{"paper_id":"49c46501-18e5-49f4-a158-72ded34653de","body_text":"1 \n \nIntegrated immunogenomic analyses of high -grade serous ovarian \ncancer reveal vulnerability to combination immunotherapy \nRaphael Gronauer 1, Leonie Madersbacher 1, Pablo Monfort -Lanzas1,2, Gabriel Floriani 1, \nSusanne Sprung4, Alain Gustave Zeimet3, Christian Marth3, Heidelinde Fiegl3, Hubert Hackl1,*  \n1Institute of Bioinformatics, Biocenter, Medical University of Innsbruck, Austria \n2Institute of Medical Biochemistry, Biocenter, Medical University of Innsbruck, Austria \n3Department of Obstetrics and Gynecology, Medical University of Innsbruck, Austria \n4Institute of Pathology, Innpath GmbH, Innsbruck, Austria \n*Correspondence: Hubert Hackl, PhD \nInstitute of Bioinformatics, Biocenter, Medical University of Innsbruck \nInnrain 80, 6020 Innsbruck, Austria \nEmail: hubert.hackl@i-med.ac.at  \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \nNOTE: This preprint reports new research that has not been certified by peer review and should not be used to guide clinical practice.\n\n2 \n \nAbstract \nBackground: High-grade serous ovarian cancer (HGSOC) remains the most lethal \ngynecologic malignancy  despite new therapeutic concepts, including poly -ADP-ribose \npolymerase inhibitors (PARPis) and antiangiogenic therapy. The efficacy of immunotherapies \nis modest, but clinical trials investigating the potential of combination immunotherapy with \nPARPis are underway. Homologous recombination repair deficiency (HRD) or BRCAness and \nthe composition of the tumor microenvironment appear to play a critical role in determining the \ntherapeutic response. \nMethods: We conducted comprehensive immunogenomic analyses of H GSOC using data \nfrom several patient cohorts, including a new cohort from the Medical University of Innsbruck \n(MUI). Machine learning methods were used to develop a classification model for BRCAness \nfrom gene expression data. Integrated analysis of bulk and single-cell RNA sequencing data \nwas used to delineate the tumor immune microenvironment and was validated by \nimmunohistochemistry. The impact of PARPi and BRCA1 mutations on the activation of \nimmune-related pathways was studied in vitro  using ovarian can cer cell lines, RNA \nsequencing, and immunofluorescence analysis. \nResults: We identified a predictive 24 -gene signature to determine BRCAness. \nComprehensive analysis of the tumor microenvironment allowed us to identify patient samples \nwith BRCAness and high  immune infiltration. Further characterization of these samples \nrevealed increased infiltration of immunosuppressive cells, including tumor -associated \nmacrophages (TAMs) expressing TREM2, C1QA, and LILRB4, as identified by further analysis \nof single-cell RNA sequencing data and gene expression analysis of samples from patients \nreceiving combination therapy with PARPi and anti -PD-1. PARPi activated the cGAS -STING \nsignaling pathway and the downstream innate immune response in a similar manner to \nHGSOC patient s with BRCAness status. We have developed a web application \n(https://ovrseq.icbi.at) and an associated R package OvRSeq, which allow for comprehensive \ncharacterization of ovarian cancer patient samples and assessment of a vulnerability score \nthat enables s tratification of patients to predict response to the mentioned combination \nimmunotherapy. \nConclusions: Genomic instability in HGSOC affects the tumor immune environment , and \nTAMs play a crucial role in modulating the immune response. Based on various datasets, we \nhave developed a diagnostic application that uses RNA sequencing data not only to \ncomprehensively characterize HGSOC but also to predict vulnerability and response to \ncombination immunotherapy. \nKeywords: High-grade serous ovarian cancer, BRCAness, PARP inhibitor, immunotherapy, \nvulnerability, RNA sequencing, tumor -associated macrophages, tumor immune \nmicroenvironment \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \n\n3 \n \nBackground \nDespite newer therapeutic concepts, ovarian cancer, particularly  high-grade serous ovarian \ncancer, is still the deadliest gynecologic malignancy, with 13,270 expected deaths in 2023 in \nthe U.S. [1]. While immunotherapy, such as immune checkpoint inhibition monotherapy (e.g., \nantibodies against PD -1 or PD -L1), has dramatically changed the therapeutic concepts of \ndifferent cancer types , especially those with mismatch  repair deficiency [2], the benefit for \novarian cancer patients with an objective response rate of approximately 10% was found to be \nrather modest [3–6]. However, poly -ADP-ribose polymerase inhibitors ( PARPis) and \nantiangiogenic therapy have improve d the survival outcomes of ovarian cancer patients \nbeyond standard care , namely, debulking surgery and platinum -based therapy [7]. \nFurthermore, a number of clinical trials of combination therapies, including immune checkpoint \nblockade, are underway [8–12]. Whereas  the recent  primary analysis of the double -blind \nplacebo-controlled ENGOT-Ov41/GEICO 69-O/ANITA phase III trial showed that the addition \nof the anti -PD-L1 antibody (atezolizumab) did not significantly improve the clinical outcome \n[12], early analysis of the MEDIOLA phase II study adding the PD -L1 inhibitor (durvalumab) \nand the angiogenesis inhibitor (bevacizumab) to a PARPi (olaparib) was promising , with an \nobjective response rate >90% for a specific patient group  with platinum-sensitive relapsed \novarian cancer harboring germline BRCA mutations [11]. \nPARP is involved in DNA damage and repair, binds to single-strand DNA breaks, and performs \nposttranslational modifications of histones and DNA -associated proteins by poly -ADP-\nribosylation, also known as parylation. PARP inhibitors trap PARP and stall the replication fork, \nwhich can subsequently cause DSBs. PARP inhibition is synthetic lethal with deleterious \nBRCA1 and BRCA2 mutations because homologous recombination repair (HRR) cannot \nrestore these double -strand breaks, introducing genome instability by nonhomologous end \njoining or leading to tumor cell death [13]. In high-grade serous ovarian cancer, approximately \n14% harbor a germline and 6% a somatic mutation in the BRCA1 or BRCA2 gene, and \napproximately 50% are HRR deficient, indicating favorable PARPi therapy [14,15]. Sequencing \napproaches such as targeted sequencing, whole exome sequencing, or whole -genome \nsequencing enable resea rchers to detect mutations in other genes involved in homologous \nrecombination repair (HRR). However, the concept of HRR deficiency (HRD) or BRCAness \ngoes beyond, as it encompasses instabilities and genomic scars, including large -scale \ntransitions, loss of  heterozygosity, telomeric allelic imbalance and specific mutational \nprocesses with uneven base substitution patterns (mutational signature 3). Several diagnostic \nassays from commercial providers for the detection of HRD have already been approved [16]. \nHowever, further efforts are undertaken to identify various biomarkers based on different \nmodalities, such as gene expression or methylation , in the context of different cancer types \n[17–21]. Deleterious BRCA1 mutations and /or PARP inhibition can trigger an immune \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \n\n4 \n \nresponse at least in part through the cGAS -STING pathway [20-24], suggesting advantages \nfor combined immunotherapies. However, biomarkers or phenotypes to predict the response \nto therapies, including PARPis and immune checkpoint blockers, are lacking. \nIn this study, we conducted comprehensive immunogenomic analyses of HGSOC using data \nfrom several patient cohorts, including a cohort from The Cancer Genome Atlas (TCGA -OV) \n(n=226), a new cohort from the Medical Univer sity of Innsbruck (MUI) (n=60), and data from \nthe TOPACIO clinical trial in ovarian cancer patients treated with niraparib and pembrolizumab \n(n=22) (Fig. S1, Table S1, S2). Integrated gene expression analysis and machine learning on \nbulk and single -cell RN A sequencing data enabled the 1) development of  a 24 -gene \nexpression classification model for BRCAness, 2) stratification of  patient samples with \nBRCAness and high immune infiltration, whereby tumor -associated macrophages proved to \nbe an important suppressive component, 3) identification of the activation of immune-related \npathways such as the cGAS-STING or JAK -STAT pathway and downstream signaling by \nPARPi and BRCA1 mutation (BRCAness), and 4) development of a diagnostic application from \nRNA sequencing dat a to comprehensively characterize HGSOC samples and predict \nvulnerability and response to combination immunotherapy. \nMethods \nPatient cohorts and datasets \nThe analysis workflow and used datasets from various cohorts are summarized in Fig. S1. \nPatient characteristics for the TCGA-OV cohort (n=226) and the new HGSOC cohort from the \nMedical University in Innsbruck (MUI) (n=60) are listed in Tables S1 and S2. RNA sequencing \ndata and clinical data for the validation cohort (Medical University of Innsbruck; MUI) were \ndeposited at https://doi.org/10.5281/zenodo.10251467. Controlled access data for whole \nexome sequencing and RNA sequencing data for the TCGA-OV cohort were obtained through \ndbGaP access permission (phs000178). Processed data (including methylation beta  values) \nand clinical data were downloaded from Firebrowse (firebrowse.org, BROAD Institute) . \nAdditional clinical data were retrieved from the supplementary data of another resource [22]. \nReads in bam format were converted into fastq  files using samtools fastq [23]. Raw RNA \nsequencing data and clinical annotations for the ICON7 cohort were downloaded from the EGA \narchive (EGAS00001003487). Single -cell RNAseq data were downloaded from the Gene \nExpression Omnibus (GEO) (GSE180661) as an annotated count matrix ( anndata-object) in \nh5ad-format. Data files f rom the TOPACIO clinical trial were retrieved from Synapse \n(https://doi.org/10.7303/syn21569629). RNA sequencing data for ovarian cancer cell lines \nwere downloaded from the gene expression omnibus (GSE120792). Data from RNA \nsequencing analysis of OVCAR 3 and UWB1.289 cancer cell lines performed in this study were \ndeposited in GEO (GSE237361). \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \n\n5 \n \nCell line experiments \nTwo epithelial ovarian carcinoma cell lines, UWB1.289 harboring a deleterious BRCA1 and \nOVCAR3 with intact BRCA1, were obtained from ATCC. OVCAR3 cells were grown in RPMI \n1640 with 0.01 mg/ml bovine insulin and 20% FBS, whereas UWB1.289 cells were grown in a \nmixture of 48. 5% MEGM Bullet Kit medium (Lonza) and 48. 5% RPMI 1640 with 3% FBS. \nViability assays were used to determine the IC50 for olaparib. Both cell lines were treated with \nolaparib or DMSO for 96 hours in four replicates.  Treated and untreated UWB1.289 and \nOVCAR3 cells were stained with indirect immunofluorescent antibodies to detect γH2AX as an \nindicator of double-strand breaks. To determine activated STING signaling, double -stranded \nDNA and its presence in the cytosol, cGAS, STING, and phosphorylated STING were detected. \nThe antibodies used are listed in Table S3. \nImmunohistochemistry analyses \nSlices of 10 selected tumor blocks were subjected to immunohistochemistry analyses \nperformed on the BenchMark ULTRA automated staining device (Ventana, Oro Valley, \nAZ/Roche, Vienna, Austria). The examined markers were CD163 for macrophages and CD8, \nPD-1, CD4, and FOXP3 for T cells. Furthermore, the markers γH2AX and STING were \nanalyzed. All antibodies used are listed in Table S4. \nRNA sequencing analyses \nRNA from cancer cell line samples was isolated from 2x106 cells each using the RNeasy Mini \nKit (Qiagen) according to the manufacturer’s  protocols. RNA quantity and quality were \nassessed using NanoDrop™ 2000c and Bioanalyzer 2100 with Agilent 6000 Nano Kit, RNA \nintegrity numbers (RIN) were between 9.3 and 9.8, and cDNA libraries were generated using \nthe QuantSeq 3' mRNA -Seq Library Prep Kit (Lexogen) according to the manufacturer’s \ninstructions. Paired-end sequencing (150 bp) was performed on a NovaSeq 6000 sequencing \ndevice at GENEWIZ/Azenta. RNA isolation from 60 fresh frozen tumor samples from th e \nHGSOC validation cohort from the Medical University of Innsbruck was conducted in a similar \nmanner at the Department of Obstetrics and Gynecology , resulting in sufficient quality (RIN \nfactors from 6.4 to 9.9) , and sequencing was performed at Novogene (Ca mbridge, UK) for \npaired end sequencing (PE150) on an Illumina NovaSeq 6000 sequencing device using \nTrueSeq (Illumina) strand-specific total RNA libraries. \nRNA sequencing data analyses \nRaw reads were  quality checked using FastQC [24], and the results were summarized with \nMultiQC [25]. Reads were mapped to the human reference genome version hg38 (GRch38) \nusing STAR (version 2.7.1) in 2pass mode [26]. Gene level expression quantification was \nperformed with featureCounts (version 2.0.0) [27] using GENCODE annotations (v36). Raw \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \n\n6 \n \ncounts were normalized using TPM (transcripts per million). RNA sequencing raw data from \nthe HGSOC cohort from the Medical  University of Innsbruck and the ICON7 cohort were \nanalyzed in the same way. For raw sequencing data of the cell lines , single-end reads were \nprocessed by trimming adapter and low-quality sequences using BBDuk with the parameters \nspecified by Lexogen. The trimmed reads were mapped to the human reference genome \nversion hg38 (GRch38) using STAR (version 2.7.9a) in 2 -pass mode. Gene level expression \nquantification was performed with featureCounts (version 2.0.0) and GENCODE annotations \n(v38). \nWhole exome sequencing analyses and variant calling \nRaw exome sequencing reads in fastq format were quality checked using FastQC , and the  \nresults were summarized with MultiQC. Reads of paired tumor and normal samples were \nmapped against the human reference genome version hg38 (GRch38) using BWA [28]. After \nmapping the aligned bam files were sorted with samtools sort, mate coordinates and insert \nsizes were added with samtools fixmate, and duplicates were removed with samtools markdup. \nFinally, index files were generated using the SAMtools index. From exome sequencing data, \nsmall nucleotide variants (SNVs) consisting of single nucleotide variants and small indels were \nassessed. For germline variants , HaplotypeCaller was used to call variants with allele  \nfrequencies of 0.5 or 1.0 in the tumor samples as well as in the normal samples, respectively. \nTo assess somatic variants in the tumor samples , four different variant callers, Mutect2 [29], \nSomaticSniper [30], Varscan2 [31], and Strelka2 [32], were used. If a variant was called by two \nof four variant callers and the variant allele frequency was ≥ 0.05 in the tumor sample and \n<0.05 in the normal sample, the variant passed filtering. Variants were annotated using VEP \n[33] with the ClinVar extension. Only pathogenic (class V) and likely pathogenic ( class IV) \nvariants were considered to affect the function of homologous recombination repair genes such \nas BRCA1 or BRCA2. Tumor mutational burden was calculated based on the number of \nnonsynonymous single nucleotide variants per megabase for each tumor sample. Neoantigen \nprediction is based on previous efforts using analyses of a combina tion of exome and RNA \nsequencing data. HLA alleles (HLA type) were estimated using OptiType [34]. Variant calls, \ntumor and paired normal alignments, and aligned RNA sequencing reads were used together \nto compute mutational haplotypes using phasing. Based on the mutational haplotypes , \npeptides with lengths between 8-11 amino acids were generated and tested for the respective \nHLA alleles with NetMHCpan-4.0 [35], whereby %rank<2 was considered a weak binder and \n%rank<0.5 was considered a strong binder. Dissimilarity to the normal human proteome (hg38) \nwas identified by the antigen.garnish package [36]. Neoantigen load was calculated for each \ntumor based on predicted weak and strong binding neoantigens – irrespective of their peptide \nlength and taking all HLA alleles (type) into account – per megabase. \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \n\n7 \n \nFunctional analysis of gene expression and the tumor immune environment \nDifferential gene expression analysis was conducted using the R package DESeq2 [37]. P \nvalues were adjusted for multiple testing based on the false discovery rate (FDR) according to \nthe Benjamini‒Hochberg method. Genes with more than a twofold change at an FDR<0.1 and \naverage expression across all sample s (baseMean>10) were considered significantly \ndifferentially expressed. To identify functional annotation and affected biological processes , \nlog2-fold change preranked gene set enrichment analyses (GSEA) [38] using hallmark and \nselected immune -related gene sets from MSigDB were performed. Overrepresentation \nanalyses for biological processes (GO) and pathways (Reactome) were performed using the \nR package ClusterProfiler separately for significantly up - and downregulated genes [39]. \nClueGO was used to build a network and group significantly overrepresented pathways, which \nare shared genes [40]. The STRING database [41] was used to identify an interaction network \nwithin the differentially expressed genes, and subnetworks were found by MCL clustering with \ninflation parameter=3. Footprint analyses of response genes of perturbed cancer sign aling \npathways were performed using PROGENy [42]. To assess tumor infiltration of immune cells, \ndeconvolution methods, i.e., quanTIseq [43] using the immunedeconv R package [44] was \napplied to bulk RNA sequencing data from tumor samples. To characterize the immune-related \nprocesses, well-described immune signatures , such as T -cell inflammation, IFN gamma \nsignature, cytolytic activity, cytotoxic T lymphocyte function, and T-cell exhaustion (Table S5), \nwere analyzed. Based on log2(TPM+1) normalized expression data , single sample gene set \nenrichment using GSVA [45] was performed for signatures with more than 10 genes. For short \nsignatures, in cluding fewer than 10 genes , the average expression of the signature was \ncalculated. The tumor-immune phenotype (infiltrated, excluded, desert) was determined based \non a previously developed classification model based on 157 genes using digital pathology \ndescribing the presence and position of CD8+ T cells relative to the center or margin of the \ntumor [46]. A random forest model based on the expression data (TPM) of these 157 gene s \nwas used to characterize samples from the TCGA and the MUI cohorts. To classify ovarian \ncancer samples into immune reactive (IMR), proliferative (PRO), differentiated (DIF) and \nmesenchymal (MES) molecular subtypes, the consensusOV R package [47] was used. The \nimmunophenoscore (IPS) and immunop henogram for all samples were determined as \ndescribed previously [48]. \nDetermination of BRCAness \nBRCAness was determined based on HRD scores [49], mutational signature 3 [50], mutations \nin homologous recombination repair pathway genes and methylation of promoter regions of \nBRCA1 and BRCA2. All high -grade serous ovarian cancer (HGSOC) samples of the TCGA \ncohort for which paired tumor and normal exome sequenci ng and matched RNA sequencing \ndata were available (n=226) were used. Samples were classified with a BRCAness phenotype \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \n\n8 \n \nwhen they had either a deleterious mutation in the homologous recombination pathway, an \novarian cancer-specific HRD score of ≥ 63 [51], a mutational signature 3 ratio > 0.25 or a \nmethylation level beta value >0.7 of the BRCA1 or BRCA2 promoter. HRD  scores were \ncalculated as the unweighted sum of the three genomic scar values, loss of heterozygosity \n(LOH) [52], telomeric allelic imbalance (TAI) [53], and large-scale state transitions (LST) [54]. \nTo compute the genomic scar values, scarHRD [55] was used on genome segmentation files \ngenerated with sequenza -utils [56]. The mutational signature 3 score was computed using \nMutationalPatterns [57]. The mutational signature 3 ratio was calculated as the ratio between \nmutational signature 3 supporting mutations and all detected mutations. \nBRCAness classification \nFirst, genes expressed in ovarian cancer cells were identified using single -cell RNAseq data. \nGenes that are expressed in at least one ovarian cancer cell were consider ed expressed in \ncancer cells. Log 2(TPM+1) normalized gene expression values of these genes in the TCGA \ndataset were then subjected to recursive feature elimination with three different machine \nlearning models ( random forest, AdaBoost and gradient boosting ) to identify the 50 most \nimportant features for each model. Genes that were among the top 50 in at least two of the \nthree models (24 genes) were then used subsequently to train a random forest classification \nmodel to discriminate between BRCAness and noBRCAness samples based on gene \nexpression data. The performance of the classifier was evaluated by analysis of the receiver \noperating characteristic (ROC) curve with 10 -fold cross -validation. The area under curve \n(AUC) was used as a performance measure. A cutoff for BRCAness (P>0.5266) was selected \nusing the Youden index. Furthermore, the classifier was tested in the independent validation \ncohort (MUI) for 29 patients with HRD information based on SNP arrays and further validation \nusing Myriad MyChoice CDx. Samplewise BRCA classification in single-cell RNA sequencing \ndata from 29 patients was performed with an optimized cutoff ( P>0.45) and based on the \nmajority of classified tumor cells. The BRCAness classifier was further validated using \nadditional somatic mut ation and RNA sequencing data from the Clinical Proteomic Tumor \nAnalysis Consortium ovarian cancer (CPTAC -OV) cohort [58]. SigMA scores representing \nmutational signature 3 were calculated using SigMA [59] as a surrogate for BRCAness. \nSingle-cell RNA sequencing analysis \nAll analyses of single-cell data were performed in Python using scanpy [54] and scvi-tools [60]. \nSince the samples were sequenced separately for sorted CD45+ and CD45 - cells, the raw \nread counts were integrated using scvi -tools with a batch  effect correction. Counts were \nnormalized to counts per million (CPM) and log2 transformed , adding a pseudocount of 1. \nQuality metrics were determined using scanpy and filtered for genes that are expressed in at \nleast one cell. The dataset was filtered fo r samples from the primary tumor (adnexal tumor \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \n\n9 \n \ntissue). Principal component analysis and nearest neighbor analyses were calculated with \ndefault settings, and clustering was performed with the Leiden algorithm. Super cell types were \nannotated as previously defined. Subtypes of T-cell and myeloid cell clusters were assigned \nbased on the expression of marker genes using published marker genes for different cell types \nand the PanglaoDB [61]. Differentially expressed genes  between clusters were calculated \nusing the Wilcoxon ranked sum test. For visualization , we used uniform manifold \napproximation and projection (UMAP) dimensional reduction. Gene expression between cell \ntypes was compared by heatmaps, violin plots, and bubb le plots. To assess ligand‒receptor \ninteractions between cancer cells and cells from the TME , CellPhoneDB [62] analysis was \nused. \nGene expression analysis \nGene expression analysis in the TOPACIO cohort was performed on the NanoString platform. \nWe used the nSolver software from NanoString (Seattle, US) to obtain normalized data. \nDifferential expression analysis was performed using the R package limma [63], and genes \nwith p<0.05 were considered differentially expressed. \nVulnerability score and maps \nVulnerability maps consist of three variables: the vulnerability score, the BRCAness score and \nthe cytolytic activity (CYT) to C1QA ratio. For the BRCAness score, the prediction probability \nfrom the random forest classifier was used. The CYT to C1QA ratio was calculated from the \nlog2 (TPM+1) values of GZMB, PRF1, and C1QA (1). \nCYT to C1QA ratio = 0.5 x (GZMB + PRF1)/C1QA (1) \nThe CYT to C1QA ratio was transformed to values between 0 and 1 using a sigmoid function \nwith softmax transformation and parameters derived from the TCGA cohort and termed C2C \n(2). \nC2C = 1/(1 + exp(- (CYT to C1QA ratio - 0.301)/0.0433))  (2) \nThe vulnerability score was defined as the weighted sum of BRCAness pro bability and C2C \n(3), whereby the weights were identified using a logistic regression model on the CYT to C1QA \nratio using log2 intensity expression values and SigMA status (mutational signature 3) data \nfrom the TOPACIO cohort and the treatment response as a binary dependent variable. \nVulnerability score = 2.597 x BRCAness probability + 1.166 x C2C (3) \nFor visualization of the vulnerability map, a two-dimensional map was created with C2C as one \ncoordinate, BRCA probability as the other coordinate, and the color-coded vulnerability score. \nStatistical analysis \nSurvival analyses were performed for both HGSOC cohorts (TCGA, MUI) for selected genes \nor immune parameters by dichotomization of patients based on the median or maximum log -\n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \n\n10 \n \nrank statistics using the R package survival. For the TCGA cohort, overall survival and survival \nstatus were derived from a clinical data resource for TCGA [22] and for the cohort from Medical \nUniversity Innsbruck from the clinical data as provided by the Department of Obstetrics and \nGynecology. Univariate Cox regression was performed , and for parameters significantly \nassociated with overall survival , multivariable Cox regression taking clinical parameters into \naccount (age, FIGO stage, grade, residual tumor) was performed. To determine the \nassociation between continuous o r binary variables , point biserial correlation analysis was \nused. For the correlation between binary variables, the Phi coefficient and chi-square test were \nused, and for the correlation between continuous variables , Pearson’s correlation coefficient \nwas used. To compare parameters between two groups, the Wilcoxon rank-sum test was used. \nWhere indicated, p values were adjusted for multiple testing based on the false discovery rate \n(FDR) according to the Benjamini‒Hochberg method. P<0.05 or FDR<0.1 were cons idered \nsignificant. \nResults \nA 24-gene signature predicts BRCAness in HGSOC patients \nBecause the response to platinum-based chemotherapies or therapy with PARP inhibitors in \novarian cancer is not limited to patients with tumors harboring BRCA1 or BRCA2 mutations, \nwe expanded the group of patients by using a genomic characterization termed BRCAness , \nwhich has very much in common with homologous recombination repair deficiency (HRD) \nstatus [64]. BRCAness status includes mutations of genes in the homologous recombination \nDNA repair pathway (HRR), genomic scars, loss of heterozygosity, telomeric allelic imbalance, \nor large-scale transitions, mutational signature 3, or promoter methylation of the BRCA1 or \nBRCA2 gene. We assessed these parameters based on whole exome sequencing dat a and \nmethylation data from the TCGA OV cohort (Fig. 1A). Very few patients harboring HRR \nmutations or BRCA1/2 promoter methylation fell below the combination of the HRD cutoff \n(HRD>63) and the MutSig3 ratio cutoff (0.25) , indicating a reasonable selection of the cutoff \nvalues (Fig. 1B). To identify BRCAness solely based on gene expression data, we developed \na machine learning classifier that can discriminate between BRCAness and non -BRCAness \nsamples using bulk and single -cell RNA sequencing data (Fig. 1A).  Recursive feature \nelimination based on multiple models resulted in a BRCAness gene expression signature with \n24 genes, which was used to train a random forest model discriminating between BRCAness \nand noBRCAness. The receiver operating characteristics (ROC) with 10-fold cross-validation \non the training dataset showed an area under the curve (AUC) of 0.91±0.04 (Fig.  1D). \nFurthermore, we demonstrated that in addition to classifying bulk RNAseq samples from the \nvalidation cohort (MUI) (Fig. 1E) with an accuracy of 0.79, an F1-score of 0.86, and a positive \nprediction value of 0.86, the classifier is also capable of classifying samples from scRNAseq \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \n\n11 \n \ndata at the sample level (Fig. 1F) with an accuracy of 0.86, an F1 score of 0.87 and a positive \nprediction value o f 0.87. There was also good agreement with a recently defined gene \nexpression-based HRDness signature including 173 up - and 76 downregulated genes [65] \nusing a single sample gene set e nrichment [45,66] derived score in the TCGA cohort as well \nas the MUI validation cohort with Spearman’s rank correlation of ρ=0.72 (P<0.001) and ρ=0.63 \n(P<0.001), respectively (Fig. S2, S3). Interestingly, six genes from our 24 -gene signature to \nclassify BRCAness (CCDC90B, CRABP2, FZD4, GPAA1, PRCP, SNRP1) were also among \nthe upregulated and two genes ( RAD17, LTA4H) among the downregulated genes. The 24 -\ngene BRCAness signature was further validated in the CPTAC -OV cohort (n=71) by \ncomparison to SigMA (mutational signature 3) with a Spearman’s rank correlation of ρ=0.43 \n(P<0.001) (Fig. S4). \nIn summary, we developed a 24-gene-based BRCAness model validated in several single-cell \nand bulk RNAseq datasets with reasonable classification performance. \nGenome instability is associated with immune-related processes \nTo identify the relationship betwee n genomic instability and the activation of the immune \nsystem, we performed correlation analyses between the BRCAness status and various \nimmune-related signatures. BRCAness could be significantly positively associated with the \nenrichment of immune -related signatures, such as those for IFNG response (rho=0.38, \np=0.004) and T-cell inflamed tumor microenvironment (rho=0.46, p=0.0014), even to a larger \nextent with high tumor mutational burden (p<0.001) and high neoantigen load (p<0.001) \n(Figure 2A, 2B). However , compared to other cancer types with defective DNA mismatch \nrepair, such as melanoma or microsatellite instable colorectal cancer, the TMB or neoantigen \nload in ovarian cancer is rather low. Thus, this is more indicative of deficient homologous \nrecombination repair. BRCAness was also associated with longer overall survival in the TCGA \ndataset (HR=0.50, 95%-CI 0.34-0.69; p<0.001 log rank test), indicating that those patients are \nmore responsive to platinum -based chemotherapy (Fig. 2C). Although this status could be \nassociated with higher CD8+ T-cell infiltration (estimated by deconvolution methods from RNA \nsequencing using quanTIseq) (HR=0.67, 95%-CI 0.47-0.93; p=0.019, log-rank test) (Fig. 2D), \nthis could not completely explain the survival advantage. Never theless, analyses of signaling \npathways by downstream target expression using PROGENy indicated for the TCGA cohort \n(n=226) as well as the MUI validation cohort (n=60) that immune -related pathways, including \nTNFa, NFkB, and JAK-STAT, were activated in the BRCAness samples (Fig. 2E, 2F). Using \nSTRING analyses, we also identified a highly connected network including various chemokines \nand interleukins and their respective receptors (CCL7, CCL11, CXCL5, CXCL9, CXCL13, \nCCR2, CCR3, CCR4, CCR8, CXCR3, and IL6 (Fig. S11)), which were significantly more highly \nexpressed in BRCAness tumors than in non -BRCAness tumors, indicating attraction and \ninteraction with various immune cells. \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \n\n12 \n \nWe observed a significant association of BRCAness with longer overall survival and a less \npronounced correlation with immune-related processes in HGSOC patients. \nPARP inhibition activates the cGAS-STING pathway in vitro \nTo study the effect of PARPis on immune activation, we performed in vitro analyses. As tumor \nmodels, an ovarian cancer cel l line with a proficient BRCA1 gene (OVCAR3) and a cell line \nwith a mutation in the BRCA1 gene (UWB1.289) were utilized. We performed RNA sequencing \nanalyses to identify differentially expressed genes between olaparib (PARPi) -treated and \ncontrol (DMSO)-treated cell lines. Significantly upregulated genes (Figure 3A, 3B, S18, S19, \nS20, S21, Data file 1) indicate activation of various processes (Fig. 3C, 3D, S22) , including \npattern recognition receptor activation, response to cytokine signaling, interferon alp ha \nresponse (type I), NFkB pathway, and cGAS-STING signaling. To further validate the results \nat the protein level, we performed immunofluorescence analyses indicating effects on gH2AX \nby mutation in the BRCA1 gene and an even stronger effect by olaparib (PARPi) treatment \n(Fig. 3E). Similarly, we observed a different activation of cGAS and STING in the BRCA1 -\ndeficient versus the BRCA1 -proficient cell model (Fig. 3F). Furthermore, using gene set \nenrichment analyses, a significant interferon alpha response was also observed in BRCAness \nsamples of both the TCGA cohort and the MUI validation cohort (Fig. 3C). \nIn summary, we observed cGAS-STING activation by olaparib treatment in vitro and an \ninterferon type I response as well as chemokine expression in HGSOC pat ient cohorts with \nBRCAness status. \nBRCAness and immune subtype stratifies HGSOC patients \nWe next focused on characterizing the presence of cytotoxic T lymphocytes and their spatial \ndistribution in the tumor, following a recent approach in which digital pathology could be linked \nto gene expression [46]. With the reported list of 157 genes and using random forest analysis, \nwe were able to divide the patients into a group with infiltrated, excluded, or desert tumor-\nimmune phenotypes. Interestingly, the excluded phenotype was associated with upregulation \nof TGFβ and high expression of markers for cancer -associated fibroblasts, such as FAP or \nPDPN, which could form a physical barrier to prevent T -cell infiltration (Fig. 4A). Although \nvarious definitions of molecular subtypes based on gene expression or copy number \naberrations have been described in the last decade, we are convinced that the immunoreactive \nmolecular subtype (IMR) is the most meaningful to delineate immunoreactivity because many \nof the immunity genes, including cytotoxic effectors, factors involved in antigen processing and \npresentation, or immune checkpoints , are highly expressed in this condition (Fig. 4A, Table \nS6). The definition of molecular subtype also includes mese nchymal (MES), proliferative \n(PRO), and differentiated (DIF) molecular subtypes [47]. To identify patients most likely to \nbenefit from the combination of PARP inhibitor therapy, where the BRCAness phenotype may \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \n\n13 \n \nbe beneficial, with immune checkpoint inhibitor therapy, where the immune-related phenotype \nmay be beneficial, we selected a group of patients with tumor BRCAness, an infiltrated tumor \nimmune phenotype, and an immune -reactive molecular subtype termed BRCAness immune \ntype (BRIT). When comparing the estimated immune cell infi ltrates in these cancer samples \nwith BRCAness cancers without immune type (noBRIT), we found that not only cytotoxic T \nlymphocytes such as CD8+ T cells were significantly more abundant (p<0.001) but also a \nnumber of suppressive immune cells (M2 macrophages  (p<0.001), regulatory T cells \n(p<0.001), myeloid-derived suppressor cells; MDSCs (p<0.001)) (Fig. 4B). Furthermore, we \ndid not observe a significant difference in overall survival between the groups (p=0.56, HR \n=0.81, 95% CI 0.42-1.60). \nThese observations underscore the importance of the suppressive immune environment and \nsuggest that suppressive immune cells may be an important factor, which is why ovarian \ncancer patients have a limited response to immunotherapy. \nTumor-associated macrophages inform therapy response \nAs the power of deconvolution approaches from bulk RNA sequencing analyses shows some \nlimitations, we took advantage of single-cell RNA sequencing analysis , allowing a more \ncomprehensive characterization of the tumor environment and evaluation of the cell interplay. \nAnalyses of more than 300 thousand cells of adnexal ovarian tumor tissue from 29 patients \nallowed a clear separation between major cell type populations by clustering and nonlinear \nprojection (UMAP) (Figure 5A). In contrast to cell ty pes from the tumor microenvironment, \ntumor cells showed a clear separation between BRCAness and noBRCness samples (Fig. 5A). \nBecause cells from the suppressive environment have a major impact, we focused on the \nmyeloid cell compartment and demonstrated tha t the majority of these cells were \nmacrophages, and we identified subpopulations based on most dominant marker genes, \nincluding CD169 (SIGLEC1) macrophages, CX3CR1 macrophages, and MARCO \nmacrophages (Fig. 5B). One described hallmark marker of tumor -associated macrophages \n(TAMs) is TREM2, which has been identified as an attractive target for cell depletion therapy \nand is being tested in an ongoing clinical trial [67]. Notably, the expression patterns of TREM2 \nand BRCAness are very similar, showing high expression in all macrophage subtypes and, to \na lesser extent , in monocytes (Fig. 5B). To search for further genes with similar expression \npatterns in myeloid subpopulations , we analyzed known tumo r-associated macrophage and \nmonocyte marker genes [68]. As indicated by this analysis, C1QA showed a similar but even \nmore pronounced expression pattern than TREM2 (Fig. 5B, 5C). C1QA was also recently \ndescribed as a surrogate marker for the CD68+CD163+ macrophage subset [69]. \nTo determine whether tumor-associated macrophages might also play a role in the response \nto combined cancer immunotherapy, we used expression data from a clinical trial (TOPACIO). \nWe analyzed in which cell types genes with different expression in responders versus \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \n\n14 \n \nnonresponders were dominantly expressed. In fact, a number of downregulated genes in \nresponders, such as LYZ, LILRB4, and ITGB2, were most highly expressed in myeloid cells \n(macrophages), LILRB4 in dendritic cells, and integrin subunit beta 2 ( ITGB2) in other cell \ntypes, such as T/NK cells (Fig. 5D, 5E). Interestingly, we identified various ligand‒receptor \ninteractions with expressed ligands in tumor cells and respective receptors expressed in tumor-\nassociated macrophage subsets using CellPhoneDB [62] (Fig. 5F). \nThe growth arrest -specific pro tein 6 (GAS6) – AXL tyrosine kinase (AXL) interaction, for \nexample, which are both associated with poor outcome, have already been evaluated in clinical \ntrials in ovarian cancer by inhibiting their interaction [70]. LILRB1 and LILRB2 expressed in \nmacrophage subsets were found to interact with the nonclassical human leukocyte antigen \nHLA-F expressed in cancer cells. VEGFA and VEGFB expressed in tumor cells potentially \ninteract with NRP1, and FLT1 is particularly expressed in endotheli al cells. Blocking \nmacrophage colony-stimulating factor CSF1 and its receptor CSF1R axis and several drugs \nthat target these factors have been under investigation [71]. \nThese observations summarized together suggest that tumor -associated macrophages may \nnot only play a role in immunotherapy alone but are also essential in informing about therapy \nresponse when combined with PARP inhibitors. \nAnalyses of an independent cohort indicate vulnerability to combination \nimmunotherapy. \nTo validate the results, we performed RNA sequencing analyses of an HGSOC cohort of \npatients from Medical University Innsbruck (n=60). Stratification of these patients resulted in \nvery similar expression patterns evident from a number of immune marker genes, which were \nhighly expressed in the BRCAness immune type patient group (BRIT) (Fig. 6A). To further \ncharacterize immune infiltrates in different patient groups , we performe d \nimmunohistochemistry analyses on ten selected samples for various markers. BRIT tumor \nsamples showed high γH2AX activity, STING activation, CD8+ T -cell infiltration, CD4+ T-cell \ninfiltration, and strong CD163+ tumor -associated macrophage populations (Fig . 6B). These \neffects were even more pronounced in one sample with no detected BRCA1 or BRCA2 \nmutation, underscoring the importance and validity of predicted BRCAness. Another tumor  \nsample with no BRCAness, a desert tumor-immune phenotype, and a differentiated molecular \nsubtype was used as a negative control, and in fact, no activity for any of the tested markers \nwas observed. To better address the potential for combination immunotherapy response, we \nagain took advantage of data from the TOPACIO trial and , based on the clinical response , \ntrained a logistic regression model and learned weights for three surrogate variables: MutSig3 \nas an indicator for BRCAness, average expression of PRF1 and GZMB as indicators for \ncytolytic activity, and expression of C1QA as an indicator for tumor -associated suppressive \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \n\n15 \n \nmacrophages. Based on the HGSOC samples from TCGA, we developed a two -dimensional \nvulnerability map, with the ratio of cytolytic activity and C1QA expression as one variable (C2C) \nand the BRCAness prediction pr obability as the other variable. The vulnerability score  is \nindicated by color (Fig. 6C). When applied to the selected examples from the validation cohort \nof the Medical University of Innsbruck, these differed significantly for areas with high \nvulnerability scores (indicating response to combination immunotherapy) compared to the \nnegative control with low vulnerability scores (Fig. 6C). Furthermore, we observed a significant \ndifference in overall survival between patients with high and low vulnerability scores (p<0.001, \nHR = 0.47, 95% CI 0.33-0.66) in the TCGA HGSOC cohort , indicating a positive association \nof a high vulnerability score with longer overall survival. For patients in the validation cohort \n(MUI HGSOC), no significant difference in overall survival (p=0.368, HR = 0.78, 95% CI 0.45-\n1.34) could be revealed. To enable the characterization of newly diagnosed HGSOC samples \nbased on RNA sequencing data , we developed an easy-to-use R package (OvRSeq), which \nallows us to not only estimate the parameters to determine the vulnerability score (and \ngenerate the vulnerability maps) but also comprehensively annotate the sample for \nBRCAness, tumor -immune phenotype, molecular subtype, estimate immune infiltrates, \nenrichment of immune -related signatures, and indiv idual marker genes. This also includes \nother clinically relevant parameters, such as the angiogenesis score we previously defined, \nwhich might be useful for the prediction of anti -VEGF therapy [72]. The web application \n(https://ovarseq.icbi.at) allows the generation of summary information as a report of individual \nsamples (Fig. S23). \nThe developed application should ultimat ely be useful to identify vulnerabilities and support \nclinical therapy decisions for high-grade serous ovarian cancer patients. \nDiscussion \nHere, we described how genomic instability in high -grade serous ovarian cancer affects the \ntumor immune environment a nd the consequences and vulnerabilities of combination \nimmunotherapy combining PARP inhibitors with immune checkpoint inhibitors such as anti -\nPD1 antibodies. A particular status in which patients respond well to PARP inhibitors and \nplatinum-based chemotherapy is given when genes of the homologous recombination repair \npathway such as BRCA1 or BRCA2 are mutated. Genomic scars are consequences of a \nhomologous recombination repair deficiency and are used to define an HRD score, often \nmeasured by established com mercial assays, which allows the assignment of  a responsive \nstatus beyond BRCA1 and BRCA2 mutations. The applicability and associated cutoff values \nfor different assays and cancer types are under discussion , as the HRD algorithm has been \nused in clinical s tudies including different cancer types , such as breast cancer and ovarian \ncancer [49,73,74]. Genomic scars are predictive but do not allow direct functional \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \n\n16 \n \ninterpretation, whereas gene expression signatures could be an alternative in this regard. Very \nfew approaches have associated gene expression with HRD status [18,19,65]. Whereas a \nsixty-gene signature [18] and a two -gene signature ( CXCL1, LY9) [19] have f ocused on \nmicroarray data, a recent approach using RNA sequencing data identified a 249 -gene \nsignature to predict HRD [65]. We observed a number of overlaps with our 24 -gene BRCA \nsignature and a high concordance of signature scores in our training (TCGA) and validation \n(MUI) cohorts, indicating the reliability of our approach. This was underscored by comparison \nwith mutational signature 3 (SigMA) in an independent cohort (Fig S3). The p erformance of \nthe BRCAness classifier is reasonable, with AUC=0.91 (10-fold cross-validation) and positive \npredictive value for validation on both bulk RNA sequencing in the validation cohort (MUI) \n(PPV=0.86) and samplewise single-cell RNA sequencing data (PPV=0.87). \nBRCAness is associated with a longer overall survival since all patients usually receive \nplatinum-based chemotherapy such as carboplatin -paclitaxel combination. Although this \nassociation was previously reported and therefore expected, it highli ghts the performance of \nthe classifier to BRCA status. There is evidence  that BRCA1/2 -mutated tumors exhibit \nsignificantly increased CD8+ TILs [75], although in breast cancer , differential modulation \nbetween BRCA1 and BRCA2 mutations in the tumor immune microenvironment has been \nfound [76]. The association between BRCAness  and several immune -regulated signatures \nwas significant but not very pronounced. We found evidence that several signaling pathways \nand processes known to modulate the immune system are activated by BRCA1 mutations or \na BRCAness-related phenotype, such as JAK-STAT signaling or an interferon type I response, \nwhich are activated by free double-stranded DNA in the cytoplasm of tumor cells via the cGAS-\nSTING pathway and affect dendritic cells [77–79]. By expression and immunofluorescence \nanalyses of ovarian cancer cell lines and by treatment with PARPi, we demonstrated that this \naxis is actually activated. Notably, the STAT3 pathway, which is activated by PARP inhibition, \nmay, however, mediate treatment resistance by promoting the polarization of protumor TAMs, \nwhich could be overcome by STING agonism [80]. STING, CSF1R, SREBP -1, and VEGFA \nmight also be targets to overcome resistance to PARPi-immunotherapy combinations [81]. The \nupregulation of many chemokines and chemokine receptors (Fig. S10, S11) indicates that \nBRCAness tumors are actively involved in immune cell attraction and interaction. For example, \nCCL5 produced by tumor cells or CXCL9 and CXCL10 also expressed by tumor -resident \nmyeloid cells determine effector T-cell recruitment to the tumor microenvironment [82,83]. We \ndetected significant upregulation of CCL5 and CXCL10 by PARP inhibition, which was also \nidentified as a downstream target of STING [77]. Another interesting chemokine that is strongly \nupregulated in cancer cell lines, particularly by olaparib treatment, is CCL20. CCL20 could be \nassociated with cancer metastasis and progression by interacting with its cognate receptor \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \n\n17 \n \nCCR6 in an ovarian cancer mouse model. However, the higher expression in the myeloid cell \ncompartment, as evident from single-cell analyses (Fig. S16), overlies the intrinsic tumor effect. \nOne of our basic hypotheses was that samples with BRCAness respond better to PARPi \ntherapy and that hot tumors with an acti vated immune milieu respond better to immune \ncheckpoint inhibition, as has been shown , for example, in melanoma for the activated IFNG \npathway [84]. However, when we compared  the BRCAness immune type (BRIT) with other \nsamples, we observed by using deconvolution approaches that suppressive cell types such as \nM2 macrophages, MDSCs, and Tregs were more abundant. In particular, tumor -associated \nmacrophages (TAMs) could be a major factor together with low mutational burden, abnormal \nneovascularization, altered metabolism, and failure to reverse T-cell exhaustion for the limited \nimmunotherapy response in ovarian cancer [85]. By using single -cell RNA sequencing data \nanalyses in adnexal cancer tissue from 29 patients, we demonstrated that myeloid cells are \nthe most abundant immune cells , and the majority were characterized as tumor -associated \nmacrophages. We could identify various subtypes of these TAMs , and based on known \nmacrophage polarization marker genes (Fig. S14), we observed a bias toward alternative (M2-\nlike) macrophages compared to classical (M1 -like) macrophages, although this classification \nis limited and may be better described as a continuum of different stages than isolated cell \ntypes. Nevertheless, a majority of these TAMs are suppressive , as i ndicated by TREM2 \nexpression. TREM2 is a promising therapeutic target for TAM depletion [71]. Inhibition of \nTREM2 has been shown to improve the anti-PD1 response in various mouse models and is \ncurrently being investigated in a clinical trial [67,86]. Another recent study underscored the role \nof TAMs and demonstrated that specifically , the Siglec-9-positive TAM subset is associated \nwith an immune-suppressive phenotype and adverse prognosis in HGSOC patients [87]. \nInterestingly, a previous work using cyclic immunofluorescence highlighted the role of \nexhausted T cells in the response to niraparib/pembrolizumab. I n responders, particularly in \nextreme responders, frequent proximity between exhausted T cells and PD -L1+ (CD163+, \nIBA1+, CD11b+) TAMs was observed [9]. Noticeably, based on the selected marker \nexpression, we observed an overlap w ith the CD169/SIGLEC-1 TAM cluster from single -cell \nRNA sequencing data analyses (Fig. S9). In addition, in patients who responded to this \ncombination therapy, we identified a number of downregulated genes that were also highly \nexpressed in TAMs, such as LYZ, LILRB4, and ITGB2. Whereas lysozyme (LYZ) is an \nantimicrobial ligand and is involved in central macrophage function and is therefore \nnonspecifically and highly expressed,  LILRB4 is an immune checkpoint on myeloid cells, \nindicating a more regulatory role. High expression of the integrin ITGB2 was previously shown \nto be associated with poor survival outcome [88], underscoring that high expression in TAMs \nis crucial. In contrast, ITGB2 is also associated with CD8+ T cells, as it encodes the beta chain \nof the LFA-1 protein, which has been shown to be essential in the assembly of the immune \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \n\n18 \n \nsynapse or to influence lymphocyte extravasation and T -cell recruitment to the tumor and is \nregulated by GDF -15 [89]. This association is also underlined by a positive s ignificant \ncorrelation of ITGB2 expression with M2 macrophages and, to a lesser extent, with estimated \nCD8+ T-cell infiltration in the TCGA cohort (Fig. S11). \nBecause stratification of patients based on gene expression in our validation cohort was very \nsimilar to the analysis on the TCGA cohort, we set out to adapt our hypothesis andalso include \nelements of the suppressive environment. Already, it was shown that regulatory T cells (Tregs) \nare an important component of the suppressive milieu and are associat ed with unfavorable \nsurvival outcomes in ovarian cancer [90,91]. We performed immunohistochemistry analyses \nusing FOXP3 and CD163 antibodies in the validation cohort and found very pronounced \nmacrophage infiltration (CD163) but hardly Treg infiltration (FOXP3) into the tumor site in some \nsamples. The results of the single -cell RNA sequencing analyses and the fact that various \nTAM marker genes were associated with poorer overall survival (Fig. S13) also suggest that \nTAMs play a more dominant role in ovarian cancer. \nWhile infiltration of various cell types from the adaptive immune system [92] and other markers, \nsuch as tumor mutational burden (TMB) [93] or IFNG signature [84], have been associated \nwith good prognosis and immunotherapy response in various cancer types, the suppress ive \nimmune environment with tumor-supportive CD68+CD163+ macrophages is becoming more \nimportant [69]. Accordingly, a signature of the immune activation ratio of CD8A/C1QA has \nbeen found to be prognostic and predictive for immunotherapy response [69]. Based on \nprevious analyses [44,69], we considered the mean PRF1 and GZMB expression as a proxy \nfor cytolytic activity as predominantly exerted by cytotoxic T lymphocytes. The specific \nexpression pattern of C1QA on TAMs was comparable to that of TREM2 but at a much higher \nlevel. Therefore, we also used the member of the complement system C1QA as a surrogate \nfor TAMs and the suppressive tumor immune environment and finally built a ratio of cytolytic \nactivity (CYT) to the expression of C1QA (C2C), indicating the pro- and antitumoral balance of \nthe immune environment. Finally, to build a predictive algorithm for combination therapy \nresponse, we included both C2C on the one hand and BRCAness on the other hand into one \nmodel. Since HRD measured with companion diagnostic tests is not able to predict all PARPi \nresponders, as shown in several clinical trials, and since PARPi treatment can activate a \nnumber of immune -related pathways even in situations with proficient HRR , which is also \nunderlined by our in vitro analyses, this model is considered to be relevant for combination \nimmunotherapy. \nOur studies have some limitations in that the training and validation patient cohorts were \nretrospective studies, and RNA sequencing was performed at a later time point.  Additionally, \nonly a limited number of patients who received combination therapy could be included ; \ntherefore, the conclusion about the predictive power for the treatment is limited and requires \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \n\n19 \n \nfurther validation in larger cohorts. One component that was  not considered in this study is \nmalignant ascites, which has been shown to contain various cell types, such as macrophages, \nmany soluble factors and cytokines, that influence the protumorigenic phenotype and promote \nmetastatic spread of HGSOC through tran scoelomic dissemination [94]. Although the \napplication is not a clinically approved software for the purpose of therapy and diagnosis, the \nvery easy-to-use application ( https://ovrseq.icbi.at) and the respective R package OvRSeq \nallow based on RNA sequencing to gain comprehensive information about the phenotype of a \ntumor sample, support clinical decisions, and stimulate further research. \nConclusions \nOur approach using RNA sequencing data to comprehensively characterize both genome \ninstability and the tumor immune environment enabled us to stratify HGSOC patients. \nHowever, further analyses indicate that suppressive tumor -associated macrophages in the \ntumor immune microenvironment may play an essential role in understanding  why \nimmunotherapy shows only a modest response in ovarian cancer, and in a similar fashion, this \napplies to combination immunotherapy , including PARP inhibitors and immune  checkpoint \nblockers. Based on various datasets, we have developed a methodology and corresponding \ndiagnostic application that uses RNA sequencing data not only to comprehensively \ncharacterize newly diagnosed HGSOC patients but also to provide information on vulnerability \nto combination immunotherapy that may inform whether the patient will respond or not. \nAbbreviations \nAUC  Area under curve \nBRCA  Breast cancer DNA repair-associated genes \nBRIT  BRCAness immune type \nCTL  Cytotoxic T lymphocytes \nCYT  Cytolytic activity \nC1QA  Complement C1q A chain \nC2C  Transformed CYT to C1QA ratio \nDIF  Differentiated molecular subtype \nEOC   Epithelial ovarian cancer \nFDR   False discovery rate \nGSEA  Gene set enrichment analysis \nGSVA  Gene set variation analysis \nHGSOC High-grade serous ovarian cancer \nHR   Hazard ratio \nHRD  Homologous recombination repair deficiency \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \n\n20 \n \nHRR  Homologous recombination repair \nIMR  Immune reactive molecular subtype \nIPS  Immunophenoscore \nLOH  Loss of Heterogeneity \nLST  Large-scale transitions \nMDSC  Myeolid-derived suppressor cell \nMES  Mesenchymal molecular subtype \nMutSig3 Mutational signature 3 \nNES  Normalized enrichment score \nPARP  Poly (ADP-Ribose) Polymerase \nPARPi  PARP inhibitor (olaparib, niraparib) \nPCA  Principal component analysis \nPD-1  Programmed cell death 1 (PDCD1) \nPRO  Proliferative molecular subtype \nPROGENy Pathway RespOnsive GENes for activity inference \nROC  Receiver operating characteristics \nTAI  Telomeric allelic imbalance \nTAM  Tumor-associated macrophages \nTCGA  The Cancer Genome Atlas \nTPM   Transcript per millions \nTregs  Regulatory CD4+ T cells \nUMAP  Uniform manifold approximation and projection \nConflict of interest statement \nAGZ reports consulting fees from Amgen, Astra Zeneca,  GSK, MSD, Novartis, PharmaMar, \nRoche-Diagnostic, Seagen; honoraria from Amgen, Astra Zeneca, GSK, MSD, Novartis, \nPharmaMar, Roche, Seagen; travel expenses from Astra Zeneca, Gilead, Roche; participation \non advisory boards from Amgen, Astra Zeneca, GSK, MSD, Novartis, Pfizer, PharmaMar, \nRoche, Seagen. CM reports consulting fees and honoraria from Roche, Novartis, Amgen, \nMSD, PharmaMar, Astra Zeneca, GSK, Seagen; travel expenses from Roche, Astra Zeneca; \nparticipation on advisory boards from Roche, Novartis, Amgen, MSD, Astra Zeneca, Pfizer, \nPharmaMar, GSK, Seagen. HH has received research funding via Catalym and Secarna. The \nauthors declare that they have no known competing financial in terests or personal \nrelationships that could have appeared to influence the work reported in this paper. \nFunding \nThis research was funded in whole, or in part, by the Anniversary Fund of the National Bank \nof Austria (OeNB) (grant number 18279 to HH). \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \n\n21 \n \nAuthor contributions \nRG conducted all computational analyses, and  LM performed all in vitro  analyses. PML \ndeveloped the R package and application. GF performed antigen prediction and deconvolution \nanalyses. SS performed immunohistochemistry analyses. AGZ served as a clinical consultant. \nCM headed the pilot study from the Medical University in Innsbruck and supervised all clinical \naspects. HF coordinated tumor samples from the biobank and their molecular analysis. HH \nconceived the study, supervised all analyses, and together with RG wrote the paper. All the \nauthors have read and approved the final manuscript. \nEthics approval and consent to participate \nFor the pilot study (validation cohort from the Medical University of Innsbruck), written informed \nconsent was needed for all patients before enrollment. The study was reviewed and approved \nby the Ethics Committee of the Medical University of Innsbruck (reference number: 1189/2019) \nand conducted in accordance with the Declaration of Helsinki. \nAvailability of data and materials \nRNA sequencing data from in vitro  experiments are available via the Gene Expression \nOmnibus (GEO) (GSE237361). RNA sequencing data and patient information of the validation \ncohort (MUI) are available at Zenodo  https://doi.org/10.5281/zenodo.10251467. The R \npackage OvRSeq is available from GitHub (https://github.com/icbi-lab/OvRSeq) under MIT \nlicense and the corresponding web application (http s://ovrseq.icbi.at). The analysis scripts \nused in this manuscript are available at GitHub (https://github.com/icbi-lab/hgsoc). \nReferences \n1.  Siegel RL, Miller KD, Wagle NS, Jemal A. Cancer statistics, 2023. CA Cancer J Clin 2023. \n73:17–48. doi: 10.3322/caac.21763 \n2.  Le DT, Durham JN, Smith KN, Wang H, Bartlett BR, Aulakh LK, Lu S, Kemberling H, Wilt \nC, Luber BS, et al. Mismatch repair deficiency predicts response of solid tumors to PD-1 \nblockade. Science 2017. 357:409–413. doi: 10.1126/science.aan6733 \n3.  Disis ML, Taylor MH, Kelly K, Beck JT, Gordon M, Moore KM, Patel MR, Chaves J, Park \nH, Mita AC, et al. Efficacy and Safety of Avelumab for Patients With Recurrent or \nRefractory Ovarian Cancer: Phase 1b Results From the JAVELIN Solid Tumor Trial. \nJAMA Oncol 2019. 5:393. doi: 10.1001/jamaoncol.2018.6258 \n4.  Varga A, Piha-Paul S, Ott PA, Mehnert JM, Berton-Rigaud D, Morosky A, Yang P, Ruman \nJ, Matei D. Pembrolizumab in patients with programmed death ligand 1 –positive \nadvanced ovarian cancer: Analysis of KEYNOTE -028. Gynecol Oncol 2019. 152:243 –\n250. doi: 10.1016/j.ygyno.2018.11.017 \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \n\n22 \n \n5.  Konstantinopoulos PA, Cannistra SA. Immune Checkpoint Inhibitors in Ovarian Cancer: \nCan We Bridge the Gap Between IMagynation and Reality? JCO 2021. 39:1833 –1838. \ndoi: 10.1200/JCO.21.00571 \n6.  Matulonis UA, Shapira -Frommer R, Santin AD, Lisyanskaya AS, Pignata S, Vergote I, \nRaspagliesi F, Sonke GS, Birrer M, Provencher DM, et al. Antitumor activity and safety \nof pembrolizumab in patients with advanced recurrent ovarian cancer: results from the  \nphase II KEYNOTE -100 study. Ann  Oncol 2019. 30:1080 –1087. doi: \n10.1093/annonc/mdz135 \n7.  Ray-Coquard I, Pautier P, Pignata S, Pérol D, González-Martín A, Berger R, Fujiwara K, \nVergote I, Colombo N, Mäenpää J, et al. Olaparib plus Bevacizumab as F irst-Line \nMaintenance in Ovarian Cancer. N Engl J Med 2019. 381:2416 –2428. doi: \n10.1056/NEJMoa1911361 \n8.  Musacchio L, Cicala CM, Camarda F, Ghizzoni V, Giudice E, Carbone MV, Ricci C, Perri \nMT, Tronconi F, Gentile M, et al. Combining PARP inhibition and i mmune checkpoint \nblockade in ovarian cancer patients: a new perspective on the horizon? ESMO Open \n2022. 7:100536. doi: 10.1016/j.esmoop.2022.100536 \n9.  Färkkilä A, Gulhan DC, Casado J, Jacobson CA, Nguyen H, Kochupurakkal B, Maliga Z, \nYapp C, Chen Y-A, Schapiro D, et al. Immunogenomic profiling determines responses to \ncombined PARP and PD-1 inhibition in ovarian cancer. Nat Commun 2020. 11:1459. doi: \n10.1038/s41467-020-15315-8 \n10.  Lampert EJ, Zimmer A, Padget M, Cimino-Mathews A, Nair JR, Liu Y, Swisher EM, Hodge \nJW, Nixon AB, Nichols E, et al. Combination of PARP Inhibitor Olaparib, and PD -L1 \nInhibitor Durvalumab, in Recurrent Ovarian Cancer: a Proof -of-Concept Phase II Study. \nClin Cancer Res 2020. 26:4268–4279. doi: 10.1158/1078-0432.CCR-20-0056 \n11.  Drew Y, Kim J-W, Penson RT, O’Malley DM, Parkinson C, Roxburgh P, Plummer R, Im \nS-A, Imbimbo M, Ferguson M, et al. Olaparib plus Durvalumab, with or without \nBevacizumab, as Treatment in PARP Inhibitor -Naïve Platinum -Sensitive Relapsed \nOvarian Cancer: A Phase I I Multi -Cohort Study. Clin Cancer Res 2023. doi: \n10.1158/1078-0432.CCR-23-2249 \n12.  Gonzalez Martin A, Rubio Perez MJ, Heitz F, Christensen RD, Colombo N, Van Gorp T, \nOaknin A, Leary A, Gaba Garcia L, Lebreton C, et al. LBA37 Atezolizumab (atezo) \ncombined with platinum -based chemotherapy (CT) and maintenance niraparib for \nrecurrent ovarian cancer (rOC) with a platinum -free interval (TFIp) >6 months: Primary \nanalysis of the double-blind placebo (pbo)-controlled ENGOT-Ov41/GEICO 69-O/ANITA \nphase III trial. Ann Oncol 2023. 34:S1278–S1279. doi: 10.1016/j.annonc.2023.10.031 \n13.  Lord CJ, Ashworth A. PARP inhibitors: Synthetic lethality in the clinic. Science 2017. \n355:1152–1158. doi: 10.1126/science.aam7344 \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \n\n23 \n \n14.  Konstantinopoulos PA, Ceccaldi R, Shapiro GI, D’Andrea AD. Homologous \nRecombination Deficiency: Exploiting the Fundamental Vulnerability of Ovarian Cancer. \nCancer Discov 2015. 5:1137–1154. doi: 10.1158/2159-8290.CD-15-0714 \n15.  Cancer Genome Atlas Research Network. Integrated genomic analyses of o varian \ncarcinoma. Nature 2011. 474:609–615. doi: 10.1038/nature10166 \n16.  Stewart MD, Merino Vega D, Arend RC, Baden JF, Barbash O, Beaubier N, Collins G, \nFrench T, Ghahramani N, Hinson P, et al. Homologous Recombination Deficiency: \nConcepts, Definitions, and Assays. The Oncologist 2022. 27:167 –174. doi: \n10.1093/oncolo/oyab053 \n17.  Takamatsu S, Brown JB, Yamaguchi K, Hamanishi J, Yamanoi K, Takaya H, Kaneyasu \nT, Mori S, Mandai M, Matsumura N. Utility of Homologous Recombination Deficiency \nBiomarkers Across Cancer Types. JCO Precis Oncol 2021. 1270 –1280. doi: \n10.1200/PO.21.00141 \n18.  Konstantinopoulos PA, Spentzos D, Karlan BY, Taniguchi T, Fountzilas E, Francoeur N, \nLevine DA, Cannistra SA. Gene Expression Profile of BRCA ness That Correlates With \nResponsiveness to Chemotherapy and With Outcome in Patients With Epithelial Ovarian \nCancer. JCO 2010. 28:3555–3561. doi: 10.1200/JCO.2009.27.5719 \n19.  Chen T, Yu T, Zhuang S, Geng Y, Xue J, Wang J, Ai L, Chen B, Zhao Z, Li Y, et al. \nUpregulation of CXCL1 and LY9 contributes to BRCAness in ovarian cancer and \nmediates response to PARPi and immune checkpoint blockade. Br J Cancer 2022. doi: \n10.1038/s41416-022-01836-0 \n20.  Kraya AA, Maxwell KN, Wubbenhorst B, Wenz BM, Pluta J, Rech AJ , Dorfman LM, \nLunceford N, Barrett A, Mitra N, et al. Genomic Signatures Predict the Immunogenicity of \nBRCA-Deficient Breast Cancer. Clin Cancer Res 2019. 25:4363 –4374. doi: \n10.1158/1078-0432.CCR-18-0468 \n21.  Hoppe MM, Sundar R, Tan DSP, Jeyasekharan AD. B iomarkers for Homologous \nRecombination Deficiency in Cancer. J Natl Cancer Inst 2018. 110:704 –713. doi: \n10.1093/jnci/djy085 \n22.  Liu J, Lichtenberg T, Hoadley KA, Poisson LM, Lazar AJ, Cherniack AD, Kovatich AJ, \nBenz CC, Levine DA, Lee AV, et al. An Integr ated TCGA Pan -Cancer Clinical Data \nResource to Drive High-Quality Survival Outcome Analytics. Cell 2018. 173:400-416.e11. \ndoi: 10.1016/j.cell.2018.02.052 \n23.  Li H, Handsaker B, Wysoker A, Fennell T, Ruan J, Homer N, Marth G, Abecasis G, Durbin \nR, 1000 Gen ome Project Data Processing Subgroup. The Sequence Alignment/Map \nformat and SAMtools. Bioinformatics 2009. 25:2078 –2079. doi: \n10.1093/bioinformatics/btp352 \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \n\n24 \n \n24.  Andrews, S. A S. FastQC:  A Quality Control Tool for High Throughput Sequence Data. \n2010.  \n25.  Ewels P, Magnusson M, Lundin S, Käller M. MultiQC: summarize analysis results for \nmultiple tools and samples in a single report. Bioinformatics 2016. 32:3047 –3048. doi: \n10.1093/bioinformatics/btw354 \n26.  Dobin A, Davis CA, Schlesinger F, Drenkow J, Zalesk i C, Jha S, Batut P, Chaisson M, \nGingeras TR. STAR: ultrafast universal RNA-seq aligner. Bioinformatics 2013. 29:15–21. \ndoi: 10.1093/bioinformatics/bts635 \n27.  Liao Y, Smyth GK, Shi W. featureCounts: an efficient general purpose program for \nassigning sequence reads to genomic features. Bioinformatics 2014. 30:923 –930. doi: \n10.1093/bioinformatics/btt656 \n28.  Li H, Durbin R. Fast and accurate short read alignment with Burrows-Wheeler transform. \nBioinformatics 2009. 25:1754–1760. doi: 10.1093/bioinformatics/btp324 \n29.  Van der Auwera GA, Carneiro MO, Hartl C, Poplin R, Del Angel G, Levy -Moonshine A, \nJordan T, Shakir K, Roazen D, Thibault J, et al. From FastQ data to high confidence \nvariant calls: the Genome Analysis Toolkit best practices pipeline. Curr Protoc \nBioinformatics 2013. 43:11.10.1-11.10.33. doi: 10.1002/0471250953.bi1110s43 \n30.  Larson DE, Harris CC, Chen K, Koboldt DC, Abbott TE, Dooling DJ, Ley TJ, Mardis ER, \nWilson RK, Ding L. SomaticSniper: identification of somatic point mutations in whole \ngenome sequencing data. Bioinformatics 2012. 28:311 –317. doi: \n10.1093/bioinformatics/btr665 \n31.  Koboldt DC, Zhang Q, Larson DE, Shen D, McLellan MD, Lin L, Miller CA, Mardis ER, \nDing L, Wilson RK. VarScan 2: somatic mutation and copy number alteration discovery \nin cancer by exome sequencing. Genome Res 2012. 22:568 –576. doi: \n10.1101/gr.129684.111 \n32.  Kim S, Scheffler K, Halpern AL, Bekritsky MA, Noh E, Källberg M, Chen X, Kim Y, Beyter \nD, Krusche P, Saunders CT. Strelka2: fast and accurate calling of germline and somatic \nvariants. Nat Methods 2018. 15:591–594. doi: 10.1038/s41592-018-0051-x \n33.  McLaren W, Gil L, Hunt SE, Riat HS, Ritchie GRS, Thormann A, Flicek P, Cunningham \nF. The Ensembl Variant Effect Predictor. Genome Biol 2016. 17:122. doi: \n10.1186/s13059-016-0974-4 \n34.  Szolek A, Schubert B, Mohr C, Sturm M, Feldhahn M, Kohlbacher O. OptiType: precision \nHLA typing from next-generation sequencing data. Bioinformatics 2014. 30:3310 –3316. \ndoi: 10.1093/bioinformatics/btu548 \n35.  Jurtz V, Paul S, Andreatta M, Marcatili P, Peters B, Nielsen M. NetMHCpan-4.0: Improved \nPeptide-MHC Class I Inte raction Predictions Integrating Eluted Ligand and Peptide \nBinding Affinity Data. J Immunol 2017. 199:3360–3368. doi: 10.4049/jimmunol.1700893 \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \n\n25 \n \n36.  Richman LP, Vonderheide RH, Rech AJ. Neoantigen Dissimilarity to the Self -Proteome \nPredicts Immunogenicity and Response to Immune Checkpoint Blockade. Cell Syst 2019. \n9:375-382.e4. doi: 10.1016/j.cels.2019.08.009 \n37.  Love MI, Huber W, Anders S. Moderated estimation of fold change and dispersion for \nRNA-seq data with DESeq2. Genome Biol 2014. 15:550. doi: 10.1186/s13059-014-0550-\n8 \n38.  Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, Paulovich \nA, Pomeroy SL, Golub TR, Lander ES, Mesirov JP. Gene set enrichment analysis: a \nknowledge-based approach for interpreting genome-wide expression profiles. Proc Natl \nAcad Sci U S A 2005. 102:15545–15550. doi: 10.1073/pnas.0506580102 \n39.  Yu G, Wang L-G, Han Y, He Q-Y. clusterProfiler: an R package for comparing biological \nthemes among gene clusters. OMICS 2012. 16:284–287. doi: 10.1089/omi.2011.0118 \n40.  Bindea G, Mlecnik B, Hackl H, Charoentong P, Tosolini M, Kirilovsky A, Fridman W -H, \nPagès F, Trajanoski Z, Galon J. ClueGO: a Cytoscape plug -in to decipher functionally \ngrouped gene ontology and pathway annotation networks. Bioinformatics 2009. 25:1091–\n1093. doi: 10.1093/bioinformatics/btp101 \n41.  Szklarczyk D, Gable AL, Nastou KC, Lyon D, Kirsch R, Pyysalo S, Doncheva NT, Legeay \nM, Fang T, Bork P, et al. The STRING database in 2021: customizable protein -protein \nnetworks, and functional characterization of  user-uploaded gene/measurement sets. \nNucleic Acids Res 2021. 49:D605–D612. doi: 10.1093/nar/gkaa1074 \n42.  Schubert M, Klinger B, Klünemann M, Sieber A, Uhlitz F, Sauer S, Garnett MJ, Blüthgen \nN, Saez-Rodriguez J. Perturbation-response genes reveal signaling footprints in cancer \ngene expression. Nat Commun 2018. 9:20. doi: 10.1038/s41467-017-02391-6 \n43.  Finotello F, Mayer C, Plattner C, Laschober G, Rieder D, Hackl H, Krogsdam A, Loncova \nZ, Posch W, Wilflingseder D, et al. Molecular and pharmacological mod ulators of the \ntumor immune contexture revealed by deconvolution of RNA -seq data. Genome Med \n2019. 11:34. doi: 10.1186/s13073-019-0638-6 \n44.  Sturm G, Finotello F, Petitprez F, Zhang JD, Baumbach J, Fridman WH, List M, Aneichyk \nT. Comprehensive evaluation of transcriptome-based cell-type quantification methods for \nimmuno-oncology. Bioinformatics 2019. 35:i436 –i445. doi: \n10.1093/bioinformatics/btz363 \n45.  Hänzelmann S, Castelo R, Guinney J. GSVA: gene set variation analysis for microarray \nand RNA-Seq data. BMC Bioinformatics 2013. 14:7. doi: 10.1186/1471-2105-14-7 \n46.  Desbois M, Udyavar AR, Ryner L, Kozlowski C, Guan Y, Dürrbaum M, Lu S, Fortin J -P, \nKoeppen H, Ziai J, et al. Integrated digital pathology and transcriptome analysis identifies \nmolecular mediators of T-cell exclusion in ovarian cancer. Nat Commun 2020. 11:5583. \ndoi: 10.1038/s41467-020-19408-2 \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \n\n26 \n \n47.  Chen GM, Kannan L, Geistlinger L, Kofia V, Safikhani Z, Gendoo DMA, Parmigiani G, \nBirrer M, Haibe-Kains B, Waldron L. Consensus on Molecular Subtypes o f High-Grade \nSerous Ovarian Carcinoma. Clin  Cancer Res 2018. 24:5037–5047. doi: 10.1158/1078-\n0432.CCR-18-0784 \n48.  Charoentong P, Finotello F, Angelova M, Mayer C, Efremova M, Rieder D, Hackl H, \nTrajanoski Z. Pan -cancer Immunogenomic Analyses Reve al Genotype -\nImmunophenotype Relationships and Predictors of Response to Checkpoint Blockade. \nCell Rep 2017. 18:248–262. doi: 10.1016/j.celrep.2016.12.019 \n49.  Telli ML, Timms KM, Reid J, Hennessy B, Mills GB, Jensen KC, Szallasi Z, Barry WT, \nWiner EP, Tung NM, et al. Homologous Recombination Deficiency (HRD) Score Predicts \nResponse to Platinum -Containing Neoadjuvant Chemotherapy in Patients with Triple -\nNegative Breast Cancer. Clin  Cancer Res  2016. 22:3764 –3773. doi: 10.1158/1078 -\n0432.CCR-15-2477 \n50.  Alexandrov LB, Kim J, Haradhvala NJ, Huang MN, Tian Ng AW, Wu Y, Boot A, Covington \nKR, Gordenin DA, Bergstrom EN, et al. The repertoire of mutational signatures in human \ncancer. Nature 2020. 578:94–101. doi: 10.1038/s41586-020-1943-3 \n51.  Takaya H, Nakai H, Takamatsu S, Mandai M, Matsumura N. Homologous recombination \ndeficiency status-based classification of high -grade serous ovarian carcinoma. Sci Rep \n2020. 10:2757. doi: 10.1038/s41598-020-59671-3 \n52.  Abkevich V, Timms KM, Hennessy BT, Potter J, Carey MS, Meyer LA, Smith-McCune K, \nBroaddus R, Lu KH, Chen J, et al. Patterns of genomic loss of heterozygosity predict \nhomologous recombination repair defects in epithelial ovarian cancer. Br J Cancer 2012. \n107:1776–1782. doi: 10.1038/bjc.2012.451 \n53.  Birkbak NJ, Wang ZC, Kim J-Y, Eklund AC, Li Q, Tian R, Bowman-Colin C, Li Y, Greene-\nColozzi A, Iglehart JD, et al. Telomeric Allelic Imbalance Indicates Defective DNA Repair \nand Sensitivity to DNA -Damaging Agents. Cancer Discov  2012. 2:366 –375. doi: \n10.1158/2159-8290.CD-11-0206 \n54.  Popova T, Manié E, Rieunier G, Caux -Moncoutier V, Tirapo C, Dubois T, Delattre O, \nSigal-Zafrani B, Bollet M, Longy M, et al. Ploidy and large -scale genomic instability \nconsistently identify basal-like breast carcinomas with BRCA1/2 inactivation. Cancer Res \n2012. 72:5454–5462. doi: 10.1158/0008-5472.CAN-12-1470 \n55.  Sztupinszki Z, Diossy M, Krzystanek M, Reiniger L, Csabai I, Favero F, Birkbak NJ, \nEklund AC, Syed A, Szallasi Z. Migrating the SNP array -based homologous \nrecombination deficiency measures to next generation sequencing data of breast cancer. \nnpj Breast Cancer 2018. 4:1–4. doi: 10.1038/s41523-018-0066-6 \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \n\n27 \n \n56.  Favero F, Joshi T, Marquard AM, Birkbak NJ, Krzystanek M, Li Q, Szallasi Z, Eklund AC. \nSequenza: allele -specific c opy number and mutation profiles from tumor sequencing \ndata. Ann Oncol 2015. 26:64–70. doi: 10.1093/annonc/mdu479 \n57.  Manders F, Brandsma AM, de Kanter J, Verheul M, Oka R, van Roosmalen MJ, van der \nRoest B, van Hoeck A, Cuppen E, van Boxtel R. M utationalPatterns: the one stop shop \nfor the analysis of mutational processes. BMC Genomics 2022. 23:134. doi: \n10.1186/s12864-022-08357-3 \n58.  McDermott JE, Arshad OA, Petyuk VA, Fu Y, Gritsenko MA, Clauss TR, Moore RJ, \nSchepmoes AA, Zhao R, Monroe ME, et al. Proteogenomic Characterization of Ovarian \nHGSC Implicates Mitotic Kinases, Replication Stress in Observed Chromosomal \nInstability. Cell Rep Med 2020. 1:100004. doi: 10.1016/j.xcrm.2020.100004 \n59.  Gulhan DC, Lee JJ-K, Melloni GEM, Cortés-Ciriano I, Park PJ. Detecting the mutational \nsignature of homologous recombination deficiency in clinical samples. Nat Genet 2019. \n51:912–919. doi: 10.1038/s41588-019-0390-2 \n60.  Gayoso A, Lopez R, Xing G, Boyeau P, Valiollah Pour Amiri V, Hong J, Wu K, Jayasuriya \nM, Mehlman E, Langevin M, et al. A Python library for probabilistic analysis of single-cell \nomics data. Nat Biotechnol 2022. 40:163–166. doi: 10.1038/s41587-021-01206-w \n61.  Franzén O, Gan L-M, Björkegren JLM. PanglaoDB: a web server for exploration of mouse \nand human single -cell RNA sequencing data. Database  2019. 2019:baz046. doi: \n10.1093/database/baz046 \n62.  Efremova M, Vento -Tormo M, Teichmann SA, Vento -Tormo R. CellPhoneDB: inferring \ncell-cell communication from combined expressio n of multi -subunit ligand -receptor \ncomplexes. Nat Protoc 2020. 15:1484–1506. doi: 10.1038/s41596-020-0292-x \n63.  Ritchie ME, Phipson B, Wu D, Hu Y, Law CW, Shi W, Smyth GK. limma powers \ndifferential expression analyses for RNA -sequencing and microarray stu dies. Nucleic \nAcids Res 2015. 43:e47. doi: 10.1093/nar/gkv007 \n64.  Lord CJ, Ashworth A. BRCAness revisited. Nat Rev Cancer 2016. 16:110 –120. doi: \n10.1038/nrc.2015.21 \n65.  Takamatsu S, Yoshihara K, Baba T, Shimada M, Yoshida H, Kajiyama H, Oda K, Mandai \nM, Okamoto A, Enomoto T, Matsumura N. Prognostic relevance of HRDness gene \nexpression signature in ovarian high-grade serous carcinoma; JGOG3025-TR2 study. Br \nJ Cancer 2023. 128:1095–1104. doi: 10.1038/s41416-022-02122-9 \n66.  Barbie DA, Tamayo P, Boehm JS, Kim SY, Moody SE, Dunn IF, Schinzel AC, Sandy P, \nMeylan E, Scholl C, et al. Systematic RNA interference reveals that oncogenic KRAS -\ndriven cancers require TBK1. Nature 2009. 462:108–112. doi: 10.1038/nature08460 \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \n\n28 \n \n67.  Binnewies M, Pollack JL, Rudolph J, Dash S, Abushawish M, Lee T, Jahchan NS, \nCanaday P, Lu E, Norng M, et al. Targeting TREM2 on tumor -associated macrophages \nenhances immunotherapy. Cell Rep 2021. 37:109844. doi: 10.1016/j.celrep.2021.109844 \n68.  Cassetta L, Fragkogianni S, Sims AH, Swierczak A, Forrester LM, Zhang H, Soong DYH, \nCotechini T, Anur P, Lin EY, et al. Human Tumor-Associated Macrophage and Monocyte \nTranscriptional Landscapes Reveal Cancer -Specific Reprogramming, Biomarkers, and \nTherapeutic Targets. Cancer Cell 2019. 35:588 -602.e10. doi : \n10.1016/j.ccell.2019.02.009 \n69.  Mezheyeuski A, Backman M, Mattsson J, Martín -Bernabé A, Larsson C, Hrynchyk I, \nHammarström K, Ström S, Ekström J, Mauchanski S, et al. An immune score reflecting \npro- and anti-tumoural balance of tumour microenvironment has major prognostic impact \nand predicts immunotherapy response in solid cancers. EBioMedicine 2023. 88:104452. \ndoi: 10.1016/j.ebiom.2023.104452 \n70.  Mullen MM, Lomonosova E, Toboni MD, Oplt A, Cybulla E, Blachut B, Zhao P, Noia H, \nWilke D, Rankin EB, et al. GAS6/AXL Inhibition Enhances Ovarian Cancer Sensitivity to \nChemotherapy and PARP Inhibition through Increased DNA Damage and Enhanced \nReplication Stress. Mol Cancer Res 2022. 20:265 –279. doi: 10.1158/1541 -7786.MCR-\n21-0302 \n71.  Truxova I, Cibula D, Spisek R, Fucikova J. Targeting tumor-associated macrophages for \nsuccessful immunotherapy of ovarian carcinoma. J Immunother Cancer 2023. \n11:e005968. doi: 10.1136/jitc-2022-005968 \n72.  Wieser V, Tsibulak I, Reimer DU, Zeimet AG, Fiegl H, Hackl H, Marth C. An ang iogenic \ntumor phenotype predicts poor prognosis in ovarian cancer. Gynecol Oncol 2023. \n170:290–299. doi: 10.1016/j.ygyno.2023.01.034 \n73.  Rempel E, Kluck K, Beck S, Ourailidis I, Kazdal D, Neumann O, Volckmar AL, Kirchner \nM, Goldschmid H, Pfarr N, et al. P an-cancer analysis of genomic scar patterns caused \nby homologous repair deficiency (HRD). NPJ Precis Oncol 2022. 6:36. doi: \n10.1038/s41698-022-00276-6 \n74.  Perez-Villatoro F, Oikkonen J, Casado J, Chernenko A, Gulhan DC, Tumiati M, Li Y, \nLavikka K, Hietane n S, Hynninen J, et al. Optimized detection of homologous \nrecombination deficiency improves the prediction of clinical outcomes in cancer. NPJ \nPrecis Oncol 2022. 6:96. doi: 10.1038/s41698-022-00339-8 \n75.  Strickland KC, Howitt BE, Shukla SA, Rodig S, Ritte rhouse LL, Liu JF, Garber JE, \nChowdhury D, Wu CJ, D’Andrea AD, et al. Association and prognostic significance of \nBRCA1/2-mutation status with neoantigen load, number of tumor-infiltrating lymphocytes \nand expression of PD -1/PD-L1 in high grade serous ovaria n cancer. Oncotarget 2016. \n7:13587–13598. doi: 10.18632/oncotarget.7277 \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \n\n29 \n \n76.  Samstein RM, Krishna C, Ma X, Pei X, Lee K-W, Makarov V, Kuo F, Chung J, Srivastava \nRM, Purohit TA, et al. Mutations in BRCA1 and BRCA2 differentially affect the tumor \nmicroenvironment and response to checkpoint blockade immunotherapy. Nat Cancer \n2020. 1:1188–1203. doi: 10.1038/s43018-020-00139-8 \n77.  Ding L, Kim H -J, Wang Q, Kearns M, Jiang T, Ohlson CE, Li BB, Xie S, Liu JF, Stover \nEH, et al. PARP Inhibition Elicits STING -Dependent Antitumor Immunity in Brca1 -\nDeficient Ovarian Cancer. Cell Reports 2018. 25:2972 -2980.e5. doi: \n10.1016/j.celrep.2018.11.054 \n78.  Bruand M, Barras D, Mina M, Ghisoni E, Morotti M, Lanitis E, Fahr N, Desbuisson M, \nGrimm A, Zhang H, et al. Cell -autonomous inflammation of BRCA1 -deficient ovarian \ncancers drives both tumor-intrinsic immunoreactivity and immune resistance via STING. \nCell Reports 2021. 36:109412. doi: 10.1016/j.celrep.2021.109412 \n79.  Zheng J, Mo J, Zhu T, Zhuo W, Yi Y, Hu S, Yin J, Zhang W, Zhou H, Liu Z. Comprehensive \nelaboration of the cGAS -STING signaling axis in cancer development and \nimmunotherapy. Mol Cancer 2020. 19:133. doi: 10.1186/s12943-020-01250-1 \n80.  Ding L, Wang Q, Martincuks A, Kearns MJ, Jiang T, Lin Z, Cheng X, Qian C, Xie S, Kim \nH-J, et al. STING agonism overcomes STAT3 -mediated immunosuppression and \nadaptive resistance to PARP inhibition in ovarian cancer. J Immunother Cancer 2023. \n11:e005627. doi: 10.1136/jitc-2022-005627 \n81.  Konstantinopoulos PA, Matulonis UA. Clinical and translational advances in ovarian \ncancer therapy. Nat Cancer 2023. 4:1239–1257. doi: 10.1038/s43018-023-00617-9 \n82.  Dangaj D, Bruand M, Grimm AJ, Ronet C, Barras D, Duttagupta PA, Lanitis E, \nDuraiswamy J, Tanyi JL, Benencia F, et al. Cooperation between C onstitutive and \nInducible Chemokines Enables T Cell Engraftment and Immune Attack in Solid Tumors. \nCancer Cell 2019. 35:885-900.e10. doi: 10.1016/j.ccell.2019.05.004 \n83.  Kandalaft LE, Dangaj Laniti D, Coukos G. Immunobiology of high -grade serous ovarian \ncancer: lessons for clinical translation. Nat Rev Cancer 2022. 22:640 –656. doi: \n10.1038/s41568-022-00503-z \n84.  Ayers M, Lunceford J, Nebozhyn M, Murphy E, Loboda A, Kaufman DR, Albright A, \nCheng JD, Kang SP, Shankaran V, et al. IFN -γ-related mRNA profile p redicts clinical \nresponse to PD-1 blockade. J Clin Invest 2017. 127:2930–2940. doi: 10.1172/JCI91190 \n85.  Johnson RL, Cummings M, Thangavelu A, Theophilou G, de Jong D, Orsi NM. Barriers \nto Immunotherapy in Ovarian Cancer: Metabolic, Genomic, and Immune Perturbations in \nthe Tumour Microenvironment. Cancers (Basel) 2021. 13:6231. doi: \n10.3390/cancers13246231 \n86.  Molgora M, Esaulova E, Vermi W, Hou J, Chen Y, Luo J, Brioschi S, Bugatti M, Omodei \nAS, Ricci B, et al. TREM2 Modulation Remodels the Tumor Myeloid Landscape \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \n\n30 \n \nEnhancing Anti -PD-1 Immunotherapy. Cell 2020. 182:886 -900.e17. doi: \n10.1016/j.cell.2020.07.013 \n87.  Wang Y, He M, Zhang C, Cao K, Zhang G, Yang M, Huang Y, Jiang W, Liu H. Siglec-9+ \ntumor-associated macrophages delineate an immunosuppressive subset with therapeutic \nvulnerability in patients with high -grade serous ovarian cancer. J Immunother Cancer \n2023. 11:e007099. doi: 10.1136/jitc-2023-007099 \n88.  Li C, Deng T, Cao J, Zhou Y, Luo X, Feng Y, Huang H, Liu J. Identifying ITGB2 as a \nPotential Prognostic Biomarker in Ovarian Cancer. Diagnostics  2023. 13:1169. doi: \n10.3390/diagnostics13061169 \n89.  Haake M, Haack B, Schäfer T, Harter PN, Mattavelli G, Eiring P, Vashist N, Wedekink F, \nGenssler S, Fischer B, et al. Tumor -derived GDF-15 blocks LFA -1 dependent T cell \nrecruitment and suppresses responses to anti -PD-1 treatment. Nat Commun 2023. \n14:4253. doi: 10.1038/s41467-023-39817-3 \n90.  Curiel TJ, Coukos G, Zou L, Alvarez X, Cheng P, Mottram P, Evdemon-Hogan M, Conejo-\nGarcia JR, Zhang L, Burow M, et al. Specific recruitment of regulatory T cells in ovarian \ncarcinoma fosters immune privilege and predicts  reduced survival. Nat Med 2004. \n10:942–949. doi: 10.1038/nm1093 \n91.  Wolf D, Wolf AM, Rumpold H, Fiegl H, Zeimet AG, Muller -Holzner E, Deibl M, Gastl G, \nGunsilius E, Marth C. The expression of the regulatory T cell -specific forkhead box \ntranscription fact or FoxP3 is associated with poor prognosis in ovarian cancer. Clin \nCancer Res 2005. 11:8326–8331. doi: 10.1158/1078-0432.CCR-05-1244 \n92.  Galon J, Costes A, Sanchez-Cabo F, Kirilovsky A, Mlecnik B, Lagorce-Pagès C, Tosolini \nM, Camus M, Berger A, Wind P, et al. Type, density, and location of immune cells within \nhuman colorectal tumors predict clinical outcome. Science 2006. 313:1960 –1964. doi: \n10.1126/science.1129139 \n93.  Chan TA, Yarchoan M, Jaffee E, Swanton C, Quezada SA, Stenzinger A, Peters S. \nDevelopment of tumor mutation burden as an immunotherapy biomarker: utility for the \noncology clinic. Annals of Oncology 2019. 30:44–56. doi: 10.1093/annonc/mdy495 \n94.  Almeida-Nunes DL, Mendes-Frias A, Silvestre R, Dinis-Oliveira RJ, Ricardo S. Immune \nTumor Microenvironment in Ovarian Cancer Ascites. Int J Mol Sci 2022. 23:10692. doi: \n10.3390/ijms231810692 \n \n  \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \n\n31 \n \nFigure legends \nFig. 1  BRCAness classification based on the expression of 24 genes. A Decision tree of \nBRCAness determination in the TCGA-OV cohort and the development of a gene expression-\nbased BRCAness classifier. B Different BRCAness parameters in the TCGA cohort compared \nbetween the HRD score and the mutation signature 3 ratio. Samples with mutated homologous \nrecombination repair pathway genes are marked in red, BRCA1/2 promoter methylation in blue \nand samples with an HRD score > 63 and /or a signature 3 ratio > 0.25 but no mutation or \nBRCA1/2 promoter methylation are marked in yellow. Samples without BRCAness are marked \nin white. C Z scores of log2(TPM+1) normalized expression of the 24 genes of the BRCAness \nsignature in the TCGA cohort as a heatmap clustered by BRCAness and non -BRCAness \nsamples. D Mean ROC curve with 10-fold cross-validation of the classifier tested on the TCGA \ndataset. E Confusion matrices with correctly and incorrectly classified instances when the \nclassifier was tested in independent test cohorts of single-cell RNA sequencing and bulk RNA \nsequencing data. \nFig 2  Association between BRCAness and immune parameters. A Results of correlation \nanalysis of selected immune signatures and BRCAness parameters in the TCGA -HGSOC \ncohort (CYT, cytolytic activity; CTL, cytotoxic T  lymphocytes; IFNG, interferon gamma \nsignature; HRR mutations, mutations in the homologous recombinatio n repair pathway; \nNeoAG load, neoantigen load; TMB, tumor mutational burden); white dots indicate significance \n(FDR<0.1). B Direct comparison of selected immune parameters between BRCAness and \nnoBRCAness samples with significant differences, Wilcoxon rank -sum test (FDR<0.1). C, D \nKaplan‒Meier curve for BRCAness and CD8 T -cell infiltration in the TCGA cohort. E, F \nWaterfall plot of normalized enrichment scores (NES) for the footprint analysis of immune -\nrelated pathways with PROGENy between BRCAness and non-BRCAness samples in the MUI \n(Medical University of Innsbruck) and TCGA cohorts. \nFig 3 Results from cell line experiments with olaparib treatment A Top up- and downregulated \ngenes for the cell lines OVCAR3 and UWB1.289 under olaparib treatment when compared to \nDMSO control. B General distribution of up- and downregulated genes after olaparib treatment \ncompared to the DMSO control in both cell lines as volcano plots. Red indicates significantly \nupregulated genes (FDR<0.1, log 2-fold change>1) , and blue indicates significantly \ndownregulated genes (FDR<0.1, log 2-fold change< -1). C Normalized enrichment score of \npathways associated with activation of the cGAS STING pathway in BRCA1 mutated (cell \nlines) and BRCAness samples (cohorts) as well as olaparib -treated cell  lines. D ClueGO \nnetwork indicating overrepresented biological processes in the olaparib-treated UWB1.289 cell \nline. E Immunofluorescence staining of the DNA damage marker γH2AX in OVCAR3 and \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \n\n32 \n \nUWB1.289 cell lines with and without olaparib treatment. Comparing the different response to \nPARPi treatment between BRCA1 mutation and wild type BRCA1 F Immunofluorescence \nstaining of cGAS, double stranded DNA (dsDNA) and STING in the OVCAR3 and UWB1.289 \ncell line comparing the difference between BRCA1 mutation and wild type BRCA1. \nFig 4 Profiles of immune parameters in the TCGA HGSOC cohort A Heatmap of z scores of \nlog2(TPM+1) expression of immune -related genes and fr action of tumor infiltrating immune \ncells assessed with quanTIseq in all samples (n=226) from the TCGA -HGSOC cohort \ncategorized by BRCAness, tumor -immune phenotype, molecular subtype and BRCA1/2 \nmutation. Furthermore, BRCAness samples are stratified into B RCAness immune type \nsamples (BRIT), which show an immunoreactive molecular subtype and an infiltrated tumor -\nimmune phenotype, and noBRIT samples , which only have BRCAness but do not fulfill the \nother two requirements. B Comparison of BRIT and noBRIT sample s showing infiltration of \nCD8+ T cells, tumor-associated M2 macrophages (TAMs), regulatory T cells (Tregs) assessed \nwith quanTIseq and myeloid -derived suppressor cells (MDSCs) assessed as the z score for \nMDSCs form the immunophenoscore (IPS). \nFig 5 Single cell analysis of 29 ovarian cancer samples A UMAP showing the different cell \ntypes in of the ovarian cancer samples and which cells and cell types are associated with \nBRCAness samples. B UMAP plots of the myeloid cell compartment showing the association \nof macrophages with BRCAness cells and the expression of the macrophage marker gene \nC1QA and the TAM marker gene TREM2 especially in cell clusters associated with BRCAness. \nC Heatmap of expression of macrophage associated marker genes in the different cell types \nin the myeloid cell compartment. D Association of genes identified as differentially expressed \nbetween responder and non -responder to PARPi -immune checkpoint inhibition  combination \ntherapy (niraparib and pembrolizumab) with different cell types and BRCAness. Genes \ndepicted in blue are downregulated in responders while genes depicted in red are upregulated \nin responders. E UMAP visualization of LYZ and LILRB4 genes highly expressed in the myeloid \ncompartment, which are associated with BRCAness and non-responders. F Distribution of the \nexpression of ITGB2 in the different cell types, which is more highly expressed in non -\nresponders and BRCAness. \nFig 6 Expression profiles in the MUI cohort, immunohistochemistry validation, and vulnerability \nmap A Heatmap of z-scores log2(TPM+1) expression of immune related genes and fraction of \ntumor infiltrating immune cells assessed with quanTIseq in all samples (n=60) from the MUI \ncohort categorized by BRCAness, tumor-immune phenotype, molecular subtype and BRCA1/2 \nmutation. B Immunohistochemistry images stained for CD8, CD4, CD163, FOXP3, γH2AX, \nand STING for three selected patients from the MUI cohort. Two BRIT samples one with a \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \n\n33 \n \nBRCA1 mutation and one without and one other sample without BRCAness, a deserted tumor-\nimmune phenotype and a differentiated molecular subtype. C Vulnerability map showing the \nratio between cytolytic activity CYT and C1QA (C2C) on the x-axis and the BRCAness score \non the y-axis coloured by the vulnerability score. The three selected samples  were mapped \nto the vulnerability map based on their CYT to C1QA ratio (C2C) and BRCAness score.\n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \n\nHRR mutation\nHRD>63\nMutSig3 ratio>0.25\nBRCA1/2 meth>0.7\nN\nN\nN\nN\nY\nY\nY\nY\nBRCA1, BRCA2\nRAD51D, RAD51C\nBLM, MRE11, ATM\nBRIP1, PALB2\nHGSOC TCGA\n     n = 226\nC>A      T>G\nMutational Sign. 3\nLOH TAI LST\nBRCA1\nBRCA2\n20 40 60 80 100\n0.0\n0.1\n0.2\n0.3\n0.4\n0.5\n0.6\nHRD\nMutSig3 ratio\nGenes expressed \nin HGSOC cells  \n(scRNA-seq) \nTCGA data set \n(bulk RNA-seq) \nMachine learning \nRecursive feature elimination \nRandom forest classification \nExpression in\n24 gene signature\n0.0 0.2 0.4 0.6 0.8 1.0\n0.0\n0.2\n0.4\n0.6\n0.8\n1.0\nFalse positive rate\nTrue positive rate\nTCGA  (10-fold cross validation)\nMean ROC\nAUC = 0.91   0.04+-\n13\n122\n2BRCAness+\nBRCAness-\nBRCAnessnoBRCAness\nSingle cell RNA-seq\n18\n53\n3BRCAness+\nBRCAness-\nBulk RNA-seq (MUI)\nTSPAN1\nTCEA3\nFGGY\nZSWIM4\nA B\nC\nD E\nBRCAness\nDCAF15\nPTDSS1\nDAPK3\nNUDT15\nCCAR2\nRAD17\nLTA4H\nMT1M\nPLGRKT\nSNRPA1\nDNAJC25\nMRS2\nUSP28\nFZD4\nCCDC90B\nPRCP\nDERL3\nNAPRT\nGPAA1\nCRABP2\nnoBRCAness\nFigure 1\nz-score\n-1.0 1.00.0-0.5 0.5\nnoBRCAness\nHRD>63, MutSig3 ratio>0.25\nHRR gene mutation (cat4, cat5)\nBRCA1/2 promoter methylation (beta>0.7)116 BRCAness- 110 BRCAness+\n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \n\nFigure 2\n< 0.001\n< 0.01\n< 0.1\nExpanded immune\nA\nInflamed\nINFG\nCD8 exhaustion\nCTL\nCYT\nCD8\nHRR mutation\nBRCAness\nHRD score\nMutSig3 ratio\nNeoAG load\nTMB\nBRCA1 methylation\nBRCA2 methylation\nT cell exclusion\nExpanded immune\nInflamed\nINFG\nCD8 exhaustion\nCTLCYTCD8\nHRR mutation\nBRCAnessHRD scoreMutSig3 ratioNeoAG load\nTMB\nBRCA1 methylationBRCA2 methylation\nT cell exclusion\nFDRCorrelation\n1.0\n0.0\n-1.0\n10k\n5k\n0k\n-5k\np=0.0014\nBRCAnessnoBRCAness\nInflamed\n0\n2\n4\n6\n8INFG\np=0.004\n8\n0\n4Neoantigen load\np<0.001\nB\nBRCAnessnoBRCAness\np<0.00130\n20\n10\n0\nTMB (mut/MB)\nC D\nTime (months)\nOverall survival (probability)\n0.0\n0.2\n0.4\n0.6\n0.8\n1.0\nlogrank p<0.001\nHR=0.50 (0.34−0.69)\nBRCAness (n=106;ev=55)\nnoBRCAness (n=119;ev=77)\nNo. at risk\nBRCAness:\nnoBRCAness:\n106\n119\n74\n68\n58\n31\n30\n14\n17\n7\n11\n5\n4\n2\n2\n0\nBRCAness \nTime (months)\nOverall survival (probability)\n0.0\n0.2\n0.4\n0.6\n0.8\n1.0\n0 20 40 60 80 100 120 140\nlogrank p=0.019\nHR=0.67 (0.47−0.93)\nhi CD8 (n=112;ev=59)\nNo. at risk\nhi CD8:\nlo CD8:\n112\n113\n72\n70\n48\n41\n26\n18\n17\n7\n12\n4\n6\n0\n2\n0\nCD8 T cell infiltration \nlo CD8 (n=113;ev=73)\n0 20 40 60 80 100 120 140\np53 TGFbMAPK\nAndrogen\nPI3K\nEstrogen\nTrailWNT VEGF EGFRHypoxia TNFaNFkB\nJAK-STAT\n0.0\n4.0\n8.0\n-4.0 NES\n0.0\n4.0\n8.0\n-4.0 NES\np53TGFb MAPK\nAndrogen\nPI3K\nEstrogen\nTrailWNT VEGFEGFR Hypoxia TNFaNFkB\nJAK-STAT\nE FBRCAness vs. noBRCAness \n(MUI)\nBRCAness vs. noBRCAness \n(TCGA)\n2.0\n-2.0\n1.0\n0.0\n-1.0\nNES\n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \n\nFigure 3\nA\nSAA1\nCASC19\nTNFRSF9\nFOSL1\nOAS2\nPECAM1\nTLR2\nCXCL8\nCXCL10\nPTX3\nKLKP1\nTNIP3\nSPRY4\nNKX3-1\nDKK1\nTNF\nPADI1\nCNTNAP1\nIL7R\nPDZK1IP1\nCXCL5\nBIRC3\nCXCL1\nNTRK2\nPTPRE\nCCL20\nTNFSF14\nFST\nSPTSSB\nALX1\nLRATD1\nUSP51\nVIT\nPRR15L\nDYSF\nGAS1\nKCNIP1\nNEK5\nKCNJ5\nADAMTS8\nDKK2\nPALMD\nGGT6\nCRB2\nLAMA4\nMETTL7A\nTSHR\nPDE11A\nSERPINF2\nWSCD2\nBMP3\nSMIM43\nCXCL10\nCCL20\nLY6E\nNTRK2\nSEPTIN3\nIFIT1\nRSAD2\nMSI1\nMX1\nIFIT3\nCMPK2\nGBP4\nXAF1\nMAP6\nOAS2\nIFI44L\nIL7R\nDYRK3\nCLDN10\nIRX3\nSLC15A3\nOASL\nSAA1\nMX2\nCXCL3\nHERC5\nLGALS9\nSMAD9\nTCIRG1\nE2F2\nCCDC88B\nSCX\nSLITRK5\nH1-1\nPBX1\nFAM86JP\nKLHL14\nWNT5A\nH1-4\nH1-3\nCACNA2D3\nFUT9\nMETTL7A\nGUCY1B1\nSSBP2\nCILP\nCAPS\nSMARCD3\nSELENBP1\nGGT6\nEGR1\nRBBP8NL\nOVCAR3 UWB1.289\nlog2 fold change\n-2.0           0.0             2.0\nOlaparib vs. DMSO\nOVCAR3 UWB1.289\nBRCA1(+/+) BRCA1(-/-)\nOVCAR3 UWB1.289\nBRCA1(+/+) BRCA1(-/-)\n�H2AX cGAS\n�H2AX\ndsDNA\nSTING\n+Olaparib (PARPi)\nINTERFERON ALPHA RESPONSE\nCORE NFKB PATHWAY\nIRF3 TARGETS\nSTING SIGNALING\nUWB vs. OVCAR3UWB\n1\nOVCAR3\n1  \nMUI\n2\nTCGA\n2  \n1 Olaparib vs. DMSO\n2 BRCAness vs. noBRCAness\nFDR<0.1\nFDR>0.1\npositively enriched\nnegatively enriched\n|NES|=3 |NES|=2 |NES|=1\nUWB1.289 Olaparib\nOVCAR3 Olaparib\nC\nD\nE FB\n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \n\nFigure 4\nB\nA\nCytotoxic\nB cells\nMacrophages M1\nCD8+ T cells\neffectors\nImmune\nresponse\nAntigen\npresentation\nImmune\nIFNG\nT cell\nTGFb\nImmune cell\n(quanTIseq)\nprocessing &\ncheckpoints\nexhaustion\nsignature\nCAF marker\ninfiltration\nBRCAness\nTumor immune phenotype\nMolecular subtype\nBRCA1/2 mutation\nBRCAness\nTumor immune\nphenotype\nMolecular\nBRCA1/2\nZ-score\nsubtype\nmutation\nFraction\nCD8A\nGZMA\nIFNG\nICAM3\nJAK3\nB2M\nCD74\nHLA-A\nHLA-E\nTAP1\nBTLA\nCD274\nCTLA4\nIDO1\nLAG3\nPDCD1\nSTAT1\nCXCL10\nHAVCR2\nTOX\nDUSP4\nTIGIT\nCD38\nBMP4\nFAP\nOGN\nPDPN\nMacrophages M2\nMonocytes\nNeutrophils\nNK cells\nCD4+ T cells\nTregs\nMyeolid DCs\nBRCAness\nnoBRCAness\nDesert\nExcluded\nInfiltrated\nUnclassified\nDIF\nIMR\nMES\nDIF\nMutation\nNo mutation\n4.0\n2.0\n0.0\n-2.0\n-4.0\n0.15\n0.10\n0.05\n0.00\nimmune type\nBRIT\nnoBRIT\nBRCAness\nBRCAness\nimmune type\n0.00\n0.05\n0.10\n0.15\n0.20\n0.00\n0.02\n0.04\n0.06\n0.08\n0.00\n0.01\n0.02\n0.03\n0.04\n0.00\n0.50\n1.00\nBRIT\nnoBRIT noBRIT\nBRIT BRIT BRIT\nnoBRIT noBRIT\n0.25\n0.75\nCD8+ T cells\nMacrophages M2 (TAM)\nRegulatory T cells (Tregs)\nMDSCs*\n*z-score\nfrom IPS\np<0.001p<0.001p<0.001p<0.001\n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \n\nFigure 5\nA\nMyeloid cells\nB cells\nCancer cells\nPlasma cells\nT/NK cells\nDendritic cells\nEndothelial cells\nFibroblasts\nMast cells\nUMAP1UMAP2\nOther macrophages\nCD169 macrophages\nOther myeloid cells\nCX3CR1 macrophages\nMyeolid dendritic cells\nMARCO macrophages\nMonocytes\nBRCAness\nnoBRCAness\nBRCAness\nnoBRCAness\nBRCAness\nMyeolid cellsCell types\nBRCAness\nUMAP1UMAP2\nBRCAness\nnoBRCAness\nMyeloid cells\nB cells\nCancer cells\nPlasma cells\nT/NK cells\nDendritic cells\nEndothelial cells\nFibroblasts\nMast cells\nTNFRSF11B\nWNT5BSIRPB2\nLYZ\nCOL4A5HDAC4TGFB2CXCL9GBP2IL18CCNOLILRB4CXCL2ITGB3HES1NFAM1OAS3ITGALGBP4NKG7CCL18FCGR2BITGB2NOTCH2IL2RBLILRB2CMKLR1\nTREM2\n8\n6\n4\n2\n0\nUMAP1UMAP2\n8\n6\n4\n2\n0\n10\nC1QA\nOther macrophages\nCD169 macrophages\nOther myeloid cells\nCX3CR1 macrophages\nMyeolid dendritic cells\nMARCO macrophages\nMonocytes\nTherapy response (R vs. NR)1: Upregulated genes, downregulated genes \nB\nC\nD\nF\nAIF1\nITGB2\nC1QC\nC1QA\nC1QB\nPLTP\nVSIG4\nTREM2\nCCL4\nCCL3\nLILRB4\nLAIR1\nPSMB9\nCD83\nCD4\nHCST\nCLEC7A\nLAP3\nLST1\nGBP1\nTNFAIP3\nFCN1\nCCL8\nHLA-DOA\nCD40\nFGD2\nIRF8\nGBP4\nDOK2\nCCL2\nGBP5\nTCN2\nTLR7\nGIMAP6\nVAV1\nFOLR2\nSIGLEC1\n7.5\n5.0\n2.5\nUMAP1\nUMAP2\nUMAP1\nUMAP2\nSEMA3FLGALS3 SEMA3CLGALS9\nE\nCancer cells\nHLA-F\nLILRB1JAML\nCD169 macrophages\nMARCO macrophages\nCX3CR1 macrophages\nOther macrophages\nOther myeolid cells\nMonocytes\nEndothelial cells\nMyeolid dendritic cells\nDendritic cells\nFibroblasts\nNK cells\nRegulatory T cells\nCD8 T cells\nCD4 T cells\nMast cells\nB cells\nPlasma cells\nNRP1\nCD320VEGFBVEGFBTHY1CD99JAG1CD58C3HLA-EVEGFAVEGFAANXA1PDGFCAPP APPJAG1IGFBP1HLA-FCSF1GAS6PDGFAVEGFA VEGFA ANXA1GAS6BST2 THY1HEBP1\nFLT1\nADGRE5\nPILRACD46CD2\nC3AR1KLRC1NRP1NRP2NRP2FPR3\nPDGFRAMERTKSORL1CD74\nNOTCH4TMEM219\nLILRB2CSF1R\nAXL\nPDGFRA1\nFLT1KDR\nLILRA4HAVCR2\nFPR3 FPR3MERTKNRP2\nADGRE2\nlog CPM\n8\n4\n0\nlog CPM\n8\n4\n0\nFDR\n<0.1\nLILRB4\n8\n6\n4\n2\n0\n10\nLYZ\nITGB2\nMyeloid cells\nB cells\nCancer cellsPlasma cellsT/NK cells\nDendritic cellsEndothelial cells\nFibroblastsMast cells\n8\n6\n4\n2\n0\n8\n6\n4\n2\n0\nUMAP1\nUMAP2\n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint \n\nFigure 6\nA\n�H2AX\nSTING\nCD8\nCD4\nCD163\nCYT/C1QA score\nBRCAness score\nBRCA1 mutation\nBRCAness\nMolecular subtype\nImmunophenotypeInfiltrated Desert\nIMR DIF\n- -\n+ + -\n+\nIMR DIF\nInfiltrated\nB\nC\nBRCAness\nTumor immune\nphenotype\nMolecular\nBRCA1/2\nZ-score\nsubtype\nmutation\nFraction\nBRCAness\nnoBRCAness\nDesert\nExcluded\nInfiltrated\nUnclassified\nDIF\nIMR\nMES\nPRO\nMutation\nNo mutation\n2.0\n1.0\n0.0\n-1.0\n-2.0\n0.15\n0.10\n0.05\n0.00\nBRIT\nnoBRIT\nBRIT\nBRIT\nB\nM1\nM2\nCD4\nCD8\nNK\n1.0\nVulnerability\nscore\n0\n3\nARG1\nCCL5\nCCR5\nITGAE\nCD163\nCD19\nCD2\nCD27\nCD274\nCD276\nCD3D\nCD3E\nCD8A\nCD8B\nCMKLR1\nCTLA4\nCXCL1\nCXCL8\nCXCL9\nCXCL10\nCXCL11\nCXCL13\nCXCR4\nCXCR6\nEOMES\nFASLG\nFGFBP2\nFOXP3\nGNLY\nGZMA\nGZMB\nGZMK\nHLA-DRA\nHLA-DRB1\nHLA-E\nICOS\nIDO1\nIFNG\nIL2RG\nIL6\nIL10\nIRF1\nITGAL\nKLRC3\nKLRC4\nKLRD1\nLAG3\nLY9\nNKG7\nPDCD1\nPRF1\nPDCD1LG2\nPSMB10\nPTPRC\nSLAMF6\nSTAT1\nTAGAP\nTBX21\nTIGIT\nTLR3\nTLR4\nVEGFA\nVTCN1\nZAP70\n0.0\n0.5\n1.0\n0.0 0.5 1.0\n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted January 10, 2024. ; https://doi.org/10.1101/2024.01.09.24301038doi: medRxiv preprint","source_license":"CC-BY-4.0","license_restricted":false}