Article https://doi.org/10.1038/s41467-022-34523-y Analysis of matched primary and recurrent BRCA1/2 mutation-associated tumors identi- fies recurrence-specific drivers Received: 23 July 2021 Jennifer B. Shah 1, Dana Pueschl1, Bradley Wubbenhorst1, Mengyao Fan1, John Pluta1, Kurt D’Andrea1, Anna P. Hubert1, Jake S. Shilan1, Wenting Zhou1, Accepted: 27 October 2022 Adam A. Kraya1, Alba Llop Guevara2, Catherine Ruan1, Violeta Serra 2, Judith Balmaña 3,4, Michael Feldman5,6, Pat J. Morin6, Anupma Nayak5, Kara N. Maxwell 6,7,8, Susan M. Domchek 6,7,8 & Check for updates Katherine L. Nathanson 1,6,8 1234567890():,; 1234567890():,; Recurrence is a major cause of death among BRCA1/2 mutation carriers with breast (BrCa) and ovarian cancers (OvCa). Herein we perform multi-omic sequencing on 67 paired primary and recurrent BrCa and OvCa from 27 BRCA1/2 mutation carriers to identify potential recurrence-specific drivers. PARP1 amplifications are identified in recurrences (False Discovery Rate q = 0.05), and PARP1 is significantly overexpressed across primary BrCa and recurrent BrCa and OvCa, independent of amplification status. RNA sequencing analysis finds two BRCA2 isoforms, BRCA2-201/Long and BRCA2-001/Short, respectively pre- dicted to be sensitive and insensitive to nonsense-mediated decay. BRCA2-001/ Short is expressed more frequently in recurrences and associated with reduced overall survival in breast cancer (87 vs. 121 months; Hazard Ratio = 2.5 [1.18–5.5]). Loss of heterozygosity (LOH) status is discordant in 25% of patient’s primary and recurrent tumors, with switching between both LOH and lack of LOH found. Our study reveals multiple potential drivers of recurrent disease in BRCA1/2 mutation-associated cancer, improving our understanding of tumor evolution and suggesting potential biomarkers. Pathogenic variants in BRCA1/2 are the most common cause of her- rearrangements, which facilitate a transition to cancer4. Carriers of editary breast and ovarian cancer, underlying 5–10% of breast and 20% pathogenic germline variants in BRCA1/2 have an increased risk of of ovarian cancers1,2. BRCA1 and BRCA2 function independently to breast and ovarian carcinomas, along with prostate and pancreatic maintain genomic integrity, with both playing critical roles in the cancers3,5. homologous recombination (HR) pathway of DNA repair3. The loss of Women with pathogenic variants in BRCA1/2 are diagnosed with BRCA1 or BRCA2 allows cells to accrue mutations and genomic high-grade serous ovarian carcinomas6. BRCA1 mutation carriers are 1 Division of Translational Medicine and Human Genetics, Department of Medicine, Perelman School of Medicine at the University of Pennsylvania, Phila- delphia, PA 19104, USA. 2Experimental Therapeutics Group, Vall d’Hebron Institut d’Oncologia, Barcelona, Spain. 3Hereditary Cancer Genetics Group, Vall d’Hebron Institut d’Oncologia, Barcelona, Spain. 4Department of Medical Oncology, Hospital Vall d’Hebron, Universitat Autònoma de Barcelona, Barcelona, Spain. 5Division of Surgical Pathology, Department of Pathology and Laboratory Medicine, Perelman School of Medicine at the University of Pennsylvania, Philadelphia, PA 19104, USA. 6Abramson Cancer Center, Perelman School of Medicine at the University of Pennsylvania, Philadelphia, PA 19104, USA. 7Division of Hematology-Oncology, Department of Medicine, Perelman School of Medicine at the University of Pennsylvania, Philadelphia, PA 19104, USA. 8Basser Center for BRCA, Perelman School of Medicine at the University of Pennsylvania, Philadelphia, PA 19104, USA. e-mail: knathans@upenn.edu Nature Communications | (2022)13:6728 1 Article https://doi.org/10.1038/s41467-022-34523-y predisposed to hormone receptor-negative breast cancers, whereas proteins in primary and recurrent BRCA1/2 mutation-associated BRCA2 mutation carriers more commonly develop hormone-positive tumors may help to identify markers of response to immunomodula- breast cancers3. Ovarian and triple-negative breast cancers tend to tory therapies. recur within a few years of diagnosis6,7. Recurrences are usually asso- To address outstanding questions about cancer recurrence in ciated with therapeutic resistance, and both recurrent ovarian cancer BRCA1/2 mutation carriers, we performed an analysis of paired primary and metastatic breast cancer are considered incurable. The standard of and recurrent BRCA1/2 mutation-associated tumors. The sample set care in BRCA1/2 mutation-associated ovarian cancers is platinum-based consisted of 67 primary and recurrent tumors from 13 BRCA1/2 chemotherapy (cisplatin, carboplatin) as first line treatment, with poly germline mutation carriers with breast cancer and 14 BRCA1/2 germline (ADP-ribose) polymerase inhibitors (PARPi) used increasingly in both mutation carriers with ovarian cancer. We performed a combination of the maintenance and metastatic settings. For BRCA1/2 mutation- whole exome, targeted, and RNA sequencing on this cohort of speci- associated breast cancer, systemic therapy is given as appropriate to mens, in addition to multiplex immunofluorescent microscopy40,41. the stage and receptor status of a patient’s disease. PARPi are Our goal was to identify genetic, transcriptomic, and immune events employed for late-stage breast cancer, and the OlympiA trial has sug- that underlie acquired therapeutic resistance. We hypothesized that gested benefit in the adjuvant setting8–10. Although tumors in BRCA1/2 differences between paired tumors would indicate mechanisms of mutation carriers may initially respond to platinums and PARPi, many tumor evolution associated with post-treatment tumor recurrence in develop resistance over time. Preclinical studies have demonstrated BRCA1/2 mutation-associated cancers. that platinum and PARPi resistance are mediated by several of the same mechanisms, such as expression of MDR1 (ABCB1) efflux pumps Results or loss of DNA damage response regulators (TP53BP1, REV7, CHD4, Cohort selection and clinical characteristics and PARP1)11–19. However, in tumors from BRCA1/2 germline mutation We studied matched primary and recurrent breast cancers from carriers, the most common identified resistance mechanism to plati- 13 subjects, nine BRCA1 and four BRCA2 pathogenic variant carriers nums and PARPi are reversion mutations affecting the mutant BRCA1/2 (Table 1). All but one BRCA2 mutation carrier with breast cancer were allele11,17,20–26. female. Nine of 13 primary breast tumors were estrogen receptor Prior studies of matched primary and recurrent BRCA1/2 expression negative (ER-), predominantly triple-negative breast cancer mutation-associated breast and ovarian cancers have been limited by (TNBC) with concordant ER- recurrences. We evaluated matched pri- small numbers (3–8 patients) or focused only on tumors resistant to mary and recurrent ovarian cancer from 14 women, 10 BRCA1 and four PARPi, not the entire breadth of frontline therapy and resistance17,27–31. BRCA2 pathogenic variant carriers. Most (9/14) primary ovarian tumors At present, most BRCA1/2 mutation carriers with breast cancer do not were diagnosed at stage III or IV. As treatment for their primary tumor, receive PARPi prior to their first recurrence. Therefore, acquired all patients received chemotherapy (Table 1). All 14 (100%) ovarian resistance mechanisms to conventional frontline breast and ovarian cancer patients received platinums and 4/14 (29%) received PARPi cancer therapies are largely unknown. In addition, clinical trials have prior to a collected recurrence. Three of 13 breast cancer patients evaluated the efficacy of immune checkpoint blockade on BRCA1/2 (23%) received platinums and a single breast cancer patient (1/13; 8%) mutation-associated breast and ovarian cancers, with variable objec- received PARPi. Breast cancer patients received hormonal therapy and tive response rates32–39. Characterizing expression of checkpoint radiation (38 patients, 54%) more frequently than ovarian cancer patients (21 patients, 21%). Additional clinical and sequencing details for each tumor are documented in Supplementary Data 1. Table 1 | Characteristics of primary/recurrent patient cohort and tumors Analysis of somatic mutations from whole-exome and targeted Breast Ovarian Total sequencing Patients 13 14 27 We used whole-exome sequencing (WES) to assess the landscape of Tumors Primary 13 14 27 somatic mutations in 67 paired primary/recurrent tumors and mat- Recurrent 18 22 40 ched germline samples from 27 BRCA1/2 mutation carriers with breast or ovarian cancer. We also re-sequenced 44 tumors and 18 respective Total 31 36 67 germline samples at high depth (>300 ×) using a custom targeted Number of recurrences 1 9 11 20 collected per patient capture of 209 genes (Supplementary Table 1). >1 4 3 7 We assessed individual gene variants with alternative allele fre- Germline mutation BRCA1 9 10 19 quency over 5% in oncogenes or tumor suppressors to identify whe- BRCA2 4 4 8 ther they were shared across primary and recurrent tumors Estrogen receptor (ER) ER+ 4 N/A 4 (Supplementary Data 2). The majority of the cohort (53/67 tumors; status of primary tumor ER- 9 N/A 9 80%) had loss of function (LoF) mutations in TP53. They were con- Stage at diagnosis I 2 1 3 cordant between the primary tumor and recurrence(s) in all but one of the patients, when limited to alternative allele fraction (AAF) of ≥5%; II 8 3 11 several tumors had additional TP53 mutations at AAF < 5% (Fig. 1a; III 2 8 10 Supplementary Fig. 1a–c and Supplementary Data 3). Patient 21 had IV 0 1 1 two different pathogenic variants in TP53: p.C141R in the primary Unreported 1 1 2 ovarian cancer and in her third and fifth recurrences; and p.P278S in Treatment receiveda,b Platinums 3 14 17 her first, second, and fourth recurrences (Supplementary Data 3). PARPic 1 4 5 Patient 13 had the same TP53 mutation (p.E285K) at 1.2% AAF in the Chemotherapy 13 14 27 primary tumor and 32.9% AAF in the recurrence. Breast tumors without Hormonal therapy 5 3 8 TP53 mutations were mostly BRCA2 mutation-associated and estrogen receptor positive (ER+) in origin. Five ovarian tumors from three Radiation 7 3 10 patients (17, 20, 25) also lacked TP53 mutations. a Treatments are limited to reflect the number of patients from whom a post-treatment recur- NOTCH1 mutations were identified exclusively in breast tumors, rence was collected. b Treatment groups are not mutually exclusive; most patients received multiple treatments. including one primary tumor, two recurrences, and one matched pair c PARPi poly (ADP-ribose) polymerase inhibitor. of primary/recurrent tumors. NF1 was mutated in both the primary and Nature Communications | (2022)13:6728 2 Article https://doi.org/10.1038/s41467-022-34523-y a b 1 2 3 Breast 4 Ovary BRCA1 5 BRCA2 Primary 6 Recurrence Intact 7 Heterozygous Loss Homozygous 8 Deletion 9 10 LOH 11 LOH Undetermined nonLOH 12 13 Patient 14 15 Mutation Type Heterozygous LoF 16 Heterozygous GoF 17 No mutation 18 19 20 21 22 23 Weight 24 0.75 25 0.50 26 0.25 0.00 27 TB A4 TS P1 U P PITCH1 PI 3C 3 AP 4 CD F 1 R K A M P2KC ABB1 M ME 3 AR 51 1 D DRA H R1 N 1 FGER2 K JA S C AK L1 EB 1 TP T X3 2 M MTK1 B 3 U T N NF1 N TCF2 Tu ut igin R or tion EN St pe St atus 1A 1B 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 1 2 3 1 2 O H K SM AD3R S8 R T C 53P5 C B D L4 A A 2 us R R R U U C R PT B1 Ty IC K AB at e Or m a O lin r M m o Somatic Mutation Mutational Signature er Tum G c HALLMARK_WNT_BETA_CATENIN_SIGNALING HALLMARK_UV_RESPONSE_DN HALLMARK_TGF_BETA_SIGNALING HALLMARK_PROTEIN_SECRETION HALLMARK_PEROXISOME HALLMARK_P53_PATHWAY HALLMARK_OXIDATIVE_PHOSPHORYLATION HALLMARK_NOTCH_SIGNALING HALLMARK_MYOGENESIS HALLMARK_MTORC1_SIGNALING HALLMARK_MITOTIC_SPINDLE HALLMARK_KRAS_SIGNALING_DN HALLMARK_HYPOXIA LoF Mutations in Primary Tumors HALLMARK_HEDGEHOG_SIGNALING LoF Mutations in Recurrences HALLMARK_GLYCOLYSIS HALLMARK_G2M_CHECKPOINT HALLMARK_EPITHELIAL_MESENCHYMAL_TRANSITION HALLMARK_E2F_TARGETS HALLMARK_DNA_REPAIR HALLMARK_COMPLEMENT HALLMARK_COAGULATION HALLMARK_BILE_ACID_METABOLISM HALLMARK_APOPTOSIS HALLMARK_ADIPOGENESIS 0.0 0.5 1.0 1.5 2.0 Normalized Enrichment Score d e Primary * Ovary Variant Classification of MutSigCV Mutations(*.maf) Mutation Type Frameshift Nonsilent Nonsynonymous SNV (REVEL > 0.5) Silent Nonsynonymous SNV (REVEL 0.5) Splicing Breast Intronic Recurrence * 0 10 20 30 10 20 30 0 Number of TP53 Mutations Number of TP53 Mutations recurrence for one breast and one ovarian cancer, and in two breast We also noted pathogenic variants in TP53BP1 (post-PARPi) and PARP1, cancer recurrences. One BRCA2 mutation carrier (Patient 10) had an each in a solitary BRCA1 mutation-associated recurrence. ER + primary breast tumor and two recurrences lacking TP53 muta- We next examined the global landscape of mutations by tumor. tions; instead, the tumor appeared to be driven by the oncogenic AKT1 Mutational signatures were dominated by Signature 3 (BRCA1/2 inac- E17K mutation42,43. Gain of function (GoF) mutations were identified in tivating mutations and defective HR) and Signatures 1A/B (aging)44 PIK3CA and TBX3 in one breast and one ovarian tumor, respectively. (Fig. 1b). Sixteen of 27 (60%) tumors had mutations associated with Nature Communications | (2022)13:6728 3 Article https://doi.org/10.1038/s41467-022-34523-y Fig. 1 | Integrated analysis of somatic mutations by whole-exome and targeted patient, with the primary tumor at the top and latest recurrence at the bottom. sequencing. a Phenotype, BRCA1/2 allele-specific loss of heterozygosity (LOH), RB1 LoF loss of function, GoF gain of function. b Mutational signatures derived from and PTEN status, and somatic mutations found in 67 paired primary and recurrent WES mutations. c Hallmark Gene Sets enriched in LoF mutations for primary and tumors (from 27 patients) sequenced by whole-exome sequencing (WES, n = 67) recurrent tumor groups (all FDR q < 0.25). d MutSigCV results for both primary and and high-depth targeted sequencing (n = 44). Display is limited to genes with ≥1 recurrent tumors (computed independently, *FDR q < 0.05 per MutSigCV results). mutation with alternative allele fraction (AAF) ≥ 0.05 from targeted sequencing and e Variants contributing to MutSigCV results in d, by tumor type. SNV single- variants with AAF ≥ 0.05 from WES. Tumors are displayed in chronological order by nucleotide variant. Signature 4 (smoking)44. In paired tumor comparisons, tumor muta- recurrences also had a lower HRD score compared to primary tumors tional burden (TMB) did not change from primary tumor to recurrence with LOH, including decreases in all three individual HRD metrics used (Supplementary Fig. 1d and Supplementary Data 4). To investigate to compute the score (Fig. 2b and Supplementary Data 4). For all three functional impact, we determined whether LoF variants accumulated tumors that first developed LOH upon recurrence, nonLOH primary in common pathways across the cohort. We ranked genes by pre- tumors had a lower HRD score than recurrences (Fig. 2b). We did not valence of LoF variants within primary and recurrent tumor groups, find any associations between LOH discordance and treatment history. then performed Gene Set Enrichment Analysis (GSEA) Preranked with Hallmark Gene Sets (Fig. 1c and Supplementary Data 5)45. LoF variants Assessment of genomic scarring and copy number variation from recurrences were enriched for sixteen gene sets not found in We next evaluated whether recurrent tumors showed more chromo- primary tumor LoF variants (all FDR q < 0.25). These gene sets high- somal instability and genomic scarring than primary tumors. Neither lighted potential differences in metabolic processes (oxidative phos- aneuploidy nor HRD scores (individual metrics or combined) were phorylation, glycolysis), apoptosis, TP53 signaling, and pathways significantly different between tumor pairs overall, although some regulating cell growth and identity (TGFβ, Notch, MTORC1, KRAS). primary and recurrent tumors did have different scores (Supplemen- To identify additional driver mutations in BRCA1/2 mutation- tary Fig. 4a–e and Supplementary Data 4). To identify commonly associated tumors, we performed MutSigCV analysis on primary and amplified and deleted genes in primary and recurrent tumors, we recurrent tumor groups independently (Fig. 1d, e and Supplementary performed GISTIC analysis, stratified by tumor type (Fig. 2c, d; Sup- Data 6)46. TP53 was the only significantly mutated gene within both plementary Figs. 4f, g and 5a–d; Supplementary Data 9)49. In a com- tumor groups (FDR q < 0.05). parison of GISTIC segments between breast and ovarian tumors, we identified intersections (>50% overlap) for sixteen deletions and eight Analysis of BRCA1/2 loss of heterozygosity amplifications (Supplementary Data 9). As the tumor types exhibited Using whole-exome and targeted sequencing, we assessed BRCA1/2 similar copy number profiles, we combined all primary and all recur- allele-specific loss of heterozygosity (LOH). First, we assessed whether rent tumors to identify shared and distinct amplifications and dele- somatic mutations in BRCA1/2 caused biallelic loss of function in our tions (Supplementary Data 10). Significant (90% CI and FDR q < 0.05) sample set of tumors (Supplementary Data 7 and Supplementary deletions were more similar between primary and recurrent tumors Data 8). We found pathogenic variants in the respective BRCA1 or than significant amplifications (Jaccard similarity index of 0.20 vs. BRCA2 genes in 20 of 67 tumors, all with very low alternative allele 0.06, Supplementary Data 10). Primary tumors had amplifications in fraction (< 2.7%), suggesting that if real, these mutations only affected MYC and CHD2 (Fig. 2c). Recurrent tumors shared these amplifications a minority of cells (Supplementary Data 8). Next, we used allele- and were also uniquely enriched for amplifications of ABCB10, another specific copy number to determine whether tumors underwent LOH ABC transporter associated with chemotherapy resistance (Fig. 2d)50. via copy number loss of their wild-type BRCA1/2 allele47. Overall, 22/27 Further, we noted recurrence-specific amplifications in TGFB2, TRAF5, (81%) primary tumors and 33/40 (82%) recurrences demonstrated LOH WNT3A, SOX4, and E2F3, suggesting that recurrences may have (Fig. 2a). Most primary/recurrent tumor pairs had concordant BRCA1/2 increased dosage of growth and lineage signal pathway genes. In allele-specific LOH status (20/27 patients; 74%). Seven patients showed recurrences, unique deletions included CASP9, CDKN2C, JUN, MUTYH, discordant LOH status when comparing their primary tumor to and RAD54L, suggesting dysregulation of cell cycle, DNA damage recurrence(s). Discordant tumor LOH status was more common in response, and apoptosis (Supplementary Fig. 4g). In addition, our BRCA2 mutation carriers, likely due to the lower prevalence of LOH subtractive analysis also identified recurrence-private deletions in within ER+ breast cancers48. KMT2C and MAD1L1. We observed discordant LOH status more frequently in breast We found PARP1 to be significantly (90% confidence interval, FDR cancers (five—38%) than in ovarian cancer patients (two—14%). Notably, q = 0.05) amplified in recurrences but not primary tumors (Fig. 2d). three of four ER+ breast cancer patients in the cohort had discordant Given the use of PARPi for treatment in BRCA1/2 mutation-associated LOH in their tumors. We noticed overlap between patients with LOH tumors, we investigated further. Over half of tumors (35/67, 52%) in the transitions and those with TP53 wild-type tumors (6/7 patients—86%). cohort had somatic gains (copy number ≥4) or amplifications (copy In three patients (6, 11, 12), the primary tumor did not have LOH, but number ≥6) in PARP1 (Supplementary Data 11). PARP1 gains and first exhibited LOH upon recurrence (nonLOH to LOH, Supplementary amplifications were present in primary ovarian tumors (43%), recur- Fig. 2a, b). Conversely, patients 10, 13, 17, and 20 showed the opposite rent ovarian tumors (32%), primary breast tumors (62%), and recurrent effect: primary tumors exhibited LOH but had at least one nonLOH breast tumors (78%) (Fig. 2e). Seventy percent of all patients (19/27) in recurrence (LOH to nonLOH, Supplementary Fig. 3a, b). This “LOH the cohort had a PARP1 gain or amplification in at least one tumor (85% reversal” was observed in two BRCA2 mutation carriers with breast of breast patients, 57% of ovarian patients). Most of these patients (11/ cancer, treated with non-platinum chemotherapy and radiation, and 19, 60%) had a PARP1 gain or amplification in both primary and two BRCA1 mutation carriers with ovarian cancer, both treated with recurrent tumors. Gains and amplifications tended to be focal and platinums. In patient 20, mutational Signature 3 was also lost, reflect- centered around the PARP1 locus, rather than arm-level, underscoring ing the change in HR proficiency. For patient 10, Signature 3 was pre- a potential selection effect (Fig. 2f)49. We also evaluated PARP1 copy sent in the primary tumor (with LOH) but not in the first recurrence number status in TCGA PanCancer cohorts of primary breast and (nonLOH) nor the final recurrence (again with LOH) (Fig. 1b). For ovarian tumors. We grouped TCGA tumors into BRCA1/2 germline patient 13, Signature 3 was observed in the recurrence without LOH, mutation-associated (gBRCA1/2) and HR wild-type (HR-WT) groups, as but not the primary with LOH. In patients 10, 17, and 20, nonLOH previously described (Supplementary Fig. 6a and Supplementary Nature Communications | (2022)13:6728 4 Article https://doi.org/10.1038/s41467-022-34523-y Data 12)48,51. Using the same methods for copy number analysis as To assess copy number variation by pathway, we next performed utilized for the primary/recurrence cohort to analyze the TCGA Level 1 GSEA Preranked on lists of genes ranked by prevalence of copy num- data, we found a high prevalence of PARP1 gains and amplifications in ber gain or loss within each tumor group (Fig. 2g and Supplementary both tumor types, with no differences based on BRCA1/2 status (Sup- Data 5). Primary tumors were enriched for gains in gene sets for UV plementary Fig. 6b and Supplementary Data 11). response, TNFα signaling via NF-κB, IFNα and IFNγ response, and angiogenesis. Recurrent tumors were uniquely enriched for gains in Nature Communications | (2022)13:6728 5 Article https://doi.org/10.1038/s41467-022-34523-y Fig. 2 | Genome-wide and gene-specific copy number variation. a Allele-specific confidence interval amplifications in all recurrences (breast and ovarian). For c and loss of heterozygosity (LOH) of BRCA1/2 in 67 paired primary and recurrent tumors, d, all highlighted genes have residual q ≤ 0.05. e PARP1 copy number by tumor in with treatment exposure. Tumors are displayed in chronological order by patient, primary/recurrent cohort. Groupwise differences in average copy number were with the primary tumor at the top and latest recurrence at the bottom. PARPi PARP determined by Kruskal–Wallis test, followed by Dunn’s test with Bonferroni cor- inhibitor. b Comparison of Homologous Recombination Deficiency (HRD) score for rection (α = 0.05, **p < 0.01). NS not significant. f Segments of copy number gains tumors with a change in LOH status from primary to recurrence (n = 14 biologically and amplifications encompassing PARP1, by patient. For e and f, total copy number independent samples, comprising 2 samples/patient from 7 patients). For one (Sequenza) was binned as follows: Deletion, CN = 0; Loss, CN = 1; Neutral, CN = 2–3; patient (Patient 6), who had multiple recurrences with LOH, the first recurrence is Gain, CN = 4–5; Amplification, CN ≥ 6. CNV copy number variation. g Hallmark Gene displayed. Boxplot elements are as follows: median, center line; box limits, first and Sets enriched in primary and recurrent tumor gains and amplifications (CN ≥ 4, top) third quartiles (spanning the IQR, interquartile range); whiskers, 1.5× IQR in each vs. losses and deletions (CN ≤ 1, bottom) (all FDR q < 0.25 per gene set enrichment direction; outliers plotted individually. c GISTIC qplot for 90% confidence interval analysis). amplifications in all primary tumors (breast and ovarian). d GISTIC qplot for 90% MTORC1 signaling genes and deletions in Hedgehog signaling genes previously described gene fusions with >5 junction-spanning reads (consistent with our finding that recurrent tumors accumulate LoF (range 5–228 reads) found in ≥3 patients. Six of 54 (11%) tumors had mutations in Hedgehog signaling genes). fusions involving genes for immunoglobulin light and/or heavy chains (IGL@, IGH@) (Fig. 4a, b and Supplementary Data 14). Fusions of Global transcriptomic analysis of primary and recurrent tumors MALAT1, a lncRNA associated with poor survival in metastasis in sev- We performed RNA sequencing (RNA-seq) on 50 primary and recur- eral tumor types, were recently observed at high frequency (13/18, rent tumors from the WES cohort, adding in recurrences from four 72%) in ovarian cancer and TNBC patients after PARPi treatment54,55. additional patients (subjects 28–31, Fig. 3a and Supplementary Data 1). We found MALAT1 fusions in one breast and two ovarian (3/54, 5.5%) We also conducted RNA-seq on 12 normal breast and fallopian tube tumors (Fig. 4b). We did not identify any fusions involving efflux pump samples from prophylactic surgeries in BRCA1/2 mutation carriers ABCB1, which have been previously associated with multidrug (subjects 32–43, Fig. 3a; Supplementary Fig. 7a and Supplementary resistance17,27. Data 1). We used hierarchical clustering to assess sample relatedness across the cohort (Supplementary Fig. 7b). In the resulting dendro- Differential BRCA2 isoform usage in primary and recurrent gram, many paired primary and recurrent tumors were not each oth- tumors er’s closest relatives (11/19 tumor pairs). In general, ER- breast tumors As mRNA isoform switching events are a well-recognized phenomenon clustered closer to ovarian tumors than ER+ breast tumors, which in in cancer, we evaluated differential transcript usage among primary turn clustered closest to normal samples. Sample relatedness tumors, recurrent tumors, and normal tissue samples56. We identified appeared to be most informed by ER status and tissue of origin. We an isoform switching event between recurrences and normal tissue, further assessed sample clustering using principal components (Sup- involving two protein-coding BRCA2 transcripts. Overall, BRCA2 gene plementary Fig. 7c). We did not observe any clustering associated with expression was significantly higher in breast and ovarian recurrent tumor origin (breast or ovarian), tumor type (primary or recurrent), or tumors compared to normal breast and fallopian tube tissue (adj. germline mutation (BRCA1 vs. BRCA2). p = 7.13 × 10−6, Fig. 4c), whereas BRCA2 was intermediately expressed in PARP1 expression was significantly increased in breast and ovarian primary tumors. The increase in BRCA2 expression was also observed recurrences compared to normal tissue (log2FC = 1.62, adj. in breast and ovarian recurrences when compared separately to mat- p = 2.52 × 10−6 and log2FC = 9.44, adj. p = 1.58 × 10−6) (Fig. 3b, c). PARP1 ched normal tissue (both log2FC ≥ 1.79, adj. p < 1 × 10−3). We found that expression was also increased in breast primary tumors (log2FC = 1.43, recurrent tumors preferentially expressed a shorter transcript adj. p = 3.26 × 10−5) (Supplementary Fig. 8a). Primary and recurrent (GRCh37.p13 ENST00000380152.3 or BRCA2-001, hereafter referred ovarian tumors both showed increased expression of MYBL2, for which to as “BRCA2-001/Short,” q = 1.7 × 10−4). In contrast, normal samples PARP1 is a putative coactivator (Fig. 3c and Supplementary Fig. 8b)52. favored a longer transcript (GRCh37.p13 ENST00000544455.1 or Primary and recurrent breast tumors showed increased expression of BRCA2-201, hereafter referred to as “BRCA2-201/Long,” q = 6.0 × 10−4) MKI67 (Ki67), which has been positively correlated with PARP1 (Fig. 4d). No significant difference in isoform usage between primary expression in other breast tumor cohorts (Fig. 3b and Supplementary tumors and normal tissues was observed. Both transcripts contain all Fig. 8a)53. Overall, PARP1 expression in all tumors exceeded the mean the functional domains, such as the BRC repeats and OB domains. PARP1 expression in normal tissue samples, an effect that did not BRCA2-201/Long has a longer 3’ UTR, predicted to be sensitive to depend on the presence of a PARP1 gain or amplification (p = 3.17 × 10−7 nonsense-mediated decay (NMD); whereas with a shorter 3’ UTR, and p = 5.01 × 10−5 for tumors with and without PARP1 gain or amplifi- BRCA2-001/Short is predicted to be NMD-insensitive, due to decreased cation, respectively, Supplementary Fig. 9). distance between the stop codon and terminal exon junction complex As the average log2(counts per million) for individual genes were (Fig. 4e)57. The differing region of the BRCA2 3’ UTR contains binding similar between breast and ovarian tumors overall (R2 = 0.934), we sites for RNA-binding proteins ELAVL1 (HuR) and PABPC1 (both FDR combined all BRCA1/2 mutation-associated tumors to identify the q < 0.01, Supplementary Fig. 10a), based on RNA immunoprecipitation major transcriptomic differences compared to normal tissues. Using sequencing (RIP-seq) experiments in human B cells (GEO Accession GSEA, we found enrichment of 25 Hallmark gene sets (all adj. p < 0.01, GSE35585, (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc= Fig. 3d, Supplementary Data 13). The identified gene sets included GSE35585). The presence of HuR binding sites in pre-mRNA introns several pathways also enriched for copy number gains (Fig. 2g): IFNα and 3’ UTR has been directly associated with increased mRNA and IFNγ signaling, MTORC1 signaling, response to UV light, and TNFa stability58. PABPC1 has been to shown to inhibit NMD in mRNA tran- signaling via NF-kB. Increased expression of these gene sets was also scripts with long 3’ UTRs59,60. identified in GSEA stratified by tumor type (Supplementary Fig. 8c, d We observed that expression of each BRCA2 isoform was mutually and Supplementary Data 13). exclusive in 64 of 66 (97%) RNA-seq samples, with increasing enrich- ment of the short isoform from the normal to primary to recurrence Analysis of gene fusions in primary and recurrent tumors groups (p = 0.001, Fig. 4f, Supplementary Data 15). Across all tumor We investigated whether gene fusions could be potential drivers in samples, expression of BRCA2-001/Short was slightly more common in BRCA1/2 mutation-associated tumors. We limited our analysis to breast tumors (10/20; 50%) than in ovarian tumors (14/34; 42%). Nature Communications | (2022)13:6728 6 Article https://doi.org/10.1038/s41467-022-34523-y a 3 b Breast Recurrences vs. Normal Breast 4 15 5 6 7 8 10 − Log10(adj.P) 9 CD300LG CENPF 12 RHAG MKI67 13 PSG8 ASPM ADIPOQ 14 5 PARP1 NEK2 15 CDK1 16 17 18 0 19 Breast 15 10 5 0 5 10 Ovary Log2(Fold change) BRCA1 20 BRCA2 c Patient Ovarian Recurrences vs. Normal Fallopian Tube Primary Recurrence 21 Normal Tissue from ZCCHC12 Mutation Carrier 15 CRISP2 RERGL Received 22 Not Received 23 MS4A8 LOH 24 LOH 25 Undetermined 10 nonLOH − Log10(adj.P) 26 27 ASPM CRISP3 28 SPC25 29 MYBL2 30 31 5 BIRC5 32 33 PARP1 34 35 36 37 0 38 15 10 5 0 5 10 39 Log2(Fold change) 40 41 42 d All Tumors vs. All Normal Tissue 43 HALLMARK_UV_RESPONSE_UP y HALLMARK_UNFOLDED_PROTEIN_RESPONSE al apy M gin n Pl ype s ot Pi n R rap um m tio tio R ri T e r Sa uta PA ia he in O Th e ad HALLMARK_TNFA_SIGNALING_VIA_NFKB at pl ue H em ss e on lin HALLMARK_SPERMATOGENESIS Ti h m m C or er G HALLMARK_REACTIVE_OXYGEN_SPECIES_PATHWAY HALLMARK_PI3K_AKT_MTOR_SIGNALING HALLMARK_P53_PATHWAY log(adj. p) HALLMARK_OXIDATIVE_PHOSPHORYLATION 2.5 HALLMARK_MYC_TARGETS_V2 5.0 7.5 HALLMARK_MYC_TARGETS_V1 Hallmark Gene Set HALLMARK_MTORC1_SIGNALING Set Size HALLMARK_MITOTIC_SPINDLE 50 100 HALLMARK_INTERFERON_GAMMA_RESPONSE 150 HALLMARK_INTERFERON_ALPHA_RESPONSE HALLMARK_INFLAMMATORY_RESPONSE NES 2.0 HALLMARK_IL6_JAK_STAT3_SIGNALING 1.8 HALLMARK_IL2_STAT5_SIGNALING 1.6 HALLMARK_GLYCOLYSIS 1.4 HALLMARK_G2M_CHECKPOINT 1.2 HALLMARK_ESTROGEN_RESPONSE_LATE HALLMARK_E2F_TARGETS HALLMARK_DNA_REPAIR HALLMARK_COMPLEMENT HALLMARK_APOPTOSIS HALLMARK_ALLOGRAFT_REJECTION BRCA2-001/Short was expressed in 8/11 (73%) breast recurrences recurrences compared to 2/7 (29%) BRCA2 mutation-associated compared to 12/23 (52%) ovarian recurrences. In general, BRCA2-001/ recurrences, but non-significant. There were no apparent associa- Short was expressed more frequently in BRCA1 mutation-associated tions with treatment history or LOH status. tumors (21/42; 50%) than in BRCA2 mutation-associated tumors (3/12; We confirmed expression of BRCA2-001/Short and BRCA2-201/ 25%). This difference was largest within the recurrence group; BRCA2- Long in primary and recurrent tumors by quantitative reverse tran- 001/Short was expressed in 18/27 (67%) BRCA1 mutation-associated scription PCR (RT-qPCR; Supplementary Fig. 10b, c and Nature Communications | (2022)13:6728 7 Article https://doi.org/10.1038/s41467-022-34523-y Fig. 3 | Global transcriptomic programs detected by RNA sequencing. ovarian tumor recurrences vs. normal fallopian tube tissue. For b and c, a positive a 66 samples used for RNA sequencing. Tumors are displayed in chronological log2(fold-change) indicates genes with increased expression in recurrent tumors. order by patient, with the primary tumor at the top and latest recurrence at the Adjusted p-values were computed based on linear modeling of mean-variance bottom. Patients 28–31 had recurrent tumors only. Patients 32–43 were BRCA1/2 trends (limma). d Hallmark Gene Sets enriched in genes with increased expression mutation carriers with no prior history of cancer or cancer treatment; their samples in primary and recurrent breast and ovarian tumors compared to normal breast and are normal tissue from breast and fallopian tube. LOH BRCA1/2 allele-specific loss of fallopian tube tissue (all adj. p < 0.05 per gene set enrichment analysis). NES nor- heterozygosity; PARPi PARP inhibitor. b Differential gene expression in breast malized enrichment score. tumor recurrences vs. normal breast tissue. c Differential gene expression in Supplementary Data 16). To further validate these results, we assessed Tumor evolution BRCA2 isoform expression in an independent cohort of 42 BRCA1/2 We present the multi-omic results from three patients for an in-depth mutation-associated primary breast and ovarian tumors (Supplemen- analysis of the relationship between primary and recurrent tumors. tary Data 1, Supplementary Fig. 11a). As in our discovery set, primary Patient 20 was a BRCA1 mutation carrier with an ovarian primary tumor tumors in the validation cohort more commonly expressed BRCA2- and four recurrences (Fig. 6a). Over 9 years during which the patient 201/Long (33/42; 79%, Supplementary Data 15). Expression of the received multiple lines of therapy, the tumor maintained BRCA1 LOH, shorter isoform BRCA2-001/Short was similarly uncommon in breast copy number gains in ABCB10/11, high aneuploidy scores (≥13), MYC vs. ovarian primary tumors (28%, 29%) and in BRCA1 vs. BRCA2 amplifications (CN ≥ 10) and PARP1 gains (CN = 4) (Supplementary mutation-associated primary tumors (28%, 31%). Fig. 3a). However, BRCA2 isoform usage evolved, with her primary Next, we tested whether expression of BRCA2-001/Short affected tumor expressing BRCA2-201/Long and recurrences expressing BRCA2- overall survival across 67 BRCA1/2 mutation carriers (from discovery 001/Short. Recurrence #2 had different genomic features (apparent and validation cohorts) for whom we had survival data and RNA nonLOH and a completely unique set of driver mutations, including a sequenced from at least one tumor (Supplementary Data 15). Survival PARP1 mutation), which suggest evolution from a distinct clone time for breast cancer patients did not have a significant association (Supplementary Fig 3b and Supplementary Data 2). The patient even- with ER status (p = 0.16), age at diagnosis (p = 0.47), BRCA1 vs. BRCA2 tually progressed on PARPi, and a BRCA1 reversion was identified germline mutation (p = 0.4), tumor stage at diagnosis (p > 0.74), or (Supplementary Fig. 15a, b). patient recurrence status (p = 0.58) (Supplementary Data 15). Out of 34 Patient 13 was a BRCA2 mutation carrier with TNBC. Her LOH breast cancer patients assessed, those with BRCA2-001/Short expres- primary yielded a nonLOH recurrence (Fig. 6b). This recurrent tumor sion in any tumor had significantly worse overall survival compared to was the only one that underwent LOH reversal without a concomitant patients whose tumors did not express this alternative isoform decrease in HRD score (Fig. 2b, left panel). The primary tumor had a (p = 0.017, HR = 2.535, 95% CI 1.179–5.45) (Fig. 4g). Median overall low-level pathogenic TP53 mutation with AAF of 1.2%, which increased survival was nearly 3 years shorter for patients whose tumor(s) to 32.9% in the recurrence, suggesting the outgrowth of a nonLOH expressed BRCA2-001/Short (87 months, 95% CI 71.2–121) compared to TP53-mutant subclone (Supplementary Data 3). Although the recur- those whose tumors did not (121 months, 95% CI 100.0–159). Breast rence shared multiple variants with the primary, multiple additional tumors expressing BRCA2-001/Short were evenly split between mutations emerged, reflective of an increase in TMB (2.1 to 12.8, Sup- patients that eventually recurred (54%) and those that did not (46%). plementary Data 4). The tumor also showed BRCA2 isoform switching Expression of this shorter BRCA2 isoform was not correlated with and a MYC amplification. The recurrence had copy number losses and overall survival in 33 ovarian cancer patients after adjustment for mutations in HR and other DNA damage response genes, including patient recurrence status (Supplementary Fig. 11b). MUTYH, RAD54L, and RAD52. RAD52 specifically is required for survival of BRCA2-deficient cells61. Aneuploidy and HRD scores all increased Analysis of DNA damage response and immune checkpoint upon recurrence, indicating continued accumulation of genomic proteins instability and mutations. We next assessed whether PARP1 copy number gains and mRNA For Patient 6, a BRCA1 mutation carrier, we evaluated one primary expression translated to high PARP1 protein expression. We per- tumor and three recurrences of TNBC (Fig. 6c). The recurrent tumor formed immunohistochemistry (IHC) for PARP1 on three tissue had allele-specific LOH and a BRCA2 isoform switch (Supplementary microarrays (TMAs) containing 23 primary and recurrent BRCA1/2 Fig. 2b). Two recurrences had copy number gains in ABCB10/11. Over mutation-associated tumors from our sequencing cohort (Supple- the course of tumor progression, recurrences underwent loss of DNA mentary Fig. 12a–d, Supplementary Data 1, Supplementary Methods). damage response and repair genes (by copy number loss or LoF PARP1 protein was highly expressed across the cohort of tumors (mean mutation). These deleterious alterations varied across recurrences but H-score 252.5, median H-score 276.4) with no significant differences by clustered into the same pathways. tumor type (Fig. 5a). As in the RNA-seq results, high PARP1 expression did not depend on the presence of a PARP1 gain or amplification Discussion (Fig. 5b). However, the only tumor with a PARP1 copy number loss had In this study, we evaluated multi-omic features of paired primary and a relatively low H-score of 112. Overall, our results from RNA-seq and recurrent tumors from 27 BRCA1/2 mutation carriers, representing 13 IHC suggest that PARP1 is highly expressed in BRCA1/2 mutation- breast cancers (primarily TNBC) and 14 ovarian cancers (mostly high- associated breast and ovarian cancers. grade serous). Patients received a ‘real world’ mix of chemotherapy Next, we assessed DNA damage markers and immune checkpoint and radiation, in addition to platinums and PARPi, prior to collected proteins using the PhenoCycler platform (Supplementary Fig. 13a–d, recurrences. We assessed HRD, aneuploidy, somatic mutations, copy Supplementary Table 2, Supplementary Methods). We found no sig- number variation, and global transcription in paired BRCA1/2 nificant differences in percent of RAD51+ tumor cells between primary mutation-associated breast and ovarian tumors, focusing on features and recurrent tumors (Supplementary Fig. 14a, b). However, we found associated with recurrence. that the percent of (tumor and stromal) RAD51+ cells trended higher in We initially assessed LOH status in matched primary and recurrent tumors expressing BRCA2-001/Short compared to those expressing tumors to determine whether BRCA1/2 deficiency varied over the BRCA2-201/Long (Fig. 5c, d). We found no differences in percent of course of tumor evolution. Seven pairs of primary and recurrent (tumor and stromal) CTLA-4+ or PD-1+ cells between primary and tumors demonstrated changes in allele-specific LOH status, more recurrent tumors (Supplementary Fig. 14c–f). commonly in breast cancer. We observed the highest frequency of Nature Communications | (2022)13:6728 8 Article https://doi.org/10.1038/s41467-022-34523-y a b MALAT1 IGH-@ 6 chr11 chr14 Breast Ovary Junction Reads BRCA1 BRCA2 Primary Recurrence 9 Received Not Received Patient Spanning Reads LOH nonLOH 21 24 f 32 0 Breast 33 0 No fusion 26 Ovary 34 0 1 fusion 2 fusions BRCA1 35 0 27 BRCA2 36 0 Normal 37 0 Received on n 38 0 py in AT ion pe n ad y s n io Pi ap io um r ig tio si ra Ty at Not Received s us R er fu fu ia O he 39 0 ut PA in e f Th 1 at M ue pl @ @ ot m Pl R m e al ss 40 0 H L AL Sa lin he on IG Ti IG m LOH M C m 41 0 er or G H LOH 42 0 c Undetermined 43 0 p=7.13x 10−6 nonLOH Total B R C A 2 Expression 3 1 (Transcripts per Million) 30 p=2.6x 10−4 ns 4 1 d Transcripts per Million 5 6 1 1 Sample Type Normal Tissue Recurrent Tumors 25 7 1 (out of Total BRCA2) 20 p=6.0x10-4 20 8 1 0.9 15 9 1 Isoform Fraction 10 12 1 p=1.7x10-4 0.6 5 13 1 10 Primary 15 1 0.3 16 1 17 1 0 0.0 18 1 19 1 g rt ue s s on ho or or 20 1 ss 1/L m m 1/S Ti 22 1 Tu Tu 20 00 al nt y 2- 23 1 2- m ar re CA CA or im ur 24 1 N BR BR Pr ec e 25 1 R 26 1 Sample BRCA2-001/Short (ENST00000380152.3) 3 2 4 2 5 2 Protein coding 6 4 NMD Insensitive 6 3 Shorter 3’ UTR 6 2 7 2 8 2 9 2 BRCA2-201/Long (ENST00000544455.1) 12 2 13 2 14 2 Protein coding 15 2 NMD Sensitive 16 2 Longer 3’ UTR 17 2 Recurrence 18 2 19 3 19 2 20 5 g BRCA2-001/Short Status Expressed in Primary and/or Recurrent Breast Tumor Not Expressed in Any Breast Tumor 20 4 20 2 21 6 1.00 Median OS 87 (95%CI 71.2 121) vs. 121 (95%CI 100.0 159) Months 21 5 p=0.017 21 3 21 2 22 2 23 2 0.75 24 2 25 2 Overall Survival Probability 27 2 28 2 29 2 30 2 0.50 31 2 ia y R her py P uta in H C latin tion on o A s T ra i n al the RP ad ap m m P um rt tio ng M rig ho Lo e O /S 1/ lin e 01 m su 20 -0 or he er Tis - A2 A2 C C 0.25 BR BR G 0.00 0 50 100 150 200 250 Time (Months) nonLOH tumors (50% of tumors, 100% of patients) in BRCA2 mutation- Each primary/recurrent pair contained shared variants with relatively associated breast cancers, as has been previously reported48. We also stable AAFs, demonstrating they arose from the same clone. However, found that LOH discordance was more common in patients with ER+ the recurrences with LOH had multiple additional somatic variants, breast cancer and those without somatic TP53 LoF mutations. Three which were shared between recurrences (if multiple were collected). breast cancers (one BRCA1, two BRCA2) developed allele-specific LOH Thus, the development of allele-specific LOH appeared to be part of upon recurrence, associated with a significant increase in HRD score. tumor progression in these cases. These data are consistent with the Nature Communications | (2022)13:6728 9 Article https://doi.org/10.1038/s41467-022-34523-y Fig. 4 | Gene fusions and isoform switching detected by RNA sequencing. computed based on linear modeling of mean-variance trends (limma). d BRCA2 a Clinical characteristics of patients in which IGH-@, IGL-@, and MALAT1 fusions isoform usage (BRCA2 isoform expression normalized to total BRCA2 expression) in were identified. Tumors are displayed in chronological order by patient, with the recurrences vs. normal samples; q-values computed using DEXSeq within iso- primary tumor at the top and latest recurrence at the bottom. “1 fusion” refers to a formSwitchAnalyzer. e BRCA2 isoforms involved in isoform switching event. NMD translocation with one other gene. LOH BRCA1/2 allele-specific loss of hetero- nonsense-mediated decay, UTR untranslated region. f BRCA2 isoform expression zygosity; PARPi PARP inhibitor. b Example of MALAT1-IGH-@ gene fusion IGV tracks by sample and group for entire RNA sequencing cohort. g Overall survival (OS) from patient 9. Junction reads (red, middle track) represent split RNA-seq reads curve for patients that expressed BRCA2-001/Short in any (primary or recurrent) used to map the fusion breakpoint. Spanning reads (black, bottom track) represent breast tumor compared to those that did not. Survival proportions and p-value paired-end reads of fragments that span, but do not directly overlap, the fusion were calculated using a Cox proportional hazards model tested for significant breakpoint. c Expression of total BRCA2 (all isoforms) in recurrences vs. normal associations with ER status, age at diagnosis, tumor stage at diagnosis, and patient samples (n = 66 biologically independent samples from 39 patients). Data are recurrent status (α = 0.05, see Methods). expressed as mean values +/– SD. NS not significant. Adjusted p-values were a b ns ns 300 300 Average PARP1 H Score by Tumor Average PARP1 H Score by Tumor 200 200 Tumor Type Primary Breast Tumor Recurrent Breast Tumor Primary Ovarian Tumor Recurrent Ovarian Tumor 100 100 0 0 Breast Breast Ovarian Ovarian Loss Neutral Gain/Amp Primary Tumor Recurrence Primary Tumor Recurrence PARP1 Copy Number Status n=3 n=4 n=8 n=8 Tumor Type c Breast Tumor d Ovarian Tumor 80 80 ns RAD51+ Cells (%) ns RAD51+ Cells (%) 60 60 40 40 20 20 0 0 rt g g rt ho on on ho /S /L /L /S 01 01 01 01 -2 -2 -0 -0 A2 A2 A2 A2 C C C C BR BR BR BR Fig. 5 | Validation of prior results by PARP1 and RAD51 protein expression. RAD51+ cells in primary (n = 6) and recurrent (n = 7) ovarian tumors based on a PARP1 nuclear positivity (average tumor H-score) by tumor type, for 23 tumors in BRCA2 transcript usage. For c and d, boxplot elements are as follows: median, tissue microarrays. NS not significant. b PARP1 nuclear positivity (average tumor H- center line; box limits, first and third quartiles (spanning the IQR, interquartile score) by PARP1 copy number status across all tumors. For a and b, groupwise range); whiskers, 1.5x IQR in each direction; outliers plotted individually. Groupwise differences were assessed by Kruskal–Wallis test, followed by Dunn’s test with differences in c and d were assessed by two-sided Wilcoxon rank sum test Bonferroni correction (α = 0.05). c Percent of RAD51+ cells in primary (n = 3) and (α = 0.05). Tumors without RNA-seq data (n = 2) or expressing both BRCA2 isoforms recurrent (n = 4) breast tumors based on BRCA2 transcript usage. d Percent of (n = 1) were excluded from this analysis. higher rates of BRCA1/2 allele-specific LOH reported for patients that in some cases, recurrences are associated with the outgrowth of a enrolled in clinical trials with metastatic breast cancer compared to nonLOH clone secondary to therapeutic pressures, which is supported those with primary tumors48,62. by the trajectory of genomic changes in patients 10 and 13. For patient Interestingly, two breast (both BRCA2) and two ovarian (both 10, one recurrence (10-2) does not demonstrate LOH and has a lower BRCA1) tumors underwent an apparent reversal of allele-specific LOH, HRD score. In contrast, the primary and second recurrence (10-3) do reflected by lower HRD scores in nonLOH recurrences. We postulate have LOH, with the primary tumor also demonstrating Signature 3. 10-1 Nature Communications | (2022)13:6728 10 Article https://doi.org/10.1038/s41467-022-34523-y Fig. 6 | Evolution of BRCA1/2 mutation-associated breast and ovarian tumors in mutation carrier with breast cancer. For a–c, display is limited to tumors sequenced three patients. a Genetic and clinical features in Patient 20, a BRCA1 mutation for this study. Copy number losses refer to genes with total copy number of 0 or 1; carrier with ovarian cancer. LOH allele-specific loss of heterozygosity. b Genetic copy number gains refer to genes with total copy number ≥4. Loss of function (LoF) and clinical features in Patient 13, a BRCA2 mutation carrier with breast cancer. mutations reported here are from high-depth targeted sequencing. AAF alternative allele fraction. c Genetic and clinical features in Patient 6, a BRCA1 Nature Communications | (2022)13:6728 11 Article https://doi.org/10.1038/s41467-022-34523-y and 10-3 share several variants at high levels not seen in 10-2, although previously50. In a subtractive copy number analysis, we also observed other variants are shared amongst the primary and both recurrences. recurrence-specific deletions of KMT2C and MAD1L1, and frequent Similarly, the recurrence in patient 13 demonstrates the same TP53 mutations across the cohort in KMT2C. Loss of KMT2C (MLL3) is asso- mutation as the primary, detected with a much higher AAF. Our ciated with progression and metastatic disease in multiple cancer observations of outgrowth of sub-clones without LOH are consistent types, including BRCA1/2 mutated breast and ovarian cancers66–70. with the findings of Imyantiov and colleagues, who found that Mitotic arrest deficient-like 1 (MAD1L1) is a component of the mitotic neoadjuvant therapy with platinums resulted in rapid selection of pre- spindle-assembly complex and repressor of TERT; its loss increases existing BRCA1-proficient cells, so that the tumors post-treatment did chromosomal instability and could allow telomere lengthening71,72. not exhibit BRCA1 LOH31,63. Our data extend that finding, suggesting Although the role of MAD1L1 in cancer has not been fully elucidated, that LOH reversal can also occur in tumors treated with non-platinum some studies suggest loss of expression is associated with metastatic chemotherapy and radiation. Of note, BRCA1 proficiency appeared to disease and poor prognosis73,74. Our results suggest that MAD1L1 is a be disadvantageous to tumors in the absence of platinum exposure, as tumor suppressor in the context of BRCA1/2 mutation-associated tumor relapses usually re-acquired BRCA1 LOH during therapy holi- tumors, loss of which is associated with recurrent disease. Ultimately, days, an effect we could not evaluate in this cohort31,63. the genetic and transcriptomic features identified as unique to recur- The etiology for the other tumors demonstrating LOH reversal is rences could represent drivers and therapeutic targets for late-stage not as straightforward to characterize. It is possible for some indivi- BRCA1/2 mutation-associated tumors. duals, differences between primary and recurrent tumors could be due We initially observed that PARP1 amplifications were statistically to the existence of a second unrelated primary tumor, not clinically significant (90% CI per GISTIC) in recurrent tumors and observed most appreciated. Although LOH status was consistent across tumors in frequently (78%) in breast cancer recurrences. We then found that patient 21, the pattern of variants suggests this phenomenon. The PARP1 was gained or amplified across all tumor groups. We identified primary tumor and two recurrences (ovary and intra-abdominal) had up to 8-fold amplifications centering on a minimal common region of the same TP53 mutation and are clearly closely related with multiple chr1:226,474,131-227,148,258, which fully encompasses only three overlapping variants. In contrast, three of the recurrences (breast, other genes (ITPKB, PSEN2, STUM) in addition to PARP1. In our and the lymph node) share a different TP53 mutation and no additional TCGA cohorts of primary tumors, we found a similar frequency of pathogenic variants, suggesting the breast tumor was in fact a second PARP1 gains and amplifications across gBRCA1/2 and sporadic HR-WT primary. For ovarian cancer patients 17 and 20, nonLOH recurrences breast and ovarian tumors. Variation in methods likely accounts for the both emerged post-platinums and did not share pathogenic variants different prevalence of PARP1 amplification reported in cBioPortal for with the primary and other recurrences. Both tumors had high purity TCGA PanCancer ovarian (5% vs. 22%) and breast (9% vs. 25%) cohorts75. and were included in high-depth targeted sequencing, so variants were We then observed (over two-fold) increases for PARP1 expression in unlikely to be missed. In these cases, it is difficult to know if nonLOH primary and recurrent breast tumors, as well as ovarian recurrences, recurrences arose from a subclone or were associated with a distinct independent of PARP1 copy number. The correlation of PARP1 copy unknown primary tumor. Thus, multiple etiologies may underlie the number with PARP1 expression approached significance, suggesting observation of allele-specific LOH reversal, supporting genomic ana- that with a larger sample set we might find an association, as has been lysis of recurrences to ensure optimal therapeutic selection in this previously observed76. Our results also agree with prior reports of high patient population. PARP1 expression in unselected ovarian and hormone receptor- We also evaluated multi-omic data to identify changes associated negative breast cancers77–79. Increased nuclear PARP1 protein has with recurrence across tumors. HRD and aneuploidy scores did not been correlated with decreased relapse-free and overall survival in change significantly with recurrence, consistent with prior findings breast cancer and AML80,81. In this study, we present evidence of PARP1 that HRD scores do not increase upon post-platinum recurrence in overexpression as a near-ubiquitous feature of both primary and BRCA1/2 mutation-associated ovarian cancer29. We also did not recurrent BRCA1/2 mutation-associated tumors, corroborated at the observe significant increases in TMB. Although a prior observation protein level. Our study supports that PARP1 copy number gains suggested that TMB increases upon recurrence post-platinum therapy represent a common (but not exclusive) route to PARP1 over- in BRCA1/2 mutation-associated ovarian cancers, the study included expression, and that such gains are common in breast and ovarian only five primary/recurrent ovarian tumor pairs17. Consistent with prior tumors regardless of BRCA1/2 status76. A highly positive correlation studies of ovarian cancers, our results suggest that BRCA1/2 mutation- between PARP1 expression and PARPi resistance previously has been associated malignancies accrue most of their genomic scarring and demonstrated in breast cancer cell lines, wherein the BRCA1-mutated mutations early in tumorigenesis, and increasing genomic instability is cell line HCC1937 had both the highest level of PARP1 expression and not a major factor driving recurrence. greatest resistance to PARPi82. Similarly, increased PARP1 expression We identified several significant differences between primary and has been correlated with PARPi resistance in AML cell lines83. Increased recurrent tumors with respect to somatic mutations, copy number levels of PARP1 protein may stoichiometrically dilute PARPi, which trap variation, and gene expression. We identified a TP53BP1 mutation fol- PARP1/2 complexes on DNA for synthetic lethality84,85. Thus, levels of lowing PARP inhibition in a BRCA1 mutation carrier, in line with prior PARP1 protein, which could be increased due to amplifications or studies demonstrating that TP53BP1 loss mediates PARPi resistance in overexpression, should be explored as biomarkers of therapeutic BRCA1-mutant cells and murine mammary tumors18,19. Several cancer- response or clinical outcomes in recurrences from patients treated associated gene sets were exclusively enriched for LoF mutations with PARPi86–89. within breast and ovarian recurrences, including those for apoptosis; We assessed transcriptomic differences in primary and recurrent glycolysis; and P53, TGFβ, Notch, and MTORC1 signaling. Hedgehog tumors and found two different isoforms of BRCA2. In silico predic- signaling genes accumulated both LoF variants and copy number tions suggest that BRCA2-001/Short and BRCA2-201/Long differ in deletions exclusively within recurrences, suggesting that Hedgehog regulation of mRNA stability and/or localization, as dictated by NMD signaling is required for primary tumorigenesis (as in sporadic breast sensitivity and regulation by RNA-binding proteins56,57. Expression of and ovarian cancers) but dispensable or dysregulated in BRCA2-001/Short was significantly more common in recurrent tumors recurrence64,65. Breast and ovarian recurrences had distinct copy and further enriched among breast tumors and BRCA1 mutation car- number profiles, despite having similar HRD and aneuploidy scores. riers. We hypothesize that the isoform switch may represent a Notably, recurrences of both type had ABCB10 amplifications, sug- mechanism by which BRCA1 mutation-associated tumors can mod- gesting a shared mechanism of chemoresistance only observed in vitro ulate BRCA2 transcript stability to retain HR function and enhance Nature Communications | (2022)13:6728 12 Article https://doi.org/10.1038/s41467-022-34523-y tumor survival. In nonLOH BRCA2 mutation-associated tumors, Methods expression of the shorter isoform could also protect mRNA expression Acquisition of tumor, germline, and normal tissue specimens from the wild-type BRCA2 allele. Our survival analysis offers support from BRCA1/2 mutation carriers for a correlation between BRCA2 isoform expression and clinical out- Patients gave written informed consent for research use of germline comes. We found that breast cancer patients lacking BRCA2-001/Short DNA, tumor specimens, and clinical data—including the publication of expression in any tumor (primary or recurrences) had significantly patient age –under IRB approved protocol at the University of Penn- improved overall survival (median 87 vs. 121 months), suggesting it sylvania and Vall d’Hebron. Participants were not compensated. Eligi- could be a prognostic biomarker. We did not observe differences in ble patients met the following criteria: (1) diagnosis of breast or ovarian survival based on isoform usage in ovarian cancer patients. In existing cancer; (2) positive genetic test result for pathogenic germline muta- literature, multiple mechanisms have been shown to restore BRCA2 tion in BRCA1 or BRCA2 from a Clinical Laboratory Improvement function in platinum and PARPi resistance, and isoform switching Amendments (CLIA)-approved laboratory; (3) available archived could represent an additional mechanism25,26,90–92. This hypothesis is germline DNA from blood or saliva; (4) available formalin-fixed par- also supported by increased RAD51 positivity (a readout of BRCA2 affin-embedded (FFPE) blocks from the primary tumor; and (5) avail- activity) among tumors expressing BRCA2-001/Short. Taken together, able FFPE blocks from at least one recurrence matching the primary our results suggest that BRCA1 mutation-associated tumors, as well as tumor. Matched recurrent tumor samples were identified using man- nonLOH BRCA2 mutation-associated tumors, could benefit from ual review of patient charts and pathology reports, supervised by KNM enhanced stability of BRCA2 mRNA. We therefore propose that BRCA2 and SMD. All clinical phenotypic data were collected through manual isoform switching represents a tumorigenic driver in BRCA1/2 chart review. Patients 1-27 met all study criteria and comprised the mutation-associated cancers. primary/recurrent tumor cohort. An additional four patients (Patients Although this cohort represents a relatively large set of paired 28–31) lacked an available primary tumor sample (meeting all but cri- primary and recurrent BRCA1/2 mutation-associated tumors, the terion 4) and were included only in groupwise RNA sequencing number of matched primary and recurrent tumors was still relatively analyses. small. As a result, we were limited in our ability to identify correlations We also identified 12 BRCA1/2 mutation carriers as a control group between genomic or transcriptomic features and treatment, which was for RNA sequencing analyses. Eligible patients met the following cri- quite heterogeneous. The cohort includes few post-PARPi recurrences, teria: (1) positive genetic test result for pathogenic germline mutation as PARPi use was relatively rare at the time of sample collection. We in BRCA1 or BRCA2 from a CLIA-approved laboratory; (2) no prior focused our efforts on identifying similarities and differences between history of cancer or cancer treatment; and (3) available FFPE blocks paired tumors across the cohort. Seven of the 27 patients had multiple from prophylactic mastectomy or salpingo-oophorectomy surgeries. recurrences. We accounted for this enrichment in our GSEA analyses, We extracted and sequenced RNA from normal breast and fallopian by counting a gene as mutated or gained/lost only once even if tube samples from these patients. Of note, fallopian tube samples were observed across common recurrences. We also limited pairwise sta- used as controls for ovarian cancer, for which the tissue of origin is the tistical comparisons to one recurrence per patient, selected at random. fallopian tube96. However, for MutSigCV and GISTIC analyses, all recurrences were included to improve our power to detect recurrence-specific events. Sample acquisition of primary tumor validation cohorts from Since we employed bulk sequencing techniques, we may not fully TCGA and Penn appreciate tumor heterogeneity associated with recurrent disease. We For validation of PARP1 copy number gains, we used WES from primary attempted to use the same formalin-fixed paraffin-embedded (FFPE) breast and ovarian tumors in TCGA cohorts (Supplementary Data 12). block and samples for multi-omic sequencing, but it was not always Level 1 WES binary alignment map (bam) files from TCGA Breast and possible due to sample availability. For the same reason, we could not Ovarian cohorts were obtained from Genomic Data Commons (https:// assess BRCA1 promoter methylation, another possible route to biallelic gdc.cancer.gov/)48,51. Tumors were excluded from analysis for any of BRCA1 loss in BRCA1 mutation-associated tumors. However, we did the following reasons: (1) failed quality control; (2) somatic BRCA1/2 confirm BRCA1 expression in all nonLOH BRCA1 mutation-associated loss as indicated by mutation, low expression, or BRCA1 promoter tumors with RNA-seq. Additionally, BRCA1 promoter methylation does methylation; or (3) germline or somatic mutations in genes related to not always abrogate BRCA1 expression93–95. Further validation of key homologous recombination (see Supplementary Data 12). TCGA results (LOH transitions, PARP1 amplifications, BRCA2 isoform switch- tumors were determined to be associated with BRCA1 or BRCA2 ing) in an independent cohort of paired primary and recurrent BRCA1/2 pathogenic variants if they had the following: (1) pathogenic BRCA1/2 mutation-associated breast and ovarian tumors will be important. variants (either known missense or loss of function, both known and Although we were able to determine that the BRCA2 isoform switching unknown); (2) germline variant allele frequency exceeding 30%; and (3) event was associated with decreased survival in two independent total allele depth exceeding 30 reads in both germline and tumor sample sets, additional cohorts are needed to confirm, and mechan- samples48,51. Tumors were grouped in gBRCA1/2 or Homologous istic studies remain to be done to confirm functionality. Recombination-Wild Type (HR-WT) groups, as visualized in Supple- Ultimately, these results suggest key biological features of mentary Fig. 6a. therapy-resistant recurrences, thereby highlighting therapeutic pos- For validation of BRCA2 isoform switching, we evaluated an sibilities for BRCA1/2 mutation carriers with cancer. We identified independent cohort of 42 primary breast and ovarian tumors from allele-specific LOH transitions in 25% of patients, underlining the patients with BRCA1/2 mutations, also consented under the University plasticity of BRCA1/2 status after treatment with DNA damaging agents. of Pennsylvania approved IRB protocol. Eligible patients had a diag- These results suggest that sequencing the most recent tumor or cir- nosis of breast or ovarian cancer with available FFPE tumor tissue and culating DNA will be most informative in planning personalized positive genetic test result for pathogenic germline mutation in BRCA1 treatment. Lastly, we found that BRCA2 isoform switching may be a or BRCA2 from a CLIA-approved laboratory. RNA extraction, sequen- pro-tumorigenic event that appears in both breast and ovarian recur- cing, and analyses were identical to those used for primary/recurrent rences and is associated with reduced breast cancer patient survival. cohort samples. Further studies are needed to assess whether these mechanisms are specific to BRCA1/2 mutation-associated breast and ovarian cancers, or Pathological review of FFPE specimens also observed in the context of other tumor types or other DNA repair FFPE tumors were collated and sectioned by the Tumor Tissue and defects. Biospecimen Bank at the University of Pennsylvania, then stained using Nature Communications | (2022)13:6728 13 Article https://doi.org/10.1038/s41467-022-34523-y hematoxylin and eosin. For tumor samples, staining was reviewed and performed using StringTie (v2.1.3b) to generate abundance files in marked by AN to identify sections with ≥70% invasive tumor. Normal Ballgown readable format100. We used Tximport (v1.16.1) to quantify tissue (RNA sequencing controls) was reviewed for no evidence of transcript abundance from t_data.ctab files, generating “lengthSca- tumor. DNA and RNA were extracted from FFPE sections or rolls of ledTPM” counts by gene101. Next, we used EdgeR (v3.30.3) to gen- 5–10 μm thickness. RNA extractions were performed within 1 week of erate, filter, and normalize counts per million (cpm)102. Briefly, gene- sectioning to minimize degradation. level transcripts per million were used to create three separate DGELists: one for all samples, one for ovarian tumor and normal Whole-exome and targeted sequencing of tumor and germ- fallopian tube samples, and one for breast tumor and normal breast line DNA samples. All DGELists were filtered to include only genes for which DNA was extracted from FFPE using standard laboratory depar- transcripts were detected in ≥2 or ≥3 samples, depending on the size affinization, proteinase K digestion, and ethanol precipitation. Germ- of the smallest biological group. Filtered DGELists were then nor- line DNA was extracted from whole blood or saliva. Germline DNA was malized using the trimmed mean of M values (TMM) method103. Fil- extracted from whole blood using sucrose-based lysis of erythrocytes, tered, normalized DGELists were used to generate filtered, followed by proteinase K digestion and ethanol precipitation from normalized cpm for each of the three comparisons. We assessed the leukocytes. Germline DNA was extracted from saliva samples using distribution of filtered, normalized cpm across the cohort to identify Oragene kits (DNA Genotek). any outlying samples (Supplementary Fig. 7a). Filtered, normalized All DNA was sheared for library preparation using a Covaris cpm were used for all cohort-level and tumor-specific RNA-seq ana- sonicator. Tumor DNA libraries were prepared using the NEBNext FFPE lyses, except for detection of isoform switching. Repair mix and NEBNext Ultra II DNA library prep kit (New England Biolabs), per manufacturer’s instructions. Germline DNA libraries were Identification, filtering, and analysis of somatic variants prepared using the NEBNext Ultra DNA library prep kit (New England We used a union of MuTect2 (v4.1.8.1), Strelka2 (v2.9.2), VarDictJava Biolabs), per manufacturer’s instructions. DNA libraries were pooled (v1.5.1), and VarScan2 (2.4.4) to call somatic variants from whole- and hybridized using SureSelect Target Enrichment System for Illu- exome and targeted sequencing of matched tumor and germline mina Multiplex Sequencing (Agilent) and associated protocols. For DNA104–107. VarDictJava and VarScan2 were used to call and filter WES, tumor and germline libraries were hybridized to SureSelect All germline variants104,107. Each patient’s germline variant calls were Exon v5, SureSelect All Exon v6+COSMIC, and SureSelect All Exon v7 checked to ensure detection of the same pathogenic BRCA1/2 variant captures (Agilent). For targeted sequencing, tumor and germline reported in their genetic test results. All variants were annotated with libraries were hybridized to a custom capture, which largely utilized ANNOVAR (October 2019 release)108. We limited our analysis of baits from the SureSelect All Exon platform (Agilent; see Supplemen- mutational signatures to variants called by MuTect2 and with alter- tary Table 1 for gene list). DNA samples, libraries and hybridization native allele depth of ≥10 reads. Mutational signatures were assessed pools were quantified using a Qubit (ThermoFisher) and fragment size separately for breast and ovarian tumor cohorts using deconstructSigs was determined using a BioAnalyzer 2100 (Agilent). WES was per- (v1.8)44,109. formed using an Illumina HiSeq 4000 and targeted sequencing was Apart from mutational signatures, all downstream analyses of performed using an Illumina NovaSeq 6000. All sequencing was per- whole-exome sequencing were restricted to exonic variants that were formed with 150 paired-end reads by the University of Pennsylvania called by and passed filtering for at least one caller and had alternative Next Generation Sequencing Core. allele read depth of ≥five reads. In order to exclude common SNPs from further analyses, we also removed somatic variants with fre- RNA sequencing of tumor and normal RNA quency in any population (PopFreqMax) ≥0.01108. Tumor mutational Tumor RNA was extracted from the same FFPE tumor blocks from burden (TMB) was calculated for each tumor using the following which DNA was extracted. Tumor and normal RNA samples were equation: TMB = ðn somatic,exonic nonsynonymous SNVs + indelsÞ110 size of WES captureðMbpÞ . Using the fil- extracted from FFPE with the RNeasy FFPE kit (Qiagen). To preserve tered variant set, we identified significantly mutated genes within RNA integrity, RNA samples were extracted within 1 week of sectioning primary and recurrent tumor groups using MutSigCV (v1.3)46. and stored at −80 °C. We used the TruSeq RNA Exome platform (Illu- Next, we identified gain of function (GoF) and loss of function mina) for library preparation and hybridization capture, per manu- (LoF) mutations in individual genes. LoF mutations were identified as facturer protocols. Sequencing was performed with 150 paired-end those that met one of the following criteria: (1) frameshift mutation, (2) reads using an Illumina HiSeq 4000 at the University of Pennsylvania nonsense mutation, or (3) nonsynonymous SNV predicted to be Next Generation Sequencing Core. pathogenic by REVEL score >0.5111. GoF mutations were defined as those that met all of the following criteria: (1) not a LOF mutation; (2) Bioinformatic analysis of DNA and RNA sequencing documentation in Catalog of Somatic Mutations in Cancer (COSMIC Fastq files from whole-exome and targeted sequencing were aligned to v84, https://cancer.sanger.ac.uk/cosmic); (3) occurring in a Tier 1 the hg19 build of the human genome using the Burrows-Wheeler oncogene as defined by the Cancer Gene Census (CGC, https://cancer. Aligner (BWA v.0.7.17-r1188)97. Various bam file processing operations sanger.ac.uk/census); and (4) matching the mutation type determined were performed using Samtools/htslib/bcftools v1.11. The resulting to be oncogenic for that gene by CGC. Supplementary Data 2 includes bam files were processed according to Genome Analysis Toolkit (GATK all LoF variants by sequencing type, regardless of alternative allele v3.7) best practices (picardtools v2.20.7)98. WES achieved a mean fraction. depth of 98× in tumor samples (median depth 87×) and 93× in germ- Pathway analysis was performed on LoF variants using Gene Set line samples (median depth 94×). Targeted sequencing achieved a Enrichment Analysis (GSEA) Preranked (v4.0.2 for Windows)45. Briefly, mean depth of 332× in tumor samples (median depth 304×) and 317× genes were ranked based on frequency of LoF mutations in primary in germline samples (median depth 316×). and recurrent tumor groups separately. A gene could be counted as Fastq files from RNA sequencing were aligned using STAR mutated only once per patient within each tumor group. This aligner (v2.7.2a) and gene annotations from the GENCODE Human approach was used to prevent skewing of results by patients with Release 19 reference assembly (https://www.gencodegenes.org/ multiple recurrences or by individual tumors accruing distinct LOF human/release_19.html)99. RNA sequencing achieved a mean depth mutations in the same gene. Ranked gene lists were used as input for of 67.6 million reads (median 55.6 million reads) for tumor and GSEA Preranked with Hallmark gene sets (1000 permutations). Gene normal samples. Alignment and quantification of transcripts was set enrichment was considered significant using a cutoff of FDR < 0.25. Nature Communications | (2022)13:6728 14 Article https://doi.org/10.1038/s41467-022-34523-y Genome-wide and gene-level analysis of copy number variation LOH. When possible, we confirmed proficient BRCA1/2 expression in We performed allele-specific copy number analysis using Sequenza nonLOH tumors with RNA sequencing. For patients with LOH transi- (v3.0), to identify segments of copy number variation (CNVs), as well as tions, we assessed the allele fraction of all BRCA1/2 germline variants ploidy and cellularity estimates47. Sequenza-derived segments were (SNPs and the pathogenic variant) across tumors to further interrogate used to calculate the components of homologous recombination LOH calls. deficiency (HRD): non-telomeric allelic imbalance (NtAI), large scale state transitions (LST), and genomic loss of heterozygosity (LOH). The Differential gene expression and gene fusion analysis same segments were used to calculate aneuploidy scores. Calculations Filtered, normalized log2cpm were used to assess sample relatedness. were performed using custom R-scripts (https://github.com/maxwell- We performed hierarchical clustering (maximum distance, average lab/HRDex)48,51. agglomeration) to generate dendrograms. Dendrograms and principal We used GISTIC2.0 (v2.0.23, Gene GISTIC mode, with default components analysis (PCA) plots were generated using the stats cutoffs) to identify significant CNVs at 90% confidence in tumor package in base R (v4.0.2). Between-group fold changes were com- cohorts49. Genes were highlighted from segments with residual puted as the difference in average log2cpm for a given gene between q ≤ 0.05. BEDtools (v2.29.2) was used to compute Jaccard statistics for groups. Using log2cpm fold changes as input, we ran GSEA with Hall- similarity between CNV sets, limited to segments of ≥50% reciprocal mark gene sets using clusterProfiler (v3.16.1) and msigdbr (v7.2.1). overlap112. BEDtools intersect was used to identify segments of 50% We next used the limma R package (v3.44.3) to model mean- overlap between significant (90% CI, FDR q < 0.05) GISTIC segments in variance separately for all-tumor, ovarian-only, and breast-only DGE- primary and recurrent tumor groups. Segments with ≥50% reciprocal Lists. Bayesian statistics were extracted from the resulting linear overlap were considered “shared” for the purposes of this analysis, models to generate adjusted p-values (Benjamini–Hochberg method) while those with <50% reciprocal overlap were considered “private” to and log2(fold-change) by gene for each comparison. Significantly one group. This subtractive analysis was performed separately for both altered genes were extracted using decideTests in limma (global amplifications and deletions, and Jaccard statistics were also com- method, adj. p < 0.05, |log2(fold-change)| >1). puted. Primary-private, recurrence-private, and shared (intersecting) Gene fusions were identified using STAR-Fusion (v1.7.0) and plots segments were annotated with RefSeq (GRCh37) genes using AnnotSV were generated using FusionInspector (v2.1.0)114. We limited our ana- (v2.3)113. Genes were considered to be deleted if >50% of the locus was lysis of gene fusions to those with ≥5 junction-spanning reads. encompassed by a deletion. Conversely, genes were only considered to be amplified if 100% of the locus was encompassed by an Assessment of differential transcript usage amplification. Abundance was re-quantified from bam files per recommendations in We also assessed genes within Sequenza-derived CNVs. Segments the Stringtie manual for novel isoform detection (http://ccb.jhu.edu/ were binned based on total (integer) copy number using the following software/stringtie/index.shtml?t=manual). Briefly, starting from initial convention: CN = 0, Deletion; CN = 1, Loss; CN = 2–3, Neutral; CN = 4–5, bam files, we ran Stringtie in merged mode (using GENCODE Human Gain; CN ≥ 6, Amplification. Binned segments were annotated with Release 19 as reference) to generate a merged gene transfer format RefSeq (GRCh37) genes using AnnotSV (v2.3) following the convention (gtf) file for the cohort. We quantified abundance from initial bams described above113. In the case of a gene spanning several different using Stringtie in Ballgown mode and the merged gtf file as reference. segments of CNV, the copy number of the gene was reported as the Differential transcript usage was assessed, and switch plots generated, minimal copy number of the entire locus. For pathway analysis of using IsoformSwitchAnalyzer (v1.10.0). The validation cohort of pri- genes subject to copy number gains, genes were ranked based on mary tumors was analyzed for isoform switching in an independent frequency of their presence in a segment of gain or amplification, use of this workflow. RIP-seq plots were generated using UCSC Gen- within primary and recurrent tumor groups. Similar to our GSEA Pre- ome Browser (hg19 assembly) with the ENCODE RNA-Binding Proteins ranked analysis of variants (see above), a gene could only be counted track corresponding to GEO Accession GSE35585. The track pictured as gained once per patient per tumor group. We then repeated this for ELAVL1/HuR in Supplementary Fig. 10a shows a peak with coordi- process to rank genes again based on frequency of copy number loss nates chr13:32,973,328-32,973,692 (average enrichment 20.30, FDR or deletion by tumor group. Ranked gene lists of “gained” genes and q = 4.2 × 10−3). The track pictured for PABPC1 in Supplementary Fig. 10a “lost” genes were used as input for GSEA Preranked with Hallmark gene shows a peak with coordinates chr13:32,973,450-32,973,693 (average sets (1000 permutations)45. Gene set enrichment was considered sig- enrichment 31.59, FDR q = 3.7 × 10−3). nificant using a cutoff of FDR < 0.25. Analysis of tissue microarray (TMA) using immunohistochem- Detection of BRCA1/2 biallelic loss istry and co-detection by indexing (CODEX) We first checked whether BRCA1/2 biallelic loss occurred via secondary The methods for TMA construction, immunohistochemistry of PARP1, somatic mutations as detected by whole-exome or targeted sequen- imaging and analysis of the CODEX, including cell segmentation and cing (sequencing metrics by BRCA1/2 exon for each method are com- marker quantification are detailed in the Supplementary Methods. piled in Supplementary Data 7). Specifically, we determined whether LoF mutations in BRCA1/2 were present at an alternative allele fraction Statistical analysis of ≥0.25. This cutoff reflected whether pathogenic secondary muta- All paired comparisons of continuous variables (HRD scores, aneu- tions were present in enough of the tumor to facilitate loss of the wild- ploidy scores, TMB) were tested for significance with two-sided Wil- type BRCA1/2 allele. Next, we assessed loss of the wild-type BRCA1 or coxon signed rank tests (α = 0.05). For patients with multiple BRCA2 allele via allele-specific copy number variation from Sequenza. recurrences, one recurrence was chosen at random for pairwise Briefly, we used tumor and germline bam files from WES for Sequenza comparisons (such that n = 27 for each group). In the primary/recur- input. Zygosity was estimated for all segments of somatic copy num- rent cohort, groupwise differences in average PARP1 copy number, ber variation based on the number of A (major) and B (minor) alleles PARP1 mRNA expression, PARP1 H-score, and RT-qPCR ΔΔCT were detected. Segments with zero B alleles (only A alleles present) were assessed by Kruskal–Wallis test, with any significant results further deemed to have undergone LOH. Conversely, if at least one B allele was assessed by Dunn’s test with Bonferroni correction (α = 0.05). We also detected, a segment was considered nonLOH (heterozygous). Tumors assessed the presence of a linear relationship between PARP1 expres- were deemed to have undergone loss of heterozygosity (LOH) if the sion and (continuous) PARP1 copy number by Spearman’s test segment containing their pathogenic BRCA1/2 germline mutation had (α = 0.05). Groupwise differences in CODEX datasets were assessed by Nature Communications | (2022)13:6728 15 Article https://doi.org/10.1038/s41467-022-34523-y two-sided Wilcoxon rank sum test (α = 0.05). In TCGA cohorts, and aneuploidy scores is available at https://github.com/maxwell- groupwise differences in average PARP1 copy number were assessed lab/HRDex. by two-sided t-tests (α = 0.05). PARP1 copy number is reported for primary/recurrent and TCGA tumors in Supplementary Data 11. References The adjusted p-value for total BRCA2 expression was generated by 1. Slavin, T. P. et al. The contribution of pathogenic variants in breast linear modeling with limma (see above). The adjusted p-values for cancer susceptibility genes to familial breast cancer risk. Breast BRCA2 isoform usage were computed using DEXSeq within iso- Cancer 3, 22 (2017). formSwitchAnalyzer. Association between sample type and BRCA2 2. Pennington, K. P. & Swisher, E. M. Hereditary ovarian cancer: isoform usage was determined by Mantel-Haenszel chi-squared test beyond the usual suspects. Gynecol. Oncol. 124, 347–353 (2012). with continuity correction (α = 0.05, 1df). 3. Roy, R., Chun, J. & Powell, S. N. BRCA1 and BRCA2: different roles Overall survival was defined as time from diagnosis to death or in a common pathway of genome protection. Nat. Rev. Cancer 12, last follow-up. Survival metrics were calculated based on manual chart 68–78 (2012). review of the electronic health record for each patient. The Penn 4. Lord, C. J. & Ashworth, A. BRCAness revisited. Nat. Rev. Cancer 16, Medicine electronic health record is indexed routinely against the 110–120 (2016). National Death Index, with a time delay of ~1 year. The survival analysis 5. Jonsson, P. et al. Tumour lineage shapes BRCA-mediated pheno- was limited to 67 patients for whom we collected survival data and types. Nature 571, 576–583 (2019). RNA-seq from at least one tumor (see Supplementary Data 15). For the 6. Matulonis, U. A. et al. Ovarian cancer. Nat. Rev. Dis. Prim. 2, 33 ovarian cancer patients, we first used a Cox proportional hazards 889–908 (2016). model to control for patient age at diagnosis, tumor stage at diagnosis, 7. Reddy, S. M. et al. Long-term survival outcomes of triple-receptor BRCA1 vs. BRCA2 mutation, and disease recurrent status as con- negative breast cancer survivors who are disease free at 5 years founding variables. We used the same confounding variables, in and relationship with low hormone receptor positivity. Br. J. Can- addition to ER status, in multivariate Cox proportional hazards analysis cer 118, 17–23 (2017). of the 34 breast cancer patients. In ovarian cancer patients, the only 8. Tutt, A. et al. Carboplatin in BRCA1/2-mutated and triple-negative confounding variable associated with survival was disease recurrent breast cancer BRCAness subgroups: The TNT Trial. Nat. Med. 24, status (a logical variable indicating whether the patient’s disease 628–637 (2018). eventually recurred). Therefore, we simplified our ovarian Cox pro- 9. Tung, N. M. & Garber, J. E. BRCA1/2 testing: therapeutic implica- portional hazards model to include recurrent status as the only con- tions for breast cancer management. Br. J. Cancer 119, founding variable. In breast cancer patients, none of the confounding 141–152 (2018). variables were significantly associated with survival, so these were 10. Tutt, A. N. J. et al. Adjuvant olaparib for patients with BRCA1- or discarded in favor of a univariate analysis. Cox proportional hazards BRCA2-mutated breast cancer. N. Engl. J. Med. 384, analyses were performed and plotted using the Survival (v3.2.7) and 2394–2405 (2021). Survminer (v0.4.8) R packages. p-values reported on the plots are 11. Lord, C. J. & Ashworth, A. Mechanisms of resistance to therapies those corresponding to the isoform expression term itself. p-values for targeting BRCA-mutant cancers. Nat. Med. 1381–1388 https://doi. the models were tested for significance by Wald test (α = 0.05, 1 df for org/10.1038/nm.3369 (2013). breast, 2 df for ovarian). 12. Konstantinopoulos, P. A., Ceccaldi, R., Shapiro, G. I. & D’Andrea, A. D. Homologous recombination deficiency: exploiting the funda- Reporting summary mental vulnerability of ovarian cancer. Cancer Discov. 5, Further information on research design is available in the Nature 1137–1154 (2015). Portfolio Reporting Summary linked to this article. 13. Xu, G. et al. REV7 counteracts DNA double-strand break resection and affects PARP inhibition. Nature 521, 541–544 (2015). Data availability 14. Guillemette, S. et al. Resistance to therapy in BRCA2 mutant cells The raw capture-targeted, WES and RNA sequencing data generated due to loss of the nucleosome remodeling factor CHD4. Genes as part of this study have been deposited in the NCBI SRA database Dev. 29, 489–494 (2015). under accession code PRJNA751555. The raw germline WES data are 15. Pettit, S. J. et al. A genetic screen using the piggybac transposon in protected due to lack of patient consent to deposit in a public haploid cells identifies Parp1 as A Mediator of Olaparib toxicity. repository. The germline WES will be made available upon request PLoS ONE 8, e61520 (2013). from the corresponding author and will made available under a Data 16. Bunting, S. F. et al. 53BP1 inhibits homologous recombination in Transfer Agreement (DTA) and transferred via FTP when the DTA is brca1-deficient cells by blocking resection of DNA breaks. Cell 141, complete. Most data transfers will be completed within a month’s 243–254 (2010). time. Access to the raw imaging data also is available upon request 17. Patch, A. M. et al. Whole-genome characterization of chemore- from the corresponding author. TCGA data are available under sistant ovarian cancer. Nature 521, 489–494 (2015). controlled-use conditions; data use limitations and the instructions 18. Johnson, N. et al. Stabilization of mutant BRCA1 protein confers for applying for access are available through dbGaP. The raw TGCA PARP inhibitor and platinum resistance. Proc. Natl Acad. Sci. USA data are accessible via the Genomic Data Commons from the 110, 17041–17046 (2013). National Center Institute (breast cancer patients from TCGA-BRCA, 19. Jaspers, J. E. et al. Loss of 53BP1 causes PARP inhibitor resistance https://portal.gdc.cancer.gov/projects/TCGA-BRCA and ovarian in BRCA1-mutated mouse mammary tumors. Cancer Discov. 3, cancer patients from TCGA-OV, https://portal.gdc.cancer.gov/ 68–81 (2013). projects/TCGA-OV) once access is made available through dbGAP 20. Goodall, J. et al. Circulating cell-free DNA to guide prostate cancer application. All data needed to evaluate the conclusions in the paper treatment with PARP inhibition. Cancer Discov. 7, 1006–1017 (2017). are present in the paper and/or the Supplementary Information. 21. Quigley, D. et al. Analysis of circulating cell-free DnA identifies Source Data are provided with this paper. multiclonal heterogeneity of BRCA2 reversion mutations asso- ciated with resistance to PARP inhibitors. Cancer Discov. 7, Code availability 999–1005 (2017). All code used for the study is available at https://github.com/nathanson- 22. Cheng, H. H., Salipante, S. J., Nelson, P. S., Montgomery, B. & lab/BRCA-PrimaryRecurrent115. Additional code for calculating HRD Pritchard, C. C. Polyclonal BRCA2 reversion mutations detected in Nature Communications | (2022)13:6728 16 Article https://doi.org/10.1038/s41467-022-34523-y circulating tumor DNA after platinum chemotherapy in a patient 42. Carpten, J. D. et al. A transforming mutation in the pleckstrin with metastatic prostate cancer. JCO Precis. Oncol. 2, homology domain of AKT1 in cancer. Nature 448, 439–445 (2007). PO.17.00169 (2018). 43. Chen, Y. et al. Effect of AKT1 (p. E17K) Hotspot mutation on 23. Swisher, E. M. et al. Secondary BRCA1 mutations in BRCA1- malignant tumorigenesis and prognosis. Front. Cell Develop- mutated ovarian carcinomas with platinum resistance. Cancer mental Biol. 8, 1–12 (2020). Res. 68, 2581–2586 (2008). 44. Alexandrov, L. B. et al. Signatures of mutational processes in 24. Norquist, B. et al. Secondary somatic mutations restoring BRCA1/2 human cancer. Nature 500, 415–421 (2013). predict chemotherapy resistance in hereditary ovarian carcino- 45. Subramanian, A. et al. Gene set enrichment analysis: a knowledge- mas. J. Clin. Oncol. 29, 3008–3015 (2011). based approach for interpreting genome-wide expression pro- 25. Banda, K., Swisher, E. M., Wu, D., Pritchard, C. C. & Gadi, V. K. files. Proc. Natl Acad. Sci. USA 102, 15545–50 (2005). Somatic reversion of germline BRCA2 mutation confers resistance 46. Lawrence, M. S. et al. Mutational heterogeneity in cancer and the to poly(ADP-ribose) polymerase inhibitor therapy. JCO Precis. search for new cancer-associated genes. Nature 499, 214–218 Oncol. 1–6 https://doi.org/10.1200/PO.17.00044 (2018). (2013). 26. Barber, L. J. et al. Secondary mutations in BRCA2 associated with 47. Favero, F. et al. Sequenza: allele-specific copy number and clinical resistance to a PARP inhibitor. J. Pathol. 229, mutation profiles from tumor sequencing data. Ann. Oncol. 26, 422–429 (2013). 64–70 (2014). 27. Brady, S. W. et al. Combating subclonal evolution of resistant 48. Maxwell, K. N. et al. BRCA locus-specific loss of heterozygosity cancer phenotypes. Nat. Commun. 8, 1231(2017). in germline BRCA1 and BRCA2 carriers. Nat. Commun. 8, 28. Meric-Bernstam, F. et al. Concordance of genomic alterations 1–11 (2017). between primary and recurrent breast cancer. Mol. Cancer Ther. 49. Mermel, C. H. et al. GISTIC2.0 facilitates sensitive and confident 13, 1382–1389 (2014). localization of the targets of focal somatic copy-number alteration 29. Patel, J. N. et al Characterisation of homologous recombination in human cancers. Genome Biol. 12, R41 (2011). deficiency in paired primary and recurrent high grade serous 50. Yasui, K. et al. Alteration in copy numbers of genes as a ovarian cancer. Br. J. Cancer 1–7 https://doi.org/10.1038/s41416- mechanism for acquired drug resistance. Cancer Res. 64, 018-0268-6 (2018). 1403–1410 (2004). 30. Paul, M. R. et al. Genomic landscape of metastatic breast cancer 51. Kraya, A. A. et al. Genomic signatures predict the immunogenicity identifies preferentially dysregulated pathways and targets. J. of BRCA-deficient breast cancer. Clin. Cancer Res. https://doi.org/ Clin. Invest. 130, 4252–4265 (2020). 10.1158/1078-0432.CCR-18-0468 (2019). 31. Sokolenko, A. P. et al. Molecular profiles of BRCA1-associated 52. Santilli, G. et al. PARP co-activates B-MYB through enhanced ovarian cancer treated by platinum-based therapy: analysis of phosphorylation at cyclin/cdk2 sites. Oncogene 20, primary, residual and relapsed tumors. Int. J. Cancer 146, 8167–8174 (2001). 1879–1888 (2020). 53. Bieche, I. et al. Variations in the mRNA expression of poly(ADP- 32. LoRusso, P. et al. Trial in progress: a phase II open-label, randomized ribose) polymerases, poly(ADP-ribose) glycohydrolase and ADP- study of PARP inhibition (olaparib) either alone or in combination ribosylhydrolase 3 in breast tumors and impact on clinical out- with anti-PD-L1 therapy (atezolizumab) in homologous DNA repair come. Int. J. Cancer 133, 2791–2800 (2013). (HDR) deficient, locally advanced or metastatic non-HER2-positive 54. Eismann, J. et al. Transcriptome analysis reveals overlap in fusion breast cancer. J. Clin. Oncol. 38, TPS1102–TPS1102 (2020). genes in a phase I clinical cohort of TNBC and HGSOC patients 33. Zhang, J., Shih, D. J. H. & Lin, S. Y. Role of DNA repair defects in treated with buparlisib and olaparib. J. Cancer Res. Clin. Oncol. predicting immunotherapy response. Biomark. Res. 8, 1–8 (2020). 146, 503–514 (2020). 34. Zhang, G. et al. Combinatorial therapy of immune checkpoint and 55. Li, Z.-X. et al. MALAT1: a potential biomarker in cancer. Cancer cancer pathways provides a novel perspective on ovarian cancer Manag. Res. 6757–6768 https://doi.org/10.2147/CMAR. treatment (Review). Oncol. Lett. 17, 2583–2591 (2019). S169406 (2018). 35. Matulonis, U. A. et al. Antitumor activity and safety of pem- 56. Vitting-Seerup, K. & Sandelin, A. The landscape of isoform brolizumab in patients with advanced recurrent ovarian cancer: switches in human cancers. Mol. Cancer Res. 15, 1206–1220 (2017). results from the phase II KEYNOTE-100 study. Ann. Oncol. 30, 57. Vitting-Seerup, K., Porse, B. T., Sandelin, A. & Waage, J. SpliceR: An 1080–1087 (2019). R package for classification of alternative splicing and prediction 36. Disis, M. L. et al. Efficacy and safety of avelumab for patients with of coding potential from RNA-seq data. BMC Bioinforma. 15, recurrent or refractory ovarian cancer phase 1b: results from the 81 (2014). JAVELIN solid tumor trial. JAMA Oncol. 5, 393–401 (2019). 58. Mukherjee, N. et al. Integrative regulatory mapping indicates that 37. Domchek, S. M. et al. Olaparib and durvalumab in patients with the RNA-binding protein HuR couples pre-mRNA processing and germline BRCA-mutated metastatic breast cancer (MEDIOLA): an mRNA stability. Mol. Cell 43, 327–339 (2011). open-label, multicentre, phase 1/2, basket study. Lancet Oncol. 21, 59. Fatscher, T., Boehm, V., Weiche, B. & Gehring, N. H. The interaction 1155–1164 (2020). of cytoplasmic poly(A)-binding protein with eukaryotic initiation 38. Hamanishi, J. et al. Safety and antitumor activity of Anti-PD-1 factor 4G suppresses nonsense-mediated mRNA decay. RNA 20, antibody, nivolumab, in patients with platinum-resistant ovarian 1579–1592 (2014). cancer. J. Clin. Oncol. 33, 4015–4022 (2015). 60. Behm-Ansmant, I., Gatfield, D., Rehwinkel, J., Hilgers, V. & Izaur- 39. Liu, Y. L. et al. BRCA mutations, homologous DNA repair defi- ralde, E. A conserved role for cytoplasmic poly(A)-binding protein ciency, tumor mutational burden, and response to immune 1 (PABPC1) in nonsense-mediated mRNA decay. EMBO J. 26, checkpoint inhibition in recurrent ovarian cancer. JCO Precis. 1591–1601 (2007). Oncol. 4, 665–679 (2020). 61. Feng, Z. et al. Rad52 inactivation is synthetically lethal with BRCA2 40. Goltsev, Y. et al. Deep profiling of mouse splenic architecture with deficiency. Proc. Natl Acad. Sci. USA 108, 686–691 (2010). CODEX multiplexed imaging. Cell 174, 968–981.e15 (2018). 62. Hodgson, D. et al. Analysis of mutation status and homologous 41. Du, Z. et al. Qualifying antibodies for image-based immune pro- recombination deficiency in tumors of patients with germline filing and multiplexed tissue imaging. Nat. Protoc. 14, BRCA1 or BRCA2 mutations and metastatic breast cancer: Olym- 2900–2930 (2019). piAD☆. Ann. Oncol. 32, 1582–1589 (2021). Nature Communications | (2022)13:6728 17 Article https://doi.org/10.1038/s41467-022-34523-y 63. Sokolenko, A. P. et al. Rapid selection of BRCA1-proficient tumor 83. Faraoni, I. et al. BRCA1, PARP1 and γH2AX in acute myeloid leu- cells during neoadjuvant therapy for ovarian cancer in BRCA1 kemia: Role as biomarkers of response to the PARP inhibitor ola- mutation carriers. Cancer Lett. 397, 127–132 (2017). parib. Biochim. Biophys. Acta 1852, 462–472 (2015). 64. Souzaki, M. et al. Hedgehog signaling pathway mediates the 84. Murai, J. et al. Trapping of PARP1 and PARP2 by clinical PARP progression of non-invasive breast cancer to invasive breast inhibitors. Cancer Res. 72, 5588–5599 (2012). cancer. Cancer Sci. 102, 373–381 (2011). 85. Pommier, Y., O’Connor, M. J. & De Bono, J. Laying a trap to kill 65. Liao, X. et al. Aberrant activation of hedgehog signaling pathway cancer cells: PARP inhibitors and their mechanisms of action. Sci. in ovarian cancers: effect on prognosis, cell invasion and differ- Transl. Med. 8, 1–8 (2016). entiation. Carcinogenesis 30, 131–140 (2009). 86. US Food and Drug Administration. FDA approves niraparib for first- 66. Sato, K. & Akimoto, K. Expression levels of KMT2C and SLC20A1 line maintenance of advanced ovarian cancer. (US Food and Drug identified by information-theoretical analysis are powerful prog- Administration, 2020). Accessed: 26th May 2021. Available at: nostic biomarkers in estrogen receptor-positive breast cancer. https://www.fda.gov/drugs/drug-approvals-and-databases/fda- Clin. Breast Cancer 17, e135–e142 (2017). approves-niraparib-first-line-maintenance-advanced-ovarian- 67. Gala, K. et al. KMT2C mediates the estrogen dependence of breast cancer cancer through regulation of ERα enhancer function. Oncogene 87. US Food and Drug Administration. FDA approves olaparib for 37, 4692–4710 (2018). germline BRCA-mutated metastatic breast cancer. (US Food and 68. Liu, X. et al. KMT2C is a potential biomarker of prognosis and Drug Administration, 2018). Accessed: 26th May 2021. Available chemotherapy sensitivity in breast cancer. Breast Cancer Res. at: https://www.fda.gov/drugs/resources-information-approved- Treat. 189, 347–361 (2021). drugs/fda-approves-olaparib-germline-brca-mutated-metastatic- 69. Na, F. et al. KMT2C deficiency promotes small cell lung cancer breast-cancer. metastasis through DNMT3A-mediated epigenetic reprogram- 88. US Food and Drug Administration. FDA approves rucaparib for ming. Nat. Cancer https://doi.org/10.1038/S43018-022-00361- maintenance treatment of recurrent ovarian, fallopian tube, or 6 (2022). primary peritoneal cancer. (US Food and Drug Administration, 70. Wang, R. et al. Multiplex profiling of peritoneal metastases from 2018). Accessed: 26th May 2021. Available at: https://www.fda. gastric adenocarcinoma identified novel targets and molecular gov/drugs/resources-information-approved-drugs/fda-approves- subtypes that predict treatment response. Gut 69, 18–31 (2020). rucaparib-maintenance-treatment-recurrent-ovarian-fallopian- 71. Li, D. et al. Mitotic arrest deficient-like 1 is correlated with poor tube-or-primary-peritoneal. prognosis in small-cell lung cancer after surgical resection. 89. US Food and Drug Administration. FDA approves talazoparib for Tumour Biol. 37, 4393–4398 (2016). gBRCAm HER2-negative locally advanced or metastatic breast 72. Oh, S., Song, Y. H., Yim, J. & Kim, T. K. Identification of Mad as a cancer. (US Food and Drug Administration, 2018). Accessed: 26th repressor of the human telomerase (hTERT) gene. Oncogene 19, May 2021. Available at: https://www.fda.gov/drugs/drug- 1485–1490 (2000). approvals-and-databases/fda-approves-talazoparib-gbrcam- 73. Kienitz, A., Vogel, C., Morales, I., Müller, R. & Bastians, H. Partial her2-negative-locally-advanced-or-metastatic-breast-cancer. downregulation of MAD1 causes spindle checkpoint inactivation 90. Sakai, W. et al. Secondary mutations as a mechanism of cisplatin and aneuploidy, but does not confer resistance towards taxol. resistance in BRCA2-mutated cancers. Nature 451, 1116–1120 (2008). Oncogene 24, 4301–4310 (2005). 91. Sakai, W. et al. Functional restoration of BRCA2 protein by sec- 74. Sun, Q. et al. Increased expression of mitotic arrest deficient-like 1 ondary BRCA2 mutations in BRCA2-mutated ovarian carcinoma. (MAD1L1) is associated with poor prognosis and insensitive to Cancer Res. 69, 6381–6386 (2009). Taxol treatment in breast cancer. Breast Cancer Res. Treat. 140, 92. Park, P. H. et al. Amplification of the mutation-carrying BRCA2 323–330 (2013). allele promotes RAD51 loading and PARP inhibitor resistance in the 75. Hoadley, K. A. et al. Cell-of-origin patterns dominate the molecular absence of reversion mutations. Mol. Cancer Ther. 19, classification of 10,000 tumors from 33 types of cancer. Cell 173, 602–613 (2020). 291–304.e6 (2018). 93. Wang, C. et al. Expression of BRCA1 protein in benign, borderline, 76. Gonçalves, A. et al. Poly(ADP-ribose) polymerase-1 mRNA and malignant epithelial ovarian neoplasms and its relationship to expression in human breast cancer: a meta-analysis. Breast Can- methylation and allelic loss of the BRCA1 gene. J. Pathol. 202, cer Res. Treat. 127, 273–281 (2010). 215–223 (2004). 77. Ossovskaya, V., Koo, I. C., Kaldjian, E. P., Alvares, C. & Sherman, B. 94. Shilpa, V. et al. BRCA1 promoter hypermethylation and protein M. Upregulation of poly (ADP-Ribose) polymerase-1 (PARP1) in expression in ovarian carcinoma - An Indian study. Tumor Biol. 35, triple-negative breast cancer and other primary human tumor 4277–4284 (2014). types. Genes Cancer 1, 812–821 (2010). 95. Garg, K. et al. BRCA1 immunohistochemistry in a molecularly 78. Bi, F. F., Li, D. & Yang, Q. Hypomethylation of ETS transcription characterized cohort of ovarian high-grade serous carcinomas. factor binding sites and upregulation of PARP1 expression in Am. J. Surg. Pathol. 37, 138–146 (2013). endometrial cancer. Biomed. Res. Int. 2013, 1–5 (2013). 96. Labidi-Galy, S. I. et al. High grade serous ovarian carcinomas ori- 79. Lheureux, S. et al. Somatic BRCA1/2 recovery as a resistance ginate in the fallopian tube. Nat. Commun. 8, 1–10 (2017). mechanism after exceptional response to poly (ADP-ribose) 97. Li, H. & Durbin, R. Fast and accurate long-read alignment with polymerase inhibition. J. Clin. Oncol. 35, 1240–1249 (2017). Burrows-Wheeler transform. Bioinformatics 26, 589–595 (2010). 80. Rojo, F. et al. Nuclear PARP-1 protein overexpression is associated 98. Van der Auwera, G. A. et al. From fastQ data to high-confidence with poor overall survival in early breast cancer. Ann. Oncol. 23, variant calls: The genome analysis toolkit best practices pipeline. 1156–1164 (2012). Curr. Protoc. Bioinforma. 43, 11.10.1–11.10.33 (2013). 81. Pashaiefar, H. et al. PARP-1 overexpression as an independent 99. Dobin, A. et al. STAR: ultrafast universal RNA-seq aligner. Bioin- prognostic factor in adult non-m3 acute myeloid leukemia. Genet. formatics 29, 15–21 (2013). Test. Mol. Biomark. 22, 343–349 (2018). 100. Pertea, M., Kim, D., Pertea, G. M., Leek, J. T. & Salzberg, S. L. 82. Gilabert, M. et al. Poly(ADP-ribose) polymerase 1 (PARP1) over- Transcript-level expression analysis of RNA-seq experiments expression in human breast cancer stem cells and resistance to with HISAT, StringTie and Ballgown. Nat. Protoc. 11, olaparib. PLoS ONE 9, e104302 (2014). 1650–1667 (2016). Nature Communications | (2022)13:6728 18 Article https://doi.org/10.1038/s41467-022-34523-y 101. Soneson, C., Love, M. I. & Robinson, M. D. Differential analyses for K.L.N.), the Gray Foundation Team Science Award (K.L.N.), the V Foun- RNA-seq: Transcript-level estimates improve gene-level infer- dation for Cancer Research BRCA1/2 Convergence Team Award (K.L.N.), ences. F1000Research 4, 1521 (2016). the Breast Cancer Research Foundation (S.M.D., K.L.N.) T32 HG009495 102. Robinson, M. D., Mccarthy, D. J. & Smyth, G. K. edgeR: a Bio- (A.K.), Miguel Servet Program (ISCIII) [CPII19/00033] (V.S.) and Asocia- conductor package for differential expression analysis of digital ción Española Contra el Cáncer (A.L.G.). gene expression data. Bioinformatics 26, 139–140 (2010). 103. Robinson, M. D. & Oshlack, A. A scaling normalization method for Author contributions differential expression analysis of RNA-seq data. Genome Biol. 11, J.B.S., K.N.M., S.M.D., and K.L.N conceived the study. J.B.S. performed R25 (2010). the majority of DNA/RNA sequencing, bioinformatic analyses, and figure 104. Lai, Z. et al. VarDict: a novel and versatile variant caller for next- preparation (with exceptions below). J.B.S. and K.L.N. primarily wrote the generation sequencing in cancer research. Nucleic Acids Res. 44, manuscript. D.P. performed all CODEX staining and imaging, and also 1–11 (2016). prepared associated figures. B.W. performed sequence alignment, var- 105. Kim, S. et al. Strelka2: fast and accurate calling of germline and iant calling, and Sequenza analysis. J.P. assisted with calculation of HRD somatic variants. Nat. Methods 15, 591–594 (2018). and aneuploidy scores and advised on numerous statistical analyses. 106. Cibulskis, K. et al. Sensitive detection of somatic point mutations K.D’A. performed DNA extraction and whole-exome sequencing of in impure and heterogeneous cancer samples. Nat. Biotechnol. 31, germline specimens. A.H. and C.R assisted with clinical data curation. 213–219 (2013). A.A.K. contributed to sample collection and protocols for RNA sequen- 107. Koboldt, D. C. et al. VarScan 2: somatic mutation and copy number cing. K.N.M. optimized sequencing protocols and initiated tumor col- alteration discovery in cancer by exome sequencing. Genome lection. J.S.S. performed DNA and RNA extractions and library Res. 22, 568–576 (2012). preparations. M.F. performed rt-PCR, which was designed by P.M. W.Z., 108. Wang, K., Li, M. & Hakonarson, H. ANNOVAR: functional annota- A.L.G., V.S., J.B., K.N.M., S.M.D., and K.L.N. all assisted in identifying tion of genetic variants from high-throughput sequencing data. patients and specimens for the cohort. M.F. supervised collection and Nucleic Acids Res. 38, 1–7 (2010). sectioning of FFPE specimens. A.N. performed pathological review of 109. Rosenthal, R., McGranahan, N., Herrero, J., Taylor, B. S. & Swanton, FFPE specimens, including IHC grading. S.M.D. offered clinical exper- C. deconstructSigs: delineating mutational processes in single tize. All authors contributing to editing the manuscript. tumors distinguishes DNA repair deficiencies and patterns of carcinoma evolution. Genome Biol. 17, 31 (2016). Competing interests 110. Wu, H.-X. et al. Tumor mutational and indel burden: a systematic S.M.D. has received honoraria from AstraZeneca. J.B. has received pan-cancer evaluation as prognostic biomarkers. Ann. Transl. honoraria from AstraZeneca and Pfizer. The remaining authors declare Med. 7, 640–640 (2019). no competing interests. 111. Ioannidis, N. M. et al. REVEL: an ensemble method for predicting the pathogenicity of rare missense variants. Am. J. Hum. Genet. Additional information 99, 877–885 (2016). Supplementary information The online version contains 112. Quinlan, A. R. & Hall, I. M. BEDTools: a flexible suite of utilities for supplementary material available at comparing genomic features. Bioinformatics 26, 841–842 (2010). https://doi.org/10.1038/s41467-022-34523-y. 113. Ronique Geoffroy, V. et al. AnnotSV: an integrated tool for struc- tural variations annotation. Bioinformatics 34, 3572–3574 (2018). Correspondence and requests for materials should be addressed to 114. Haas, B. J. et al. Accuracy assessment of fusion transcript detec- Katherine L. Nathanson. tion via read-mapping and de novo fusion transcript assembly- based methods. Genome Biol. 20, 213 (2019). Peer review information Nature Communications thanks the other 115. Shah, J. & Wubbenhorst, B. Analysis of matched primary and anonymous reviewer(s) for their contribution to the peer review of recurrent BRCA1/2 mutation-associated tumors identifies this work. recurrence-specific drivers. BRCA-PrimaryRecurrent. https://doi. org/10.5281/zenodo.7076481 (2022). Reprints and permissions information is available at http://www.nature.com/reprints Acknowledgements The authors thank the patients and their families for participating in this Publisher’s note Springer Nature remains neutral with regard to jur- study. We would also like to thank Lauren Schmucker, Caitlin Feltcher, isdictional claims in published maps and institutional affiliations. Amy Ziober, Danielle Sharpe, Li-Ping Wang, Dr. Sean Carlin, and Heather Symecko for help with acquisition and staining of pathological speci- Open Access This article is licensed under a Creative Commons mens. We would like to thank Liza M. Dorfman for her assistance with Attribution 4.0 International License, which permits use, sharing, whole-exome sequencing and for her well-written protocols. We would adaptation, distribution and reproduction in any medium or format, as like to thank Christina Cruz for her efforts transferring sequencing data long as you give appropriate credit to the original author(s) and the from Vall D’Hebron to the University of Pennsylvania. This work was source, provide a link to the Creative Commons license, and indicate if supported by the Tumor Tissue and Biospecimen Bank and the Next changes were made. The images or other third party material in this Generation Sequencing Core at the University of Pennsylvania (and in article are included in the article’s Creative Commons license, unless particular, we appreciated the help of Dr. Jonathan Schug). We would indicated otherwise in a credit line to the material. If material is not also like to thank Akoya Biosciences for their technical expertize included in the article’s Creative Commons license and your intended regarding CODEX analyses. Lastly, the RNA sequencing pipeline used use is not permitted by statutory regulation or exceeds the permitted for this study was based on content from Dr. Dan Beiting’s DIY Tran- use, you will need to obtain permission directly from the copyright scriptomics class at the University of Pennsylvania (Spring 2021), for holder. To view a copy of this license, visit http://creativecommons.org/ which we thank him as well. Figure 6 and Supplementary Fig. 15 were licenses/by/4.0/. created using a licensed version of BioRender. This work was supported by the Basser Center for BRCA at the University of Pennsylvania (S.M.D., © The Author(s) 2022 Nature Communications | (2022)13:6728 19