Integrated genomics point to immune vulnerabilities in pleural mesothelioma

Pleural mesothelioma is an aggressive malignancy with limited effective therapies. In order to identify therapeutic targets, we integrated SNP genotyping, sequencing and transcriptomics from tumours and low-passage patient-derived cells. Previously unrecognised deletions of SUFU locus (10q24.32), observed in 21% of 118 tumours, resulted in disordered expression of transcripts from Hedgehog pathways and the T-cell synapse including VISTA. Co-deletion of Interferon Type I genes and CDKN2A was present in half of tumours and was a predictor of poor survival. We also found previously unrecognised deletions in RB1 in 26% of cases and show sub-micromolar responses to downstream PLK1, CHEK1 and Aurora Kinase inhibitors in primary mesothelioma cells. Defects in Hippo pathways that included RASSF7 amplification and NF2 or LATS1/2 mutations were present in 50% of tumours and were accompanied by micromolar responses to the YAP1 inhibitor Verteporfin. Our results suggest new therapeutic avenues in mesothelioma and indicate targets and biomarkers for immunotherapy.

Copy number alterations (CNAs) analysis shows recurrent CDKN2A, RB1 and SUFU deletion and RASSF7 amplification. We analysed the SNPs for CNAs using the GISTIC program 14 . The program estimates genomic boundaries for recurrent CNA events and assigns statistical significance after false discovery rate (FDR) corrections.
Deletion of the CDKN2A region is extensive in many malignancies and a full locus map (Fig. 1b) revealed large deletions to also be present in mesothelioma. The map revealed a frequent deletion of the closely neighbouring Type I Interferon (IFN) genes, as we have first reported 16 and has later been confirmed 17 (Fig. 1b and Supplementary File 2_Table 6): 38/118 patients (32%) had predicted homozygous IFN Type I deletion and 24/118 (20%) had heterozygous deletion. The median survival of patients with co-deletion of CDKN2A and IFN Type I genes was not statistically different to CDKN2A deletions alone (8.3 months compared to 10.7, P = 0.6; Supplementary File 2_Table 6).
We observed frequent deletions at multiple other loci ( Fig. 1d-f, Supplementary File 1_Table 3 and Supplementary File 2_Table 5B for the extensive list of copy-number coordinates for each sample). The most common novel deletion was the RB1 locus on 13q14.2 in 31/118 patients (26%). The RB1 tumour suppressor is activated downstream of CDK4 and CDK6 18 and its deletion predicts a mechanism for resistance to CDK4/6 inhibitors.
Deletion of a locus on 10q24.32 containing SUFU (Suppressor of Fused), ARL3 (ADP Ribosylation Factor Like GTPase 3) and TRIM8 (Tripartite Motif Containing 8) was observed in 25/118 tumours (21%). SUFU inhibits activation of GLI transcription factors in the Hedgehog pathway, which is known to be disordered during mesothelioma carcinogenesis 19,20 . TRIM8 acts as a tumour suppressor inducing cell cycle arrest in a TP53 dependant manner, and as an oncogene activating NF-kβ and TNF-α 21 . It is involved both in immunity and cancer 22 .
Previously unrecognised regions of amplification (Figs. 1a, 2, Supplementary File 1_Table 3 and Supplementary File 2_Table 5B for the extensive list of copy-number coordinates for each sample) included a locus on 11p15.5, amplified in 39/118 tumours (33%) that contained RASSF7 and miR-210 (Fig. 1g). When up-regulated RASSF7 controls cell growth and apoptosis in different tumours 23 , and functions as an oncogene in NSCLC, interacting with MST1 to dysregulate Hippo signalling 24 .
We sought replication of the novel CNAs in 98 tumours from the landmark study of Bueno et al. 5 which carried out WES and concurrent RNA-sequencing. We confirmed amplification of RASSF7 and deletion of RB1 and SUFU, each of which correlated with its transcript abundance (Supplementary File 1_Fig. 4

and Supplementary File 2_Table 7).
Deletion of SUFU locus, component of Hedgehog signalling pathway, is associated with downregulation of immune related genes. Deletion of the SUFU locus in 25/118 tumours (21%) was associated with marked upregulation of the Patched 2 tumour suppressor (PTCH2) ( Table 1). Ptch2 overexpression has been observed in Sufu knockout mice and is indicative of aberrant Hedgehog signalling 20 . Hedgehog pathways are activated in mesothelioma patients, in the absence of obvious mutations 19 . Also, upregulated were NHS, HOXA7 and TRPS1, each of which regulate tissue differentiation (Table 1).
We found a SMO (Smoothed) inhibitor (Vismodegib, GDC-0499) to be inhibitory in only one PMCC (Table 2). Vismodegib has previously been shown to be effective in modulating mesothelioma tumour and Figure 1. Mapping of copy number alterations in subjects with pleural mesothelioma. (a) Statistically significant regions of amplification and deletion from GISTIC analysis of 118 subjects. Peak regions which pass both the G-Score ( Gistic score in ver.2.0 is defined as the negative logarithm of the probability of observing a candidate copy-number segment with given amplitude and frequency, provided the background copy-number alteration rate) and q-bound (< 0.05) threshold cut-offs are shown for deletions (blue) and amplifications (red) (see also  27 , suggesting beneficial effects of Hedgehog inhibition would be worthy of investigation in mixed cellular cultures of human mesothelioma. SUFU deletion unexpectedly correlated with downregulation of prominent T-cell genes (Table 1), including ODF3B; the killer-cell receptor KLRD1 (CD94); and HSH2D, a target of T-cell activation. Downregulated monocyte/macrophage and dendritic cell markers included IL4R, SCIMP, SIGLEC1 (CD169), CLEC10A, and CR1 (Complement C3b/C4b receptor 1).
We confirmed these results in the large independent dataset from Bueno et al. 5 , confirming that abundances of Hedgehog pathway transcripts SUFU, PTCH1 and PTCH2 correlated with KLRD1 and CR1 (Supplementary File 1_Table 5).
Hippo pathway is deregulated in 50% of mesothelioma patients due to RASSF7 amplification and NF2, LATS1/2 mutations. Hippo monitors external factors that shape tissue structure 28 . NF2 recruits core Hippo signalling pathway members (LATS1/2) to inhibit activation of the transcriptional cofactors YAP1 and TAZ 29 . RASSF7 also regulates Hippo pathways, and its overexpression promotes phosphorylation and nuclear translocation of YAP1 24 . We found RASSF7 amplification in 39 mesotheliomas, NF2 mutations in 24, LATS2 mutations in 6 and LATS1 in 2, so that non-overlapping lesions in Hippo pathways were present in 52/121 mesotheliomas (43%) and a further 9 mesothelioma had more than one lesion (total 50%) (Fig. 2). WNT5B transcription, which we found increased in sarcomatoid tumours, may also induce YAP/TAZ activation through non-canonical pathways 30 .
We did not find significant differences in transcript abundances when comparing RNA-sequencing derived transcriptomes for RASSF7 amplifications to other tumours; or for lesions in Hippo signalling genes (NF2, LATS1, LATS2) singly or combined; or for mesotheliomas with or without SETD mutations. We did not detect RASSF7 amplification in any of 19 primary cell lines examined by WGS and SNP array (Supplementary File 1_Fig. 5), which may reflect selection in culture for mesothelioma genotypes that grow independently of a fibroblast matrix.  We found two NRAS mutations at known oncogenic RAS hotspots (G12V and Q61H). Both mutations were found in sarcomatoid subtype tumours that did not have alterations in CDKN2A, BAP1 or NF2. In the TCGA-Meso cohort, only one patient with biphasic histology had a G12C KRAS hotspot mutation with a concurrent TP53 mutation but without alterations in other mesothelioma drivers, similar with our cohort. In the Bueno et al. 5 cohort also only one patient with sarcomatoid subtype had a Q61K NRAS mutation but this was detected simultaneously with a BAP1 frameshift mutation. We also identified three other RAS pathway related genes by WES: a NF1 stop mutation (c.6439C > T, p.Q2147*), a splice site RASA1 mutation (c.829-1_858.del) and a HRAS in-frame deletion (c.187_189del, p.E63del).
TP53 mutations carried a worse prognosis compared with TP53 wild-type counterparts (mean OS 5.7 vs. 13.6 months, P = 0.0005), as previously described 5 . We did not detect significant associations of other mutations with survival.
We detected a missense germline mutation localized in the UCH domain of BAP1 from one patient with epithelioid subtype (Supplementary File 1_Fig. 3b). In other subjects, single deleterious germline mutations were found in MSH5 and MSH6 (representing the mismatch-repair (MMR) pathway), RB1, SETD6 and BRCA2. BAP1 is the most common mutated gene in mesothelioma and is associated with up-regulation of RET. We explored the effects of genetic alteration by comparing RNA-sequencing samples with and without specific genetic alterations. When compared to other tumours, BAP1 mutations or deletions were associated (P adjusted < 0.05) with up-regulation of the RET proto-oncogene 31 and NNAT. NNAT overexpression is associated with poor outcome in multiple cancers 32 (Supplementary File 1_Table 6). We replicated the negative association of BAP1 with RET to be present also in the Bueno et al. 5 (r = −0 .32, P = 2.2E−06) and TCGA-Meso 4 (r = − 0.45, P = 1.3E−06) datasets.

Mutational signatures 1 and 3 are prevalent in mesothelioma.
A median of 31 non-synonymous somatic mutations per tumour exome were present in the 21 WES paired samples, consistent with the low rate observed by Bueno et al. 5 . We observed a similar low tumour mutational burden in the 77 paired samples that underwent targeted capture sequencing (Supplementary File 1_Fig. 3i).
One patient (NCMR035) had a hypermutated tumour (167 somatic mutations from WES) (Supplementary File 1_Fig. 6), accompanied by a frame-shift deletion in MSH6 (p.Phe1104LeufsX11) and a frame-shift insertion in PALB2 (p.Met1049AspfsX4). PALB2 encodes a protein that recruits BRCA2 and RAD51 at the site of doublestrand breaks 33 and plays a critical role in homologous recombination repair.
The mutation spectrum was characterized by C > T transitions, in both WES and TC-NGS panel data (Supplementary File 1_Fig. 3 and 3a respectively), consistent with earlier reports 4,5 . Analysis of mutational signatures 34 Signature 3 is indicative of DNA damage and failed breakpoint repair 35 . In other cancers, signature 3 mutations often accompany biallelic inactivation of BRCA1 or BRCA2, where the inability to repair DNA predicts good responses to platinum therapy. Mesothelioma responds poorly to such therapies, and we hypothesise that signatures of DNA damage may follow the actions of asbestos in the progenitor neoplastic cell.
WNT5B is higher expressed in sarcomatoid tumours. We did not see any significant associations between common lesions and histological subtypes. RNA-sequencing however revealed differential transcription between histologies (Supplementary File 1_Table 4). As reported previously 5 , WNT5B had higher expression in sarcomatoid and mixed histology tumours compared to epithelioid tumours. Other genes significantly upregulated (P adjusted < 0.001) in non-epithelioid tumours included GPR176 which acts as a circadian pacesetter 36 , and known adverse factors for other cancers such as IGF2BP1, CCBE1, HS3ST3A1, TRAM2 and SERTAD2.
High level of VISTA is frequent in epithelioid mesothelioma and its expression level correlates with Hedgehog and immune pathway components. We tested how the most frequent genomic alterations, BAP1 mutation and CDKN2A deletion, were translated at protein levels. We stained a subset of 28 tumours (Fig. 3a) with antibodies against BAP1 and MTAP (as a potential surrogate marker for CDKN2A deletion) along with PD-L1, VISTA, Ki-67 and an antibody for mitotic count (Fig. 3b-e). BAP1 staining revealed general or focal loss in 17/28 (61%) of cases, which only partially associated with BAP1 mutation or deletion (P = 0.01) (Fig. 3f), as previously suggested 37 . There was significant difference when comparing MTAP H-score between CDKN2A/MTAP deleted and wild-type samples (P = 0.001) (Fig. 3g)  www.nature.com/scientificreports/ correlated with copy number burden (P = 0.03, r = 0.42) and with MTAP score (P = 0.04, r = − 0.39), consistent with disordered cellular division accompanying CDKN2A deletion. Checkpoint inhibitors targeting PD-1 and its ligand PD-L1 cause marked tumour regression in some patients with mesothelioma 3 . However, PD-L1 is expressed at a low level, if at all, in most mesothelioma cases and its status imperfectly predicts response to immune checkpoint inhibitors 3 .
High-level staining of the alternative immune-checkpoint protein VISTA (V-domain Ig suppressor of T cell activation) 38 has been observed in epithelioid mesothelioma, and implies a better prognosis 4,39 . We confirmed a high level of VISTA by IHC in our samples (Fig. 3a), and in RNA-sequencing data replicable associations were seen between VISTA and SUFU, PTCH1, PTCH2, KLRD1 and CR1 (Supplementary File 1_Table 5).
Drug-testing shows that primary cell models of mesothelioma are sensitive to cell cycle targeted drugs. We explored potential therapeutic pathways that had been suggested by our genomic findings by determining the half maximal inhibitory concentration (IC 50 ) of selected compounds with three PMCC 12 that had been whole-genome sequenced (Supplementary File 1_ Fig. 5). We assessed by Western blots if deletions or mutations of the main mesothelioma drivers were translated to protein levels ( Supplementary File 1_Fig. 7a). For comparison, we included an immortalized mesothelioma cell line (H2052), a lung adenocarcinoma cell line (A549), and a transformed normal mesothelial cell line (Met-5A). We exposed cells to a range of drug concentrations (0.0005 to 50 micromolar (µM)) using as controls DMSO treated cells (Supplementary File 1_Fig. 7b-g).
As we had observed mutation signatures associated with repair of DNA double-strand breaks (COSMIC mutational signature 3), we tested two PARP inhibitors (Niraparib and Olaparib) that are effective in homologous repair deficiency 40 . Despite reports suggesting utility in mesothelioma 41 , we did not see a consistent inhibition of primary cell growth ( Table 2). Our results are supported by a recent report by Fennell et al. 42 that tested Rucaparib, a third PARP inhibitor, in BAP1 or BRCA1 defective mesothelioma patients. Although the clinical trial achieved its endpoints, the author suggested that there is no correlation between BAP1 or BRCA1 expression and patients' objective response to Rucaparib.
Another tested drug in our screen, Palbociclib, PD0332991, a CDK4/CDK6 inhibitor, that is supposed to target the most frequent deletion in mesothelioma, CDKN2A, showed only minor effects in our cell lines (Table 2). RB1 deletion has recently been shown to confer a robust and selective vulnerability to drugs that target DNA damage checkpoint (CHEK1) and chromosome segregation proteins such as Polo-Like-Kinase 1 (PLK1) 43 . CHEK1 is over-expressed in mesothelioma 44 , and RNAi screens have shown mesothelioma lines to be sensitive to CHEK1 and PLK1 knockdown. RB1 deficient tumours are hyper-dependent on Aurora kinase B (AURKB) for survival 45 , and AURKB inhibitors are efficacious against RB1 deficient lung cancers at non-toxic doses 45 . Consistent with these observations, we found micromolar (µM) to sub-µM responses when treating with an Aurora Kinase inhibitor (Alisertib, MLN8237); an inhibitor of PLK1/2/3 that induces G2/M arrest and apoptosis (Volasertib, BI6727) 46 ; and a CHEK1/2 inhibitor that abrogates the G2/S checkpoint (AZD7762).
The involvement of Hippo pathways in our results supports previous suggestions that YAP1 axis inhibition may be used in mesothelioma therapy 29 . We found µM IC 50 responses to the YAP inhibitor Verteporfin in all mesothelioma cell lines (Table2). Notably, immortalised Met-5A mesothelial cells that are not deficient in NF2 also responded.

Discussion
In our study we have extended previous genomic analyses by testing copy number aberrations (CNAs) through SNP genotyping arrays together with WES and targeted capture sequencing. We found genetic lesions to be enriched in RB1/cell-cycle, Hippo and Hedgehog pathways, and identified two major immunological influences. We assessed vulnerabilities of mesothelioma tumours by testing drugs that target altered pathways or their members in whole genome sequenced primary mesothelioma cells.
The most frequent genomic alteration in our subjects was deletion of the CDKN2A locus on 9p21.3, found in 60% of the analysed samples. This deletion predicts a worse OS than tumours without the deletion 47 . We observed that tumours with this deletion had a higher copy number burden compared with CDKN2A wild type patients, consistent with cell cycle dysregulation.
Hippo pathway activation was observed in more than 50% of mesothelioma tumours while Hedgehog pathway activation, as SUFU deletion, was seen in 21% of the tumours.
Cdkn2a deletion and Hedgehog and Hippo pathway activation have been observed in murine models of asbestos exposure well before tumour development 7,8,19 . Our results and the remarkable consistency of genetic lesions in mesothelioma in humans 4,5 and in mice 7,8 suggest a hypothesis that recurrent mesothelioma breakpoints and mutations occur in regions of chromatin that have been accessed during the inflammatory response to asbestos. Activation of Hedgehog might also contribute to the stroma-rich microenvironment that characterises mesothelioma tumours.
Our finding of RB1 deletions in 34% of tumours with CDKN2A deletion makes responses to CDK4/CDK6 antagonists less likely and underpins our finding that the CDK4/CDK6 inhibitor Palbociclib had marginal effects on primary cell survival. However, we note a previous publication has reported mesothelioma cell lines to be sensitive to this compound 48 and further investigations are indicated.
We therefore tested compounds downstream of RB1 and showed that RB1 defective primary cells responded well (irrespective of CDKN2A deletion) to an Aurora Kinase inhibitor (Alisertib, MLN8237); an inhibitor of www.nature.com/scientificreports/ PLK1/2/3 that induces G2/M arrest and apoptosis (Volasertib, BI6727) and a CHEK1/2 inhibitor (AZD7762). These findings encourage the clinical investigation of these or related compounds, and a Phase II Trial of Alisertib in salvage malignant mesothelioma is currently under way (NCT02293005). We identified a recurrent novel amplification of RASSF7 in 31% of tumours. Taken with other Hippo pathway members (NF2, LATS1 and LATS2), 50% of tumours had at least one lesion of this pathway. Our testing of primary cells revealed micromolar responsiveness of mesothelioma to the YAP inhibitor Verteporfin, although it did not seem to depend on the presence of NF2 or other Hippo mutations. These results are consistent with recent studies that have shown Verteporfin to be effective in vitro against mesothelioma cells 49,50 .
BAP1 is the archetypal mesothelioma gene 51 and was mutated in 31% of tumours and deleted in 33%. We gained some insight into its function by comparing transcriptome abundances between BAP1 mutation/deletion and BAP1 wild type tumours, where we found replicated up-regulation of the RET proto-oncogene. These results suggest tumour suppressor activities of BAP1 beyond deubiquitination 52 . It may be of interest that RET inhibitors are effective in RET-driven NCSLC and thyroid cancers 53 .
An important finding of our study was deletion of the SUFU locus on chromosome 10q24.32 in 21% of tumours. However, we did not find SUFU deletions in primary cells and Vismodegib, a Hedgehog inhibitor, was efficient in only one primary cell line. Beneficial effects of Vismodegib on tumour-stromal interactions have previously been shown 27 , and a role for Hedgehog pathways in mesothelial-matrix interactions (as opposed to simple driving of cell division) is further suggested by our findings of the upregulation of PTCH2, GJB2, NHS and HOXA7 in SUFU deleted tumours (Table 1). These results encourage the speculation that Vismodegib may be of clinical use to modify mesothelioma fibrosis. Together with SUFU, two other genes were deleted at the same locus, ARL3 and TRIM8, the latter being involved in innate immunity.
A striking novel finding in RNA-sequencing expression data of tumours with SUFU locus deletion was the downregulation of T-cell and antigen-presenting cell genes (Table 1). Although unexpected, these findings were strongly replicated in other data (Supplementary File 1_Table 5) and are consistent with the known central function of Hedgehog signalling in T-cells at the immunological synapse 54,55 .
High levels of VISTA, an alternative checkpoint inhibitor, have previously been reported in mesothelioma and confer a better outcome 4,39 . We confirmed the strong staining for VISTA by IHC and found that VISTA abundance strongly correlated with other SUFU-affected immune-synapse genes. By contrast, PD-L1 staining was generally weak in the tumours. A small molecule inhibitor against VISTA (CA-170) 56 is currently in a Phase I clinical trial (NCT02812875), and our findings may provide biomarkers as well as a stimulus further to investigate VISTA therapeutic blockade in mesothelioma.
In the same context, Vismodegib might be considered as an adjuvant to immunotherapy in the presence of SUFU deletion . It will be relevant to test if aberrant Hedgehog immune signalling is detectable in other malignancies.
It may also be of interest that the Type I Interferon genes on 9p21.3 were deleted in 52% of all mesothelioma. Interferons induce complex pro-inflammatory responses within tumour cells as well as in accessory immune cells 57,58 . Homozygous deletion of IFN genes is associated with poor response to CTLA4 blockade in patients with malignant melanoma 59 . Historically, administration of IFNA2 to patients with mesothelioma has occasionally induced complete regression 60,61 . Supplementary, early stage trials suggest that intra-pleural infection with viral vectors containing IFNA2 62 or IFNB1 63 induce inflammation and encourage beneficial mesothelioma responses, suggesting an adjuvant role for interferons in therapy.
The association of CDKN2A deletion with higher copy number burdens might indicate a beneficial effect of immune checkpoint inhibitors, but the co-deletion of IFN type I could enhance tumour cell evasion of immune surveillance. On the other hand, the deletion of IFN genes may encourage the use of oncolytic virus as therapies 17 . These alternatives could be explored in immunocompetent murine models of mesothelioma.
The results of our investigations should be interpreted in the light of several limitations. We were not powered to investigate determinants of fibrosis, an important variable feature of mesothelioma. Nevertheless, we examined tumours with tumour extent down to 20% as estimated by ASCAT analysis, compared to TCGA analysis of mesothelioma that was confined to tumours with only > 70% mesothelioma cells 4 , providing a reference for one extreme of the range.
The dose of asbestos exposure is a known determinant of fibrosis, but not of mesothelioma 64 . Sixty percent of our cases reported working with asbestos, but we did not have accurate estimates of exposure from a detailed occupational history or from fibre counts in unaffected lungs to include exposure as a covariate in our analysis.
Calling the presence of small or uncommon amplifications and deletions in heterogenous tumour samples may be problematic. The amplification segment sizes that we observed in the Infinium SNP array (with ~ 3 Kb spacing between SNPs) were on average seven times smaller than deletion segments (35.1 Kb vs. 265.8 Kb), and will have been harder to detect in the uneven coverage of WES. Nevertheless, the novel CNAs are detectable in the large Bueno et al. dataset 5 , each showing significant associations with the abundance of their transcribed RNA.
We have shown that genomic findings correlate with protein levels in histological sections of FFPE. Notably, we confirmed a high level of VISTA staining 4,39 which our transcriptomic data shows to be related to Hedgehog aberration. We were not powered to test systematically for the determinants of histological subtypes or histological features which may be important in clinical decision making. Therefore, there is an unmet need for investigations of the relationship between histology and genomic features identified in this and other studies.
We found distinctive transcriptome changes for some common lesions (BAP1 and SUFU deletion), but not for lesions in the RB1 or NF2 pathways. This may be due to lack of power, or possibly to differences in gene expression that result from either acceleration or braking of cellular division.
We tested tool therapeutic compounds as monotherapy in patient-derived low-passage mesothelioma cells. We selected primary cells and cell lines for the drug screen to include the main histological subtypes of mesothelioma and to cover the main drivers of mesothelioma. Although the cell lines contained the most common www.nature.com/scientificreports/ mutations, some CNAs, such as RASSF7 amplification were not found. We speculate that they may have been lost by weaning of pure mesothelioma cultures from other cell types. Compounds were chosen for testing to target the pathways shown by our genomic analysis. The known interactions of mesothelioma with stroma suggest better understanding of a wider range of drug effects may come from 3D models that include fibroblasts and immune cells.

Conclusions
In conclusion, our analyses suggest roles for Aurora Kinase, PLK, CHEK and YAP inhibitors in the treatment of mesothelioma growth. IFN Type I and SUFU deletions as biomarkers may guide more effective immunotherapies. VISTA inhibition may directly modify immune recognition of mesothelioma, and an adjuvant role in immunotherapy seems possible for Hedgehog inhibitors. The involvement of Hippo and Hedgehog signals and the intense fibrosis seen clinically assert a central role for tumour-matrix interactions in the pathogenesis of mesothelioma and suggest therapeutic avenues beyond tumour cell killing.  In vitro drug testing. Patient-derived primary cells, Meso-27T, Meso-33T and Meso-70T were obtained from the MRC Toxicology Unit, University of Cambridge, UK. Commercial cell lines NCI-H2052 (sarcomatoid mesothelioma), A549 (lung adenocarcinoma) and Met-5A (normal mesothelial, SV40 transformed) previously obtained from ATCC were gifted from the MRC Toxicology Unit. Original establishment of the primary cells was as previously described 12 . Additional information can be found in Supplementary File 1_Supplementary Materials and Methods. All primary cells and cell lines were maintained in RPMI-1640 growth media supplemented with L-glutamine (2 mM), penicillin (100 U/ml), streptomycin (100 μg/ml) and 10% FBS at 37 °C in 5% CO 2 . Eight drugs were investigated, based on our results. These were Niraparib (MK-4827, HY-10619, MedChem Express), Olaparib (HY-10162, MedChem Express), Palbociclib, PD0332991 (A8316, ApexBio), Alisertib (MLN8237, S1133, Selleckchem), Volasertib (BI6727, S2235, Selleckcehm), Vismodegib (GDC-0499, S1082, Selleckchem), AZD7762 (S1532, Selleckchem) and Verteporfin (SML0534, Sigma Aldrich). All drugs were diluted in DMSO and aliquots maintained at -20 °C. Drug aliquots were freeze-thawed no more than three times. For all experiments, controls consisted of DMSO-alone treated primary cells or cell lines.

Methods
Cells were seeded in 96-well plates (4 × 10 3 cells/well) 24 h prior to drug treatments. Each line was treated for six days (except for drug PD0332991 where treatment was 3 days) with a range of concentrations from 0.0005 to 50 μM. Cell viability was measured with MTS assay (CellTiter 96 ® AQueous One Solution Cell Proliferation Assay, Promega) on a plate reader (Tecan). License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.