Artificial intelligence–enhanced biomarker analysis of WNT, TGF-β, and PI3K pathway alterations in FOLFOX-treated early-onset colorectal cancer in high-risk populations.

Abstract

236 Background: Early-onset colorectal cancer (EOCRC; <50 years) is rising rapidly, with the steepest increases among Hispanic/Latino (H/L) populations, yet most genomic studies emphasize non-Hispanic White (NHW) cohorts. Standard therapy for CRC includes FOLFOX (folinic acid, fluorouracil, oxaliplatin), but its molecular impact in EOCRC is poorly defined. The WNT, TGF-β, and PI3K pathways are key regulators of CRC biology and resistance, but their prognostic and treatment-specific associations across ancestry and age remain unclear. We investigated alterations in these pathways as candidate biomarkers of response and outcome in FOLFOX-treated CRC, integrating artificial intelligence (AI) for multi-parameter analyses. Methods: Somatic mutation and clinical data from 2,515 CRC patients across public datasets were analyzed. Patients were stratified by age (EOCRC vs. late-onset [LOCRC]), ancestry (H/L vs. NHW), and FOLFOX status. Tumors were derived from clinical sequencing of primary or metastatic samples. WNT (APC, RNF43, AXIN1/2, TCF7L2, CTNNB1, AMER1), TGF-β (SMAD2/4, TGFBR2, BMPR1A), and PI3K (PIK3CA, PTEN, AKT isoforms, regulators) alterations were coded as categorical. Mutation prevalence was compared using Fisher’s exact or chi-square tests. Overall survival (OS) was the primary endpoint, assessed by Kaplan–Meier and Cox models. Analyses were exploratory and unadjusted. In addition, AI platforms (AI-HOPE, AI-HOPE-WNT, AI-HOPE-TGFβ, AI-HOPE-PI3K) automated cohort construction, stratification, and survival analysis. Results: WNT alterations were common and dominated by APC. In H/L EOCRC, FOLFOX-treated tumors had lower CTNNB1 (5.5% vs. 17.3%, p = 0.04) and RNF43 (5.5% vs. 19.2%, p = 0.02) compared with untreated; NHW LOCRC treated cases showed reduced AXIN1, AXIN2, RNF43, and TCF7L2. TGF-β alterations occurred in 28–39% of H/L and 23–31% of NHW, with SMAD4 predominating. BMPR1A was enriched in FOLFOX-treated EO H/L (5.5% vs. 1.1% EO NHW, p = 0.027), while SMAD2 and TGFBR2 were higher in untreated NHW LOCRC. TGF-β alterations predicted worse OS in FOLFOX-treated EO H/L ( p = 0.029), with SMAD4 mutations less frequent in EO H/L vs. NHW (2.7% vs. 13.9%, p = 0.013). PI3K alterations were heterogeneous; in EO NHW FOLFOX-treated patients, they predicted reduced OS (n = 124 vs. 251; p = 0.0008). AI interrogation highlighted exploratory candidates (INPP4B, RPTOR) in EO H/L without enrichment. Conclusions: This AI-enhanced biomarker study reveals pathway-specific, ancestry- and treatment-dependent associations in EOCRC. WNT alterations were linked to favorable survival in NHW, TGF-β alterations to poor outcomes in FOLFOX-treated EO H/L, and PI3K alterations to poor survival in EO NHW. These findings highlight the heterogeneity of CRC and the promise of AI tools for advancing precision oncology.