Protein-metabolite association studies identify novel proteomic determinants of metabolite levels in human plasma

Abstract

Although many novel gene-metabolite and gene-protein associations have been identified using high-throughput biochemical profiling, systematic studies that leverage human genetics to illuminate causal relationships between circulating proteins and metabolites are lacking. Here, we performed protein-metabolite association studies in 3,626 plasma samples from three human cohorts. We detected 171,800 significant protein-metabolite pairwise correlations between 1,265 proteins and 365 metabolites, including established relationships in metabolic and signaling pathways such as the protein thyroxine-binding globulin and the metabolite thyroxine, as well as thousands of new findings. In Mendelian randomization (MR) analyses, we identified putative causal protein-to-metabolite associations. We experimentally validated top MR associations in proof-of-concept plasma metabolomics studies in three murine knockout strains of key protein regulators. These analyses identified previously unrecognized associations between bioactive proteins and metabolites in human plasma. We provide publicly available data to be leveraged for studies in human metabolism and disease.

Keywords: GWAS; functional genomics; genome-wide association study; genomics; metabolomics; multi-omics; pathway discovery; proteomics.

Figure 1.. The integration of human plasma proteomic, metabolomic, and genomic profiling data for pathway discovery.

Flow diagram detailing the experimental pipeline and main results from the integration of plasma proteomic, metabolomic, and genomic profiling datasets in the Jackson Heart Study (JHS), Multi-Ethnic Study of Atherosclerosis (MESA), and Health, Risk Factors, Exercise Training and Genetics Study (HERITAGE Family study). WGAS = whole genome association study, GWAS = genome wide association study.

Figure 2.. Protein-metabolite correlations in human plasma.

Pearson correlation coefficients were calculated for every pairwise protein-metabolite combination within the JHS, MESA, and HERITAGE Family study using age- and sex-adjusted, log-normalized, and standardized protein and metabolite levels. (A) A subset of proteins and metabolites were positively (red) or negatively (blue) correlated with an FDR-adjusted q-value ≤ 0.05 in each study, and in a meta-analysis of the three studies. (B) Ninety-seven percent of the meta-analyzed age- and sex-adjusted protein-metabolite correlations remained significant with an FDR-adjusted q-value ≤0.05 when further adjusted for BMI, and 87% remained significant when additionally adjusted for eGFR. (C) The magnitude and directionality of the meta-analyzed age- and sex-adjusted protein-metabolite correlations were consistent across studies. (D) Visualization of the protein-metabolite correlations using a heat map demonstrated distinct patterns of associations between individual proteins and members of specific classes of metabolites. Individual metabolites were ordered according to RefMet class along the y-axis, and proteins were allowed to order using a hierarchical cluster analysis along the x-axis. The magnitude and directionality of the correlation coefficient for each protein-metabolite association is depicted by color, as indicated in the legend. (E) The statistical significance (−log(p-value)) of the correlation between each metabolite and four representative protein hormones is shown.

Figure 3.. Protein correlations are significantly enriched for specific metabolite classes.

(A) A volcano plot demonstrates that the most significantly correlated metabolites with plasma levels of APOE protein were members of the lipid metabolite class. (B) To evaluate this enrichment for lipids more quantitatively, an enrichment score (ES) was computed and visualized as the maximum point (marked with ●) of a running sum statistic that increases proportionally with each lipid and decreases with non-lipids along the ranked list of metabolite correlations with plasma levels of APOE. (C) 241 proteins were significantly enriched for correlations with plasma lipids (FDR-adjusted q-value ≤ 0.05). In the top panel, the ES tracings are shown for each individual protein, and the ten proteins with the highest enrichment scores are listed. In the bottom panel, proteins are ordered in descending order of calculated enrichment scores (x-axis), and the number of proteins significantly enriched for correlations with lipids (FDR-adjusted q-value ≤ 0.05) is indicated with the vertical red line. Similar enrichment analyses are shown for Amino and Organic Acids (D), Carbohydrates (E), and Nucleic Acids (F). The complete enrichment analysis dataset is available in Supplemental Table 5.

Figure 4.. Mendelian Randomization analyses identify putative causal protein-to-metabolite associations in humans.

Proteins with at least one pQTL in cis (located within 1Mb) of the protein coding gene that could be used as an instrumental variable (IV) in Mendelian Randomization (MR) analyses are depicted in blue (a). The proteins with available cis instruments were distributed evenly across the genome (b). pQTLs used in IVs were generally located near the transcriptional start site (TSS) of the protein coding gene (c). A volcano plot depicts the 224 significant MR associations between 52 proteins and 146 metabolites with an FDR-adjusted q-value ≤ 0.05 (d). The three proteins with the most significant MR metabolite associations are depicted by distinct shapes described in the figure legend.

Figure 5.. Protein-to-metabolite causal associations predicted by Mendelian Randomization analyses in humans experimentally validate in murine knockout models.

Plasma metabolomics was conducted on C57BL/6 murine knockout (KO) strains for CD36 (n=6), APOE (n=6), and ACY1 (n=6) and compared to wild type (WT) controls (n=8). Scatterplots depict the number of predicted protein-to-metabolite MR associations in humans (with q≤0.1) that validated in each murine model (with p≤0.05, highlighted in color), as well as the concordance in directionality of these associations (a, c, e). The position on the x axis represents the p-value of the predicted MR association between each protein and metabolite level in the human studies. The position on the y axis represents the p-value of the difference in metabolite level between KO and WT mice. Metabolites on the right half of the scatterplots are predicted to be positively associated with each protein by MR, whereas metabolites on the left half are predicted to be inversely associated with each protein. Similarly, metabolites on the top half of the scatterplots were higher in KO vs WT animals, whereas metabolites on the bottom half were lower in KO vs. WT animals. The northwest and southeast quadrants of each scatter plot are shaded to depict consistent directionality between the MR predictions and KO experiments. The levels of each metabolite that were significantly different in KO vs. WT animals (p≤0.05) are shown as fold changes compared to WT animals (b, d, f). * indicates P<0.05, and ** indicates P<0.01 by students two tailed t-test.