Cross-Platform Evaluation of a Tumor-Budding-Associated Gene Signature in Colorectal Cancer.

Spatial transcriptomics has identified AIF1, LIF, MRC1, and TNC as enriched in tumor budding (TB) regions of colorectal cancer. We investigated whether this TB associated gene module could be detected in bulk transcriptomic data. Pairwise gene co-occurrence was analyzed in bulk RNA-seq (TCGA-COADREAD, n=547) and single-cell RNA-seq (GSE178341, n=26,057 cells) using Fisher's exact tests with Benjamini-Hochberg correction. In bulk RNA-seq, pairwise fold-enrichments were weak (1.16 to 2.72), with AIF1 to LIF showing no significant association (FE=1.16, FDR=0.31).

In single-cell data, the same pairs showed 31 to 184.5 fold enrichment (all FDR < 8e-38). Total UMI counts provided no evidence of doublet artifacts explaining these co-detection patterns. Exploratory enhancer-promoter interaction analysis in HCT116 cells further suggested these genes are controlled differently. GOT1 and MDH1, two malate-aspartate shuttle genes, showed lower expression in late-stage CRC (FDR less than or equal to 0.010).

A bulk-derived risk score showed a modest association (Cox HR=1.21, p=0.021). Overall, TB-associated co-expression was more clearly detected at single-cell than bulk resolution, highlighting the importance of transcriptomic resolution when studying tumor-microenvironment-associated gene signatures.