Residual disease as a therapy-remodeled ecosystem: spatial niches, tumor persistence, and relapse in solid tumors

Residual disease comprises the malignant burden that remains after treatment, including low-volume lesions and clinically occult reservoirs. Surviving cells encounter a tissue environment altered by cell death, inflammation, repair, stromal activation, vascular injury, and metabolic stress. These changes can support spatially restricted persistence, but co-localization alone does not establish a protective niche. This narrative review links therapy-associated tissue remodeling to the location and state of residual cells and to the timing of potential relapse-directed intervention. We distinguish molecular residual disease, treatment-tolerant persister states, dormancy, and stable resistance, and separate selection of pre-existing populations from experimentally demonstrated state induction. Across chemotherapy, radiotherapy, targeted therapy, endocrine therapy, and immunotherapy, we distinguish human observational or trial evidence from mechanistic findings in mouse models, organoids, and cell culture. We organize niche evidence into spatial enrichment, association with a residual-cell state, selective perturbation, and an effect on reactivation or relapse. Examples illustrate dose-dependent radiation responses, multifactorial stromal protection, context-dependent immune surveillance, and collateral metabolic vulnerabilities. Longitudinal spatial profiling, lineage analysis, spatial pharmacology, and clinically informative circulating tumor DNA studies offer complementary information; molecular detection does not identify a particular tissue niche or directly demonstrate viable residual cells. Four therapeutic windows are proposed as research strategies rather than a validated clinical classification. Their utility depends on the interaction being required, the drug reaching the relevant region, and acceptable effects on normal repair and tumor-restraining functions. Limited paired human sampling and the compressed timescale of experimental models remain major constraints on translation.