PBRM1 Deficiency Confers Synthetic Lethality to DNA Repair Inhibitors in Cancer

Polybromo 1 (PBRM1), a tumor suppressor gene encoding the BAF180 protein, is a specific subunit of the polybromo-associated BAF (PBAF) complex, which is one of the three classes of SWItch/sucrose non-fermentable (SWI/SNF) chromatin remodeling complexes (1). PBRM1 contains six bromodomains, which recognize acetylated lysine histone residues (2–4), and is involved in preserving genome and chromosomal stability by maintaining centromeric cohesion during mitosis (5). During interphase, PBRM1 also facilitates replication re-priming downstream of stalled replication forks (RF; ref. 6) and promotes DNA repair by mediating transcriptional silencing at DNA double-strand breaks (DSB), via ATM- (7) and cohesin-dependent processes (8). In addition, PBRM1 influences the antitumor immune response (9), notably by mediating resistance to T-cell-dependent killing in preclinical cancer models (10).

PBRM1 is one of the most frequently altered genes in cancer. Deleterious PBRM1 mutations are found in 28% to 55% of clear cell renal cell carcinomas (ccRCC), where they are an early, driver event (11) that occurs subsequent to VHL alteration. When evaluated by IHC, loss of PBRM1 expression is seen in 57% to 84% of ccRCC, particularly in advanced disease (12–15). Several other aggressive malignancies also harbor PBRM1 defects, including 11% to 59% of chordomas, 12% to 23% of cholangiocarcinomas, 7% to 20% of mesotheliomas, 12% of endometrial carcinomas, and 3% of non–small cell lung cancers (NSCLC; refs. 16–18).

There are currently no personalized medicine approaches that target PBRM1-defective cancers, an area of unmet medical need (19, 20). We show here that clinical PARPi cause synthetic lethality (SL) with PBRM1 defects, thereby representing a novel, readily-testable, precision medicine-based therapeutic strategy for PBRM1-defective cancers.

U2OS, H1299, A549, RCC-MF, and RCC-FG2 cell lines were purchased from ATCC. 786-O and A498 cell lines were kindly provided by Dr. Sophie Gad-Lapiteau (Gustave Roussy Cancer Campus, Villejuif, France). HAP1 PBRM1-isogenic cells were purchased from Horizon Discovery. U2OS, A549, A498 cells were cultured in high-glucose DMEM supplemented with 10% FBS. H1299 and 786-O cells were cultured in RPMI1640 medium supplemented with 10% FBS. RCC-MF and RCC-FG2 cells were cultured in RPMI1640 medium supplemented with 2 mmol/L L-glutamine and 10% FBS (and 4.5 g/L glucose for RCC-MF). HAP1 cells were cultured in Iscove's modified Dulbecco's medium (IMDM) supplemented with 10% FBS. All cells were grown at 37°C and 5% CO₂. Mycoplasma testing was performed monthly using the MycoAlert Mycoplasma Detection Kit (Lonza). All cell lines were short tandem repeat typed using StemElite ID (Promega) to confirm identity. Further details on cell lines, including their histological and genetic backgrounds, as well as technical information on their use in the study are shown in Supplementary Tables S1 and S2.

The PARP inhibitors olaparib (AZD-2281), rucaparib (PF-01367338), and talazoparib (BMN-673), and the ATR inhibitors berzosertib (VE-822) and ceralasertib (AZD6738) were purchased from Selleck Chemicals. Inhibitor stock solutions were prepared in DMSO and stored in aliquots at −80°C. Mitomycin C (MMC), 5,6-dichlorobenzimidazole 1-β-D-ribofuranoside (DRB), iodo-deoxyuridine (IdU), and 5-chloro-2′-deoxyuridine (CldU) were obtained from Sigma-Aldrich. PicoGreen was purchased from Thermo Fisher Scientific.

PBRM1 gene knockout was performed in U2OS, 786-O, A498, H1299, and A549 cell lines using a CRISPR/Cas9-based gene editing approach. Cells were targeted using the Edit-R CRISPR/Cas9 gene engineering protocol (Horizon), according to the supplier's instructions. The 5′-TTCATCCTTATAGTCTCGGA-3′ sgRNA sequence was used to generate a frameshift deletion in exon 3 of PBRM1. Cells were transfected in T25 flasks with sgRNA and Cas9 plasmid, using Lipofectamine 2000 (Thermo Fisher Scientific). Several rounds of transfection were performed to obtain optimal knockout efficiency. PBRM1 expression was monitored in the transfected pool at each transfection cycle by Western blot analysis. Once expression was no longer detected, cells were plated onto 96-well plates for clonal isolation using the limiting dilution method. Colonies were recovered and profiled for PBRM1 expression by Western blot analysis.

Small-molecule and siRNA screens were performed as described previously (21). The small-molecule inhibitors and siRNA targets used in these screens are listed in Supplementary Tables S3 to S5. See also Supplementary Materials and Methods for further details.

Cell survival assays were conducted as described previously (21). See also Supplementary Materials and Methods for details.

Cells were grown in 10 cm dishes and exposed to 10 μmol/L olaparib or DMSO for 48 hours to reach 70% confluence at the time of harvesting. For RFs labeling, cells received pre-warmed medium containing 100 μmol/L CldU and were incubated at 37°C, 5% CO₂ for 30 minutes. Cells were then rinsed three times with prechilled PBS and incubated with 100 μmol/L IdU for 30 minutes. Cells were collected in cold PBS, counted, and adjusted to 50,000 cells per 50 μL PBS on ice. Plugs were generated by adding 50 μL of prewarmed 1% law-melting point agarose to the cells; the resulting 100 μL mix was gently homogenized and quickly transferred into a casting mold, and incubated for 1 hour at 4°C to solidify. Subsequent steps were performed as described previously (22). For the analysis, initiation, termination, and cluster patterns of replicative forks were considered to measure fork velocity.

For quantification of γH2AX, RAD51, RPA foci, and micronuclei, cells were seeded in 96-well plates (Greiner Bio-One, #655090) and treated with the indicated drugs. Cells were then fixed in 4% paraformaldehyde for 20 minutes at RT, washed twice with PBS, and permeabilized with 0.2% Triton X-100 in PBS for 10 minutes. Subsequent labeling, imaging, and image analysis steps were performed as described previously (23).

Animal experimentation was carried out according to ARRIVE guidelines, regulations set out in the UK Animals (Scientific Procedures) Act 1986, and in line with a UK Home Office project license held by CJL and approved by the ICR Animal Welfare and Ethical Review Body (AWERB). 786-O PBRM1-WT or -KO chunks (between 1×1 mm and 2×2 mm) were subcutaneously implanted into the flank of 5-week-old female NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ mice, with a take rate of 87%. When the tumor reached the 40 to 50 mm³ threshold size, mice were randomized into vehicle (control) or treatment groups (PBRM1-WT: N = 11 vehicle, N = 11 talazoparib; PBRM1-KO: N = 10 vehicle, N = 9 talazoparib). Mice received 0.25 mg/kg talazoparib or vehicle daily for 6 weeks per os. Tumor size was monitored twice weekly during 4 weeks using calipers.

Apart from the in vivo experiment, no statistical methods were used to predetermine sample size and experiments were not randomized. The investigators were not blinded during xenograft experiments. Unless otherwise stated, all graphs show mean values with error bars (SD); 95% confidence intervals were used and significance was considered when *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001; ns, not significant.

The proteomics and DRIP-Seq datasets are publicly available at Pride (http://www.ebi.ac.uk/pride) under the accession number PXD017826, and at Array Express (https://www.ebi.ac.uk/arrayexpress/), under the accession number E-MTAB-8837, respectively.

Additional methods details are available in Supplementary Materials and Methods. All uncropped and unedited images of the blots included in this study are available in Supplementary Fig. S20.

To uncover clinically actionable vulnerabilities associated with PBRM1 deficiency, we performed three orthogonal drug sensitivity and/or genetic perturbation screens (Fig. 1, Supplementary Figs. S1A–S1F). First, we used mouse embryonic stem (mES) cells to evaluate the effect of Pbrm1 gene silencing on sensitivity to a library of 80 small-molecule inhibitors, including those that target oncogenic kinases, DNA repair or cell cycle proteins (Fig. 1A; Supplementary Table S3). Because loss of ARID2 (another PBAF-specific subunit required for the appropriate incorporation of PBRM1 into the complex) leads to PBRM1 downregulation (1, 10, 24, 25), we carried out a parallel small-molecule inhibitor sensitivity screen using Arid2 gene silencing (Fig. 1A–C). After siRNA transfection, cells were exposed to increasing concentrations of the 80 small-molecule inhibitors for five days, after which, cell viability was assessed using CellTiter-Glo. Triplicate screens were conducted and AUC values were calculated for each small-molecule inhibitor transfected either with control, nontargeting siRNA (siCtrl), siPbrm1 or siArid2. Using the ratio of AUC values in siCtrl vs. siPbrm1-transfected cells, we identified 23 compounds with an AUC ratio >0.1, including three clinical PARPi (talazoparib, rucaparib, and olaparib; ranked #1, #7, and #10, respectively in terms of Pbrm1 selectivity; Fig. 1D and F; Supplementary Fig. S1A). Arid2 silencing also sensitized cells to PARPi (talazoparib, olaparib, and rucaparib; ranked #1, #4, and #12, respectively; Fig. 1E and G; Supplementary Fig. S1B). As loss of ARID2 results in PBRM1 downregulation and misincorporation into PBAF (1, 10, 24, 25), whereas loss of PBRM1 expression leads to ARID2 upregulation (Supplementary Figs. S1G and S1H; refs. 10, 26), we reasoned that both the PBRM1/PARPi and the ARID2/PARPi synthetic lethal effects could operate via impaired PBRM1 function, caused either directly by PBRM1 mutation, or indirectly, via ARID2 deficiency.

In a second approach, we conducted a parallel small-molecule inhibitor screen in isogenic PBRM1-wild-type (WT) and knockout (KO) HAP1 cells (in which PBRM1 had been inactivated by CRISPR/Cas9 gene editing). Both cell lines were exposed to a library of 167 small-molecule inhibitors (Supplementary Fig. S1C; Supplementary Table S4). We identified 37 small-molecule inhibitors with a PBRM1 AUC ratio >0.1, including two clinical PARPi (olaparib and veliparib) and two ATR inhibitors (ATRi; VE-822 and VE-821; Supplementary Fig. S1D).

In a third screen, we evaluated the SL between PBRM1 gene silencing and silencing of 446 genes implicated in DNA repair and chromatin remodeling (Supplementary Fig. S1E; Supplementary Table S5), using a previously described competition-based screening approach in U2OS cells (27). This screen identified synthetic lethal effects between PBRM1 silencing and siRNAs targeting either PARP3, PARP4, PARP1, or PARP2 (shPBRM1 vs. shCTRL Z-score difference of −1.47, −0.96, −0.48, and −0.37 respectively, scoring #7, #32, #122, and #152 of all siRNAs; Supplementary Fig. S1F).

Taken together, these three orthogonal screens, performed in distinct cell line models, suggested that PBRM1 defects caused sensitivity to PARPi, and that these effects were somewhat cell type-independent.

To validate the synthetic lethal effects of PARPi, we used three chemically-distinct clinical PARPi in five PBRM1-isogenic systems and one PBRM1-non-isogenic system (Fig. 2; Supplementary Figs. S2A–S2H and S3A–S3J, Supplementary Table S1). These included the HAP1 model described earlier where we found that HAP1 PBRM1-KO cells were significantly more sensitive to rucaparib, olaparib or talazoparib when compared with PBRM1-WT isogenic cells (Fig. 2A; ≈10-fold difference in SF₅₀, P < 0.0001, two-way ANOVA; Supplementary Figs. S2A and S2B). To assess whether these findings could be reproduced in cell lines derived from solid tumors, we created four new PBRM1-isogenic models using CRISPR/Cas9 gene editing in 786-O (ccRCC), U2OS (osteosarcoma), H1299 or A549 (NSCLC) cell lines (Supplementary Figs. S2C and S2F and S3A and S3E). In these models, we found that PBRM1-KO cells were more sensitive than PBRM1-WT cells to three different clinical PARPi, an effect that was observed using either short-term or long-term PARPi exposure and in both 2D and 3D cultures (Fig. 2B–F; Supplementary Figs. S2C–S2H and S3A–S3G). To further investigate the generality of this effect, we evaluated PARPi sensitivity in a molecularly diverse, non-isogenic, panel of ccRCC cell lines. PBRM1-proficient 786-O and A498 cells, which are homozygous or heterozygous for VHL mutations, respectively, and PBRM1-deficient RCC-MF and RCC-FG2 cells, which are VHL wild-type and mutant, respectively (Supplementary Fig. S3H). Consistent with the observations made in isogenic systems, loss of PBRM1 expression in ccRCC cell lines was associated with increased sensitivity to PARPi (Fig. 2G; Supplementary Figs. S3H–S3J). We noted that the magnitude of the synthetic lethal effects varied from model to model, suggesting that variables other than PBRM1 (such as distinct genetic backgrounds) might also modulate the phenotype.

We next sought to assess whether PARPi sensitivity was directly due to the deficiency in PBRM1. Because defects in ARID1A (a cBAF-specific DNA-binding subunit) and SMARCA4 (a SWI/SNF ATPase and helicase catalytic subunit) have been linked to increased PARPi sensitivity (28–30), we first evaluated ARID1A and SMARCA4 expression in PBRM1-isogenic cells. Immunoprecipitation of SMARCC2 (a core subunit of both cBAF and PBAF complexes) followed by Western blotting confirmed that ARID1A, SMARCA4, and its paralog SMARCA2, were expressed and incorporated into SWI/SNF complexes even in the absence of PBRM1 (Fig. 2H; Supplementary Figs. S1G and S1H and S4A and S4B). Consistent with previous work (10, 26), we also noted an increase in ARID2 expression in PBRM1-KO cells. We also found that in the SMARCA4-mutant H1299 and A549 cell lines, the PBAF-specific subunits ARID2 and BRD7 coimmunoprecipitated with SMARCA2, even in the absence of PBRM1 (Supplementary Figs. S4C–S4D); reciprocally, SMARCA2 coimmunoprecipitated with ARID2 and BRD7 (Supplementary Figs. S4E–S4F). This was consistent with recent reports suggesting that SMARCA2 can compensate for SMARCA4 deficiency, thereby allowing the formation SWI/SNF complexes when SMARCA4 is absent (1, 31). To further confirm that PARPi sensitivity was directly linked to PBRM1 deficiency, we re-expressed PBRM1 in 786-O PBRM1-KO cells (“PBRM1-rescue model”) using a doxycycline-inducible PBRM1 cDNA (Fig. 2I). PBRM1 re-expression restored PARPi resistance (Fig. 2J), thereby establishing a causative link between PBRM1 loss-of-function and PARPi sensitivity. Finally, we evaluated the effect of silencing PBRM1 by shRNA, as partial loss of protein expression might elicit a distinct phenotype to mutation of PBRM1. Using multiple shRNA constructs targeting different PBRM1 sequences (Supplementary Fig. S4G), we found that shPBRM1-transfected cells were significantly more sensitive to olaparib than shCTRL-transfected cells (Supplementary Figs. S4H and S4I; e.g., at 1 μmol/L olaparib, P = 0.0006 and 0.0036 with two different shPBRM1; two-way ANOVA). Together, these data suggested that the enhanced sensitivity to PARPi was directly linked to PBRM1 deficiency.

To assess the therapeutic potential of PARPi in PBRM1-deficient tumors, we evaluated the antitumor effect of talazoparib in mice bearing established xenografts derived from either 786-O PBRM1-KO or -WT cells (Fig. 2K). Consistent with previous reports (32) and the known tumor suppressive function of PBRM1, we found that PBRM1-KO xenografts grew faster in vivo than their isogenic PBRM1-WT counterparts (Fig. 2L, M). Compared with the drug vehicle, talazoparib treatment impaired the growth of PBRM1-KO xenografts, as measured by tumor volume (Fig. 2L, M; P = 0.0001, one-way ANOVA) and tumor weight (Fig. 2N; ≈2-fold change, P = 0.0012, one-way ANOVA). Conversely, talazoparib treatment had no effect on PBRM1-WT xenografts. Talazoparib also selectively increased the percentage of phosphorylated H2AX (γH2AX)-positive cells in PBRM1-KO xenografts collected 28 days after treatment initiation (Supplementary Figs. S5A and S5B), suggesting that the latter accumulated more DNA damage when exposed to PARPi.

Sensitivity to PARPi has previously been seen in tumor cells with an ongoing DNA damage response and/or preexisting DNA damage repair (DDR) defects, notably those affecting the homologous recombination (HR) pathway (5–7). To investigate whether PBRM1-defective cells exhibited biomarkers of an ongoing DNA damage response, we measured nuclear γH2AX foci in three genetically distinct PBRM1-isogenic models. In the absence of exogenous DNA damage, PBRM1-KO cells exhibited a higher number of γH2AX foci, compared with PBRM1-WT cells (Fig. 3A and B; P < 0.0001, Wilcoxon–Mann–Whitney test). Because persisting DNA damage often results in an increased tumor mutational burden (TMB) in cancer (33), we evaluated the TMB of PBRM1-defective tumors. Using whole-exome sequencing (WES) data from the Tumor Cancer Genome Atlas (TCGA), we found that PBRM1-mutant tumors from 11 cancer histologies, including ccRCC (KIRC), had a significantly higher TMB than PBRM1-wild-type tumors (P = 0.0009; Wilcoxon–Mann–Whitney test, Supplementary Figs. S6A and S6; Supplementary Tables S6 and S7). The analysis of TCGA RNA-Seq data also revealed a negative correlation between PBRM1 mRNA expression and TMB in several cancer types, including ccRCC (Supplementary Figs. S6B–S6E; KIRC, P < 0.0001, Wilcoxon–Mann–Whitney test). This suggested that our preclinical observation that PBRM1-deficient cells accumulate DNA damage might also occur in human tumors.

To investigate whether some PBRM1-dependent DDR defects might explain the PBRM1/PARPi SL, we exposed three PBRM1-isogenic cell lines to PARPi and monitored the presence of γH2AX, RAD51, and RPA foci, which are markers of DNA damage, HR, and single-stranded DNA (ssDNA), respectively. We found a concentration-dependent increase in the number of γH2AX foci after exposure to either olaparib or talazoparib, an effect that was more pronounced in PBRM1-KO cells (Fig. 3C and D; Supplementary Figs. S7A–S7D and S8A–S8D). We also found that the pre-existing number of RAD51 and RPA foci was increased in PBRM1-KO cells, and that this phenotype was further exacerbated by PARPi exposure (Fig. 3E and F; Supplementary Figs. S7E–S7J and S8E–S8I). We also noted that: (i) mRNA and protein levels of the key HR genes BRCA1, BRCA2, and RAD51 were similar in PBRM1-WT and -KO cells (Supplementary Figs. S9A and S9B); (ii) PBRM1-KO cells were as sensitive as PBRM1-WT cells to ionizing radiation (Supplementary Figs. S10A–S10C) and to the DNA DSB-inducing agents etoposide (Supplementary Figs. S10D–S10F) and doxorubicin (Supplementary Figs. S10G–S10I); and (iii) the number of loss-of-heterozygosity (LOH) events, a correlate genetic marker of HR deficiency, was not increased in PBRM1-mutant tumors from the TCGA dataset (Supplementary Fig. S11; ref. 34). Taken together, these results suggested that PBRM1-KO cells are unable to effectively process the DNA damage caused by PARPi, despite no obvious HR defect.

PARP1, when trapped on DNA by PARPi, impairs RF progression and elicits replication stress (35). This triggers the recruitment of ATR via ATRIP, which binds RPA-coated ssDNA at stalled RF. ATR subsequently activates the replication stress response, notably via CHK1 and ATM signaling. Collectively, these events prevent RF collapse and the formation of cytotoxic DSBs. Having previously seen that PBRM1-KO cells had preexisting elevated levels of γH2AX and RPA foci (see above), we assessed RF speed in PBRM1-isogenic cells by DNA fiber combing (Fig. 3G) and found a significant ≈30% decrease in RF velocity in PBRM1-KO cells (P < 0.0001, Wilcoxon–Mann–Whitney test; Fig. 3H, I). We also found increased expression and phosphorylation of ATR, ATM, and CHK1 in 786-O PBRM1-KO cells following cell-cycle synchronization using double-thymidine block, a phenotype that was exacerbated by olaparib exposure (Fig. 3J). Similar observations were also made in the HAP1, U2OS, and H1299 PBRM1-isogenic cells (Supplementary Figs. S12A–S12C). Because our initial HAP1 small-molecule inhibitor screen also identified ATRi as being synthetic lethal with PBRM1 defects (Supplementary Fig. S1D), we recapitulated this observation in PBRM1-isogenic cells derived from solid tumors (Supplementary Figs. S12D–S12G). Together, these findings suggested that loss of PBRM1 leads to increased replication stress, which may contribute to PARPi or ATRi sensitivity in PBRM1-KO cells (36).

To investigate the consequences of the replication stress observed in PBRM1-deficient cells, we assessed the formation of micronuclei in response to PARPi (37–39) and found that, in the absence of exogenous DNA damage, the proportion of micronuclei-positive cells was significantly higher in PBRM1-KO cells (Fig. 3K and L; ≈ 2-fold change, P < 0.0001, two-way ANOVA). Exposure to olaparib or talazoparib exacerbated this effect (Fig. 3K and L; Supplementary Figs. S13A–S13C), to a degree that was at least equivalent to that caused by the clastogen mitomycin C (used here as a positive control). To investigate the potential sources of micronuclei, we performed time-lapse confocal video-microscopy using an mCherry-labeled histone H2B PBRM1-isogenic cells. We found that PBRM1-KO cells exhibited a delayed metaphase plate formation, an increased average mitosis duration (45 vs. 70 minutes in PBRM1-WT and -KO cells, respectively, P < 0.0001, two-way ANOVA; Supplementary Fig. S14A; Supplementary Videos S1–S4) and a significantly higher number of anaphase bridges and lagging chromosomes compared with PBRM1-WT cells (Supplementary Figs. S14B–S14D). Together, our data suggested that loss of PBRM1 leads to increased DNA damage, replication stress, and micronuclei formation, effects that were exacerbated by exposure to PARPi.

R-loops are three-stranded nucleic acid structures that form when an RNA strand invades double-stranded DNA within the chromatin, resulting in an RNA:DNA hybrid and a displaced non-hybridized region of ssDNA. R-loops occur naturally during replication and transcription, but their accumulation is associated with increased DNA damage, notably under conditions of replication stress (40, 41). Having seen that PBRM1-KO cells exhibited markers of replication stress, we hypothesized that PBRM1 deficiency may cause this by increasing the burden of R-loops.

To assess this, we used RNA:DNA hybrid dot blot analysis on genomic DNA isolated from 786-O PBRM1-isogenic cells using the hybrid-specific S9.6 antibody in the presence or absence of RNase H. This revealed a ≈2-fold increase in RNase H-sensitive R-loops in PBRM1-KO cells, when compared with PBRM1-WT cells (Fig. 4A, B; P = 0.0113; Welch t test). To identify R-loop-enriched sequences in PBRM1-KO cells, we used genome-wide DNA:RNA immunoprecipitation sequencing (DRIP-Seq) in duplicate samples of 786-O and H1299 PBRM1-isogenic cells, cultured in the presence or absence of PARPi (Supplementary Fig. S15A). Principal component analysis (PCA) confirmed high reproducibility between duplicates (Fig. 4C, Supplementary Fig. S15B). Input and RNase H-treated conditions were used to evaluate the specificity of S9.6-associated peaks. We found that PBRM1-KO cells displayed a ≈2-fold higher number of RNase H-sensitive DRIP-Seq peaks compared with their PBRM1-WT counterparts (Fig. 4D–F, Supplementary Fig. S15C). When analyzing the genomic localization of DRIP-Seq peaks, we found that 90% of the peaks identified in PBRM1-WT cells were also present at the same genomic locus in PBRM1-KO isogenic cells. Strikingly, PBRM1-KO cells exhibited numerous additional peaks that were not detected in PBRM1-WT cells (Fig. 4E, F), notably in introns or intron-promoter spanning regions (Fig. 4G; Supplementary Figs. S15D, S16A and S16B), suggesting an increased formation or defective resolution of R-loops when PBRM1 was absent. Exposure to PARPi did not modify the total number of genomic loci with DRIP-Seq peaks in PBRM1-WT or -KO models (Supplementary Figs. S16C–S16F) but did increase the total R-loop burden in PBRM1-KO cell lines at consensus peaks (defined as peaks present in at least two independent samples from a given isogenic model; Fig. 4H; Supplementary Fig. S17A). Taken together, these data suggested that PARPi exposure might stabilize or prevent the resolution of preexisting R-loops in PBRM1-KO cells, rather than leading to the formation of new R-loops.

To investigate the potential role of R-loop in PARPi sensitivity, we overexpressed RNase H1, an endonuclease responsible for R-loop degradation, in PBRM1-isogenic cells (Fig. 4I). Overexpression of RNase H1 in H1299 PBRM1-KO cells partially restored talazoparib resistance, an effect that was less pronounced in PBRM1-WT cells (Fig. 4J). As R-loops are transcriptional by-products (41), we also assessed whether silencing transcription in PBRM1-KO cells could restore resistance to PARPi. We exposed PBRM1-isogenic cells to olaparib in the presence or absence of DRB (an inhibitor of RNA polymerase II-mediated transcription elongation) and found that DRB reduced the sensitivity of PBRM1-KO cells—but not PBRM1-WT cells—to PARPi (Fig. 4K), whereas both PBRM1-WT and -KO cells displayed similar sensitivity to DRB when used as a monotherapy (Supplementary Fig. S17B).

To explore the cause of R-loop accumulation in PBRM1-KO cells, we generated mass spectrometry proteomic profiles of HAP1 PBRM1-isogenic cells. We found a significant downregulation of more than 20 proteins involved in R-loop resolution, including RNase H1, SETX, DHX9, XRN2, and BLM [Log₂(FC) = −0.460, FDR = 0.0083; Log₂(FC) = −0.366, FDR = 0.0006; Log₂(FC) = −0.197, FDR = 0.0019; Log₂(FC) = −0.193, FDR = 0.0024; Log₂(FC) = −0.476, FDR = 0.0002, respectively; Supplementary Fig. S17C]. Using TCGA data, we further explored whether such a correlation also existed in human tumors. Consistent with our findings in HAP1 cells, we found that mRNA expression of several R-loop processing factors, including SETX, DHX9 and XRN2, correlated with PBRM1 mRNA expression in the TCGA ccRCC and NSCLC cohorts (Spearman correlation coefficient r = 0.72, 0.47, 0.40, all P < 0.0001, for SETX, DHX9, and XRN2 in KIRC, respectively; Supplementary Figs. S17D–S17F). Using the 786-O “PBRM1-rescue” isogenic model, we also found that 786-O PBRM1-KO cells exhibited decreased SETX expression, which was reversed by PBRM1 re-expression (Supplementary Fig. S17G).

To more specifically assess the potential significance of SETX downregulation in PBRM1-deficient cells, we used a CRISPR/Cas9-mediated activation (CRIPSRa) approach to drive SETX expression in PBRM1-isogenic cells (Supplementary Figs. S18A and S18B), and evaluated the effect of this genetic perturbation on PARPi sensitivity. Using two different single-guide RNAs targeting the SETX promoter region, we found that SETX overexpression led to increased resistance of H1299 PBRM1-KO (but not PBRM1-WT) cells to both olaparib and talazoparib (Supplementary Figs. S18C–S18E). These observations suggested that loss of PBRM1 is associated with a downregulation of several factors involved in R-loop processing, which might play a role in the accumulation of these structures in PBRM1-KO cells.

The cytosolic DNA sensing cyclic GMP-AMP synthase/stimulator of interferon genes (cGAS/STING) pathway detects cytosolic DNA (often the by-product of viral infection, DNA damage, or replication stress) and in turn activates type I interferon innate immune signaling (42). PARPi-induced activation of cGAS/STING following micronuclei formation and the generation of cytosolic DNA, is now recognized as an important component of PARPi-mediated synthetic lethal effects (23, 43–45). Recent reports also suggest that R-loops may contribute to cGAS/STING activation (40, 46). Moreover, the cGAS/STING pathway plays an essential role in the therapeutic effect of immune checkpoint inhibitors (ICI; refs. 47, 48), a standard-of-care in ccRCC and NSCLC, two histologies where PBRM1 is frequently mutated. Given this, we assessed whether PARPi or ATRi activated cGAS/STING signaling in PBRM1-defective ccRCC and NSCLC cells, as PARPi are known to do in BRCA-mutant tumor cells (23, 45, 49). To assess this, we measured three sequentially-occurring phenotypes that are linked to PARPi-mediated cGAS/STING pathway activation and antitumor immunity: (i) the accumulation of cytosolic DNA in the form of micronuclei; (ii) the phosphorylation of TANK-binding kinase 1 (TBK1) and interferon regulatory factor 3 (IRF3); and (iii) the induction of a type I interferon response (23, 43, 44).

Consistent with our previous observations (Fig. 3K and L), preexisting levels of micronuclei were higher in PBRM1-KO cells compared with PBRM1-WT cells (Fig. 5A; Supplementary Figs. S19A and S19B). Exposure to PARPi or ATRi resulted in a concentration-dependent accumulation of micronuclei in both H1299 and 786-O PBRM1-isogenic cells, an effect that was more pronounced in PBRM1-KO cells (Fig. 5A–C). Western blot analyses also revealed a concentration- and PBRM1-dependent increase in TBK1 and IRF3 phosphorylation levels in H1299 isogenic cells exposed to PARPi or ATRi (Fig. 5D; Supplementary Figs. S19C and S19D). Using RT-qPCR, we also found that VE-822 exposure caused a significantly more profound induction of CCL5 and CXCL10 mRNA expression in H1299 PBRM1-KO cells compared with PBRM1-WT cells, by more than 2- and 4-fold, respectively (P < 0.0001; Welch t test; Fig. 5E and F). Intriguingly, such selectivity was not observed in the 786-O PBRM1-isogenic model, in which CCL5 mRNA expression was only significantly induced in the PBRM1-WT cell line at the VE-822 concentration that caused a significant increase in micronuclei number (Fig. 5C; Supplementary Figs. S19E and S19F). This prompted us to explore the expression of cGAS and STING in these models, as loss of cGAS or STING has recently been described as an important and independent determinant of cancer cells' immunogenicity in other DDR-defective contexts (48). We found that cGAS protein expression was decreased in 786-O PBRM1-KO cells compared with PBRM1-WT cells (Supplementary Figs. S19G and S20), suggesting that defective cGAS/STING signaling might underly the absence of cell-autonomous immunity activation in the 786-O PBRM1-KO cell line, despite a significant micronuclei accumulation upon PARPi and ATRi exposure. To further investigate whether expression of cGAS and STING was necessary for ATRi-induced type I interferon response, we silenced CGAS and STING-coding genes by siRNA, and measured CCL5 and CXCL10 mRNA expression levels upon ATRi exposure. Silencing of CGAS and STING abrogated CCL5 and CXCL10 mRNA induction in both H1299 and 786-O PBRM1-isogenic models, thereby supporting a direct role for the cGAS/STING pathway in mediating ATRi-induced type I interferon response (Fig. 5E and F; Supplementary Figs. S19E and S19F). Altogether, these findings suggested that cytosolic DNA sensing in tumor cells via cGAS/STING could be an essential determinant of the cell-autonomous immunomodulatory potential of PARPi and ATRi in PBRM1-deficient ccRCC and NSCLC cells.

PBRM1 defects are frequent in human cancers. There is currently no approved precision medicine-based approach to specifically target deficiency in this PBAF-specific subunit. Here, we show that clinical PARPi and ATRi are synthetic lethal with PBRM1 defects. These genetic vulnerabilities, which we reproduced in multiple isogenic and nonisogenic models from various genetic backgrounds, appear to be cell type-independent, suggesting that these findings might have translational utility in several cancer types where PBRM1 is mutated.

The three distinct forms of mSWI/SNF complexes (cBAF, PBAF, and noncanonical BAF), assemble following an ordered, modular pathway, and have different functions and composition (31). PBRM1 is a specific subunit of the PBAF complex and, interestingly, it is the last subunit to be incorporated into PBAF (1). Notably, PBRM1 deletion has no effect on PBAF assembly (1), which supports that the PBRM1/PARPi SL is directly linked to PBRM1 loss, and is in line with our findings (Fig. 2H–J). These also suggest that the PBRM1/PARPi SL is distinct from other synthetic lethalities that have been recently described between PARPi or ATRi, and the cBAF-specific ARID1A subunit or the ATPase SMARCA4 subunit (29, 50–52). Therefore, to our knowledge, our manuscript describes, for the first time, that defects in a PBAF-specific subunit (PBRM1) associate with R-loop accumulation, and lead to increased sensitivity to PARPi and ATRi. This has important clinical implications, given that PBRM1 defects occur in a clinically-distinct patient population than ARID1A or SMARCA4 defects, where the use of PARPi or ATRi has not previously been proposed.

In Fig. 6, we propose a working model to explain our findings. Exposure to PARPi or ATRi causes increased replication stress. In the case of PARPi, this is likely due to the accumulation of trapped-PARP1 lesions, whereas in the case of ATRi, this might be because of the relative inability of tumor cells to effectively respond to replication stress caused by endogenous factors. When PBRM1 function is intact (Fig. 6A), the induced replication stress remains limited, possibly due to PBRM1-mediated RF repriming (6) and ATM-dependent transcriptional silencing around DSBs (7). This limited replication stress associates with residual levels of DNA damage and a moderate formation of micronuclei, which, in itself, is insufficient to activate the cGAS/STING signaling response. In PBRM1-deficient tumor cells (Fig. 6B), impaired DDR (Fig. 3; refs. 6–8) and higher R-loop burden (Fig. 4) likely contribute to enhance replication stress and DNA damage (41). Exposure to PARPi or ATRi in the presence of a PBRM1 defect (Fig. 6C) further exacerbates replication stress, promoting DNA damage accumulation and micronuclei formation. This triggers a cell-autonomous type I interferon response, in a cGAS/STING-dependent manner (Fig. 5). Upon prolonged exposure to PARPi or ATRi, the extent of DNA damage eventually reaches levels that are no longer compatible with cell survival (Fig. 2).

Alternative or additional mechanisms might also contribute to the PBRM1/DNA repair inhibitor SL. R-loops induce a DDR that can be mediated by ATM and/or ATR (40). Because PBRM1 participates in ATM-dependent transcriptional repression at DSBs (7), it is possible that in the absence of PBRM1, the resulting transcriptional silencing defects favor the accumulation of R-loops. The partial reversal of PARPi sensitivity observed when transcription is pharmacologically silenced may support this first alternative model. It is also possible that PBRM1-defective cells present increased transcriptional stress, which is further exacerbated by PARPi. Indeed, R-loop-dependent transcriptional stress can cause DNA damage by sequestering BRCA1 at sites of stalled RNA polymerase II (53), where BRCA1 recruits SETX to prevent DNA damage (54, 55). As SETX is also recruited at sites of DNA DSBs in transcriptionally active loci (56), a transcriptional stress-dependent sensitivity to DNA repair inhibitors represents a second alternative model. This secondary model is consistent with the previously described role of PBRM1 in regulating the transcription of stress response genes and in mediating cytoprotective effects against endogenous oncogenic, replicative stresses (57, 58), or exogenous stresses (59). A third alternative model to explain the PBRM1/DNA repair inhibitors SL may be that PBRM1, which contributes to RF repriming through PCNA recruitment (6), might also interact with other chromatin remodelers that are ATR substrates and promote RF repair, such as the SWI/SNF-family member SMARCAL1 (60).

We and others previously described that PARPi have cell-autonomous immunomodulatory properties, which activate the cGAS/STING signaling cascade in DDR-defective genetic contexts (23, 43, 44, 61). However, very little is known about the immunomodulatory properties of ATRi. Here, we describe that ATRi can trigger a cell-autonomous type I interferon response through cGAS/STING activation. In line with our results, it was recently suggested that ATRi enhance interferon responses in combination with radiotherapy (62). Because several clinical ATRi are currently being developed, notably in combination with ICI (NCT02264678), assessing whether ATRi-dependent immunomodulation also occurs in genetic contexts where ATRi elicit SL (50, 51) could have direct clinical implications.

Interestingly, PBRM1 loss-of-function has recently been associated with enhanced tumor cell sensitivity to IFNγ and increased type I interferon response (10). PBRM1 defects were further linked to better survival upon anti-PD-1 and anti-CTLA-4 therapy in mice (10), as well as enhanced efficacy of anti-PD-(L)1 therapy in patients with ccRCC (9, 63). Controversially, other studies reported that Pbrm1 inactivation resulted in a non-immunogenic tumor microenvironment favoring resistance to anti-PD-1 therapy in murine ccRCC models (64), and that PBRM1 mutation did not correlate with clinical benefit in patients treated with ICI (18, 64, 65). Investigating whether the inconstant sensitivity of PBRM1-deficient tumors to ICI results from variable cytosolic DNA sensing capacities within tumor cells, notably that of the cGAS/STING pathway as recently described in mismatch repair deficient contexts (48), is an important question that warrants further exploration.

In conclusion, our findings shed light on the genetic vulnerabilities associated with loss of PBRM1 related to its role in maintaining genome integrity. We are currently exploring the PARPi plus anti-PD-L1 combination therapy in patients with ccRCC, in an investigator-initiated academic phase II study developed on the basis of our preclinical findings (EudraCT No. 2018–001744–62).

C. Massard reports other support from Abbvie, Aduro, Agios, Amgen, Argen-X, Astex, Astellas, AstraZeneca, Aveo Pharmaceuticals, Bayer, Beigene, Blueprint, BMS, Boeringer, Celgene, Chugai, Chugai, Clovis, Daiichi Sankyo, Debiopharm, Eisai, Eos, Exelixis, Forma, Gamamabs, Genentech, Gortec, GSK, H3 Biomedecine, Incyte, Innate Pharma, Ipsen, Janssen, Kura Oncology, Kyowa, Lilly, Loxo, Lysarc, Lytix Biopharma, Medimmune, Menarini, Merus, MSD, Nanobiotix, Nektar Therapeutics, Novartis, Octimet, Oncoethix, Oncopeptides AB, Orion, Pfizer, Pharmamar, Ierre Fabre, Roche, Sanofi, Servier, Sierra Oncology, Taiho, Takeda, Tesaro, and Xencor outside the submitted work. J. Soria reports other support from Hookipa Pharmaceuticals, Relay Therapeutics, AstraZeneca, Gritstone Oncology, and Daiichi Sankyo outside the submitted work. E. Deutsch reports grants and personal fees from Roche-Genentech, AstraZeneca, Merck Serono, Boehringer, BMS, and grants and MSD outside the submitted work. C.J. Lord reports grants from CRUK and Breast Cancer Now during the conduct of the study; is a named inventor on patents describing the use of DNA repair inhibitors and stands to gain from their use as part of the Institute of Cancer Research “Rewards to Inventor” scheme; has received research funding from Astra Zeneca, Merck KGaA, Artios, Pfizer, and consultancy and/or advisory fees from Artios, Astra Zeneca, MerckKGaA, Tango, and GLG; is a shareholder of OviBio and Tango outside the submitted work. S. Postel-Vinay reports that as part of the Drug Development Department (DITEP) SPV, she is principal/sub-investigator of Clinical Trials for Abbvie, Adaptimmune, Aduro Biotech, Agios Pharmaceuticals, Amgen, Argen-X Bvba, Arno Therapeutics, Astex Pharmaceuticals, AstraZeneca Ab, Aveo, Basilea Pharmaceutica International Ltd., Bayer Healthcare Ag, Bbb Technologies Bv, Beigene, Blueprint Medicines, Boehringer Ingelheim, Boston Pharmaceuticals, Bristol Myers Squibb, Ca, Celgene Corporation, Chugai Pharmaceutical Co, Clovis Oncology, Cullinan-Apollo, Daiichi Sankyo, Debiopharm, Eisai, Eisai Limited, Eli Lilly, Exelixis, Faron Pharmaceuticals Ltd., Forma Tharapeutics, Gamamabs, Genentech, Glaxosmithkline, H3 Biomedicine, Hoffmann La Roche Ag, Imcheck Therapeutics, Innate Pharma, Institut De Recherche Pierre Fabre, Iris Servier, Janssen Cilag, Janssen Research Foundation, Kura Oncology, Kyowa Kirin Pharm. Dev, Lilly France, Loxo Oncology, Lytix Biopharma As, Medimmune, Menarini Ricerche, Merck Sharp & Dohme Chibret, Merrimack Pharmaceuticals, Merus, Millennium Pharmaceuticals, Molecular Partners Ag, Nanobiotix, Nektar Therapeutics, Novartis Pharma, Octimet Oncology Nv, Oncoethix, Oncopeptides, Orion Pharma, Ose Pharma, Pfizer, Pharma Mar, Pierre Fabre, Medicament, Roche, Sanofi Aventis, Seattle Genetics, Sotio A.S, Syros Pharmaceuticals, Taiho Pharma, Tesaro, Xencor, and grants from Roche, Institut Roche, Merck KGaA, AstraZeneca outside the submitted work. No disclosures were reported by the other authors.

R.M. Chabanon: Conceptualization, data curation, formal analysis, validation, investigation, methodology, writing–original draft, writing–review and editing. D. Morel: Conceptualization, data curation, formal analysis, validation, investigation, methodology, writing–original draft, writing–review and editing. T. Eychenne: Data curation, investigation, methodology, writing–review and editing. L. Colmet-Daage: Data curation, investigation, methodology, writing–review and editing. I. Bajrami: investigation, methodology, writing–review and editing. N. Dorvault: Data curation, investigation, methodology, writing–review and editing. M. Garrido: Data curation, investigation, methodology, writing–review and editing. C. Meisenberg: Investigation, writing–review and editing. A. Lamb: Data curation, investigation, methodology, writing–review and editing. C. Ngo: Data curation, investigation, methodology, writing–review and editing. S.R. Hopkins: investigation, writing–review and editing. T.I. Roumeliotis: Investigation, methodology, writing–review and editing. S. Jouny: Investigation, methodology, writing–review and editing. C. Hénon: Writing–review and editing. A. Kawai-Kawachi: Writing–review and editing. C. Astier: Writing–review and editing. A. Konde: Investigation, methodology, writing–review and editing. E. Del Nery: Writing–review and editing. C. Massard: Writing–review and editing. S.J. Pettitt: Writing–review and editing. R. Margueron: Writing–review and editing. J.S. Choudhary: Writing–review and editing. G. Almouzni: Resources, writing–review and editing. J.-C. Soria: Resources, writing–review and editing. E. Deutsch: Resources, writing–review and editing. J.A. Downs: Resources, writing–review and editing. C.J. Lord: Conceptualization, resources, supervision, funding acquisition, methodology, project administration, writing–review and editing. S. Postel-Vinay: Conceptualization, resources, supervision, funding acquisition, methodology, writing–original draft, project administration, writing–review and editing.

This work was funded by programme grants to SPV from Institut National de la Santé et de la Recherche Médicale (INSERM ATIP-Avenir Programme Funding), Site de Recherche Intégrée sur le Cancer SOCRATE-2 (INCa-DGOS-INSERM_12551), Cancéropôle Ile-de-France (2017–1-EMERG-72), and Association pour la Recherche contre le Cancer (PGA1 RF 20190208576), as well as programme grants to CJL from Cancer Research UK (CRUK Programme Funding C30061/A24439) and Breast Cancer Now as part of programme funding to the Breast Cancer Now Toby Robins Research Centre. R.M. Chabanon received award funding from Fondation Bettencourt-Schueller, Institut Servier, Fondation des Treilles and Cancéropôle Ile-De-France. D. Morel received funding from the INSERM ITMO Cancer grant. This study represents independent research supported by the National Institute for Health Research (NIHR) Biomedical Research Centre at The Royal Marsden NHS Foundation Trust and The Institute of Cancer Research, London. The views expressed are those of the author(s) and not necessarily those of the NIHR or the Department of Health and Social Care. The authors would like to thank Dr. Marjorie Drac and Dr. Etienne Schwob, as well as the Genomic Vision team for performing some DNA combing experiments related to this project. The authors thank Dr. Damien Drubay for his help with the statistical analyses.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.