Potent immunogenicity in BRCA1‐mutated patients with high‐grade serous ovarian carcinoma

Abstract High‐grade serous ovarian carcinomas (HGSOCs) were among the tumours with an unsatisfactory outcome of immune checkpoint inhibitors (ICIs). It is imperative to develop feasible biomarker for identifying responsive candidates and guiding precise immunotherapy for HGSOC patients. Here, we analysed genomic data of patients with HGSOCs to depict their immunological phenotype of tumour microenvironment (TME) and figure out the major determinants of immunogenicity. In comparison with other solid tumours, we observed the lowest levels of PD‐L1, total mutation burden (TMB) and cytolytic molecules in HGSOCs. Surprisingly, TMB is not certainly positively related to tumour immune response as it failed to predict the response to ICIs in a considerable portion of patients in previous clinical trials. By a machine learning approach in search of biomarkers for immunotherapy implications for HGSOCs, we identified the ten most dominant factors determining the immunogenicity of HGSOCs. Interestingly, we found that BRCA1 mutated tumours presented a potent immunogenic phenotype, independent of TMB, meeting the criteria of both our dominant factors and the determinants of immunogenicity established before. Our findings provide evidence that BRCA1‐mutation may be served as a predictive biomarker in guiding ICI therapies for the patients with HGSOCs.


| INTRODUCTION
Cancer immunotherapy, such as Nivolumab and Pembrolizumab, has been recently identified as a promising treatment across a number of solid tumour types, and its clinic licensing progressively expedited as standard-of-care for patients. 1 However, in comparison with other FDA-approved tumour types, the observed clinic benefits from immune checkpoint inhibitors (ICIs) for patients with high-grade serous ovarian cancers (HGSOCs) remain unsatisfying. 2 In a phase I trial evaluating the anti-tumour efficacy of Nivolumab, despite limitations of the small cohort size, the objective response rate (ORR) was 15% and disease control rate was 45% in patients with platinum-resistant ovarian cancer. 3 In KEYNOTE-28 trial which exploring the activity of Pembrolizumab in several solid tumours, outcome of ovarian cancer have been disclosed-the ORR was 11.5%, and only 23.1% showed tumour shrinkage from baseline for patients failed to prior chemotherapy. 4 Such response rate is fuelling an urgent need to develop novel biomarkers that enable the efficient selection of patients with greater likelihood to benefit from the ICI treatment.
Growing evidence support the critical role of immunophenotype of tumour microenvironment (TME) in predicting therapeutic response to immunotherapies. Assessment by immunohistochemistry staining of PD-L1 expression is a logical biomarker to filter populations with favourable prognosis to anti-PD-1 or anti-PD-L1 therapies. 5 Besides the PD-L1 expression, patients with higher expression of T-effector interferon-c signature, 6,7 expanded T cell repertoire and higher tumour-infiltrating lymphocyte density were seen with improved disease control. 8,9 All of the above characteristics were considered because of the immunogenicity of TME. The immunogenicity was initiated by the somatic mutations in cancer cells, which likely produced a spectrum of tumour-specific neoantigens to provoke the cytolytic activity. Thus, theoretically, the total mutation burden (TMB) should be able to determine the patients' responsiveness to ICIs. Consistently, ICIs showed to be more effective against tumour types with higher mutation load, such as non-small cell lung cancer and melanoma, implying a favourable clinical prognosis correlated with an increased capacity to generate neoantigens for immune eradication. However, for patients with HGSOCs, whether the TMB is in parallel with their immunogenicity and what aspects in TME determine the immunogenicity in specific remain less clear.
Here, we aimed to characterize the tumour immune microenvironment of HGSOCs by analysing public accessible data on The Cancer Genome Atlas (TCGA). By a machine learning approach, a panel of dominant parameters in determining the immunogenicity was identified. Besides, by comparing the presentation of dominant parameters among specific mutations subgroups, BRCA1-mutated tumours were found with higher levels of potent immunotherapyresponsive signatures, providing a rational for the ICIs treatment for BRCA1-mutation carriers with HGSOCs.

| Gene signatures and scoring for infiltration/ activity levels with ssGSEA
To determine the degree of immune cell infiltration in tumours, we used a previously described and validated computational technique. 10 In brief, this method applies expression-based gene signa-

ory T cells, central memory T cells and CD8 + T cells. This algorithm
has been orthogonally validated in samples studied with immunofluorescence staining, with high rates of concordance.

| Random forest classification
A random forest classification approach based on a multitude of decision trees was introduced. 782 parameters were input to separate tumours with high cytolytic activity from those with low cytolytic activity, and during the procedure, to sort these parameters according to their importance in determining the cytolytic activity.

| Lower TMB, PD-L1 expression and cytolytic molecules in HGSOCs
Pre-existent evidence supported that PD-L1 expression and TMB could be a manifestation of the responsiveness to ICIs in solid tumours. 11,12 To evaluate these immune phenotypes of HGSOC, we employed the genomic data of HGSOC from TCGA public database.
Genomic data from 6 other solid tumour types, which had been licensed by FDA to receive the anti-PD-1/PD-L1 therapy (SKCM, skin cutaneous melanoma; KIRC, kidney renal clear cell carcinoma; LUSC, squamous cell carcinoma of the lung; LUAD, lung adenocarcinoma; HNSC, head and neck squamous cell carcinoma; BLCA, bladder carcinoma) were also retrieved. As illustrated in Figure 1A, the established predictive biomarkers, both TMB and PD-L1 expression presented the lowest level in HGSOC among the tumour types studied.
Besides TMB and PD-L1 expression, the infiltration of cytotoxic T lymphocyte (CTL) and their major mediator interferon-c (IFNc) were also identified enriched in patients responsive to immunotherapy. 13,14 Thus, we then analysed the relative infiltration level of activated CD8 + T cells by the evaluation of expression profiling data as described previously. 15

| TMB alone could not determine the magnitude of cytolytic immune response
Increasing evidence has demonstrated TMB as the predicative biomarkers across a variety of solid tumours, including non-small-cell lung cancer (NSCLC) and melanoma. 17,18 Additionally, patients with deficient mismatch repair (dMMR), as a surrogate of higher TMB, showed more than 50% of ORR in eighteen solid tumour types. 19 Therefore, we hypothesize that TMB might be the indicator of high immunogenicity for HGSOC as well. To test this, cytolytic activity (CYT) (calculated as the geometric mean of GZMA and PRF1, a scor Previous studies have proven that, besides the CYT, immune cell infiltration in tumour microenvironment(TME) is also closely tied with the level of immune activity. 21 Thus, we analysed the association of TMB and immune cells infiltration. To this end, single-sample gene set enrichment analysis (ssGSEA) was employed to evaluate the cellular infiltration landscape using RNA-sequencing data from bulk tissue for individuals, a method which transformed the transcriptomic expression data (TPM) into normalized scores depicting the relative abundance of specific cell types. 10,16 By an unsupervised algorithm, HGSOC samples can be categorized into 3 clusters based on their spectrum of immune cell infiltration, designated as infiltration high, medium and low ( Figure 2B). Coincide with our former observations, TMB level showed no statistic differences among the 3 groups ( Figure 2C). Consistently, immune cell infiltrations were neither significantly distinct ( Figure 2D) when we divided the samples into TMB High and Low groups (median TMB as cut-off).
Our above results demonstrated that TMB failed to reflect the immunogenicity of HGSOCs whether inspecting from the magnitude of CYT or immune cells infiltration. These results suggest that TMB might not be a valid predictive biomarker for HGSOC immunotherapies.

| Dominant immunological factors in optimizing cytolytic activity beyond TMB
Following the observation that no significant correlation between TMB and CYT in HGSOC, we aim to determine elements that   Figure 3C). These results indicated that the immunophenotype of our machine learning categorization clearly distinguished, in other word, differentiating namely immunological "hot" and "cold" tumour microenvironment.
To identify the most important determinant factors discriminating immunological "hot" or "cold" HGSOCs, we calculated the important score (measured as the mean decrease in accuracy) overall cross-validated predictions. We found that the top ten vital features are the expression of CD96 molecule, the pathway of immunoregulatory interactions between lymphoid and non-lymphoid cells, activated

| Higher immunotherapy-responsive signatures for patients harbouring BRCA1-mutation
Specific oncogenic alterations led to substantial different sensitivity to immunotherapies. 22 Thus, we wondered if certain genomic muta-      To discover novel biomarkers for ICI therapies, HGSOC samples were classified according to a machine learning approach on the basis of levels of cytolytic activity. Through this method, we also filtrated the top ten key factors that matters in this classification. Top 3 factors are CD96, the pathway of immunoregulatory interactions between lymphoid and non-lymphoid cells and activated CD8 + T cell.
CD96 may play a role in the adhesive interactions of activated T cells and NK cells during the late phase of the immune response. It may also function in antigen presentation. Activated CD8 + T cell had been proven to be an important prognostic factor in ovarian cancers and many other tumour types. 26 Interestingly, we found that BRCA1-mutated tumours expressed a high level of the factors we emphasized in the random forest classification leaning to a more immune-active phenotype and also showed impressively consistent performance in the panel of determinants of immunogenicity in solid tumours. 8 And this implication is independent of TMB. To explain the phenomenon, the role of these proteins of BRCA1 or BRCA2 in the maintenance of genome integrity has received the most attention. Both of them are crucial for the process of DNA repair by homologous recombination (HR). 27 That leads to the discussion of the association between BRCA mutations and immunity, which is under investigated at present.
Tumours with BRCA1 or BRCA2 mutations had increased immune infiltrates compared with high-grade serous without mutations. 28 Strickland et al 29 reported immunohistochemistry studies demonstrating that BRCA1/2-mutated tumours exhibited significantly increased CD3 + and CD8 + tumour-infiltrating lymphocytes (TILs) compared to HR proficient (tumours without alterations in HR genes) tumours. Another investigation also showed the presence of intraepithelial cytotoxic T lymphocyte (CTL) also significantly correlates with loss of BRCA1. 30 Animal experiments are undergoing mainly focusing on combined therapies. 31 Recent work in a TP53 À/À BRCA1-mutant murine breast cancer model indicates that double blockade with 2 immune checkpoint inhibitors increases the number of tumour-infiltrating lymphocytes and overall survival after DNA damaging chemotherapy. 32 Another animal experiment showed that CTLA-4 antibody, but not PD-1/PD-L1 blockade, synergized therapeutically with the Poly (ADP-ribose) polymerase (PARP) inhibitor, resulting in immune-mediated tumour clearance and long-term survival in the majority. 31 These findings suggest an approach to enhance the impact of immune checkpoint blockade in BRCA-mutated tumours.
Reports also demonstrated that BRCA2 mutated tumours are more sensitive to chemotherapies (mainly platinum-based treatment) rather than BRCA1 mutated tumours. 11 More work in the field of mechanism of BRCA gene mutation altering sensitivity to chemotherapy and immunotherapy should be taken out to guide treatment combination or to specify the rules of patient's selection with more details.
Several next-generation sequencing techniques have been developed to detect BRCA1 and BRCA2 mutations 33,34 and thanks to the clinical application of PARP, an effective and reliable detection of these mutations in formalin-fixed and paraffin embedded (FFPE) tissue samples has recently been developed. 35 With the development of mutation detecting techniques, the clinical affirmation of our research will be possible although a large cohort will be needed because of the low mutation rate of BRCA in the whole heterogenic colony of ovarian cancer. It is hoped that in the future, our effort could balance the efficient biomarkers tests, comprehensive disease information and maximized patients' benefits.

CONFLI CT OF INTEREST
The authors confirm that there is no conflict of interests.