Somatic mutations in homologous recombination pathway predict favourable prognosis after immunotherapy across multiple cancer types


 Background: Homologous recombination (HR) pathway was recently implicated in modifying the tumor immune microenvironment and thus might serve as a candidate biomarker of immune checkpoint inhibitor (ICI) treatment. We aimed to comprehensively explore the clinical and molecular significance of HR mutations in ICI treatment across multiple cancer types.Methods: We analyzed the clinical and genomic data of 1,752 ICI-treated patients and 4,605 non-ICI-treated patients from cBioPortal, TCGA, and ICGC databases. The impacts of HR mutations on tumor mutations burden (TMB), neoantigen load (NAL), microsatellite instability (MSI), immune molecular characteristics, and multi-omics events were further investigated.Results: HR mutations corelated with improved prognosis in ICI-treated bladder cancer (BLCA), colorectal cancer (CRC), and non-small-cell lung cancer (NSCLC), while in non-ICI cohorts, HR mutations were not associated with prognosis and even suggested unfavorable prognosis. Multivariable Cox analysis demonstrated HR mutations were an independent predictive factor for ICI treatment. Moreover, HR mutations could accurately predict high level of TMB, NAL, and MSI, and displayed the more prevalent incidence than TMB-high, NAL-high, and microsatellite instability-high (MSI-H), which indicated that HR mutations might be an ideal surrogate for TMB, NAL, and MSI estimation. HR mutations were also proven to correlate with the tumor microenvironment, immunity characteristics, immune checkpoints profiles, and substantial multi-omics alteration events.Conclusions: HR mutations are predictive of improved clinical outcomes in BLCA, CRC, and NSCLC treated with ICI instead of non-ICI, and might be a promising surrogate for TMB, NAL, and MSI estimation in ICI treatment.

patients in BLCA, CRC and NSCLC ( Figure 1O-Q). To further examine the impacts of HR mutations on patients with traditional treatments in BLCA, CRC and NSCLC, we enrolled other 11 non-ICI cohorts. Interestingly, there was no survival difference between two phenotypes in most cohorts (Figure 2A-I). Also, patients with the HR-Mut phenotype had relatively lower but not significant OS in Chenlung adenocarcinoma ( Figure 2J), and significantly lower in thoracic PDX (MSK, provisional) (PDX)-NSCLC (Figure 2K), which were opposite to the above results in ICI cohorts. Thus, HR mutations are indicative of favourable prognosis in BLCA, CRC and NSCLC treated with ICI instead of non-ICI.
By analysing the DNA sequencing data from MSK and TCGA pan-cancer cohorts, we found the HR-Mut phenotype had predominantly more single-nucleotide polymorphisms, insertion or deletion and tumour mutation burden (TMB) ( Figure 3A,B). The proportion of TMB-high in the HR-Mut phenotype was also substantially higher (Figure 3C-J). Receiver operating characteristic curve (ROC) analysis revealed that HR mutations could accurately predict TMB-high (Figure 3K-N and Figure S2A-D). Using the neoantigens data from 5935 patients in TCGA dataset, 5 we further explored the association between HR mutations and neoantigen load (NAL). We found patients with HR Mut possessed dramatically more NAL ( Figure S3), and the proportion of NAL-high was also remarkably higher in the HR-Mut phenotype ( Figure S4A-D). ROC analysis showed that HR phenotype could accurately assess tumour NAL ( Figure S4E-H). Given the latent interactions between HR pathway and mismatch repair (MMR) pathway, 6 we further revealed the association between two DNA damage repair pathways in different cancer types. 7 As showed in Figure S5A, there were significant HR/MMR comutation patterns in tumours. Therefore, we speculated that HR mutations may have a potential connection with  Subsequently, gene set enrichement analysis (GSEA) was applied to decipher the underlying mechanisms in terms of cancer Hallmarks, and 22 pathways dramatically enriched ( Figure 4A). The HR-Mut subtype was mainly correlated with cell cycle and interferon response, while the HR-WT subtype enriched plenty of stromal activationrelated terms ( Figure 4A). To further explore the tumour microenvironment (TME), the "ESTIMATE" R package was employed to infer the fraction of immune and stromal components. 8 The HR-Mut phenotype scored better in immune score and displayed lower stromal score and tumour purity ( Figure S6A). There was no significant difference in the correlations between immune and stromal scores in the two HR phenotypes, and the tumour purity was lower as the immune and stromal scores were increasing in both subgroups ( Figure 4B).
In cancer-immunity cycle, 9 we found four key immuneactive steps including T cell immunity, tumour antigenicity, trafficking and infiltration, and recognition of tumour cells were significantly enhanced in the HR-Mut phenotype. Meanwhile, the HR-Mut phenotype also displayed the substantial enhancement of checkpoint expression, implying an immune-hot but suppressive microenvironment ( Figure 4C and Figure S6B). To gain more detailed insights into this issue, we applied TIMER to quantify

F I G U R E 4
Tumour microenvironment (TME) and immunity characterization of the homologous recombination (HR) phenotypes. (A) GSEA analysis of cancer Hallmarks between the HR-Mut and HR-WT phenotypes. Green points represent pathways significantly related to HR-Mut tumours, orange points represent pathways significantly related to HR-WT tumours, and grey points represent pathways with no significance between the two phenotypes. (B) Scatterplots between stromal and immune scores with tumour purity gradient were shown, and its correlation coefficient was indicated by each phenotype. The colour grading corresponds to the tumour purity, indexed as shown on the colour bar at the top right of the panel. (C) Radar plots showed that the immunogram patterns of the two phenotypes. The axes of the radar chart were generated with the median immunogram score for the HR-Mut and HR-WT phenotypes, respectively. (D) The TIMER algorithm showed the infiltration difference of six immune cell populations including B, CD4+ T, CD8+ T, neutrophil, macrophage and myeloid dendritic cells between two phenotypes. (E) The CIBORSORT algorithm showed the proportion of 22 immune cells in the HR-Mut and HR-WT phenotypes. (F) Differences in the distribution of 27 immune checkpoint molecules between two phenotypes. ns p > 0.05; *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001 the infiltration abundance of different immune cell populations. 10 The infiltration abundance of CD4+ and CD8+ T cells was higher in HR-Mut tumours, which further demonstrated T cell enrichment was a dominant factor in the TME of HR-Mut tumours ( Figure 4D). Meanwhile, the deconvolution algorithm CIBERSORT validated these results, and presented that the HR-Mut phenotype had superior infiltration of activated nature killer cells, while the HR-WT phenotype harbored more resting cells such as resting dendritic cells and resting CD4+ T memory cells ( Figure 4E and Figure S6C). In addition, we observed the expression level of CD274, PDCD1, and CTLA4 was superior in the HR-Mut subtype, which may explain its better response to ICI treatment ( Figure 4F and Figure  S6B). There were also some other molecules that were significantly up-regulated in HR-Mut tumours, such as FGL1, IDO1, LAG3 and TNFRSF18. Meanwhile, CD40LG, HHLA2 and NT5E were highly expressed in HR-WT tumours, which may be the potential targets to improve the prognosis of patients with the HR-WT phenotype.
In conclusion, HR mutations are predictive of improved clinical outcomes in BLCA, CRC and NSCLC treated with ICI instead of non-ICI. HR mutations could accurately predict TMB-high, NAL-high and MSI-H, suggesting that HR mutations may be a promising surrogate for TMB, NAL and MSI estimation. HR mutations are also associated with the TME, immunity characteristics and immune checkpoints profiles. Our findings lay a foundation for the clinical application of HR mutations in cancer immunotherapy.