Integrative Proteomics and Machine Learning Identify SLC27A2 as a Candidate Biomarker and Potential Mediator of...

Background: Pyrotinib, an irreversible pan-HER tyrosine kinase inhibitor, has demonstrated substantial clinical efficacy in patients with HER2-positive breast cancer (BC). However, intrinsic and acquired resistance remain important challenges limiting therapeutic benefit, and reliable biomarkers for predicting pyrotinib response are currently unavailable. This study aimed to identify molecular determinants associated with pyrotinib resistance and uncover their underlying mechanisms. Methods: Pre-treatment tumour samples from an exploratory discovery cohort of 12 patients with HER2-positive BC receiving pyrotinib-containing neoadjuvant therapy were analysed by proteomic profiling. Differentially expressed proteins (DEPs) between the pathological complete response (pCR) and non-pCR groups were integrated with weighted gene co-expression network analysis and protein-protein interaction network analysis to identify candidate proteins. The prognostic relevance of the candidate genes was subsequently evaluated using 127 machine-learning strategies across three independent BC cohorts. Models were ranked according to the mean area under the receiver operating characteristic curve (AUC) across two evaluation cohorts, and SHapley Additive exPlanations (SHAP) analysis was performed separately in both cohorts to prioritise a candidate for subsequent investigation. In vitro and in vivo experiments were then conducted to evaluate the biological role of the prioritised candidate and its association with pyrotinib sensitivity. Finally, the association between pre-treatment SLC27A2 expression and pCR was evaluated in an independent, non-overlapping retrospective cohort of 103 patients receiving pyrotinib-containing neoadjuvant therapy. Results: Exploratory proteomic profiling of 12 pre-treatment tumour samples identified 617 DEPs between the pCR and non-pCR groups. Among 127 machine-learning strategies used to evaluate the prognostic relevance of the candidate genes, the glmBoost-random forest model achieved the highest mean AUC across the two evaluation cohorts (mean AUC = 0.678). SHAP analysis showed that SLC27A2 ranked second in GSE16446 and first in GSE48390 according to mean absolute SHAP values, supporting its prioritisation for subsequent investigation. Functional experiments showed that SLC27A2 promoted proliferation, migration, invasion, and epithelial-mesenchymal transition in HER2-positive BC cells. SLC27A2 knockdown enhanced pyrotinib sensitivity in vitro. In the xenograft experiment using female BALB/c nude mice, both SLC27A2 knockdown and pyrotinib treatment reduced tumour growth, and a significant interaction between the two factors was observed for endpoint tumour weight (p for interaction = 0.041). Mechanistically, SLC27A2 knockdown reduced lipid accumulation and PPARα expression, whereas pharmacological activation of PPARα partially attenuated the increase in pyrotinib sensitivity induced by SLC27A2 knockdown. Clinical validation further showed that high-pre-treatment SLC27A2 expression was independently associated with a lower likelihood of achieving pCR after pyrotinib-containing neoadjuvant therapy (OR = 0.10, 95% CI: 0.03-0.31, p < 0.001). Conclusions: SLC27A2 is a candidate factor associated with BC prognosis and reduced pyrotinib sensitivity in HER2-positive BC. Preclinical findings suggested that PPARα-related fatty acid metabolism may contribute to the association between SLC27A2 and pyrotinib response, while clinical validation showed that high-pre-treatment SLC27A2 expression was independently associated with a lower likelihood of achieving pCR following pyrotinib-containing neoadjuvant therapy. These findings support SLC27A2 as a candidate response-associated biomarker and potential therapeutic target, although further mechanistic investigation and external clinical validation are required.