Genetic Architecture of the Cardiovascular Risk Proteome

Abstract

Background: We recently identified 156 proteins in human plasma that were each associated with the net Framingham Cardiovascular Disease Risk Score using an aptamer-based proteomic platform in Framingham Heart Study Offspring participants. Here we hypothesized that performing genome-wide association studies and exome array analyses on the levels of each of these 156 proteins might identify genetic determinants of risk-associated circulating factors and provide insights into early cardiovascular pathophysiology.

Methods: We studied the association of genetic variants with the plasma levels of each of the 156 Framingham Cardiovascular Disease Risk Score-associated proteins using linear mixed-effects models in 2 population-based cohorts. We performed discovery analyses on plasma samples from 759 participants of the Framingham Heart Study Offspring cohort, an observational study of the offspring of the original Framingham Heart Study and their spouses, and validated these findings in plasma samples from 1421 participants of the MDCS (Malmö Diet and Cancer Study). To evaluate the utility of this strategy in identifying new biological pathways relevant to cardiovascular disease pathophysiology, we performed studies in a cell-model system to experimentally validate the functional significance of an especially novel genetic association with circulating apolipoprotein E levels.

Results: We identified 120 locus-protein associations in genome-wide analyses and 41 associations in exome array analyses, the majority of which have not been described previously. These loci explained up to 66% of interindividual plasma protein-level variation and, on average, accounted for 3 times the amount of variation explained by common clinical factors, such as age, sex, and diabetes mellitus status. We described overlap among many of these loci and cardiovascular disease genetic risk variants. Finally, we experimentally validated a novel association between circulating apolipoprotein E levels and the transcription factor phosphatase 1G. Knockdown of phosphatase 1G in a human liver cell model resulted in decreased apolipoprotein E transcription and apolipoprotein E protein levels in cultured supernatants.

Conclusions: We identified dozens of novel genetic determinants of proteins associated with the Framingham Cardiovascular Disease Risk Score and experimentally validated a new role for phosphatase 1G in lipoprotein biology. Further, genome-wide and exome array data for each protein have been made publicly available as a resource for cardiovascular disease research.

Keywords: APOE; cardiovascular genomics; genome-wide analysis; proteomics; systems biology.

Figure 1. Relative impact of heritable and clinical factors on plasma CVD risk proteins

The percent inter-individual variation explained by genetic (top SNP and other genetic factors from genome-wide profiling), clinical factors (as shown), or unexplained factors is shown for each measured protein. Reference lines indicate 20% variability explained by either genetic (left side) or clinical (right side) factors.

Figure 2. Genome-wide study of the plasma CVD risk proteome

A) The significance of associations between measured SNPs and the 156 plasma proteins associated with FRS. The x-axis depicts the physical order of the genome and the y-axis depicts the P-value (−log10) of the SNP-protein association. Each color depicts an individual protein. The y-axis is truncated at 1×10⁻⁶⁰ for clarity. The minimum calculated –log10P was 1.8×10⁻³⁰⁷ (association between rs3816018 at 5q32 and levels of platelet-derived growth factor receptor beta). B) Ideogram demonstrating pQTLs derived from GWAS and exome array analyses. Overlapping CVD risk loci from consortium studies are shown (Bonferroni significance P ≤ 0.05/120 ≤ 4.2 × 10⁻⁴). Ideogram generated using NCBI Genome Decoration Page.

Figure 3. Validation of genome-wide and exome array pQTLs in MDCS

A) pQTLs demonstrated similar magnitude and direction of effect between the FHS discovery analyses and MDCS validation analyses. The estimated beta coefficient of each of the 120 pQTLs derived in FHS genome-wide analyses (x-axis, main frame) is plotted against the estimated beta coefficient of the pQTL in MDCS (y-axis). pQTLs that validated with Bonferroni-adjusted levels of significance in MDCS are shown in red (P ≤ 4.2 × 10⁻⁴); pQTLs that validated with nominal significance (P ≤ 0.05) are shown in orange. Reference lines demonstrate perfect concordance between discovery and validation cohorts. The inset displays a similar analysis for the 13 loci identified by exome array single variant analyses (circles; Bonferroni-adjusted level of significance P = 3.8 × 10⁻³) and 28 variants identified by exome array burden testing analyses (crosses) (Bonferroni-adjusted level of significance P = 1.8 × 10⁻³). B) pQTLs are shown as a function of distance from the coding gene for the associated plasma protein in genome-wide analyses (left panel) and exome array single variant analyses (right panel). pQTLs were considered to be in cis to the coding gene for the associated protein when they were located ≤ 1 Mb from the coding gene transcriptional start site. pQTLs were considered to be in trans if the associated coding gene was located on the same chromosome ≥ 1Mb or on a different chromosome (“inter-chrom”). The statistical strength of validation of each pQTL is shown as in panel A.

Figure 4. The identification and experimental validation of a novel central regulator of ApoE

A) Pleiotropic variants (ordered by physical location in the genome on the x-axis) were mapped against the number of distinct associations with CVD risk proteins (y-axis). A locus near the PPM1G gene was identified that was associated with five plasma CVD risk proteins, including ApoE. B) Knockdown of endogenous PPM1G in the Hep G2 cell line resulted in a significant reduction in endogenous ApoE expression, as measured by RT-PCR, but had no effect on levels of Ubc (negative control). C) Knockdown of endogenous PPM1G in Hep G2 cells also significantly reduced endogenous ApoE accumulation in the culture media, as measured by ELISA, but had no effect on culture media levels of MIF (negative control). P-values represent the statistical significance of paired, two-sample t-tests.