NGS-based Multi-Factor Prediction Model of Platinum Sensitivity for Epithelial Ovarian Cancer
Shu Tong,Hong Zheng,Yunong Gao,Nan Zhang,Hongguo Wang,Chao Zhang,Changxi Wang,Zaixian Tai,Yuting Yi,Jin Li
DOI: https://doi.org/10.1016/s0090-8258(21)01053-2
IF: 5.304
2021-01-01
Gynecologic Oncology
Abstract:Objectives: Platinum sensitivity is the most important predictor of prognosis and response to further treatment in ovarian cancer, while it is still roughly prejudged in line with the interval of progression free survival in clinical practice. Realizing individualized precision medicine require refining predictive model that could enable selection of patients who would benefit from platinum-based or non-platinum containing therapy. The study aims to identify and integrate novel biomarkers for prediction of platinum sensitivity in ovarian cancer. Methods: 60 patients with epithelial ovarian cancer receiving standard initial treatment were recruited from 2016 to 2019. According to interval of progress free survival (PFS), 44 patients (PFS>6 months) and 16 patients (PFS≤6 months) were defined as platinum-sensitive and platinum-resistant recurrence, respectively. We performed next generation sequencing (NGS) analysis of the entire genomic features of ovarian tumor tissue samples obtained from debulking surgery. Multiple genomic variation were compared between platinum sensitive and resistant patients, including single-nucleotide variant (SNV), single base or indel signature, focal copy number variation (CNV), mutational clonality, whole-genome duplication (WGD), and co-variation in DNA damage repair (DDR) pathways. Collectively, a multi-factor-based prediction model was trained and validate with Lasso regression. Results: Analysis of single base signature revealed that the contribution of SBS10a in platinum-sensitive patients was significantly higher than that of platinum-resistant patients (P<0.05). Consistently, patients with SBS10a signature exhibited higher platinum sensitivity (P<0.01) and longer PFS than SBS10a negative patients (15.8m vs 10.1m, P<0.05). Focal CNVs involved in DDR pathways which believed to play a pivotal role in platinum response were emphatically investigated. The results demonstrated the incidence of FAM175A LOH in platinum-sensitive patients was significantly higher than that of platinum-resistant patients (P<0.01), meanwhile the median PFS of patients with FAM175A LOH was significantly longer than patients with wild-type FAM175A (16.5m vs 9.2m, P<0.05). Moreover, patients with WGD showed higher proportion of platinum resistance (P<0.01) and shorter median PFS than patients with normal chromosomal ploidy (9.1m vs 16.47m, p<0.05). Unlucky, there was no marked difference in the distribution of overall SNV or DDR related SNV with top frequency between platinum sensitive and resistant cases. Patients with clonal DDR mutations or co-variation in DDR pathways exhibited higher proportion of platinum response and longer PFS than patients without these variations, but the differences were not statistically significant. By integrating of SBS10a, FAM175A LOH and WGD which showed significant differences in platinum sensitivity in univariate analyses, a lasso regression model was employed to train and validate with all patients. A promising performance with the AUC value 0.864 was observed in ROC curve, suggesting the predictive performance of this model was favorable. Graphical Abstract View Large Image Figure Viewer Download Hi-res image Conclusions: Our study successfully established a prediction model containing three novel biomarkers FAM175A LOH, SBS10a mutation and WGD through in-depth mining of NGS, which has an extremely potential of predictive value for platinum sensitivity in epithelial ovarian cancer. However, it requires larger cohort for further validation. Platinum sensitivity is the most important predictor of prognosis and response to further treatment in ovarian cancer, while it is still roughly prejudged in line with the interval of progression free survival in clinical practice. Realizing individualized precision medicine require refining predictive model that could enable selection of patients who would benefit from platinum-based or non-platinum containing therapy. The study aims to identify and integrate novel biomarkers for prediction of platinum sensitivity in ovarian cancer. 60 patients with epithelial ovarian cancer receiving standard initial treatment were recruited from 2016 to 2019. According to interval of progress free survival (PFS), 44 patients (PFS>6 months) and 16 patients (PFS≤6 months) were defined as platinum-sensitive and platinum-resistant recurrence, respectively. We performed next generation sequencing (NGS) analysis of the entire genomic features of ovarian tumor tissue samples obtained from debulking surgery. Multiple genomic variation were compared between platinum sensitive and resistant patients, including single-nucleotide variant (SNV), single base or indel signature, focal copy number variation (CNV), mutational clonality, whole-genome duplication (WGD), and co-variation in DNA damage repair (DDR) pathways. Collectively, a multi-factor-based prediction model was trained and validate with Lasso regression. Analysis of single base signature revealed that the contribution of SBS10a in platinum-sensitive patients was significantly higher than that of platinum-resistant patients (P<0.05). Consistently, patients with SBS10a signature exhibited higher platinum sensitivity (P<0.01) and longer PFS than SBS10a negative patients (15.8m vs 10.1m, P<0.05). Focal CNVs involved in DDR pathways which believed to play a pivotal role in platinum response were emphatically investigated. The results demonstrated the incidence of FAM175A LOH in platinum-sensitive patients was significantly higher than that of platinum-resistant patients (P<0.01), meanwhile the median PFS of patients with FAM175A LOH was significantly longer than patients with wild-type FAM175A (16.5m vs 9.2m, P<0.05). Moreover, patients with WGD showed higher proportion of platinum resistance (P<0.01) and shorter median PFS than patients with normal chromosomal ploidy (9.1m vs 16.47m, p<0.05). Unlucky, there was no marked difference in the distribution of overall SNV or DDR related SNV with top frequency between platinum sensitive and resistant cases. Patients with clonal DDR mutations or co-variation in DDR pathways exhibited higher proportion of platinum response and longer PFS than patients without these variations, but the differences were not statistically significant. By integrating of SBS10a, FAM175A LOH and WGD which showed significant differences in platinum sensitivity in univariate analyses, a lasso regression model was employed to train and validate with all patients. A promising performance with the AUC value 0.864 was observed in ROC curve, suggesting the predictive performance of this model was favorable. Our study successfully established a prediction model containing three novel biomarkers FAM175A LOH, SBS10a mutation and WGD through in-depth mining of NGS, which has an extremely potential of predictive value for platinum sensitivity in epithelial ovarian cancer. However, it requires larger cohort for further validation.