Polygenic scores for complex traits are associated with changes in concentration of circulating lipid species

Abstract

Understanding perturbations in circulating lipid levels that often occur years or decades before clinical symptoms may enhance our understanding of disease mechanisms and provide novel intervention opportunities. Here, we assessed if polygenic scores (PGSs) for complex traits could detect lipid dysfunctions related to the traits and provide new biological insights. We constructed genome-wide PGSs (approximately 1 million genetic variants) for 50 complex traits in 7,169 Finnish individuals with routine clinical lipid profiles and lipidomics measurements (179 lipid species). We identified 678 associations (P < 9.0 × 10-5) involving 26 traits and 142 lipids. Most of these associations were also validated with the actual phenotype measurements where available (89.5% of 181 associations where the trait was available), suggesting that these associations represent early signs of physiological changes of the traits. We detected many known relationships (e.g., PGS for body mass index (BMI) and lysophospholipids, PGS for type 2 diabetes and triacyglycerols) and those that suggested potential target for prevention strategies (e.g., PGS for venous thromboembolism and arachidonic acid). We also found association of PGS for favorable adiposity with increased sphingomyelins levels, suggesting a probable role of sphingomyelins in increased risk for certain disease, e.g., venous thromboembolism as reported previously, in favorable adiposity despite its favorable metabolic effect. Altogether, our study provides a comprehensive characterization of lipidomic alterations in genetic predisposition for a wide range of complex traits. The study also demonstrates potential of PGSs for complex traits to capture early, presymptomatic lipid alterations, highlighting its utility in understanding disease mechanisms and early disease detection.

Fig 1. Study design.

Overview of the study design and analytical approach, along with the details of the diseases and biomarkers for PGS estimation and lipids included in the study, are illustrated.

Fig 2. Associations of PGSs with plasma lipidome.

(A) Number of lipids associated with PGSs for diseases and biomarkers. The bars are colored based on the lipid classes of the lipid species. (B) Number of PGSs associated with each lipid. The bars are colored based on the PGS category. Only the lipids associated with at least 5 PGSs are shown. BMI, body mass index; Clinical: routine clinical lipid measures; CE, cholesteryl ester; CER, Ceramide; DAG, Diacylglycerol; LPC, lysophosphatidylcholine; LPE, lysophosphatidylamine; PC, phosphatidylcholine; PC O-, phosphatidylcholine-ether; PE, phosphatidylamine; PI, phosphatidylinositol; SM, sphingomyelin; SHBG, sex hormone-binding globulin; TAG, triacylglycerol. The data underlying this figure may be found in S1 Data.

Fig 3. Potential of PGSs for complex traits in capturing early lipid alterations.

(A) Validation of PGS-lipid associations using disease status or actual measures of biomarkers. The scatter plot in the upper panel shows the validation of PGS-lipid associations with the actual measures. The x-axis represents the changes (beta in standardized unit) in lipids per standard deviation (SD) increase in the PGS. The y-axis represents the corresponding changes in lipids per SD increase in actual biomarker measures or with disease status. Each dot on the plot represents a lipid with at least one significant PGS-lipid association. The scatter plot in the lower panel shows the comparison of Pheno-lipid associations with the PGS-lipid associations. The x-axis represents the changes (beta in standardized unit) in lipids per standard deviation (SD) increase in actual biomarker measures or with disease status. The y-axis represents the corresponding changes in lipids per s.d. increase in the PGSs. Each dot represents a lipid with at least 1 significant Pheno-lipid association. (B) Exemplar relationship between genetic risk and lipid concentrations. The left side of the upper and lower panels shows the risk of type 2 diabetes and venous thromboembolism respectively, in the FinnGen participants in different percentile categories with 40–60 percentile as reference. The black dotted lines mark the odds ratio of 1 for reference groups (40–60 percentile). The right side shows the median levels of the most strongly associated lipids species for the 2 PGSs-PC O-18:2;0/18:2;0 and PC 18:0;0_20:4;0, respectively, in different percentiles in the GeneRISK cohort. The 95% confidence intervals are shown as error bars. The black dotted lines mark the median levels of the lipids in the reference groups (40–60 percentile), whereas the red dotted lines show the median levels in the full cohort. The data underlying this figure may be found in S1 Data.

Fig 4. Lipidomic signatures of PGSs for complex traits.

The Manhattan plots show the associations of selected PGSs with lipids. The lipids are grouped and colored by the lipid classes they belong. Upright triangle denotes positive effect on lipid and upside-down triangle represent negative effect. The dotted horizontal line represents the threshold for significant associations at P < 9.0 × 10⁻⁵. BMI, body mass index; CE, cholesteryl ester; CER, ceramide; Chol, free cholesterol; DAG, diacylglycerol; LPC, lysophosphatidylcholine; LPE, lysophosphatidylamine; PC, phosphatidylcholine; PC O-, phosphatidylcholine-ether; PE, phosphatidylamine; PE O-, phosphatidylamine-ether; PI, phosphatidylinositol; SM, sphingomyelin; TAG, triacylglycerol. The data underlying this figure may be found in S1 Data.

Fig 5. Sharing and differences in the lipidomic alterations in genetic risks across the traits.

(A) The heatmap shows the pattern of associations of lipids across the PGSs. The light blue shade represents negative effect (beta <0) with P < 0.05, dark blue represents negative effect with P < 9.0 × 10⁻⁵. The pink shade represents positive effect (beta >0) with P < 0.05, red represents positive effect with P < 9.0 × 10⁻⁵. White shade denotes no association (P > 0.05). Only the PGSs and lipids with at least 10 and 5 significant associations (P < 9.0 × 10⁻⁵), respectively, are shown. BMI, body mass index; SHBG, sex hormone-binding globulin. (B) Association of PGS for type 2 diabetes with lipid indices suggesting altered D6D and ELVOL6 activities. The median levels of the lipid indices representing D6D and ELVOL6 activities are plotted for individuals in different percentiles in the GeneRISK cohort. The 95% confidence intervals are shown as error bars. The black dotted lines mark the median levels of the lipids in the reference groups (40–60 percentile), whereas the red dotted lines show the median levels in full cohort. (C) Association of SM 34:2;2 and SM 36:2;2 with the PGSs. The change (beta in standardized unit) in lipids per standard deviation (SD) increase in PGSs with 95% confidence intervals are shown. The data underlying this figure may be found in S1 Data.