Pan-Cancer Prediction of Immunotherapy Benefit Using DNA Damage Response and Repair Gene Signature: A Model Development and Validation Study

Background: The predictive capability of the current immunotherapy biomarkers is still unsatisfying. Although alterations in DNA damage response and repair (DDR) genes were shown to be a potential immunotherapy biomarker in certain cancer types, no reliable DDR gene panel was available and whether DDR gene alterations are associated with outcomes to immune checkpoint blockades (ICBs) across cancer types and drug classes is unknown.Methods: In this study, we used the publicly available genomic data from the training set of 1630 patients with eleven different types of cancer treated with ICBs in Memorial Sloan-Kettering Cancer Center (MSKCC) to analyze for the mutational status across a panel of 43 DDR genes covered in all three versions of MSK-IMPACT. Then, we developed a ten-DDR-gene score (DDRScore) by use of LASSO Cox regression followed by stepwise regression analysis through bidirectional elimination, based on association between mutational status of DDR genes and overall survival after ICB treatment. The predictive value of DDRScore for immunotherapy benefit was validated in six previously published independent validation cohorts (three melanoma cohorts and three non-small lung cancer cohorts), among which, 844 patients were assessed for progression-free survival (PFS), 633 were assessed for overall survival (OS), and 659 were evaluated for objective response rate (ORR). We used Cox regression to test the association between DDRScore and clinical outcomes to ICBs and tumor mutational burden (TMB) for comparison. The χ2 test was used to test the difference of response rate between the DDRScore-high (DDRScore-H) and non-DDRScore-H groups. The main outcomes were OS and PFS; the secondary outcome was ORR.Findings: In the training cohort, higher DDRScore (cut-points stratified by cancer types) was associated with better OS (hazard ratio [HR] 0.54 [0.43-0.67]). Additionally, an association between higher DDRScore and improved OS was found in most cancer histologies and all drug classes. The HRs for tissue TMB (tTMB) (≥10 muts/Mb) and tTMB (top 20% within each histology) were 0.57 (0.48-0.67) and 0.64 (0.53-0.77), respectively, for OS in the training cohort as a whole, and tTMB associated with these two cut-points failed to reach statistical significance in predicting OS benefit from ICB combinations. In the validation cohorts, pooled analysis demonstrated that DDRScore using the same cut-points of training cohort was associated with PFS (0.69 [0.57-0.83]), OS (0.67 [0.52-0.85]) and ORR (p=0.01). Although higher tTMB with a fixed cut-off (tTMB ≥10 muts/Mb or blood TMB ≥16 muts/Mb) or stratified by cancer types (top 20% within histologies) correlated with better PFS (HRs were 0.78 [0.66-0.92] and 0.68 [0.56-0.83]) in the pooled analysis of validated cohorts, they performed worse predicting OS benefit (HRs were 0.87 [0.71-1.08] and 0.86 [0.68-1.10]). Finally, HR for combination of DDRScore and TMB with a fixed cut-off for OS in the training cohort was 0.40 (0.30-0.53), and HRs for the combination biomarker in the validation cohorts were 0.61 (0.48-0.78) and 0.61 (0.45-0.83) for PFS and OS, respectively.Interpretation: DDRScore seemed to a reliable and feasible approach to guide ICB treatment. The predictive value of DDRScore was stable across cancer types and drug classes, and outperformed that of TMB in most situations. The combination of DDRScore and TMB can provide more accurate prediction of immunotherapy benefit.Funding Information: This study was supported by National Natural Science Foundation of China (No. 81672738) and Guangzhou Science and Technology Foundation of China (No.201704020095).Declaration of Interests: The authors declare no competing interests.