Genomic characterization reveals distinct mutational landscapes and therapeutic implications between different molecular subtypes of triple-negative breast cancer

Triple-negative breast cancer (TNBC) has high heterogeneity, poor prognosis, and limited treatment success. Recently, an immunohistochemistry-based surrogate classification for the “Fudan University Shanghai Cancer Center (FUSCC) subtyping” has been developed and is considered more suitable for clinical application. Seventy-one paraffin-embedded sections of surgically resected TNBC were classified into four molecular subtypes using the IHC-based surrogate classification. Genomic analysis was performed by targeted next-generation sequencing and the specificity of the subtypes was explored by bioinformatics, including survival analysis, multivariate Cox regression, pathway enrichment, Pyclone analysis, mutational signature analysis and PHIAL analysis. AKT1 and BRCA1 mutations were identified as independent prognostic factors in TNBC. TNBC molecular subtypes encompass distinct genomic landscapes that show specific heterogeneities. The luminal androgen receptor (LAR) subtype was associated with mutations in PIK3CA and PI3K pathways, which are potentially sensitive to PI3K pathway inhibitors. The basal-like immune-suppressed (BLIS) subtype was characterized by high genomic instability and the specific possession of signature 19 while patients in the immunomodulatory (IM) subtype belonged to the PD-L1 ≥ 1% subgroup with enrichment in Notch signaling, suggesting a possible benefit of immune checkpoint inhibitors and Notch inhibitors. Moreover, mesenchymal-like (MES) tumors displayed enrichment in the receptor tyrosine kinase (RTK)-RAS pathway and potential sensitivity to RTK pathway inhibitors. The findings suggest potential treatment targets and prognostic factors, indicating the possibility of TNBC stratified therapy in the future.


Overall characterization of the SXCH cohort
Most of the SNVs observed in the SXCH cohort were missense, followed by frameshift deletions and nonsense mutations (Fig. 2A).The majority of the variants were SNPs (Fig. 2B).Furthermore, almost half of the SNVs were found to be C > T, consistent with the trend seen in the TCGA TNBC cohort (Fig. 2C).SNPs can be classified as transitions and transversions.In our cohort, transitions occur more frequently than transversions (Fig. 2D).Of note is that patients aged over 50 had a higher TMB (Fig. 2E).Analysis of pairwise mutual exclusivity and co-occurrence indicated that CDH23/ARID1A, NOTCH1/BRCA2, NOTCH1/ATRX, FANCD2/BRCA2, FANCD2/LRP1B, ATRX/BRCA2, ALK/PTEN showed significant co-occurrence (p < 0.05), and TP53 showed exclusivity with CYP2D6 (p < 0.01), indicating a multiplicity of interactions among driver genes in TNBC (Fig. 2F).With regard to SV, thirteen gene fusions were observed over the 71 samples, of which three were trans-chromosomal fusions (Table S5).The most frequent fusions (2/13) were seen in JAK1.The JAK1-EPB41 fusion and the JAK1-AK4 (intergenic) fusion were identified, indicating driver gene JAK1 may have a significant impact on the development of TNBC (Fig. 2G).To assess somatic CNVs, 18 arm-level and 57 focal CNVs were found in the 71 TNBC samples (Table S6).
Mutations in the dominant cancer-associated genes that were significantly linked with DFS were analyzed.The KM analysis showed that patients with mutations in AKT1 had markedly shorter DFS (p = 0.0019), and similar results were found for patients with CYP2D6 (p = 0.003) and BRCA1 (p = 0.018) mutations (Fig. 2H).Further multivariate Cox regression showed that the presence of mutations in AKT1 and BRCA1 were independent predictors of prognosis (Fig. 2I).The mutations in both genes in the SXCH cohort were annotated using lollipop plots (Fig. S1).

Genomic profiles of molecular subtypes of TNBC
The mutational landscape varied substantially across the molecular subtypes (Fig. 3A). Figure 3B illustrates H&E staining and IHC for the four subtypes (Fig. 3B).Overall, mutations in TP53 were seen most frequently in all four subtypes while mutations in PIK3CA occurred most frequently in LAR patients compared with the other subtypes.The BRCA2 gene showed more frequent mutation (29%) in IM (p < 0.001), while mutations in BRCA1 (40%) were more often seen in MES than in the other subtypes (p < 0.001).VHL mutations showed an

Clonal and subclonal structural analysis of four molecular subtypes
We also compared the MATH scores and VAF in two dimensions, namely, mutational heterogeneity and the frequencies of allelic mutations.We found that although the MATH scores did not differ significantly between the four subtypes, the VAF values did (Fig. 6A).Pyclone analysis was then performed, defining clusters with the highest cell prevalence as clones and others as subclones.6B).These findings demonstrate specific mutational differences between the four subtypes.The clonal and subclonal mutational profiles for the different patients are shown in Fig. 6C.Additionally, a significant association was observed between CN and the maximum VAF (Fig. 6D).
Pathway enrichment analysis was undertaken to explore the functions of the clonal and subclonal genes (Fig. 6E).This showed that clonal genes in LAR were significantly enriched in ERBB2 signaling, while subclonal genes were associated with transcriptional regulation by TP53.Enrichment of subclonal genes in BLIS suggested that patients with this subtype may be resistant to platinum drugs.Several classical DNA repair pathways were enriched in the IM, such as homology-directed repair through homologous recombination repair, and DNA double-strand-break repair.Clonal genes in MES were associated with both apoptosis and cellular senescence, while subclonal genes were enriched in choline metabolism (Fig. 6E).These differences in functional enrichment between the four subtypes suggest an association between subtypes and the evolution of clonal patterns.The VAF values differed between immune phenotypes but not in other clinical groups, suggesting that VAF may influence the immune phenotype (Figs.6F-G, S6A).Thus, we investigated the effects of these genetic mutations underlying these differences (Fig. S6B).

Mutational signature distribution and the prognostic effect of signature 9
Mutational signatures have been utilized for multiple purposes, including comprehending tumor development, identifying gene alterations associated with mutational processes, and, most importantly, serving as biomarkers for predicting treatment response 16 .The NGS data were analyzed to extract mutational signatures, based on the COSMIC database (Table S8).C > T was found to have a greater frequency (Fig. 7A).Signature 20 was mainly observed in patients with the LAR subtype, while signature 8 was mostly seen in patients with the MES subtype (Fig. 7B), suggesting that LAR-subtype tumors may be associated with mismatch repair (MMR) deficiencies, while MES-subtype tumors may be associated with homologous recombination (HR) deficiencies (Table S8).Remarkably, signature 19 was only observed in patients with the BLIS subtype, although its function is not  www.nature.com/scientificreports/curve indicated that patients lacking signature 9 had a more favorable prognosis (Fig. 7C).The differences in mutations between patients with and without signature 9 are shown in Fig. S7B.The top five signatures with the greatest weight (signatures 3, 4, 7, 11, and 23) were investigated (Fig. S7B).Furthermore, the relationship between mutation signature and clonality measures (the cluster number) was explored.The results showed that patients who underwent signature 24 had a greater cluster number compared to patients who lacked signature 24, suggesting that signature 24 is more active in tumors with a large number of subclonal mutations (Fig. 7D).
In addition, the "SomaticSignatures" package was also used to explore the mutational signature of TNBC.The results showed that when the number of signatures exceeded 9, the approximation of the residual sum of squares (RSS) and explained variance did not improve significantly with each additional signature, suggesting that TNBC could be categorized into 9 mutational signatures (Fig. S8A-C).Figure S8D illustrates the relative contribution of each signature in different molecular subtypes.Signature S8 was found to be more abundant in AR subtype, while signatures S5 and S9 contributed more in MES subtype.The biological significance behind each signature needs to be further explored in the future.

TNBC Subtypes based on mutational signature and CNVs
Alterations in DNA, such as CNVs and somatic mutations, act as "drivers" promoting tumor growth.They also contain footprints associated with specific biological processes related to tumor growth.We used k-means and consensus clustering, leading to the identification of four clusters based on COSMIC mutational signatures (Fig. 8A).These were mutation subtype 1 (sig11-LS) which was dominated by signature 11, mutation subtype 2 (sig7-LS) which included mainly signature 7, mutation subtype 3 (sig23-LS), including signature 23, and mutation subtype 4, which showed no specifically dominant signature (mixed).Furthermore, four clusters based on CNV peaks were identified (Fig. 8B; Table S9), namely, the CNV subtype 1, with frequent Chr6q22.1 amplification (6q22.1amp),CNV subtype 2, with frequent Chr8q24.21 amplification (8q24.21amp),CNV subtype 3, showing frequent Chr7q21.2 amplification (7q21.2amp),and CNA subtype 4, with low chromosomal instability (low-CIN).To elucidate the relationships between molecular subtypes and mutational signatures, we investigated the specific associations among molecular, CNV, and mutation subtypes.This showed that most of the BLISsubtype samples (74%) were classified as 8q24.21ampand low-CIN, while the IM and MES subtypes were rarely classified as 8q24.21amp(Fig. 8C).The potential prognostic value of these findings was investigated, revealing that molecular subtypes, CNV, and mutation subtypes were not reliable predictors of DFS (Fig. 8D).The specific associations between molecular subtypes, CNV, and mutation subtypes with clinical groups were examined, and it was found that clinical groups did not predict prognosis reliably (Fig. S9).

Therapeutic implications of TNBC Subtypes.
In total, 70 (98.5%)patients were found to have at least a single clinically relevant genetic alteration.Alterations that could be targeted, together with possible therapies, are shown in Fig. 9A and Supplementary Table S10.

Discussion
Treating primary TNBC as a single disease has been found to result in poor therapeutic response 3 .The proposal of molecular subtypes and stratified treatment is an inevitable trend for future research in TNBC 17 .Distinct from Lehmann subtyping and Burstein subtyping, which mainly include European and American populations, the FUSCC subtyping are more suitable for describing the biological characteristics of TNBC patients in East Asia 7,8,17 .The emergence of FUSCC subtyping provides direction for the treatment of TNBC in the clinic.With the need for more portable classifiers in the clinic, an IHC-based surrogate method to FUSCC subtyping have emerged.Previous studies have shown a large agreement between IHC-based and mRNA-based classification (overall Cohen's κ coefficient = 0.678) 14 .Of these, both classification methods showed the highest agreement in categorizing TNBC into LAR subtypes (Cohen's κ coefficient [κ] = 0.821; 95% CI, 0.733-0.908) 14.In addition, the percentage of samples classified as the same subtype by both classification methods was 76.7%, suggesting that most of the molecular features and therapeutic significance of the mRNA subtypes are retained in the corresponding IHC subtypes.The application of IHC detection of AR has been extensively studied.Clinically, the AR signaling pathway has the potential to be targeted in these LAR tumors 18 .CD8 is a widely studied immune marker, and its expression correlates with patient prognosis and tumor response to immunotherapy 19 .High expression of FOXC1 correlates with invasive tumor biological behaviors and poor prognosis 20,21 .Therefore, it is sufficient to consider clinical interventions based on the classification of IHC.
In contrast to whole genome sequencing (WGS) and whole exome sequencing (WES), targeted NGS focuses on the region of interest, achieving ultra-high sensitivity and accuracy with less data volume, while removing the interference of redundant data and enabling rapid screening of variant loci 22 .With low sequencing cost and deep sequencing depth, targeted NGS has tremendous potential in clinical applications.In this study, IHC was used to classify tumor samples from 71 Chinese TNBC patients into four subtypes and then targeted DNA-NGS was performed on all samples, to determine the relationships between the tumor genetics and the clinical features.This study is the first to explore the association between targeted NGS results, survival outcomes, and the evolution of the TNBC subtypes.
The overall genomic characteristics of TNBC were described in our study.The most prevalent somatic mutations were found to be in TP53 and PIK3CA, with mutations present in 86% and 17% of cases, respectively, followed by BRCA1 (13%), RB1 (13%), and ATM (10%), It is worth mentioning that the observed frequencies of TP53 and PIK3CA mutations are consistent with previously reported results 17 .Moreover, in the present study, AKT1 and BRCA1 mutations were shown to be independent prognostic factors, however this result needs to be further validated by the larger sample sizes included.Dysregulated PI3K and AKT signaling is considered one of the most common oncogenic changes in TNBC 23 .Recent randomized clinical trials have shown that AKT inhibitors in combination with first-line chemotherapy drugs prolong both progression-free survival (PFS) and overall survival (OS), and this benefit is more pronounced in patients with PIK3CA/AKT1/PTEN-altered TNBCs 24,25 .It is suggested that AKT1 mutations are potential targets for the treatment of TNBC patients.In addition, BRCA1 plays a role in several cellular pathways that maintain genomic stability 26 .Gaceb et al. reported that BRCA1 pathogenic variants were associated with poor prognosis, which is consistent with our findings 27 .Notably, PARP inhibitors are effective for treating tumors with BRCA1/2 mutations 28 , suggesting that PARP inhibitors may be useful for treating TNBC patients.
The TNBC subtypes showed distinct genetic features and potentially actionable alterations.Patients with the LAR subtype showed high levels of PIK3CA mutations, with predominant enrichment in PI3K oncogenic pathways.This suggests that the LAR subtype may be sensitive to inhibitors of the PI3K pathway, such as PI3K/ AKT/mTOR inhibitors.Randomized clinical trials of inhibitors targeting this pathway for the treatment of TNBC www.nature.com/scientificreports/are currently underway.The mTOR inhibitors temsirolimus (DAT) and everolimus (DAE) were found to have notable objective responses in mesenchymal TNBC patients with PI3K pathway aberrations 29 .The results for alpelisib, a selective PI3Kα inhibitor, were also encouraging, showing significant efficacy against TNBC together with manageable toxicity when combined with nab-paclitaxel 30 .Patients with the LAR subtype of TNBC who received the PI3Kα-specific inhibitor taselisib plus enzalutamide showed improved clinical benefit in comparison with other TNBC subtypes 18 , indicating that the PI3K/AKT/mTOR pathway may be responsible for the biological changes in patients with the LAR subtype.The BLIS subtype was observed to have the greatest number of CNVs, suggesting a high degree of genomic instability.Most BLIS samples (74%) were found to be 8q24.21ampand low-CIN.In addition, signature 19 was found only in the BLIS subtype, providing direction for subsequent studies on potential molecular features.In this study, no agents were found to potentially treat patients with the BLIS subtype, which may account for the poor prognosis of this subtype.
The IM subtype was enriched with BRCA2 mutations and MLL3 mutations.Mutations in the epigenetic regulator MLL3 were observed to be key drivers of the hybrid epithelial/mesenchymal phenotype and metastasis in breast cancer 31 .Loss of MLL3 in vivo improved response to lapatinib in breast cancer 32 .It is suggested that MLL3 could be a potential therapeutic target for patients in IM subtype.The subtype was also characterized by a predominance of mutations in the Notch pathway.Notch activation leads to proinflammatory cytokine production and the presence of tumor-associated macrophages in the tumor microenvironment 33 .Preclinical studies have observed reductions in Notch signaling after treatment with G9, a small-molecule USP9x inhibitor, together with remodeling of the tumor-associated immune environment and decreased tumor growth, with minimal toxicity 34 .It is suggested that patients in the IM subtype are potentially sensitive to Notch inhibitors.Notably, the present study revealed that patients with the IM subtype were all in the PD-L1 ≥ 1% subgroup and thus might be responsive to ICIs, which block immunosuppressive receptors, such as cytotoxic T lymphocyte antigen-4 (CTLA-4) and PD-1, to improve the cytotoxicity and proliferation of TILs 35 .The use of PD-1/ L1 inhibitors together with chemotherapy has been found to be relatively successful in TNBC patients [36][37][38] , suggesting that ICIs may be an appropriate choice for patients with the IM subtype.
Tumors of the MES subtype were observed to be enriched in the RTK-RAS pathway and thus may have potential sensitivity to RTK-pathway inhibitors.Regorafenib, a multi-RTK inhibitor, has been shown to inhibit TNBC cell metastasis by targeting the SHP-1/p-STAT3/VEGF-A axis 39 .A small-molecule inhibitor cabozantinib (XL184) was also observed to significantly reduce multicellular invasive outgrowths in preclinical TNBC models 40 .This provides direction for future clinical trials targeting personalized therapy for MES patients.
There are several limitations to this study.First, the low proportion of MES subtypes may be related to the rarity of MES subtypes.This issue could be clarified using larger patient cohorts.Second, the study used only genomic data.Multi-omic data would allow a more comprehensive and in-depth investigation of TNBC heterogeneity and molecular features.Lastly, the study was retrospective and descriptive, and the results require confirmation in clinical settings.

Conclusions
To sum up, the findings caution against the use of one-size-fits-all management of patients with TNBC.The TNBC subtypes were found to have unique molecular characteristics and, consequently, require specific therapeutic management.These findings assist the elucidation of the mechanisms underlying disease progression and thus contribute to its clinical management.Nevertheless, further prospective studies are required for further validation.

Patients and samples
Patients with TNBC were retrospectively enrolled.The inclusion criteria were (1) aged at least 18 years, with confirmation of TNBC according to the guidelines of the American Society of Clinical Oncology-College of American Pathologists, (2) had not undergone neoadjuvant therapy, and (3) had provided written informed consent.Patients with multicentric or bilateral lesions were excluded.A total of 71 patients with TNBC who had undergone surgery at the Department of Breast Surgery at Shanxi Cancer Hospital between January 1, 2017, and December 31, 2019, were included.All patients received four to six cycles of anthracycline-based chemotherapy postoperatively.Resected specimens were fixed in formalin and paraffin-embedded (FFPE) and sections were evaluated separately by two experienced pathologists (RQL and JL).In the event of no consensus on the diagnosis, the case was reviewed by a third pathologist (HXM), and a majority vote was obtained.This study was approved by the institutional review board of Shanxi Cancer Hospital (Shanxi, China) (Approval Number: 2019075).We confirmed that all experiments were performed in accordance with relevant guidelines and regulations.

IHC
An automated immunohistochemistry system was used to conduct IHC analyses of FFPE samples through a two-step approach in which a primary antibody was applied, followed by the application of a polymeric conjugate consisting of multiple secondary antibodies directly linked to a dextran backbone.All of the antibodies used for IHC analyses and data interpretation are summarized in Table S1

Targeted NGS and alteration identification
Targeted NGS of 1021 genes was performed on 71 FFPE sections of TNBC specimens.DNA was extracted from the tissue using FirePureTM FFPE gDNA Extraction Kit and ultrasonicated (UCD-200, Diagenode, Seraing, Belgium) with fragment selection using Hieff NGS DNA selection beads.The DNA was quantified using a Qubit 2.0 Fluorometer with a Quanti-IT dsDNA HS Assay Kit (Thermo Fisher Scientific, MA, USA) and libraries were prepared using a custom apturing probe (IDT, IA, USA).Samples with library concentrations ≥ 20 ng/μL were sequenced using Paired-End 100 bp (PE100) with the Geneplus-2000 sequencing platform (Geneplus, Beijing, China).The raw reads were then filtered with fastp (version 1) for removal of (a) reads with adaptors, (b) reads with N base proportions > 10% of the total lengths (unsure bases), and (c) single-end reads with low-quality base proportions > 50% of the total lengths (Phred scores < 5).Samples having ≥ 80% high-quality reads (Phred scores > 30) were retained for sequencing.After sequencing, the reads were compared with the hg19 reference human genome using the default parameter of BWA 0.6.2.Duplicates were identified by unique identifiers (UIDs) and positions of templates to avoid the introduction of errors by sequencing or PCR.GATK (V4.1.4.1) was used to identify single nucleotide variants (SNVs) and small insertions and deletions (InDels), together with performing quality control.Mutect 2.0 was used to analyze somatic SNVs and InDels in tumor tissues and copy number variations (CNVs) were identified by CNVKit.The self-developed algorithm NCsv (0.2.3) was used for the assessment of structural variation (SV) 41 .Synonymous variants, documented germline variants in matched gDNA and dbSNP, and variants with population frequencies > 1% in the Exome Sequencing Project were removed.

Genome characteristics and prognostic analysis
SNV data were used as input data.The "oncoplot" function in the "maftools" package in R was used for visualization of the mutation landscape and clinical features of the patients 43 .In addition, the "somaticInteractions" function was used for the identification of co-occurring or mutually exclusive genes using pairwise Fisher's exact tests.The patients were then allocated to wild-type or mutant-type gene groups based on the presence of nonsynonymous mutations and Kaplan-Meier (KM) analysis was used to compare disease-free survival (DFS).Multivariable Cox proportional hazard models were used to calculate the hazard ratios and 95% confidence intervals (CIs).Two-tailed p-values < 0.05 were considered statistically significant.

Mutated pathway analysis
SNV data were used as input data and gene sets associated with pathways related to the DNA damage response (DDR) were acquired Wang et al. 44 .Pathways were considered mutated if they contained DDR-related genes with nonsynonymous mutations.The rates of mutations in these pathways were then evaluated and compared.The "OncogenicPathways" function in "maftools" was used for determining the numbers and proportions of mutations within ten known oncogenic pathways in TCGA.The enrichment of mutated cancer-related genes was analyzed using ClusterProfiler (version 3.12.0) 45.The mutated genes and pathways were further analyzed using the KEGG and REACTOME databases [46][47][48][49] with the calculation of p-values according to the hypergeometric distribution with correction of the false discovery rate (FDR) by the Benjamini and Hochberg method.

Mutant-Allele Tumor Heterogeneity (MATH) scores and PyClone analysis
MATH scores are quantitative assessments of tumor heterogeneity and calculate the VAF distribution width 50 .The MATH scores for the tumor samples were determined using the "maftools" package in R. The clonal population structures of the tumors were analyzed using PyClone 51 which assesses clonal structures through the grouping of SNVs with comparable frequencies together.Clusters with large mean cancer cell fractions (CCFs) were considered clonal.

Mutational signature analysis
Mutational signatures from the Catalogue of Somatic Mutations in Cancer (COSMIC) were used.These utilize the proportions of the different types of substitutions, namely, C > A, C > G, C > T, T > A, T > C, and T > G, and their trinucleotide contexts, namely, the nucleotides preceding and following the mutated bases 52 .The combinations of COSMIC signatures were determined using the "deconstructSigs" package in R using an iterative approach 53 .Specifically, the appropriate structure of the input data was constructed using the "mut.to.sigs.input"method within the "deconstructSigs" package.The "whichSignatures" method was then used to identify the COSMIC signatures in the samples, as well as the contribution of individual signatures to the overall mutational spectrum.
Furthermore, R package "SomaticSignatures" uses non-negative matrix factorization (NMF) to identify de novo discovery of mutational signatures that are present within the different TNBC subtypes, and their contribution to each tumor's mutational spectrum 54 .The 3-nucleotide mutation context of each SNV is extracted using the "mutationContext" method in the "SomaticSignatures" package.This function compares the locus of each SNV with the corresponding reference genome BSgenome.Hsapiens.UCSC.hg19 to identify the 3′ and 5′ of the SNV.Next, the frequency of each of the 96 alteration types was calculated using the "motifMatrix" method."AssessNumberSignatures" was used to determine how many signatures we expected to recognize.

CNV analysis
Analysis of CNVs was performed with GISTIC2.0 using default parameters (with a confidence level of 0.99).Visualization of the results was performed using "maftools" in R 55 .

Mutational signature-based unsupervised clustering and CNV-based unsupervised clustering
Mutational signature clustering was determined by k-means clustering, using the "kmeans" R function.The "weight" of each signature, calculated by "deconstructSigs" was used as the input data for the samples.CNV-based clustering was determined by k-means and consensus clustering using the R package "ConsensusClusterPlus" to calculate the optimal number of subtypes.Input data for the samples were the "actual copy change given" of each "peak region" obtained from the file "all_lesions.conf_99.txt".Enrichment of CNVs in the CNV-based subtypes was confirmed using Kruskal-Wallis tests.

Analysis of putative clinically relevant alterations
Somatic SNVs/Indels and CNVs were analyzed by Precision Heuristics for Interpreting the Alteration Landscape (PHIAL) software (version 1.0.R) with default parameters and database.

Statistical analysis
Student's t-test, analysis of variance, and the Kruskal-Wallis test were utilized to compare continuous variables and ordered categorical variables whilst Pearson's chi-square test and Fisher's exact test were employed for comparison of unordered categorical variables.Survival curves were constructed using the Kaplan-Meier product limit method and compared with the log-rank test.A Cox proportional hazard regression model adjusting for available prognostic clinical covariates was performed to calculate HR and 95% CIs.The p values were adjusted to FDR using the Benjamini-Hochberg procedure in multiple comparisons.All analyses were performed using R packages version 3.4.2.

Figure 2 .Figure 3 .
Figure 2. Overall characterization of the SXCH cohort.(A) Variant classifications in the SXCH cohort.(B) Variant types in the SXCH cohort.(C) SNV classes in the SXCH cohort.(D) Boxplot showing frequencies of transitions and transversions in the SXCH cohort.(E) TMB values in relation to age in the SXCH cohort.(F) Pairwise mutual exclusivity and co-occurrence in the SXCH cohort.(G) Gene fusions in patients with TNBC.Gene fusions are seen between genes corresponding to the ends of each curve.(H) Kaplan-Meier analysis of DFS in patients with and without mutations in AKT1, CYP2D6, and BRCA1.(I) Forest plot showing results of the multivariate Cox proportional-hazard regression of AKT1 and BRCA1 mutations and clinical features.Ti, transition; Tv, transversion; DFS, disease free survival; TMB, tumor mutation burden."*" represents that p-value < 0.05; "**" represents that p-value < 0.01, and "***" represents that p-value < 0.001.

Figure 4 .Figure 5 .Figure 6 .Figure 7 .
Figure 4. Significantly altered somatic CNVs in TNBC subtypes.(A) Venn diagram showing significant arm-level and focal CNVs in TNBC subtypes.(B) Significant somatic CNVs in the LAR, BLIS, IM, and MES subtypes.The green line represents the threshold Q-value = 0.25.Significantly amplified focal peaks are shown in red and significantly deleted focal peaks are shown in blue.(C) Heatmap showing regions with the greatest CNV frequency.Columns represent individual samples.The 71 samples in the cohort are shown according to their molecular subtype.Numbers alongside the heatmap indicate mutation frequencies.Subtype proportions are shown on the right of the percentage stack histogram.CNV, copy number alterations; Amp, amplification; Del, deletion; LAR, luminal androgen receptor; BLIS, basal-like immune-suppressed; IM, immunomodulatory; MES, mesenchymal-like.

Figure 8 .
Figure 8. Subtyping of TNBC with mutation signature and CNV data.(A) Clustering based on mutation signatures.The heatmap shows the contribution of mutation signatures to the four mutation subtypes.(B) Clustering of CNVs shown by GISTIC peaks.Heatmap showing log2 copy-number ratios in the genome.(C) Associations between molecular and mutation subtypes (left), molecular and CNV subtypes (middle), and CNV and mutation subtypes (right).(D) Associations between molecular (left), mutation (middle), and CNV (right) subtypes with DFS.Cox regression analysis with adjustment for age was used.Hazard ratios are shown with 95% CIs.LAR, luminal androgen receptor; BLIS, basal-like immune-suppressed; IM, immunomodulatory; MES, mesenchymal-like.

Figure 9 .
Figure 9. Overview of clinically relevant mutations in 71 TNBC samples.(A) Landscape of mutated genes and their potential therapeutic applications in TNBC.Molecular subtypes are indicated by color.(B) Numbers of clinically relevant mutations in TNBC molecular subtypes.(C) Top 15 genes with greatest numbers of clinically relevant mutations.(D) Proportions of patients potentially benefiting or insensitive to therapies specific for subtype-specific TNBC therapy.LAR, luminal androgen receptor; BLIS, basal-like immune-suppressed; IM, immunomodulatory; MES, mesenchymal-like.
. Staining was conducted as per the Roche automated IHC instrument (USA).All IHC protocols and antibodies used herein were included in a standard IHC panel used in the pathology laboratory.