Deciphering the genetic landscape of obesity: a data-driven approach to identifying plausible causal genes and therapeutic targets

Abstract

Objectives: Genome-wide association studies (GWAS) have successfully revealed numerous susceptibility loci for obesity. However, identifying the causal genes, pathways, and tissues/cell types responsible for these associations remains a challenge, and standardized analysis workflows are lacking. Additionally, due to limited treatment options for obesity, there is a need for the development of new pharmacological therapies. This study aimed to address these issues by performing step-wise utilization of knowledgebase for gene prioritization and assessing the potential relevance of key obesity genes as therapeutic targets.

Methods and results: First, we generated a list of 28,787 obesity-associated SNPs from the publicly available GWAS dataset (approximately 800,000 individuals in the GIANT meta-analysis). Then, we prioritized 1372 genes with significant in silico evidence against genomic and transcriptomic data, including transcriptionally regulated genes in the brain from transcriptome-wide association studies. In further narrowing down the gene list, we selected key genes, which we found to be useful for the discovery of potential drug seeds as demonstrated in lipid GWAS separately. We thus identified 74 key genes for obesity, which are highly interconnected and enriched in several biological processes that contribute to obesity, including energy expenditure and homeostasis. Of 74 key genes, 37 had not been reported for the pathophysiology of obesity. Finally, by drug-gene interaction analysis, we detected 23 (of 74) key genes that are potential targets for 78 approved and marketed drugs.

Conclusions: Our results provide valuable insights into new treatment options for obesity through a data-driven approach that integrates multiple up-to-date knowledgebases.

Fig. 1

Overview of the data-driven integrative approach. We extract 28,787 obesity-associated SNPs from publicly available GWAS results (top panel) and systematically prioritize 74 plausible key obesity genes, by utilizing a series of bioinformatics tools and genomic and transcriptomic evidence (middle panel). We then explore major biological mechanisms of obesity from the key obesity genes, highlighting 23 potential candidates that are useful for the development of obesity therapeutics (low panel)

Fig. 2

Relationship and classification of key obesity genes identified through network analyses. A Schematic illustration of a protein-protein interaction (PPI) subset involving 74 key obesity genes, where thicker edges indicate stronger data support. Of these, 37 red nodes represent newly reported genes that have not been functionally validated for obesity. B For the 37 functionally validated known genes, the heatmap shows their involvement in five phenotypic groups i.e., appetite, fat, size, lipid, and glucose, reported in the literature; presence by dark blue and absence by light blue

Fig. 3

Representative results for enrichment analyses of key obesity genes. A Lists of the top 10 significantly enriched GO terms from biological processes (top), molecular functions (middle), and cellular components (low), respectively. B Schematic illustration of pairwise relationships between top 20 significantly enriched KEGG pathways, where darker and larger nodes indicate more significantly enriched and larger gene sets and thicker edges represent more overlapped genes

Fig. 4

Schematic illustration of 23 key obesity genes and 78 FDA-approved drugs. Genes highlighted in blue are validated, while those in red are not for functional relevance to obesity. These drugs are further classified into two groups based on their experimental evidence; promotion of weight gain (pink) and weight loss (light green) in case of over-expression of the corresponding gene product

Fig. 5

Effectiveness of workflow against negative-control and benchmarking analyses. We evaluate the effectiveness of our workflow by comparison with negative control and benchmarking analyses. Genes overlapping between the mice knockout database and the drug-gene interaction database are counted. The thick border represents our method, with 10 trials conducted for negative control analyses. The number and percentage of genes that overlapped in negative control analyses are shown as median and standard deviation