Discovery and Evaluation of Biomarkers for Triple-Negative Breast Cancer Subtypes Uncovers Patient Stratification and Targeted Therapeutic Strategies

Triple-negative breast cancer (TNBC) is the most heterogeneous and aggressive subtype of breast carcinoma, defined by the absence of clinical biomarkers and the lack of targeted therapies. Despite numerous clinical trials, patient stratification remains suboptimal, limiting the identification of effective treatment strategies. In this study, we aimed to identify biomarkers exclusively expressed in the basal mammary epithelial compartment to refine TNBC subclassification. Through computational analysis of single-cell RNA sequencing data, we defined a set of basal identity genes, which were subsequently validated by immunohistochemistry in two independent TNBC cohorts. This approach enabled the identification of a novel TNBC subgroup, termed true basal TNBC (tB-TNBC), associated with poorer prognosis and distinct molecular features. To uncover therapeutic vulnerabilities in this subgroup, we conducted a high-throughput screen of 3,200 FDA-approved compounds in breast cancer cell lines classified by basal marker expression. This analysis identified dasatinib as a promising candidate with selective activity against tB-TNBC models. Furthermore, TAGLN emerged as a strong predictive biomarker of dasatinib response, with functional studies confirming its role in modulating drug sensitivity. Altogether, these findings support the clinical utility of basal markers for TNBC stratification and highlight a targeted treatment opportunity for tB-TNBC patients. Overall design: Total RNA was extracted using the NZY miRNA Isolation & RNA Clean-up Kit (MB13402, NZYtech) following the manufacturer's protocol. The purity and concentration of the RNA from each sample were assessed using an Agilent TapeStation with an RNA ScreenTape Analysis Kit. mRNA enrichment was subsequently performed using polyA-tail-connected magnetic beads and oligos. This was followed by double-stranded DNA synthesis and polymerase chain reaction (PCR) amplification using specific primers. The PCR products were subjected to thermal denaturation to produce single-stranded DNA, which was cyclized into a circular DNA library using bridge primers. Sequencing was conducted using the DNBSEQ platform (BGI Genomics Co.). Raw sequencing data were preprocessed using SOAPnuke software (BGI Genomics Co.). This involved removing reads with adaptor contamination, more than 5% N content, and low-quality reads (where more than 20% of the bases had a quality score below 15). The resulting "Clean Reads" were saved in FASTQ format. RNA-seq data were analyzed using the Galaxy workbench platform [33], adhering to specific recommendations [34]. Quality control and trimming of the reads were performed using MultiQC [35] and Cutadpt [36]. The reads were then mapped to the reference genome (Hg38, human genome build 38) using STAR [37]. From the mapped sequences, the number of reads per annotated gene was counted using featureCounts [38]. Subsequently, DESeq2 [39] was used to normalize the read counts and identify differentially expressed genes. Functional enrichment analysis of differentially expressed genes was performed using GSEA [40]