Aggregation and Contextualization of Murine Investigations Improves Discovery of Significant Human Atherosclerotic Cardiovascular Disease Associations

Preclinical murine models are foundational for investigating pathobiologic mechanisms. Despite decades of preclinical studies, the promise of linking specific genes identified in these studies to human disease mechanisms and novel therapies remains largely unfulfilled.^1,2 One prominent example is atherosclerosis, the leading cause of death in humans.³ Unlike wild-type mice, mice with total body deletion of apolipoprotein-E (ApoE-KO) or low density lipoprotein receptor (LDLR-KO) develop atherosclerotic plaques. After more than 10,000 peer-reviewed publications, preclinical findings have had a modest impact on the clinical management of human atherosclerosis.¹

We reported a proof-of-concept study describing the Preclinical Science Integration and Translation (PRESCIANT) method that uses a systems biology approach to study the broader biologic context of gene perturbations modulating atherosclerosis in preclinical models and their relevance to human disease.⁴ We apply PRESCIANT to compare the fidelity of ApoE-KO and LDLR-KO models to identify human disease associations and test whether translation of preclinical findings is enhanced by integration and biologic contextualization.

Methods

The data for this study are available from the corresponding author upon reasonable request.

Identification and characterization of genes from murine studies

PubMed was queried for all atherosclerosis studies using the ApoE-KO or LDLR-KO mouse models published in Arteriosclerosis, Thrombosis, and Vascular Biology (ATVB) and Circulation between 1995 and 2020. The search methodology, results, data extraction procedures, and variable availability are described previously.^4,5 Abstraction from 991 ApoE-KO and 544 LDLR-KO manuscripts yielded 1,114 and 735 records, respectively, with each record corresponding to a single experiment (Figure 1A). Applying exclusions⁴ resulted in a final dataset of 527 ApoE-KO and 318 LDLR-KO experiments for which a single perturbation was made and the impact on atherosclerotic plaque size was measured (Figure 1A). This mapped to 401 unique genes which were input into Ingenuity Pathway Analysis (QIAGEN) core analysis function.

Figure 1.

(A) Data extraction process and distribution of interventions from manuscripts exploring atherogenesis in apolipoprotein E-knockout (ApoE-KO) and low density lipoprotein receptor-knockout (LDLR-KO) mouse models. (B) The number of genes tested in each mouse model that map to 3 top pathways. (C) The top 10 canonical pathways (by significance) determined by analysis of genes studied in ApoE-KO, LDLR-KO, and the combined dataset. The Z-score represents the predicted magnitude of pathway activation (orange) or inhibition (blue). (D-E) Genetically predicted gene expression of individual genes was tested for association with human atherosclerosis. For each candidate gene set, the number of genes associated (p<0.05) with human atherosclerosis (black bar) was compared with the distribution of associations from 100,000 permutations of randomly selected gene sets of equal size (histogram). (D) Gene sets derived from human homologs of murine genes tested in the ApoE-KO, LDLR-KO or both models demonstrate no significant enrichment for association with human atherosclerosis. (E) Gene sets based on pathway contextualization of genes directly tested in murine models show greater, but non-significant, enrichment (top/middle rows). When all pathway genes are considered, including those not evaluated in mice, there is significant enrichment for associations with human atherosclerosis (bottom row). (F) The significance of gene set enrichment, defined by the probability that the number of genes associated with human atherosclerosis would be observed by chance (X-axis), increases with biologic contextualization. This appears independent of gene set size. The vertical dashed red line indicates a p-value for enrichment of 0.05.

Evaluation of human orthologs in a clinical population

The Vanderbilt Institutional Review Board deemed this project exempt as non-human subjects research. As described previously, associations between the genetically predicted gene expression (GPGE) of human orthologs of candidate genes from preclinical analyses and clinical atherosclerotic cardiovascular disease (ASCVD), defined as a composite of phecodes, were evaluated in a large population (N=88,660) from the Vanderbilt University Medical Center BioVU biobank.⁴ Enrichment for associations with human ASCVD was quantified for gene sets with increasing biologic contextualization, including: 1) all genes studied in mice, 2) directly studied genes from enriched pathways, and 3) all genes from enriched pathways irrespective of direct evaluation in mice.

Statistical analyses

Logistic regression, adjusting for age (at phecode for cases and last medical touch for controls), sex, and 10 principal components of genetic ancestry, was used to test for association between the GPGE of each gene and the composite ASCVD outcome.⁴ We performed gene-set enrichment analyses and determined statistical significance based on 100,000 permutations.⁴

Results

Integrated analysis of genes tested in two preclinical models identifies convergent biologic pathways associated with atherosclerosis

The top 10 canonical pathways, defined by strength of association p-value, were identified for each dataset. In addition to atherosclerosis itself, pathways overrepresented in mouse atherosclerosis gene sets included inflammation (e.g., neuroinflammation, role of pattern recognition receptors, osteoarthritis) and fibrosis (e.g., hepatic fibrosis signaling) pathways. The plurality of genes (>80%) were tested in a single mouse model, with only 78 genes tested in both (Figure 1A). Despite modest overlap in genes studied (Figure 1B), the top pathways identified in model-specific and combined analyses were similar (Figure 1C), including the top two activated pathways (neuroinflammation and hepatic fibrosis signaling) and the top inhibited pathway (LXR/RXR signaling).

Human Association Studies

No significant enrichment for human associations was observed among human orthologs of directly evaluated murine genes, whether unselected (Figure 1D) or restricted to top enriched pathways (Figure 1E, top/middle). In contrast, inclusion of all genes from strongly associated murine pathways, irrespective of whether they were directly studied in mice, resulted in significant enrichment for associations with human ASCVD (Figure 1E, bottom). Combining all results revealed that increasing biological contextualization into pathways resulted in progressively greater significance of enrichment for associations with human ASCVD, particularly when including genes not tested in mice (Figure 1F). A sensitivity analysis using an ulcerative colitis outcome (phecode-555.2) demonstrated no enrichment (empiric p>0.77 for top pathways), suggesting results were not due to non-specific inflammation.

In summary, aggregation and pathway-based biologic contextualization of preclinical atherosclerosis data identified gene sets with high enrichment for associations with human disease. Neither the ApoE-KO or the LDLR-KO mouse model demonstrated superior fidelity for a broad definition of human ASCVD. This integrated preclinical data analysis further highlights the importance of lipid handling, fibrosis, and inflammation pathways in atherosclerosis and identified novel gene associations beyond what was directly testing preclinically. Further work to define clinical associations specific to vascular bed and disease acuity are needed. Methodologically, this approach may inform the selection of preclinical models, protocols, and endpoints and add a new component to interpretation and prioritization of preclinical data to improve translation.

Non-standard abbreviations and acronyms

ApoE

Apolipoprotein-E

ASCVD

Atherosclerotic cardiovascular disease

ATVB

Arteriosclerosis, Thrombosis, and Vascular Biology

GPGE

Genetically predicted gene expression

KO

Knockout

LDLR

Low density lipoprotein receptor

PRESCIANT

Preclinical Science Integration and Translation