Of mice and men: transcriptome-based clustering reveals molecular heterogeneity among mouse atherosclerosis models and partially reflects the diversity of human plaques

Abstract

Abstract Background Mice are frequently used as model organisms to study atherosclerosis, but many treatments effective in mice do not successfully translate to humans. In a recent study, we clustered 654 human plaques based on transcriptomic data and described the categorisation into five distinct plaque types. We now hypothesise that the reasons for the failures of translating results from mouse models can be partially explained by the fact that mouse models do not reflect all the phenotypic variability we observe in human plaques. Methods We used transcriptomic data from the aortic samples (from arch to mid abdomen) of 102 different strains of inbred mice (n = 366). Those mice are F1 progenies from the Hybrid Mouse Diversity Panel (HMDP) crossed with C57BL/6J carrying dominant ApoE-Leiden and CETP transgenes. The gene expression data were normalised and batch effect was removed and clustered by a shared nearest neighbour modularity optimization-based clustering algorithm. Differential gene expression between the clusters was performed to reveal the difference in the biology of the clusters. The clusters were linked to physiological measurements and correlated to human transcriptomic-based plaque clusters. Results Transcriptional profiles of atherosclerotic lesions in mice could be categorised into four distinct clusters with specific gene expression fingerprints. The different clusters have a different activity for some pathways, including smooth muscle cell related pathways and metabolic pathways. Comparison with human clusters revealed that most mice strains (n = 287, 78%) correlated best with human plaque type #3 which is enriched in both inflammatory and fibrosis-related genes. This cluster represents about 20% of all human lesions but was found to be associated with severe cardiovascular events. A few mouse strains showed the highest correlations with two other human plaque types (e.g. BUB/Bnj for plaque type #1 and LG/J for plaque type #4), while the human plaque types #0 and #2 were much less reflected in analysed mouse models. Interestingly inflammatory pathways, important drivers in differentiating human clusters, only showed minor differences through all mice clusters. Conclusions Transcriptomic profiling of atherosclerotic lesions reveals that mice associate only partly with the human plaque phenotypes. The best associations were observed with those human plaques that correlate with a more severe clinical phenotype. The existence of different types of lesions observed in humans should be taken into account when choosing a mouse model to study atherosclerosis.