WES data of 578 ovarian cancer patients were obtained from the TCGA database (accessed in August 2022). To study the relationship of DDR deficiency and Pt sensitivity, patients who underwent Pt-based chemotherapy and with 1) detailed clinical information on Pt sensitivity (either sensitive or resistant) and response to Pt-based chemotherapy (based on RECIST 1.1 criteria); 2) bi-allelic alteration status in HRD genes characterized previously in a publication (Riaz et al., 2017) were included in our study. Eventually, a total of 196 ovarian cancer patients from the TCGA database were enrolled in this study. Detailed clinicopathological information of all included patients was listed in Supplementary Table S1, and these samples were collected prior to any treatment. We used the WES data and clinicopathological information of 196 ovarian cancer patients from the TCGA database to construct the Pt-score subsequently. The validation cohort which including 60 ovarian cancer patients with WES data, was downloaded from public database (CNGBdb accession: CNP0001937).

Bi-allelic alterations in HR genes considering germline mutations and small fragment insertions and deletions (Indels), somatic mutations, and loss of heterozygosity (LOH) were obtained from a previous publication (Riaz et al., 2017). To be more precise, four types of bi-allelic inactivation including a) germline pathogenic mutations with LOH; b) germline pathogenic mutations with heterozygous somatic pathogenic mutations; c) somatic pathogenic mutations with LOH; d) somatic pathogenic mutations with heterozygous somatic pathogenic mutations were included in our study. Additionally, somatic single nucleotide variations (SNVs) and Indels of ovarian cancer patients were collected from the TCGA database. Variants with high frequency among unrelated individuals were firstly filtered out as previously reported (Li et al., 2017), then further filtered in a way similar to the method demonstrated in previous research (Angus et al., 2019). More specifically, alternations including nonsense mutations, missense mutations, mutations on splice sites and in frame/frameshift indels were initially collected. As for missense mutations, only those matching two or more of the following in silico functional criteria: 1) prediction score of 0–0.05 with SIFT (Sorting Intolerant from Tolerant) method; 2) “possibly damaging” or “probably damaging” in Polymorphism Phenotyping-2 (Polyphen2) algorithm or 3) reported in COSMIC database with a FATHMM-MKL (Functional Analysis through Hidden Markov Models) score larger than 0.5 (accessed in August 2020) were retained. We also quantified mutational signatures for the derivation of more comprehensive information. Trinucleotide altering patterns of unfiltered somatic SNVs and fragment changing patterns of unfiltered Indels in our cohort were matched to the signatures described in the COSMIC database using YAPSA software (version 3.11). Exposing contribution of each signature in samples was quantified and only signatures contributing to more than 1% alterations were illustrated and considered in the subsequent analyses.

Apart from mutational alternations, arm-level copy number loss was simultaneously collected in our comparative analyses due to the lack of studies related to the role of arm-level LOH in Pt sensitivity prediction. As for the detection of arm-level copy number loss, fragment-level copy number information from ABSOLUTE in TCGA database was obtained and only loss covering at least 90% of chromosome arm length were collected. We defined four types of arm-level LOH: 1) both copy numbers on major and minor allele equaled to 0; 2) loss of heterozygosity (LOH) when major and minor allele copy numbers equaled to 1 and 0 respectively; 3) copy neutral LOH (CNLOH) when major and minor allele copy numbers equaled to 2 and 0 respectively; 4) LOH at high ploidy when minor allele copy number equaled to 0 whereas major allele copy number was larger than 2. These LOH characteristics and mutational features were gathered for further confirming the prioritization of Pt sensitivity associated genetic features.

HRD bi-allelic alterations, somatic mutations, arm-level LOH, and other genetic features were respectively compared between patients stratified by Pt sensitivity. To reduce possible heterogeneity, we only selected genes with the highest mutational frequency together with mutations in well-characterized genes involved in DDR pathway (Knijnenburg et al., 2018). A catalogue containing all analyzed DDR genes was listed in Supplementary Table S2. Only the features showing significantly different proportion between two groups and demonstrated high correlation with patients’ survival were selected as candidates for further model construction. The candidate features were eventually integrated as Pt score employing the LASSO regression model. The predictive performance of the LASSO-Pt score was trained on the TCGA-sourced data and verified in an external dataset, and then compared with the HRDscore derived by the HRDetect method.

Two-sided Mann-Whitney and Fisher’s exact tests were performed on GraphPad Prism (version 7.01) or R (version 3.6.1) to quantify distribution differences. Log-rank tests were employed to compare the Kaplan-Meier survival curves of ovarian cancer patients on GraphPad Prism (version 7.01). LASSO model and the corresponding AUC curves were generated using R software (version 3.6.1). The predictive performance of LASSO-Pt score and HRDscore was compared by DeLong’s test. Statistical significance was defined as p < 0.05 for the statistical analyses.

Regarding the recent revitalization of demand in Pt sensitivity prediction for ovarian cancer, several factors including homologous recombination status, specific somatic mutations and ITH demonstrated association with Pt sensitivity, albeit the limited specificity (Luo et al., 2020). Given that bi-allelic inactivation might contribute to the loss of functions in certain genes and provide unprecedented information, we obtained the bi-allelic mutational data of HR-related genes in 196 ovarian cancer patients from TCGA database and explored their association with Pt sensitivity. As shown in Figure 1A, only germline pathogenic mutations with LOH and somatic pathogenic mutations with LOH were identified in HR-related genes in the curated cohort. Genes including BRCA1 (22/196, 11.2%) and BRCA2 (20/196, 10.2%) exhibited the highest bi-allelic mutational frequency, in line with the reported findings (Riaz et al., 2017).

As previously stated, considering the theoretical associations between DDR mechanism and Pt-based chemotherapy response, we further accessed all the somatic SNVs in these 196 patients from TCGA database and scrutinized the alterations in all DDR-related genes collected from a previous study (details in Materials and Methods). Recurrently mutated driver genes were also encompassed in our research, given that they were indicated to play essential roles in cell survival and Pt sensitivity. After filtration, TP53 (163/196, 83.2%), FAT3 (14/196, 7.1%), RYR2 (10/196, 5.1%) and MAP2 (6/196, 3.1%) demonstrated highest mutational frequency (Figure 1B). Besides, sporadic alternations were observed among DDR-related genes including ERCC6 (4/196, 2.0%), PTEN (3/196, 1.5%) and RIF1 (3/196, 1.5%), most of which showed an enrichment in Pt-sensitive patients.

Analyses on copy number variation (CNV) have provided an insightful mechanism to decipher critical conundrums for Pt sensitivity in ovarian cancer. The connection between long-segment copy number loss and HRD endorsed its application in Pt therapeutic sensitivity prediction. However, arm-level loss was habitually disregarded since chromosome arm aneuploidy was often caused by mechanisms other than HRD. To assess whether loss of copy number at arm level was associated with patients’ response to Pt-based chemotherapy, the CNV data of these 196 patients were also accessed through TCGA database. The somatic copy number alteration (SCNA) was called using ABSOLUTE and any loss of copy number at the arm-level was exhibited in Figure 1C. These results showed that chromosomal arm-level loss was common among patients, including 13q (76/196, 38.8%), 17p (67/196, 34.2%), 17q (57/196, 29.1%), 4q (33/196, 16.8), 6q (31/196, 15.8%), and others. At the same time, these changes had been considered in the subsequent analyses to explore their potential correlations with Pt sensitivity (Figure 1C).

To study whether these genetic alterations (Figure 1) are associated with Pt sensitivity, the information of patients’ response to Pt-based chemotherapy and disease progression were assessed. The result showed that a significantly higher proportion of patients with BRCA1/2 biallelic inactivation (85.4%) was observed to be sensitive to Pt-based chemotherapy compared to wild-type BRCA1/2 patients (65.2%) (p = 0.013) (Table 1). Similarly, a significantly higher proportion of patients with bi-allelic inactivation in HR-related genes (81.1%), somatic mutations in DDR pathways (100%), somatic mutations in genes involved in microtubule functions (100%) or arm-level copy number loss in chromosome 4p (93.8%) was observed sensitive to Pt-based chemotherapy compared to their counterparts (65.0%, 67.4%, 67.0% and 67.2% respectively) (p = 0.036, 0.019, 0.006, and 0.025 respectively) (Table 1). Patients with arm-level copy number loss in either chromosome 5q or 4p were also found significantly enriched in the Pt-sensitive group, with a lower p-value (0.003). However, other frequently mutated genes or pathways, including somatic mutations in TP53, FAT3, RYR2, pathways engaged in calcium channel activities, and arm-level copy number loss in 13q, 17p, 17q and others, were not found associated with patients’ sensitivity to Pt-based chemotherapy. No significant association could be observed between any genetic alterations and patients’ pathological response as well, which was defined as whether patients achieved complete response or partial response, but borderline significance was found between copy number loss of either chromosome 5q or 4p and patients’ pathological response (p = 0.084) (Table 1).

Considering certain mutations might be enriched in patients with specific clinicopathological status, we also looked at the association between these genetic alterations and clinicopathological features such as patients’ age at diagnosis and tumor histological grades. As a result, no significant association could be observed between genetic alterations and clinicopathological features (Table 1). Given that BRCA1/2 were well-accepted biomarkers to indicate HRD to date, we analyzed the association between these genetic alterations and BRCA1/2 bi-allelic inactivation, trying to see whether these genetic alterations were related to HRD. The result showed that all genetic alterations which showed a strong association with Pt sensitivity were uncorrelated with BRCA1/2 mutations, indicating that these alterations were biomarkers in predicting patients’ response to Pt sensitivity and independent to HRD (Table 1).

In addition, the survival analysis demonstrated that copy number loss of chromosome 4p/5q indicated significantly improved disease-free survival (DFS) in ovarian cancer patients who received Pt-based chemotherapy (Figure 2A), suggesting these copy number alterations might play roles in disease short-term progression, recurrence, or metastasis, while no significant difference could be observed in DFS of patients with copy number loss of chromosome 4p compared to their wild type counterparts (Figure 2B). Nonetheless, patients harboring BRCA1/2 biallelic inactivation, biallelic inactivation in the HR pathway, or more than one somatic mutation in DDR-related genes exhibited longer overall survival (OS) time in comparison with their counterparts (Figures 2C–E), indicating their associations with better prognosis in ovarian cancer patients who received Pt-based chemotherapy.

In addition to studying whether the genetic alterations are associated with Pt sensitivity in ovarian cancer patients, we also looked at the association between mutational signature and patients’ disease progression. The exposing score of each SBS signature was calculated by YAPSA software. A significant difference was observed in the absolute SBS39 exposing score and the relative ID6 exposing score between the responding group and the non-responding group, both showed higher exposing contribution in the responding group (p = 0.038 and 0.059 for absolute SBS39 exposing score and relative ID6 exposing score, Figures 3A, B). Survival analysis demonstrated that patients with higher absolute SBS39 or relative ID6 exposing scores showed significantly improved OS (p = 0.001 and 0.003 respectively) (Figures 3C,D). Suggesting SBS39 and ID6 might play important roles in long-term survival. Previously, a high SBS3 exposing score was shown to be one of the genetic features in patients with HRD. However, in our result, no significant difference could be observed in SBS3 exposing scores between the responding group and the non-responding group. Furthermore, we also analyzed whether the exposing scores of these SBS were associated with HRD in our cohort. The result showed that SBS3 and ID6 were significantly associated with BRCA mutations, indicating the correlations between these two mutational signatures and HRD, while SBS39 was observed independent of HRD (Figure 4).

TAI, LST, and LOH scores of patients who treated with Pt-based chemotherapy were previously reported to be stringently associated with HRD and predictive of disease progression. In our cohort, these genetic features that contributed to the HRD score were significantly associated with patients’ sensitivity to Pt-based chemotherapy and disease progression (Figure 5). We then analyzed the ability of these scores in predicting patients’ sensitivity and disease progression after receiving Pt-based chemotherapy in all 196 patients.

HRD bi-allelic alterations, somatic mutations in DDR pathways, arm-level LOH, mutational signatures and other genetic features that showing significantly different prevalence between the Pt responding group and the non-responding group were screened out to establish a biomarker for Pt sensitivity. Notably, patients who were sensitive to Pt-based chemotherapy were observed with significantly higher LASSO-Pt score than patients who were resistant to Pt-based chemotherapy in both TCGA dataset (Figure 6A, p < 0.0001) and CNGBdb dataset (Figure 6C, p = 0.0109).

Genetic features were integrated into an Pt score employing the LASSO regression model. The predictive ability of the LASSO-Pt score was then further validated in all 196 ovarian cancer patients. Previously, HRD score gauged using HRDetect method was shown with satisfactory predictive ability of Pt sensitivity in ovarian cancer patients (Sztupinszki et al., 2021). We then compared the performance of the LASSO-Pt score against the HRD score in predicting Pt sensitivity in our cohort. The result showed that the LASSO-Pt score was observed with significantly improved sensitivity and specificity than the HRD score (DeLong’s Test, p < 0.001, Figure 6B).

To validate the reliability of LASSO-Pt score, a public (CNGBdb accession: CNP0001937) dataset of 60 ovarian cancer patients with Pt treatment information was downloaded. After training, an AUC value of 0.744 was obtained with CNGBdb dataset, suggesting potential predictive value of these variations in clinical setting (DeLong’s Test, p < 0.003, Figure 6D).