For Ruijin Cohort 1, we retrospectively screened consecutive breast cancer patients treated at the Comprehensive Breast Health Center, Ruijin Hospital, Shanghai Jiao Tong University School of Medicine (RJBC-CBHC) from January 2012 to March 2019. Patients who met the following eligibility criteria were included: 1) invasive breast cancer, 2) pathologically diagnosed with TNBC, 3) available formalin fixed paraffin-embedded (FFPE) tissues. Exclusion criteria were as follows: 1) male breast cancer, 2) de novo stage IV, 3) incomplete immunohistochemistry (IHC) information.

For Ruijin Cohort 2, TNBC patients newly diagnosed and treated in RJBC-CBHC from June 2020 to January 2023 were prospectively recruited. Patients were eligible if they successfully received HRR genotyping. All tumor size, lymph node status, comorbidities, and adjuvant therapy strategies were permitted. Patients with metastatic or recurrent disease were excluded.

Archival FFPE blocks were selected from the biobank at the Department of Pathology, Shanghai Ruijin Hospital. Elaborate clinical data were retrieved from Shanghai Jiao Tong University Breast Cancer Database (SJTU-BCDB). All patients were given informed consent and our study was approved by the Ethical Committees of Ruijin Hospital, Shanghai Jiao Tong University School of Medicine (Clinical Trial Ethics Approval Number: 2016; 5–3). All procedures were in accordance with the 1964 Helsinki declaration and its later amendments.

At least two experienced pathologists (A. Li and M. Ruan) from the Department of Pathology, Ruijin Hospital, Shanghai Jiao Tong University School of Medicine, contributed to the tumor histopathological analysis. The IHC testing was used to determine the status of estrogen receptor (ER), progesterone receptor (PR), human epidermal growth factor receptor-2 (HER-2) and proliferation index (Ki-67). ER and PR positivity were defined as no less than 1% stained nuclei, as was described in our previous publications (23, 24). Ki67 ≥ 30% was classified as high-expression. HER-2 status was classified as “HER-2 Low” if IHC HER-2 1+ or HER-2 2+/Fluorescence in situ hybridization (FISH) negative, and HER-2 0 patients were defined as “HER-2 negative”.

Patients from Ruijin Cohort 1 went through regular outpatient follow-ups or follow-up calls, once every three months within the first two year after surgery, once every six months through the third to the fifth year, and once every year ever since, in light of the American Society of Clinical Oncology Guidelines (ASCO guidelines) (25). Overall survival (OS) was defined as the time interval between surgery and death. Recurrence-free survival (RFS) was defined as the time interval between surgery and the event of local recurrence, distant metastasis or death. For patients with no events, OS and RFS were define as the time interval between surgery and the last follow-up date (Feb 12^th 2023).

Genomic DNA from FFPE blocks was extracted, purified, and quantified using the MagPure FFPE DNA LQ Kit (Magen). A commercially available targeted AmoyDx HANDLE HRR NGS Panel covering 32 genes (17 HRR genes: ATM, ATR, BARD1, BRCA1, BRCA2, BRIP1, CHEK1, CHEK2, FANCA, FANCL, MRE11, NBN, PALB2, RAD51B, RAD51C, RAD51D, RAD54L; other cancer predisposition genes: AR, BRAF, CDH1, CDK12, ERBB2, ESR1, HDAC2, HOXB13, KRAS, NRAS, PIK3CA, PPP2R2A, PTEN, STK11, TP53) was used for next-generating sequencing analysis (Amoy Diagnostics, Xiamen, China). Prepared libraries were sequenced on the Illumina MiSeq system (Paired-End Reads, 2×150 cycles) and analyzed on the AmoyDx ANDAS Data Analyzer (Amoy Diagnostics, Xiamen, China) to accurately detect single nucleotide variants (SNVs) and short insertions/deletions (InDels) with detection sensitivity at variant allele frequency (VAF) ≥ 5%. The average coverage depth was 543X, and the results were further manually filtered to ensure that no false positives were reported.

The avidin-biotin-peroxidase complex method was adopted to perform immunohistochemistry (IHC) staining, as previously described (26, 27). After deparaffinization, rehydration and antigen retrieval, FFPE slides were incubated with CD8 (BD Pharmingen, US) and PD-L1 (Abcam, UK) primary antibodies overnight at 4°C. The secondary antibodies (Abcam, UK) were then applied and incubated at 37°C for an hour. After that, the slides were rinsed with phosphate buffer saline (PBS), stained with 3,3′-diaminobenzidine (DAB), and counterstained with hematoxylin. All specimens were independently and blindly evaluated by two expert pathologists (A. Li and M. Ruan). For CD8 staining, the number of positive cells was calculated in a 0.5 mm diameter cylinder and expressed as the mean value of the five fields per sample (cells per spot) (26). Individual values were used for correlation and survival analyses. Immunotype, as previously described (28), including “inflamed”, “immune excluded” and “immune desert”, was defined based on CD8⁺ T cell infiltration pattern at the tumor core and stroma. “Inflamed” represents diffuse CD8⁺ T cell infiltration in the tumor. “Immune excluded” is characterized by CD8⁺ T cell infiltration limited only to the invasive margin of the tumor. “Immune desert” stands for the absence of CD8⁺ T cell infiltration in the tumor. Tumor tissues with PD-L1⁺ tumor cells over 1% are considered PD-L1 positive. For combined survival analysis of HRR mutation and immune infiltration, patients were categorized into three groups: Group I, patients without HRR mutation and a low level of T cells; Group III, patients with HRR mutation and with a high level of T cells; Group II, the remaining patients (the average number of CD8⁺ T cells as cut-off value).

Data from the Cancer Genome Atlas (TCGA), the Molecular Taxonomy of Breast Cancer International Consortium (METABRIC), and the Memorial Sloan Kettering Cancer Center database 2018 (MSKCC 2018) was utilized to compare mutational frequencies. TCGA and MSKCC data was downloaded from the cBioPortal (29) (http://cbioportal.org/, files “TCGA, Firehose Legacy”, “MSK, Cancer cell 2018”). A total of 105 TNBC patients from the TCGA database and 225 TNBC MSK data were analyzed. METABRIC was retrieved as described previously (30), and 320 TNBC patients were used for analysis.

The data distribution was characterized by frequency tabulation and summary statistics. Differences in categorical data was assessed through the Chi-square test or Fisher’s exact test. Kaplan-Meier curves and the Log-rank test were used to compare unadjusted survival between study groups. Cox proportional hazards models were used to evaluate hazard ratios across subgroups and to adjust for patients’ clinicopathological and therapeutic parameters. Two-sided P < 0.05 was considered statistically significant. All statistical analyses were performed using Stata version 16.1 (StataCorp LP, College Station, Texas).

A total of 238 and 196 TNBC patients were included in the Ruijin Cohort 1 and Ruijin Cohort 2, respectively. Public reference cohorts retrieved from cancer genome databases consist of 320 TNBC patients from METABRIC, 105 from TCGA, and 225 from MSKCC. All cohorts were used for mutation rate analysis. The pooled cohort (n = 434) of Ruijin Cohort 1 and Ruijin Cohort 2 was used for clinicopathological characteristics analysis, from which a subset of 383 patients with follow-up data was used for survival and predictive value of carboplatin treatment efficacy analysis ( Figure 1 ).

Patient characteristics according to mutation status are shown in Table 1 . Patients below 50 years accounted for 46.31% of the total cohort. Breast or ovarian cancer history was found in 63 (14.52%) patients, whereas 31 (7.14%) patients had a family history of other cancers. Twenty-three (5.30%) patients had at least one more primary malignancy other than breast cancer. Invasive ductal carcinoma (IDC) was observed in 84.56% of the study cohort, whereas the proportion of patients with a histological grade of III and a Ki-67 level above 30% were 67.74% and 76.04%, respectively. And HER-2 low status was seen in 55.07% of patients ( Table 1 ).

Tumor mutations in 17 HRR genes were detected in 21.89% (95/434) patients ( Figure 2A ). In Ruijin Cohort 1, BRCA1 was the most frequently mutated HRR gene (10.08%), followed by BRCA2 (5.88%), ATR (2.52%), BARD1 (2.10%) and PALB2 (2.10%) ( Table 2 ). As for Ruijin Cohort 2, BRCA1 was also the most frequently mutated HRR gene (11.22%), followed by ATM (2.04%) and BRCA2 (1.53%) ( Table 2 ). Notably, BRCA1 was the most frequently mutated HRR gene across all cohorts (except MSKCC 2018), with a mutation rate ranging from 2.67% to 11.22% ( Figures 2A–D ). In MSKCC 2018, the most frequently mutated HRR gene was ATM (5.33%). The pathological variation (PV) prevalence of BRCA1 was similar across public controls (2.67% to 2.86%), but was much higher (10.60%) in the study cohort, whereas the PV prevalence of BRCA2 in our study (3.92%) and MSKCC 2018 (4.44%) greatly outnumbered that of TCGA (0.95%) and METABRIC (0.63%) ( Figure 2E ).

Mutations in other cancer predisposition genes were detected in 59.91% (260/434) of patients. Both in Ruijin Cohort 1 and 2, TP53 was the most frequently mutated panel gene (Ruijin Cohort 1: 61.76%; Ruijin Cohort 2: 64.80%), followed by PIK3CA (Ruijin Cohort 1: 23.53%; Ruijin Cohort 2: 15.82%), and PTEN (Ruijin Cohort 1: 7.98%; Ruijin Cohort 2: 8.67%) ( Table 2 ). TP53, PIK3CA, and PTEN were also the most highly mutated gene across all cohorts, ranging from 55.88% to 81.33%, 6.67% to 23.53% and 3.81% to 8.89% of mutation rate in TCGA, METABRIC and MSKCC 2018, respectively ( Figures 2A–E ).

Locations of mutations and domains in proteins encoded by BRCA1, BRCA2, and other predisposition genes are shown by lollipop structures ( Figure 3 ). The mutation hotspot of BRCA1 lay at protein domain p.I1824Dfs*3, which accounts for 6.52% (3/46) of BRCA1 mutations; and no mutation hotspot was detected for BRCA2. Mutations of TP53 mainly located at p.R273H/C/L (6.57%, 18/274). The most frequently mutated protein domain of PIK3CA was p.H1047R/L (70.11%, 61/87). PTEN mutations most frequently occurred at p.R130*/Q (5.56%, 2/36). And ERBB2 mutations mainly located on p.V777L (23.08%, 3/13). Further correlation analysis between HRR gene mutations indicated several co-mutation patterns, including PALB2 and RAD51B, NBN and RAD54L, ATR and RAD54B, ATR and RAD51L, FANCA and NBN, CHEK2 and PPP2R2A, and PPP2R2A and RAD54L (all odds ratios > 5.0, all P < 0.001) ( Supplementary Figure 1A and Figure 1B ).

In 434 patients from the pooled cohort of Ruijin Cohort 1 and Ruijin Cohort 2 with early TNBC, BRCA tumor mutation was associated with menopausal status (mutation rate 18.32% in premenopausal women versus 10.78% in postmenopausal women, P = 0.035), family history of breast/ovarian cancer (mutation rate 23.81% in women with family history versus 12.67% in women without family history, P = 0.016) and high Ki-67 levels (mutation rate 16.97% in high Ki-67 group versus 5.77% in low Ki-67 group, P = 0.005). TP53 mutation was associated with high Ki-67 levels (mutation rate 68.18% in high Ki-67 group versus 47.12% in low Ki-67 group, P < 0.001). A significantly higher PIK3CA mutation rate was found in patients above 50 years (mutation rate 24.89% in patients above 50 years versus 14.43% in patients below 50 years, P = 0.007) and postmenopausal women (mutation rate 26.29% in postmenopausal women versus 12.87% in premenopausal women, P < 0.001), and was also associated with low tumor grade (mutation rate = 24.49% in grade I-II tumors versus 17.01% in grade III tumors, P = 0.049), HER-2 low status (mutation rate = 26.78% in HER-2 low tumors versus 11.79% in HER-2 negative tumors, P < 0.001) and low proliferation indices (mutation rate = 29.81% in low Ki-67 group versus 16.97% in high Ki-67 group, P = 0.004). No association between tumor mutation and secondary primary tumor was found in the study cohort ( Table 3 ). Multivariate analysis further showed that breast/ovarian cancer family history (OR = 2.15, 95%CI 1.10–4.19, P = 0.025) and high Ki-67 levels (OR = 2.91, 95%CI 1.20–7.07, P = 0.018) were two independent predictive factors for BRCA mutation. A high Ki-67 level is also an independent predictive factor for TP53 mutation (OR = 2.41, 95%CI 1.53–3.77, P < 0.001). Moreover, HER-2 low status was an independent predictive factor for PIK3CA mutation (OR = 2.37, 95%CI 1.37–4.08, P = 0.002) ( Supplementary Table 1 ).

After a median follow-up of 44.73 months, eighty-six (22.45%) out of 383 patients with follow-up data experienced with locoregional recurrence and (or) distant metastasis, whereas a total of 46 (12.01%) patients died. Univariate analysis indicated that 5-year RFS (79.88% versus 70.32%, P = 0.030), but not OS (88.99% versus 82.46%, P = 0.230) was better in HRR gene mutated patients compared with their wildtype counterparts ( Figure 4A ; Supplementary Figure 2 ). Yet after adjusting for age, tumor location, tumor grade, cT stage, cN stage, Ki67 level, lymphovascular invasion status, and carboplatin administration status, no difference in RFS (Adjusted P = 0.070) and OS (Adjusted P = 0.318) was observed between HRR gene mutated and non-mutated patients ( Supplementary Figures 3A, B ).

A total of 174 (45.43%) patients received carboplatin-containing adjuvant chemotherapy. Within the carboplatin treatment group, RFS events were found in 19 (10.92%) patients, while death occurred in 6 (3.45%) patients. In the rest 209 (54.57%) patients treated without carboplatin, locoregional recurrence and (or) distant metastasis was observed in 67 (32.06%) patients, whereas death was seen in 40 (19.14%) patients. Subgroup analysis further indicated that in patients treated with carboplatin chemotherapy, no survival difference was seen between the HRR genes mutation and non-mutation groups in 5-year RFS (93.84% versus 86.34%, P = 0.105) and OS (96.88% versus 96.39%, P = 0.630) ( Figure 4B ). And in subgroup without carboplatin treatment, no significant difference was found in terms of 5-year RFS (66.44% versus 52.74%, P = 0.104) and OS (80.97% versus 69.29%, P = 0.275) between HRR gene mutation carriers and non-carriers ( Figure 4C ).

In the study cohort, a lower percentage of invasive ductal carcinoma (P = 0.026) and a higher frequency of unknown tumor grade (P = 0.003) was seen in the carboplatin treatment group than in the non-carboplatin treatment group, whereas Ki67 levels were higher in the carboplatin treatment group than in the non-carboplatin treatment group (P < 0.001). No difference was observed in other clinicopathological parameters ( Supplementary Table 2 ). The 5-year RFS (88.17% versus 56.06%, P < 0.001) and OS (96.47% versus 72.15%, P < 0.001) in patients treated with carboplatin chemotherapy were significantly higher than those without ( Supplementary Figure 4A ). In HRR gene mutation carriers, superior 5-year RFS (93.84% versus 66.44%, P = 0.006) and OS (96.88% versus 80.97%, P = 0.030) were observed in the carboplatin treatment group compared with the non-carboplatin group ( Supplementary Figure 4B ). Patients without HRR gene mutations showed better 5-year RFS (86.34% versus 52.74%, P < 0.001) and OS (96.39% versus 69.29%, P < 0.001) when carboplatin was added to the treatment regimen ( Supplementary Figure 4C ). Furthermore, in multivariate analysis, carboplatin administration acted as an independent protective factor for RFS (HR = 0.23, 95%CI 0.13–0.38, P < 0.001) and OS (HR = 0.15, 95%CI 0.06–0.37, P < 0.001) in early TNBC ( Supplementary Figures 3A, B ).

Across TP53, PIK3CA, and PTEN-PIK3CA pathway mutation and non-mutation subgroups, RFS and OS for early TNBC were significantly higher in patients treated with carboplatin chemotherapy than without regardless of gene mutation status (All Interaction P > 0.05) ( Figures 5A, B ; Supplementary Figures 5C–F ). Nevertheless, survival benefit from carboplatin chemotherapy was only seen in BRCA mutation non-carriers. In BRCA mutation carriers, survival benefit from carboplatin chemotherapy was only observed in RFS (P = 0.046), but not OS (P = 0.077) ( Supplementary Figures 5A, B ).

In Ruijin Cohort 1, tumor specimens from 162 patients successfully achieved CD8 and PD-L1 immune staining and evaluation ( Figure 6A ). Forty-six patients were immune inflamed, 39 were immune exclude, and 77 were immune desert, respectively. Of 31 HRR gene mutated patients, 21 (68%) were immune inflamed, 5 (16%) were immune excluded, and 5 (16%) were immune desert, while in 131 HRR gene wildtype patients, 25 (19%) were immune inflamed, 34 (26%) were immune excluded, and 72 (55%) were immune desert (P < 0.001) ( Figure 6B ). Twenty-one patients were BRCA mutation carriers, consisting of 14 (67%) immune inflamed, 2 (10%) immune excluded, and 5 (23%) immune desert immunotypes, whereas within 141 noncarriers, the distribution of inflamed, immune excluded and immune desert immunotypes were 32 (23%), 37 (26%) and 72 (51%), respectively (P < 0.001) ( Figure 6C ). The level of immune infiltration and CD8⁺ T cell count in HRR gene or BRCA mutation carriers were significantly higher than their wildtype counterparts (both P < 0.001) ( Figures 6D, E ). Eighty one percent of HRR gene mutated tumors expressed PD-L1, which was 53% in HRR gene wildtype tumors (P < 0.001) ( Figure 6F ). In BRCA mutation carriers and non-carriers, 76% and 56% tumors expressed PD-L1 (P = 0.003), respectively ( Figure 6G ). Kaplan-Meier curves found that patients with both HRR gene mutation and high CD8⁺ T cell counts (average count as cut-off value) had the best RFS and OS, whereas patients without HRR gene mutation and low CD8⁺ T cell counts had the worst RFS (P < 0.001) and OS (P = 0.019) ( Figures 6H, I ).