Non-invasive radiogenomic mapping of the SMARCAL1-driven ferroptotic niche is associated with longitudinal MRD-negative surveillance in early-stage NSCLC.

Post-operative molecular residual disease (MRD) detection using circulating tumor DNA (ctDNA) has been established as a sensitive biomarker for early recurrence risk stratification in early-stage non-small cell lung cancer (NSCLC). However, the biological heterogeneity of patients MRD-negative at baseline remains incompletely characterized, and a proportion subsequently experience MRD conversion and radiographic recurrence. We investigated whether pre-operative radiomic features could non-invasively capture the spatial organization of a ferroptosis-enriched, immune-active tumor microenvironment associated with sustained post-operative MRD negativity. In this single-center translational cohort nested within the prospective CTONG 2201 trial (NCT05457049), 71 patients with completely resected stage IB-IIIA NSCLC, MRD-negative at two post-operative landmarks, were followed under dynamic observation without immediate adjuvant therapy. A Rad-Score was derived from 1,432 pre-operative contrast-enhanced CT radiomic features using a pre-specified five-layer pipeline. Whole-exome sequencing, bulk RNA-seq with CIBERSORTx, and 10x Visium spatial transcriptomics in 14 tumors (five High, four Intermediate, five Low Rad-Score; 36,318 quality-controlled spatial spots) characterized molecular correlates. Multiplex immunofluorescence (n = 40 across Rad-Score quartiles), immunohistochemistry (n = 60), CRISPR-Cas9 knockout of SMARCAL1 with wild-type and helicase-dead rescue in A549/H1299 cell lines, and primary human CXCR3+ CD8+ T cell chemotaxis assays provided protein-level and functional validation. External validation was performed in two independent cohorts (TCIA NSCLC-Radiogenomics, n = 178; Lung3, n = 89) with locked-model application, following Harrell optimism-corrected internal validation (B = 1,000 bootstrap replicates with full-pipeline resampling). Nine stability-validated radiomic features (selection frequency Π ≥ 0.60, all; ≥ 0.80 for 7/9) constituted the Rad-Score. High Rad-Score patients had longer MFS (HR = 0.32, 95% CI 0.18-0.58; P < 0.001), with apparent C-index 0.812 attenuating to optimism-corrected 0.798 and externally validated 0.783 (NSCLC-Radiogenomics) and 0.741 (Lung3). The Rad-Score remained independent after adjustment for 13 clinical covariates including ECOG PS, Charlson Comorbidity Index, and smoking pack-years (adjusted HR = 0.66, P = 0.022), outperformed five established comparators (CALGB 9761-type, Kratz 14-gene, ctDNA composite, Bindea immunoscore, clinical+TMB) in decision curve analysis across the clinically meaningful threshold range 0.15-0.75. High Rad-Score tumors exhibited elevated TMB (P < 0.001, BH-FDR q < 0.001) and APOBEC signatures (SBS2, SBS13: q < 0.001; SBS4 non-significant after FDR), enriched CD8+ T cells and M1 macrophages (q < 0.01), and contiguous ferroptotic niches with monotonically increasing activity across the Rad-Score continuum (Spearman ρ = 0.87). SMARCAL1 knockout conferred resistance to erastin-induced ferroptosis (cell viability 78 ± 6% in sgSMARCAL1 cells versus 31 ± 4% in sgScramble controls, P < 0.001) and IFN-γ-induced CXCL9/10 secretion (62%/57% reduction) - phenotypes rescued by wild-type but not helicase-dead SMARCAL1 - and functionally attenuated CXCR3+ CD8+ T cell transmigration (58% reduction), with CXCR3 antagonist AMG 487 producing >90% blockade. In this externally validated, functionally interrogated translational cohort, a pre-operative CT-derived Rad-Score is associated with sustained MRD-negative surveillance, provides a non-invasive surrogate for a SMARCAL1-driven ferroptotic-immune niche, and requires prospective validation Non-small cell lung cancer; Molecular residual disease; Radiogenomics; Spatial transcriptomics; Ferroptosis; SMARCAL1; CXCL9/10-CXCR3 axis. ClinicalTrials.gov, identifier NCT05457049.