AI-based B2SC identifies pre-existing TOP2A/E2F-active populations driving resistance to T-DXd in HER2+ breast cancer

Background: Trastuzumab deruxtecan (T-DXd) has significantly improved outcomes for HER2-positive (HER2+) breast cancer. However, resistance emerges in some patients and remains a critical clinical challenge. Increasing evidence suggests that resistance may arise from pre-existing tumor cell heterogeneity rather than acquired de novo. Dysregulation of DNA replication and repair pathways has been implicated in resistance to DNA-damaging agents. Topoisomerase II A (TOP2A), a key DNA topology regulator, is frequently overexpressed in aggressive breast cancers and associated with poor prognosis. We hypothesized that a resistant-like TOP2A-high subpopulation pre-exists before T-DXd treatment and can be detected not only by single-cell profiling but also through our AI-based Bulk-to-Single-Cell (B2SC) model. Methods: Whole-transcriptome analysis compared T-DXd-resistant (TDXd-R) HER2+ breast cancer cells with their parental counterparts. Public single-cell RNA-sequencing (scRNA-seq) datasets from six HER2+ breast cancer cell lines and eight HER2+ patient tumors were analyzed to identify pre-existing resistant-like populations. Bioinformatic analyses identified molecular signatures and predicted synergistic therapeutic strategies. Experimental validation was performed using TDXd-R cell lines. In silico drug perturbation assessed therapeutic effects on the TOP2A-high population. WittGen's B2SC was applied to bulk transcriptomic datasets to infer hidden subpopulations. Results: TDXd-R HER2+ breast cancer cells showed consistent upregulation of TOP2A and E2F-driven cell-cycle programs, with enrichment of DNA repair pathways. scRNA-seq revealed a distinct TOP2A-high cell population present at baseline across all examined HER2+ breast cancer cell lines and patient tumors prior to T-DXd treatment. Gene regulatory network analysis demonstrated elevated E2F-regulon activity governing DNA synthesis and mitotic progression. Application of the B2SC to bulk data identified a corresponding TOP2A-high subpopulation, indicating that this resistance-associated state is detectable even in the absence of single-cell data. In silico drug perturbation analysis suggested persistence of resistance-associated programs in TOP2A-high cells following TOP1A inhibition. Based on molecular signatures, we selected potential synergetic agents-including Lucanthone and a CDK2 inhibitor-to target this population. Conclusion: These findings indicate that resistance to T-DXd in HER2+ breast cancer is associated with enrichment of a pre-existing TOP2A-high, E2F-active cell population. This transcriptional state is identifiable using both scRNA-seq and B2SC-applied bulk transcriptomic inference and may serve as a biomarker of intrinsic resistance. The data further support the rationale for combination strategies to prevent or overcome resistance to T-DXd.