An Icariin-Informed Nine-Gene Transcriptomic Signature Identifies Mesenchymal–Stromal Recurrence Risk in Resected Colorectal Cancer: External Validation and Single-Cell Localization

Abstract Background Recurrence after curative resection of colorectal cancer is stratified largely by stage. Icariin, a prenylated flavonoid glycoside from Epimedium spp., defines a target space not yet examined as a source of recurrence-associated transcriptomes. Methods Icariin targets intersected with tumour-versus-normal expression (TCGA-COAD, GSE39582) gave an 84-gene candidate pool; a pre-specified 100-combination screen (10 selectors × 10 models) in GSE14333 ( n = 226 , 50 events) produced a fixed nine-gene signature. Four conditions fixed before external testing — patient-level independence, a depositor-coded disease-free-survival (DFS) event, the same platform, and the largest event count — selected GSE17536 ( n = 145 , 36 events). Results The signature (TIMP1/ SPARC/CLDN1 positive; PCNA/ACADM/ DPEP1/MMP3/CASP7/MMP12 negative) stratified DFS in GSE17536 (C-index 0.734; HR 2.30 per SD, P = 2.3 × 10 − 8 ) and improved on stage, age and sex ( ΔC + 0.085 , 95% CI 0.023–0.169; likelihood-ratio P < 0.001 ), but under-estimated absolute risk (observed-to-expected 1.90). Pooling the eight deduplicated cohorts (181 DFS events) gave a random-effects HR of 1.53 per SD (95% CI 1.13–2.07, P = 0.013 ; I 2 = 69 %; prediction interval 0.71–3.31). High-risk tumours were mesenchymal/stromal and immune-suppressive; the signal localized mainly to fibroblasts and endothelium through SPARC and TIMP1. A protocol-matched random-pool control placed the signature at the 88th percentile ( P = 0.118 ), and the stage-dominated TCGA-COAD transfer added no incremental value. Conclusion The nine-gene signature is an externally validated relative-risk stratification biomarker derived from an icariin-associated candidate space; all results are in silico and await prospective and experimental validation.