Combined homologous recombination repair deficiency and immune activation analysis for predicting intensified responses of anthracycline, cyclophosphamide and taxane chemotherapy in triple-negative breast cancer

The HRD score was calculated based on the number of specific lesions in the genome, including loss-of-heterozygosity (LOH), large-scale transitions (LST) and the number of telomeric allelic imbalances (ntAI). Following Thorsson et al.’s research on TCGA data [26], we extracted the HRD scores of TNBC patients. Somatic point mutational signatures were determined with the deconstructSigs R package [27] by using the COSMIC signatures as a mutational-process matrix. The WES-based HRDetect scores were calculated using the lasso logistic model by fitting multiple predictors related to HRD, including the HRD score, the contribution of major mutational signatures in breast cancers (such as signature 1, signature 3, signature 6 and signature 20) and the insertion/deletion ratio [28]. The weights of the whole exome–specific model were trained on 560 artificial whole exomes [29].

We acquired DNA methylation data (Illumina Human Methylation 450) from the UCSC Xena database. The transcription start site (TSS) information of the human reference genome (GRCh38) was obtained from Ensembl. The promoter region was defined as the 1500 bp upstream and 500 bp downstream of the TSS. We identified a robust probe (cg13782816) for the BRCA1 gene located in the promoter region through the gene symbol annotation information. BRCA1 promoter methylation (epigenetic silencing) was defined as a cg13782816 methylation level exceeding 0.9.

The R package CIBERSORT [30] was used to calculate the level of immune cell infiltration in TNBC patients. The IFN-γ score of the TNBC patients was calculated using the Tumor Immune Dysfunction and Exclusion (TIDE) online analysis tool [31]. The immune molecular and cellular characteristics of TCGA BRCA patients were obtained from a previous study [26], including tumour lymphocyte infiltration (TLI) score, regulatory macrophages (Mregs), TCR/BCR Shannon, TCR/BCR richness and TGF-β response and neoantigens. Tumour mutation burden (TMB) was defined as the total number of nonsynonymous single nucleotide and indel variants.

We downloaded all the pathways of Collection 2 (C2) and their included gene sets from MSigDB (v7.2) [32]. Patients with the ACT-S&HR-P subtype served as the case group, other patients served as the control group and the R package DESeq2 was used for differential expression analysis. Gene set enrichment analysis (GSEA) was performed using the R package clusterProfiler. We acquired the gene sets of both immune cell types and core biological pathways from previous studies [33, 34] and calculated the activity of these pathways/immune cells utilizing the R package GSVA. We obtained sets of immunostimulators and immune inhibitors based on known studies [35, 36] and explored the differences in the expression of these genes between patients with ACT-S&HR-P and other TNBCs.

Two different methods were used to calculate the immune score (IS) of each TNBC patient based on the gene expression levels. (1) Due to the significantly higher activities of immune response-related pathways (including interferon and immune checkpoint blockade-related pathways, as well as CD8 effector T cells) in the ACT-S&HR-P subtype, we calculated the IS according to the sum of the activities of these pathways (Additional file 1: Table S2). (2). The IS was also calculated using the average expression of prognostic immune markers in breast cancer [37].

After grouping TNBC samples (HR deficiency and HR proficiency) according to HRD status and performing differential expression analysis using the R package DESeq2, we identified differentially expressed genes (DEGs) using FDR ≤ 0.05 and fold change ≥ 2 or ≤ 1/2 as the threshold. Univariate Cox regression analysis was performed in the FFI of TNBC patients based on the expression levels of all DEGs. Finally, 15 HRD-related prognostic factors (HRD expression signature) were determined after removing redundant factors using lasso logistic regression. Among those signature genes, 4 were strongly overexpressed (FDR ≤ 0.05, fold change ≥ 2), and 11 were downregulated (FDR ≤ 0.05, fold change ≤ 1/2) in HR-deficient compared to HR-proficient cases.

Considering the transcription levels of the HRD expression signature and the hazard ratio calculated by Cox regression analysis, the prognostic score (PS) of upregulated factors (all hazard ratio < 1, protective factor) and downregulated factors (all hazard ratio > 1, risk factor) were computed as follows:

where HR_j represents the hazard ratio of upregulated (or downregulated) factor j in the Cox model. Exp_ij represents the expression levels (log-transformed) of upregulated (or downregulated) factor j in sample i. We calculated the HRD-related prognostic score (HRDPS) of patients based on the PS of both upregulated and downregulated factors as follows:

where PS_i, up represents the prognostic score of upregulated factors in sample i, and PS_i, down represents the prognostic score of downregulated factors in sample i.

R Project (version 4.02) for statistical computing was used in this study. The nonparametric Wilcoxon rank-sum test was used to explore the statistical significance between discrete variables (such as HRD status and ACT response status) and continuous indicators (such as immune cell infiltration and pathway activities). The nonparametric Kruskal-Wallis test was used for comparisons among multiple groups, with Benjamini and Hochberg false discovery rate (FDR) correction. For some comparisons (activated NK cells, M0 macrophages and activated mast cells), we additionally performed a combinatorial method that Wilcoxon’s rank-sum test with continuity correction combined 10,000 iterations. Fisher’s exact test was used to examine the relationships between HRD status and HRR-related gene mutations, BRCA1 promoter methylation and the effect on the response to ACT chemotherapy. Clinical outcomes and FFI were compared using the log-rank test. Kaplan-Meier graphs were plotted using standard methodologies. A Cox proportional hazards model was used to calculate the hazard ratios and corresponding 95% confidence intervals (CIs) with adjustments for age, tumour stage and disease stage. Statistical significance was set at two-tailed P < 0.05.

We acquired data for 83 TNBC patients who underwent ACT chemotherapy after tissue sample collection based on the dates given for sampling and first treatment (Fig. 1). Consistent with a previous study [38], most TNBC patients belonged to the basal subtype (88%; Fig. 2A). As expected, TNBC patients carrying mutations in TP53 (84%), PTEN (11%) and BRCA1/2 (8% and 7%, respectively) were relatively frequent among breast cancer (BC) patients in the PanCancer Atlas [39] (Fig. 2A, Additional file 1: Fig. S1A). In contrast, the oncogene PIK3CA (11%) was less prone to the mutation in TNBCs than all BC patients.

Biomarkers for homologous recombination repair deficiency in cancers have attracted great interest from researchers [8, 40]. HR deficiency was defined as either a deleterious tumour BRCA1/2 (tBRCA) mutation or a predefined HRD score ≥ 42 [8], which was determined in 57.8% (48/83) of TNBC patients (Fig. 2A, Table 1). Our results showed that most TNBC patients with BRCA1/2 mutations presented HRD scores^high (≥ 42) and dominant mutational signature 3 (SBS3) activity (≥ 0.3) (both 8/11; Fig. 2B). Indeed, the patients with HR deficiency showed significantly higher SBS3 exposure than patients with HR proficiency (P = 4.3e−04, Wilcoxon rank-sum test; Fig. 2C). In addition to BRCA1/2, several important components of the HRR pathway, such as RAD54 family members (including RAD54B and RAD54L), DNA polymerase members (including POLD1, POLH and POLQ) and PARP1, PALB2 and TP53BP1, occurred mainly in the HR deficient samples (29/48, P = 3.04e−04, Fisher’s exact test; Fig. 2D, Additional file 1: Fig. S1B). Consistent with previous studies, we also found that HR-deficient patients showed a higher proportion of BRCA1 promoter methylation and HRDetect score^high (16/48, P = 0.018, Fisher’s exact test; Figs. 2A, E) [28, 41]. These results indicated that HRD could be characterized by mutations in HRR-related genes, SBS3 exposure and BRCA1 promoter hypermethylation.

Accumulated evidence has shown that HRD is associated with a better prognosis for patients with a variety of solid tumours [42, 43]. Whether a better prognosis is related to HRD in TNBC patients who received ACT chemotherapy has not been well characterized. Our results revealed that compared with patients with HR proficiency, patients with HR deficiency showed significantly better overall survival (OS; P = 0.0063, log-rank test; Fig. 3A) and disease-specific survival (DSS; P = 0.023, log-rank test; Additional file 1: Fig. S2A) after treatment with ACT. Specifically, the 5-year OS rate for HR-deficient patients was 98%, while that for HR-proficient patients, it was only 61% (Fig. 3A). Multivariate Cox regression analysis showed that HR deficiency was an independent protective factor associated with prolonged patient OS (P = 0.002, log-rank test; Additional file 1: Fig. S2B) and DSS (P = 0.014, log-rank test; Additional file 1: Fig. S2C) in TNBC after adjusting for clinical factors including age, AJCC stage and TNM stage.

Furthermore, to explore whether HR deficiency can indeed benefit TNBC patients’ response to ACT treatment, we determined the ACT chemotherapy failure-free interval (FFI) of patients based on the period from the end of treatment to tumour progression/recurrence or death (Fig. 3B). Our results showed that HR deficiency was correlated with durable response to ACT chemotherapy (P = 0.046, log-rank test; Fig. 3C). The 5-year FFI rate for HR-deficient patients was 75%, while that for HR proficiency, it was only 52%. Cox regression analysis showed that HR deficiency was a significant protective factor for FFI in TNBC patients with a hazard ratio of 0.16 (95% CI 0.044–0.55, P = 0.004; Fig. 3D), which improved the response interval to ACT chemotherapy. In addition, we found that HR-deficient patients tended to be more sensitive to ACT chemotherapy (54.2% for sensitive, 37.1% for resistant), compared with HR-proficient patients (37.1% for sensitive, 62.9% for resistant; P = 0.074, Fisher’s exact test; Fig. 3E).

Unrepaired DNA damage, especially HRD, modulates the tumour immune microenvironment through a range of molecular and cellular mechanisms [15, 16]. Low-dose doxorubicin and cyclophosphamide chemotherapy may stimulate anticancer immune responses and promote a more favourable tumour microenvironment [17, 18]. We speculated that the impact of HRD on the immune microenvironment was related to the ACT response in TNBC patients. As expected, we found that the effect of HRD on immune cell infiltration showed significant differences in distinct ACT response groups (Fig. 4, Additional file 1: Fig. S3). For example, in the ACT-sensitive (ACT-S) group, NK cells showed higher infiltration in HR-proficient (HR-P) samples (P = 0.0079, Wilcoxon rank-sum test, same below; Fig. 4A). In contrast, M0 macrophages and mast cells presented higher infiltration levels in HR-deficient (HR-D) samples (P-values were 0.02 and 0.012, respectively; Additional file 1: Fig. S3AB). Interestingly, differences in these immune cells were not observed in the ACT-resistant (ACT-R) group (Fig. 4A, Additional file 1: Fig. S3AB). Additionally, in the ACT-sensitive group, we found that regulatory macrophages (Mregs) were significantly activated in HR-proficient patients (P = 0.0087, Wilcoxon rank-sum test; Fig. 4B). These patients showed T cell/B cell receptor (TCR/BCR) repertoire diversity (P-values of 0.041 and 0.019, respectively; Fig. 4C, Additional file 1: Fig. S3C) and TCR/BCR richness (P-values of 0.036 and 0.018, respectively; Additional file 1: Fig. S3DE). Diversified TCR/BCR receptors are the basic attributes of an effective immune system, allowing T/B cells to target multiple types of endogenous or exogenous antigens [44, 45]. However, we did not find corresponding results in the ACT-resistant patient group (Fig. 4C, Additional file 1: Fig. S3CDE).

Studies have suggested that neoantigens presentation in tumours promotes the release of IFN-γ from tumour-infiltrating lymphocytes (TILs), and the released IFN-γ upregulates PD-L1 expression in immune cells and tumours [46, 47]. We found that in the ACT-sensitive group, HR-proficient samples showed higher TIL infiltration (P = 0.0071, Wilcoxon rank-sum test, same below) and IFN-γ activity (P = 0.012) compared to HR-deficient samples (Fig. 4D, E). Among HR-proficient patients, those sensitive to ACT also showed higher TIL scores (P = 0.079, Wilcoxon rank-sum test, same below) and stronger IFN-γ activity (P = 0.009) than ACT-resistant patients (Fig. 4D, E). The patients with ACT-S&HR-P (sensitive to ACT and HR proficiency) were associated with higher IFN-γ activity (P = 0.011, Wilcoxon rank-sum test; Additional file 1: Fig. S4A) and lower TGF beta response (Additional file 1: Fig. S4B), implying an intense immune response in this subtype. Furthermore, for the known cancer immunotherapy biomarkers, we found that their expression levels were correlated with ACT response and HRD status (Fig. 4F, Additional file 1: Fig. S4CD). For example, significantly higher expression levels of CTLA-4, PD-1 and PD-L1 were found in ACT-S&HR-P patients (P < 0.05, Wilcoxon rank-sum test; Fig. 4G–I). In particular, patients with HR deficiency showed higher TMB (P = 0.0031, Wilcoxon rank-sum test; Additional file 1: Fig. S4E) and neoantigen levels (P = 0.0018, Wilcoxon rank-sum test; Additional file 1: Fig. S4F), which may be related to the fact that HRD exacerbates DNA DSBs, thereby promoting genome instability and causing the release of molecular antigens [48].

The clinical and translational data indicated that short-term doxorubicin treatment may increase the likelihood of a response to PD-1 blockade in TNBC [17]. Considering the results of our study, we postulated that ACT-S&HR-P patients may benefit from immune checkpoint blockade (ICB) therapy. To test this postulate, we performed GSEA on the C2 pathways from MSigDB (v7.2) (the “Methods” section). We found that the genes that were upregulated in ACT-S&HR-P patients were enriched in multiple immune response-related pathways, such as interferon-gamma signalling (NES = 2.53, FDR < 0.001; Fig. 5A, Additional file 1: Fig. S6A), interferon signalling (NES = 2.30, FDR < 0.001; Additional file 1: Fig. S5A) and type II interferon signalling IFN-γ (NES = 2.36, FDR < 0.001; Additional file 1: Fig. S5B). In addition, ICB-related pathways, including cancer immunotherapy by PD-L blockade (NES = 2.41, FDR < 0.001), the CTLA-4 pathway (NES = 2.55, FDR < 0.001) and CD28 family costimulation (NES = 2.45, FDR < 0.001), were related to the upregulated genes in this subtype (Fig. 5B, Additional file 1: Fig. S5CD). In particular, natural killer cell-mediated cytotoxicity (NES = 2.05, FDR < 0.001; Fig. 5C) and antigen processing and presentation (NES = 2.63, FDR < 0.001; Additional file 1: Fig. S5E) were also enriched to the upregulated genes in this subtype.

Consistent results were found using pathway activity. For example, cancer immunotherapy by CTLA-4/PD-1 blockade was activated in ACT-S&HR-P patients (mean differences > 0, P < 0.05, Wilcoxon rank-sum test, the same below; Fig. 5D, Additional file 1: Fig. S6B). Similarly, the immune response-related pathways mentioned above also showed significant activation in this subtype, including interferon-gamma signalling (mean difference = 0.29, P = 0.0078), NK cell pathway (mean difference = 0.31, P = 0.011) and antigen processing and presentation (mean difference = 0.28, P = 0.0035) (Fig. 5D, Additional file 1: Fig. S6B). The JAK-STAT signalling pathway plays critical roles in the coordination of the immune system, especially for cytokine receptors, and it can regulate the polarization of helper T cells [14]. Our results showed that the genes upregulated in ACT-S&HR-P patients were enriched to the JAK-STAT signalling pathway (NES = 2.58, FDR < 0.001; Additional file 1: Fig. S5F). Additionally, the IL2-STAT4 pathway (NES = 2.62, FDR < 0.001) and IL2-STAT5 pathway (NES = 2.25, FDR < 0.001) showed correlations with the upregulated genes in this subtype (Additional file 1: Fig. S6A). Additionally, the pathways related to the inflammatory response, including the IL2-STAT4/5 pathways presented higher activities in this subtype (Fig. 5D, Additional file 1: Fig. S6B). These results suggested that ACT-S&HR-P patients exhibit surprisingly elevated immune activities by activating the immune response pathways.

By analysing the types of immune cells [33], we found that among HR-proficient patients, both innate immune cells (such as activated DCs [aDCs], mast cells, macrophages and natural killer (T) cells [NKs/MKTs]) and adaptive immune cells (such as T helper 1 [Th1], CD8+ T central memory [Tcm], CD8+ T effector memory [Tem] and CD4+ Tem cells) were activated only in the ACT-sensitive group (P < 0.05, Wilcoxon rank-sum test, the same below; Fig. 5E, Additional file 1: Fig. S6CD). Similarly, the core biological pathways, including immune checkpoint (P = 0.027), CD8 T effector (P = 0.011) and antigen processing machinery pathways (P = 0.027), were also activated in ACT-S&HR-P patients (Additional file 1: Fig. S6EF). Additionally, analysing both immune cell infiltration and differential expression profiling revealed that the ACT-S&HR-P subtype was enriched for both immune-activated cells and immunostimulators. For instance, immunostimulatory cells, such as M1 macrophages, NK cells and dendritic cells, showed significantly higher activity in ACT-S&HR-P patients (Fig. 5F). However, the number of immune-suppressive cells was not elevated in this subtype (Fig. 5G). Expression profiling demonstrated that immunostimulators such as CD40, CD86 and ICOS were significantly overexpressed in this subtype (Fig. 5H). The mechanisms by which ACT-S&HR-P patients show a stronger immune response likely involve the recruitment of immune-activated cells. In particular, our results revealed that immune inhibitors, especially IDO1, were significantly elevated in ACT-S&HR-P patients (Fig. 5I). This provides a valuable reference for additional reasonable immune checkpoint blockade therapeutics for TNBC patients with the ACT-S&HR-P subtype.

The above findings implied that HRD and immune cell activity might synergistically affect the ACT response; thus, we wondered whether the combination of HRD and immune activation could improve the ACT chemotherapy response. We computed the immune score (IS) of patients based on the immune response pathways (the “Methods” section) and annotated the patients with the highest 25% IS as being positive for IS (IS+). Our study showed that combined positivity (i.e. positivity for HRD, IS, or both) was significantly associated with clinical benefit (P = 1.9e−04, log-rank test; Fig. 6A) with a hazard ratio of 0.037 (95% CI 0.0048–0.29, P = 0.002, Additional file 1: Fig. S7C) and prolonged DSS of patients (P = 0.018, Additional file 1: Fig. S7AB). More importantly, we found that the patients with combined positivity had a longer ACT failure-free interval (P = 0.013, log-rank test; Fig. 6B). After adjusting for clinical factors, the combined positivity was found to be a significantly independent prognostic factor (HR = 0.21, 95% CI 0.064–0.67, P = 0.009; Fig. 6C). The results were consistent with the known prognostic immune markers of breast cancer (Additional file 1: Table S3, Fig. S7D-F). Incorporating IS into Cox models fit with age, tumour stage, and age and tumour stage improved the predictive accuracy of FFI (P < 0.002, likelihood ratio test; Fig. 6D), which highlights the importance of the combination of HRD status and immune activities in ACT chemotherapy. Additionally, we found that the prognostic efficacy of combined status (AUC = 0.91) was better than that of HRD status alone (AUC = 0.83) or clinical factors alone (AUC = 0.61) (Fig. 6E). These results indicated the necessity of combining HRD status with tumour immunity, which improves the efficacy of identifying ACT responders in TNBC.

To further validate those findings in an independent dataset of TNBC cases, we developed an HRD expression signature that predicts ACT response (FFI) in TCGA TNBC cohorts (the “Methods” section). We identified 15 genes that were associated with FFI, including 4 that had a better ACT response and 11 that had a worse response in HR deficiency than in HR-proficient cases (Fig. 7A, Additional file 1: Table S4). The HRD expression signature showed excellent performance in reflecting the genomic HRD status applied to TNBC patients, as demonstrated by a receiver operating characteristic (ROC) curve with an AUC of 0.89 (Fig. 7B), which was superior to other types of breast cancer samples, including all BC patients (AUC = 0.81) and BC patients except TNBC (AUC = 0.77).

We demonstrated that the combination of the HRD-related prognostic score (HRDPS) and IS was an effective way to predict TNBC patients who may achieve pCR to ACT chemotherapy, showing better prognosis in independent validation sets (Additional file 2: Table S5) [21, 22, 24]. Our results indicated that the patients with combined positivity had longer distant relapse-free survival (DRFS; P = 3.8e−3, log-rank test; Fig. 7C). After adjusting for clinical factors, combined negativity was a significant independent risk prognostic factor in TNBC patients with a hazard ratio of 6.4 (95% CI 2.38–17.1, P < 0.001) compared to combined positivity (Additional file 1: Fig. S8A). However, there was no statistical significance when using HRDPS alone, although patients with HRD-positive had better DRFS (P = 0.062, log-rank test; Additional file 1: Fig. S8B). In addition, the patients with combined positivity had higher pCR rates of ACT in two independent validation sets (54.5% for Hess et al. TNBC, 44.4% for GSE1998 TNBC) compared with combined negativity cases (20% for Hess et al. TNBC, 26.9% for GSE1998 TNBC; Fig. 7D, E).

Furthermore, we analysed the impact of HRDPS, IS and combined status on the prognosis of TNBC patients who were treated with chemotherapy [23]. The results showed that the combination of HRD status and tumour immune activation showed a strong correlation with patient OS (P < 0.0001, log-rank test; Fig. 7F). The patients with positivity for both factors showed the longest OS compared with those with other statuses (P < 0.0001 compared with both negative, P = 0.001 compared with HRDPS-positive only and P = 0.012 compared with IS-positive only, log-rank test; Fig. 7F). Multivariate Cox regression showed that the patients who were negative for both (HR = 2.3 95% CI 1.26–3.9, P = 0.005) and HRDPS-positive only (HR = 2.3 95% CI 1.32–4.2, P = 0.004) showed a significantly worse prognosis compared with patients who were positive for both (Fig. 7G).

Similarly, we acquired consistent results in two additional validation sets of TNBC patients who received ACT intervention [22, 25]. For example, the patients with positivity for both HRDPS and IS showed the longest DRFS (GSE25055; P = 0.007 compared with IS-positive only; P = 0.041 compared with HRDPS-positive only; P = 0.053 compared with both negative, log-rank test; Additional file 1: Fig. S9A) and the best survival outcomes for DSS (Chin et al.; P = 0.037 compared with HRDPS-positive only; P = 0.049 compared with both negative, log-rank test; Additional file 1: Fig. S9B). These results demonstrated that the combination of HRD and tumour immune activation can indeed contribute to the ACT chemotherapy response and clinical outcomes of TNBC patients.