 h 4 h 6 h B IP CON IP AS 0 h 4 h AS AS 0 h 4 h 150 kda 76 kda CDC45 225 kda 76 kda CDC45 31 kda PCNA 31 kda PCNA 102 kda RAD21 102 kda RAD21 150 kda SMC3 150 kda SMC3 225 kda 150 kda C IP D RNAi: sicon sicon AS Early S-phase 150 kda 38 kda 150 kda RAD51 PALB2 sicon 76 kda 31 kda 102 kda 225 kda CDC45 PCNA RAD21 si-3 PI Figure 6 and associate with cohesion proteins and replication complex. (A) Western blot analysis of immunoprecipitated material or whole cell lysates () from AS or synchronous 293T cultures. For synchronous cultures, cells were arrested at the G 1 /S checkpoint by aphidicolin treatment (18 h) and then released into the cell cycle by the removal of aphidicolin (t ¼ 0 h). Samples were taken for IP at 0, 4 and 6 h after release as shown. Western blots were probed with antibodies detecting human endogenous, CDC45, PCNA, RAD21, SMC3 and as shown. (B) Western blot analysis of immunoprecipitated material or from AS or synchronous 293T cultures as in (A). Samples were taken for IP at 0 and 4 h after release as shown. Western blots were probed with antibodies detecting human endogenous, CDC45, PCNA, RAD21, SMC3 and as shown. (C) Western blot analysis of immunoprecipitated material or from synchronised, early S-phase 293Tcultures as in (A). Cells were transfected with sirna as shown, 48 h prior to aphidicolin treatment and samples taken for analysis at 4 h after release from aphidicolin treatment. Western blots were probed with antibodies as shown. See also Supplementary Table SI. (D) PI FACS profiles of 293T cells treated with sirna silencing. Cells were transfected with sirna 48 h prior to aphidicolin treatment. FACS analysis was performed 4 h after release from aphidicolin treatment. Figure source data can be found in Supplementary data. No. cells with expression of the oestrogen receptor (ER) in breast tumours. Low expression correlated with the ERnegative phenotype (P ¼ 0.019), and was observed in 64.7% (22/34) of ER-negative cases. This correlation concurred with the frequency of expression in breast cancer molecular subtypes as defined by a validated immunohistochemical surrogate (Nielsen et al, 2004); low expression was most frequent in the tumours with a basal-like/triple-negative phenotype (67% (14/21)), although this did not reach statistical significance (Po0.1), less frequent in HER2-amplified tumours (44% (11/25)) and least frequent in tumours with a luminal phenotype (30% (45/105)) (Figure 7C). A summary of the correlations between expression and other clinicopathological markers is summarised in Supplementary Table SIII. We also assessed the correlation between transcript levels and breast cancer subtype. Analysis of microarray data (http://www.ebi.ac.uk/micro array-as/ae/ accession number: E-TABM-543) from a separate cohort of 55 sporadic tumours confirmed that reduced expression (P ¼ 0.009) in tumours correlated with the basal-like phenotype based on a validated immunohistochemical surrogate (Nielsen et al, 2004), when compared with HER2-positive and luminal subtypes (Supplementary Figure S6B). Remarkably, is located directly adjacent to human on chromosome 13 (Supplementary Figure S6C). This is a region often subject to genomic loss in breast and ovarian cancers. The close proximity of to raises the possibility that these genes are lost together following a gross genomic deletion at 13q13. However, low expression of in both the 160 invasive and 55 sporadic tumours did not correlate with genomic loss as assessed by array CGH & 2012 European Molecular Biology Organization The EMBO Journal VOL 31 NO 5 2012 1169

interacts with and is required for genomic integrity A B 60 n=101 D i iii ii iv HCC38 unmethylated sequence for CpG1 and CpG2 CpG1 CpG2 % Of cases with the cohorts % Of cases with the cohorts expressing low expressing low C 40 n=45 n=12 20 0 Grade I Grade II Grade III Breast tumour grade 80 n=21 60 n=25 40 n=101 20 HCC1143 methylated sequence for CpG1 and CpG2 0 Basal HER2+ Luminal CpG1 CpG2 Breast subtype E CpG1 CpG2 CpG3 CpG4 CpG5 CpG6 HCC1569 F 2.5 P =0.0137 2.0 SUM225 HCC38 HCC1937 SUM149 APRIM expression 1.5 1.0 0.5 HCC1315 0.0 Unmethylated Methylated HCC1428 HCC1954 HCC1143 SUM190 Unmethylated: Methylated: 1170

interacts with and is required for genomic integrity (data not shown and Supplementary Figure S6D (http:// www.rock.icr.ac.uk/ Project_ID ¼ 118)). To investigate this observation, we compared mrna levels in the panel of invasive breast tumour samples with the corresponding IHC expression scores. This analysis revealed no correlation (P ¼ 0.183) between and expression levels, suggesting that these two genes are perhaps independently regulated in these tumours (Supplementary Figure S6E). Since variations in expression in breast tumours did not correlate with copy number, one possible mechanism of lowered levels of could be epigenetic silencing of the promoter. Since variations in expression in breast tumours did not correlate with copy number, we investigated whether had undergone epigenetic silencing by promoter methylation, a common mechanism of transcriptional control. While we were unable to examine methylation in the same tumour panel described above, we did assess methylation in basal-like breast cancer cell lines. MethPrimer was used to design primer pairs that spanned the two CpG islands of and methylation-specific PCRs (MSPs) were performed on 20 basal-like breast cancer cell lines. Only CpG island 1, which contains six CpGs, was identified as a methylated region in some of the cell types. The position of the primer pair used (MSPF and MSPR) is depicted in Supplementary Figure S6F and the MSPs using these primer pairs are shown in Supplementary Figure S6G. To confirm that these MSPs represent methylated CpGs, we bisulphite sequenced 10 of the basal-like cell lines. The small cluster of CpGs were amplified by PCR using primer pairs BSF and BSR (Supplementary Figure S6F) and the products were TOPO-cloned. Eight individual colonies were sequenced using BSR (Figure 7D and E) and the proportion of clones with methylation-protected CpGs were calculated and are depicted in a methylation map illustrated in Figure 7E and Supplementary Figure S6G. When quantitative RT PCR (qrt PCR) was performed for mrna transcript, we noted that the methylation status correlated with mrna expression (P ¼ 0.0137) (Figure 7F), suggesting that promoter methylation may be a mechanism for silencing in breast tumours. Our dissection of function suggested that loss of expression could sensitise tumour cells to agents that cause DNA damage (Figure 3; Supplementary Figure S3B and C). To investigate the clinical significance of these findings, we compared expression in breast tumours with the response to anthracycline, a drug which is thought to elicit a therapeutic effect by causing DNA damage. In ERnegative tumours, low expression correlated with a statistically significant longer disease-free survival (DFS) in patients treated with adjuvant anthracycline-based chemotherapy (whole cohort P ¼ 0.126, ER-negative P ¼ 0.0027, ER-positive P ¼ 0.576; Supplementary Figure S7Ai, Bi and Ci). Similar associations with outcome were seen for low expressing ER-negative patients when examined for metastasis-free survival (MFS) (Supplementary Figure S7Aii, Bii and Cii) and breast cancer specific survival (BCSS) (Supplementary Figure S7Aiii, Biii and Ciii). A multivariate survival Cox Hazard model analysis demonstrated that reduced expression was independently associated with outcome in this cohort of patients treated with adjuvant anthracycline-based chemotherapy (Supplementary Table SIV). Cumulatively, these results suggest that low expression correlates with outcome in patients following chemotherapy in particular subtypes of breast cancer. Since the intercalating agent, anthracycline generates DNA damage, this correlation is possibly a consequence of the role of in HR. Discussion We have identified as a protein that interacts with the breast and ovarian cancer tumour suppressor. has been previously characterised as a cohesin-associated protein and has been shown to be dysregulated in cancer (Beckmann et al, 1996; Edwards et al, 1998; Harada et al, 2001; Zhang et al, 2008). We show here that modulates HR, is present at DSBs and influences the response to DNA-damaging agents. Importantly, analysis of the / interaction suggests that variants with unassigned functional significance modulate the / interaction. and likely sit within a network of proteins with known functions in DNA replication, repair and cohesion. appears to play a major role in regulating the interaction between and proteins such as CDC45, PCNA and RAD21 as well as having an influence on the association between and the key regulators of HR, RAD51 and PALB2. Finally, the clinical significance of our observations is underscored by the association of expression levels with pathological grade in breast cancer and clinical response to DNA-damaging chemotherapy. Our data provide the first evidence of a direct linkage between elements of the cohesin complex and the tumour suppressor. Previous reports have implicated elements of the cohesin complex and associated proteins in DNA repair. For example, the absence of a fully functional cohesin complex is essential for DSB repair after DNA replication in budding yeast (Sjogren and Nasmyth, 2001) and in human cells elements of the cohesin complex are recruited to DSBs and are essential for competent HR (Potts et al, 2006). A possible model is that cohesin enables HR by holding sister chromatids together when DSBs occur. This localisation of sister chromatids should allow strand invasion and DNA Figure 7 expression correlates with response to chemotherapy. (A) Representative micrographs of expression in invasive breast cancers; (i, ii) illustrate normal levels of expression with (i) generating a score of 9 and (ii) a score of 8; (iii, iv) show reduced levels of expression with (iii) generating a score of 6 and (iv) a score of 2. (i iv) (haematoxylin/dab 100). For distribution of scores see Supplementary Figure S5A. (B) Correlation between high grade (grade III) and low expression (two-tailed t-test: Po0.005). (C) Low expression also correlates with basal breast cancer phenotype (two-tailed t-test: Po0.05). (D) The chromatograms illustrate the sequence of unmethylated CpG1 and CpG2 seen in all HCC38 clones (top) and the methylated sequence of CpG1 and CpG2 seen in most HCC1143 clones (bottom). (E) A graphical representation of the proportion of methylated CpGs detected for each of the 10 basal cell lines. (F) A box and whisker diagram illustrating that detectable mrna levels are significantly lower in the methylated lines, when compared with unmethylated lines (P ¼ 0.0137, Student s t-test) as assessed by RT PCR. & 2012 European Molecular Biology Organization The EMBO Journal VOL 31 NO 5 2012 1171