Tissue gene expression analysis approach in a context of high technical and biological heterogeneity

Background Immune expression profiling in colorectal lesions may provide insights into the origins of antitumor immunity and senescence. Optimal approaches for analyzing samples with lower quality RNA from molecularly diverse lesions are lacking. Therefore, we developed a NanoString nCounter-based approach for quality control (QC), normalization, and differential expression (DE) analysis, optimized for FFPE samples in contexts of high biologic heterogeneity. Methods The approach incorporates a colon specific positive control gene set (11 genes) to minimize sample exclusions. We evaluated three normalization methods Removal of Unwanted Variation (RUVg), NanoStringDiff (NSDiff), and nSolver using a 277 gene immune panel to compare 100 samples, including sessile serrated lesions (SSLs) (n = 25), tubulovillous and villous adenomas (TVs) (n = 27), and tubular adenomas (TAs) (n = 48) We assessed Type I error rates, computational efficiency, and gene significance via FDR-corrected q-values. Results Incorporating the colon-specific QC set reduced sample exclusions by 63% compared to standard methods (13 vs 35 sample exclusions). All three normalization approaches identified DE genes between SSLs and TAs (e.g., TFF1, MUC5AC, MUC6). For TVs vs. TAs, only RUVg and NSDiff detected significant DE genes, revealing wide-spread under-expression of innate and adaptive genes. While NSDiff labeled twice as many significant genes as RUVg, suggesting greater sensitivity, it also exhibited higher Type I error rates and increased computational demand. Conclusions RUVg achieved a balance between computational efficiency and low Type I error, while NSDiff was more sensitive but computationally demanding and exhibited higher Type I error. Our approach provides a robust framework for profiling immune genes in heterogeneous lesions.