Epigenetic clusters identified in Hispanic/LatinX colorectal cancers reveal novel molecular subtypes

Introduction: Colorectal cancer (CRC) incidence and mortality are rising among Hispanic/LatinX (HL) individuals, yet they remain underrepresented in genomic studies. As a result, it is unknown whether HL CRCs harbor distinct epigenetic or biological features. We addressed this gap by profiling HL CRC tumors using genome-wide DNA methylation and integrating these data with matched transcriptomic and genomic datasets. Methods: We generated DNA methylation profiles for 104 HL CRC tumors using the Illumina EPIC v2 array. RNA-seq (n=92) and whole-exome sequencing (WES; n=89) were available for integrative analyses. TCGA non-Hispanic White (NHW) CRCs served as a reference cohort. CIMP status was assigned using promoter methylation values for eight canonical CIMP markers (MLH1, IGF2, CACNA1G, NEUROG1, RUNX3, SOCS1, CRABP1, CDKN2A). Unsupervised clustering of the top most variable CpGs was used to define methylation subtypes. Pathway interpretation was performed by summarizing CpGs into gene-context scores across cancer related pathways. Matched RNA-seq and WES data were used to characterize transcriptional and mutational features of each subtype and validate DNA methylation pathway interpretation. CMS classifications, MSI status were assigned for comparison. Results: CIMP patterns in HL CRCs differed markedly from NHW tumors in TCGA. Only 1% of HL MSI tumors were CIMP-H (vs. 7.1% in NHW MSI CRCs), and 11% were CIMP-0 (vs. 3.4% in NHW), indicating a decoupling of MSI and CIMP-H status. Unsupervised clustering identified six novel methylation-defined subtypes (HL-C1 to HL-C6) with distinct immune, stromal, metabolic, and epigenetic features. These subtypes also differed by mean age at diagnosis and global methylation: HL-C1 (58.7y, beta=0.493), HL-C2 (51.2y, beta=0.358), HL-C3 (60.1y, beta=0.428), HL-C4 (51.9y, beta=0.222), HL-C5 (52.8y, beta=0.460), and HL-C6 (44.5y, beta=0.608). Integration with RNA-seq confirmed concordant pathway activation across overlapping samples. Somatic mutations further distinguished subtypes: TP53 mutations were reduced in HL-C6 (37.5% vs. ~66% in other clusters), while MLH1 mutations were enriched (12.5%). CMS assignments did not recapitulate these six HL-specific subtypes. Conclusion: HL CRCs display molecular features not captured by existing CIMP or CMS frameworks. The distinct methylation subtypes identified here offer new insight into HL CRC biology and provide a foundation for refining future molecular studies in this population.