Discovery of a Systemic Immune-Inflammatory Axis Bridging the Neuroendocrine System and Breast Tumor Microenvironment via MMD-Regularized Cross-Tissue Latent Alignment

Abstract

The breast tumor immune microenvironment (TIME) is predominantly studied as a localized phenomenon. However, emerging evidence suggests tumor immune evasion may be systemically linked to master neuroendocrine regulatory centers. Investigating this systemic axis is computationally intractable due to the impossibility of sampling paired brain and breast transcriptomes from living patients. Here we present a novel in silico framework utilizing MMD-regularized cross-tissue latent alignment to bridge unpaired transcriptomic profiles from GTEx neuroendocrine tissues and the TCGA-BRCA cohort. By projecting high-dimensional RNA-seq data into a shared topological space via a Domain Adaptation Autoencoder optimized with Maximum Mean Discrepancy ([L]MMD), we established a robust mathematical bridge across tissues. Interrogation of the aligned 128-dimensional latent space isolated a dominant systemic immune-inflammatory signature rather than canonical endocrine hormone amplification. Dimension 31 of the aligned space was anchored by T-cell receptor variable chains (TRAV35, TRAV8-4, TRBV2), immunoglobulins (IGHG1, IGHGP), and the lipid-inflammatory mediator PLA2G2D, orthogonally linking pituitary neuroinflammation to a "hot" breast TIME. Independent cross-platform validation on the METABRIC microarray cohort (n = 1,980) recovered the same underlying mechanism via latent axis rotation, capturing the systemic acute-phase secretome: PLA2G2A, LBP, SAA1, and CXCL17. These findings suggest a population-level latent correspondence between neuroendocrine inflammatory programs and breast tumor immune phenotypes, revealing candidate targets for liquid biopsy and systemic immunotherapeutic investigation that warrant prospective experimental validation.