Integrative Omics: Harnessing the Proteome to Maximize the Potential of the Genome

Great strides have been made in dissecting the genetics of cardiovascular disease (CVD) through genome-wide association studies and whole exome/genome sequencing, although the full heritability remains to be elucidated. Furthermore, biological insights into the associations of common variants with complex traits are often lacking. Although static DNA variation is at the core of variability between humans, viewed in isolation it can provide only a snapshot of the complex landscape of the dynamic, temporal evolution of CVD.

One powerful approach to attempt to harness this dynamic variability is to study intermediate phenotypes that are more proximal to genomic variation than complex clinical end points. Although they can be difficult to measure accurately given the temporal nature of CVD, when analyzed on the foundation of the static genome, integrated studies can identify genomic loci that influence disease through these intermediate biomarkers. This principle underlies quantitative trait loci (QTL) studies, where genetic variation is integrated with other omics data (eg, transcriptomics, metabolomics, proteomics). Once these associations are established, the genetic variants and biomarkers can then be studied in a focused manner to determine what, if any, causal relation they have to the disease of interest. In addition, if done at the genome-wide level, these studies provide a valuable resource to identify possible functional effects of disease-associated genetic variants. The CVD community is just beginning to embark on such integrative studies that capitalize on the large amount of molecular data now available in CVD cohorts.

In this issue of Circulation, Benson et al¹ do just this, elegantly harnessing the power of the proteome to further our understanding of CVD genetics. Integrating genome-wide and exome array genetic data with proteomic data profiled by using a novel aptameter-based multiplex assay, they identify protein QTL (pQTL) in the Framingham Heart Study (FHS). Specifically, they performed pQTL analysis in the FHS Offspring cohort (n=759) on 156 proteins that they had previously identified as being associated with the Framingham CVD 10-year risk score.² They then performed validation studies in the MDCS (Malmö Diet and Cancer Study) (n=1421). Their results highlight the multifaceted output that can be extracted from population-based studies. First, taking advantage of the family-based structure of FHS, they show that a large amount of variance (>20%) in the majority protein levels (≈92%) can be explained by heritable factors, whereas the clinical components of the Framingham CVD risk score explain the same degree of variance in only 16% of the proteins. Genome-wide association studies of the 156 proteins identified 120 pQTLs that reached genome-wide significance in FHS, with 75 validating in MDCS. Four of the pQTLs had been identified as CVD risk loci in the large CARDIoGRAMplusC4D (Coronary Artery Disease Genome-wide Replication and Meta-analysis) consortium, and several additional pQTLs had been identified as risk loci in other studies. They identified another 41 pQTLs in the exome array analysis, which was restricted to variants that were likely to have a functional effect (eg, nonsynonymous, stop-altering or splice-altering substitutions), with 27 validating in MDCS. As a means to determine potential causality, they performed Mendelian randomization of the data, demonstrating 8 examples where proteins were likely causally associated with clinical risk factors. Last, they looked for genetic variants that were associated with multiple different protein levels, suggesting that these loci might report on central pathways mediating CVD risk. They found 3 of these pleotropic variants near the PPM1G, FOXP1, and FAM20A genes. The variant near PPM1G, which was associated with apolipoprotein E levels in this study, had previously been associated with circulating total cholesterol levels.³ Experiments in Hep G2 human liver hepatocellular carcinoma cell lines showed that knockdown of PPM1G resulted in decreased ApoE gene expression and decreased protein accumulation in the culture media. Although further experiments are needed to dissect the full molecular genetics and biology of these relationships, they do provide initial proof-of-principle mechanistic studies to confirm the statistical associations seen from this discovery study.

Many of the identified genetic variants were in cis (defined in this study as <1 mega base pairs) with the coding gene for the proteins and were in linkage disequilibrium with mutations in the coding genes that could plausibly directly impact circulating levels. Thus, these newly identified variants provide insight into the genetic regulation of protein abundance. In addition, several new trans variants (distant from the coding gene of the associated protein) were identified. These pQTLs provide important insight into previously unappreciated regulatory networks mediating CVD risk. For example, they found a pQTL for galectin-3–binding protein, a protein that has previously been associated with several CVD phenotypes, that is a nonsense mutation residing within the macrophage scavenger receptor type I protein (MSR1) gene, thus suggesting that variation in the MSR1 gene may modulate the macrophage lectin-binding activity of galectin-3–binding protein. Further experimental validation is needed to test this hypothesis.

The proteomics platform used in this study uses an emerging DNA-aptamer immunoaffinity technology. The power of this medium-scale platform allows scale evaluation of a larger number of proteins than afforded by traditional assays, while providing less complex analytics for protein identification and annotation required by unbiased mass spectrometry–based proteomics. This and other such medium-scale proteomics platforms are now available and, as demonstrated by this study, will likely have a significant impact on the discovery of novel biomarkers and molecular pathways in CVD and other diseases. It is important to note that these platforms provide a snapshot of the global proteome (1129 proteins), and that this study only examined a subset of those that were associated with the Framingham CVD risk score. Newer generations of the platform provide coverage of >4700 proteins⁴; pQTL studies using this expanded data set would describe even more of the genetic architecture underlying the cardiovascular risk proteome. Aptamer-based and -related proteomic technologies also have unique caveats that must be considered, because there is variability in the binding affinity and specificity for different proteins. For example, in this study, proteins represented by aptamers with lower binding affinity/specificity could result in pQTLs that account for less of the variance in protein levels. As this group has previously done, aptamer specificity for top protein hits can be validated using mass spectrometry.²

Overall, this important integrated omics study provides a valuable pQTL atlas as a publicly available resource to further study the genetic architecture that mediates intermediate markers of CVD risk and to identify potential new therapeutic targets. As with many functional genomics studies, these statistical associations serve as only the beginning of our understanding of their significance. Mendelian randomization approaches may point toward causality of variants on clinical or intermediate risk factors; however, as the authors did for their PPM1G variant, functional validation will be necessary to confirm the proposed impact of other variants on protein levels and function. This is particularly important because identified genetic variants may not directly mediate protein levels but rather be in linkage disequilibrium with true functional variants. It is important to note that the proteomic and genotyping primary data have been uploaded to the publicly available dbGAP. Such resource sharing is critical to allow other investigators to continue to explore the associations described here and to perform the necessary functional validation; the authors should be lauded for their commitment to data sharing. Moreover, the data can be combined with similar data sets through meta-analyses to increase the power to detect new pQTLs, and be used for integration of other molecular data in FHS, such as metabolomics,⁵ to approach a true systems biology perspective of the CVD risk landscape.