Abstract

The objective of this study was to explore how dietary interventions manifest themselves at the level of gene co-expression in several tissues in cattle using a robust approach based on modules which underpin complex integrative frameworks. For this purpose, we used the public available RNA-seq dataset GSE62938 which contained transcriptomic profiles of subcutaneous adipose tissue, perirenal adipose tissue and longissimus dorsi muscle from animals fed either forage and concentrate based diets. These tissues are metabolically very different systems, and dietary related transcriptional responses likely vary with tissue and depth of response. After preprocessing and normalization, each tissue was considered separately in order to accommodate the strong tissue specific expression differences. Principal component analysis demonstrated that tissue origin was the principal source of global transcriptional variation and PC1 accounted for approximately 32–50% of the total variance between tissues. This was followed by application of non-negative matrix factorization (NMF) as a biclustering method in its role as an engine for the discovery of coherent gene- and sample-modules. using a data driven rank selection identified k = 4 modules for both adipose tissues and k = 5 modules for skeletal muscle with consensus analyses showing stable, reproducible module structures despite repeated model initializations. NMF identified orthogonal tissue specific gene and sample modules with variable sample loading patterns. ANOVA permutation analysis revealed that dietary impacts were not uniformly distributed throughout the transcriptome, but rather were concentrated in a set of modulated modules. In subcutaneous adipose tissue one module explained 18–22% variance in the module loadings (P_perm < 0.01) and in longissimus dorsi muscle a similar transcript sparsity pattern explained 12-15% variance of the block pairwise similarities (P_perm < 0.05). Other modules were weakly or non-significantly correlated with dietary treatment. In conclusion, we proposed an NMF-based reusable biclustering framework for multi-tissue transcriptomic analyses in dietary intervention. By focusing condition-specific transcriptional responses to coherent gene–sample modules rather than the whole transcriptome, the proposed framework enables the discovery of slight, tissue-specific dietary effects with potential biological and metabolic consequences.