Linking Genetics and Proteomics: Gene-Protein Associations Built on Diversity

Medical applications from big data science emerge predominantly from the discovery of associations, for example, between a genomic variant and a specific disease. Although the starting point of such studies is merely observational, we have been witnessing unprecedented advances in big data science during the last 15 years, which also allow mechanistic inferences. Indeed, identification of hundreds of genetic variants showing disease associations helped better map the complex genetic architecture of many cardiovascular diseases. On the basis of such big data, Mendelian randomization studies can inform which risk markers are causal risk factors, whereas polygenic risk scores can be used to predict disease risk. Genetic associations, however, will benefit from other layers of molecular information, in particular with regard to prioritizing variants for further functional investigations. Transcriptomics, proteomics, and metabolomics will be essential to explore dysregulated pathways that ultimately cause disease. Such integrated big data sets may also have commercial value, ie, decisions on drug developments can be guided by linking potential targets to causal pathways depicted from gene variants to dysfunctional proteins. Indeed, the promise of precision medicine builds on such insights leading to differentiation of patient subgroups that benefit from different therapeutic strategies.¹

Article, see p 357

Big data science has been enabled by high-throughput methods for data generation and novel computational approaches to analyze large datasets. Genomics has been at the forefront of these developments and now expands into whole exome or even whole genome sequencing in large cohorts. Transcriptomics is moving toward single-cell analysis. Proteomics and metabolomics are about to complete the information on the key molecular layers of biological systems at the large scale.

In terms of the cohorts for big data science, an in-depth and accurate phenotypic assessment along with long-term follow-up have proven to be most valuable. And, of course, size matters. Not withstanding the importance of studying rare diseases, the principle “bigger is better” holds true for improving the resolution for the identification of disease associations. Only at scale challenges in big data science, such as adjustment for multiple testing and confounding, can be addressed successfully.

However, human diversity provides another opportunity to search for associations between genomic markers and phenotypic variability.² In fact, genetic epidemiological research has long focused on rather homogeneous cohorts in which numerous genes may lack functionally relevant variants. Given the stringent statistical thresholds in big data science, functionally relevant associations might have been overlooked. Thus, to take advantage of the gene pool of the global population—rather than predominantly of individuals of White ancestry—may further improve big data science in biomedical research (Figure).

• Download figure
• Download PowerPoint

The cradle of human evolution is the African continent, between 200 000 and 300 000 years ago. Starting about 70 000 to 100 000 years ago, the first modern humans migrated first northbound and subsequently spread around the globe, carrying with them a subset of the African gene pool. During these periods of migration and the associated environmental challenges, some genetic variation was lost. Therefore, African populations have been proposed to be particularly informative for investigation of human variation that arose among the early hominid predecessors between 100 000 and 5 million years ago.² It was hypothesized that the number of ancient variants and the selection pressures they survived may yield particularly rewarding insights into complex traits in all populations.

Other genomic variation was subsequently added, eg, the DNA of Neanderthals was only incorporated after modern humans left Africa.³ Traces of Neanderthal DNA found in a human genome suggest that the first modern humans were interbreeding with our now-extinct relatives during migration. As a consequence, the full genetic and ultimately phenotypic diversity of the human population needs assessment of all ethnic groups. For example, the strongest genetic risk for coronary artery disease resides at the 9p21 locus—but is predominately found in Western Europeans.⁴ In fact, this risk haplotype is largely absent in the Black population. ⁵ On the other hand, common variants in APOL1 (apolipoprotein-L1), which protect against sleeping sickness but increase risk of renal failure, are predominately found in sub-Saharan populations.⁶

An article in this issue of Circulation contributes to mapping the Black map of genome-proteome associations.⁷ Katz and colleagues integrated whole genome DNA sequencing and proteomic profiling of ≈1300 proteins in the plasma of individuals of Black descent. They report 569 associations between genomic variants and plasma proteins that reach a Bonferroni-adjusted significance level.⁷ One in 3 proteins found in the plasma was quantitatively modulated by genetic variants, of which about two-thirds were in cis, ie, in the vicinity of respective genes, and one-third was in trans.⁷ Overall, about a third of the variability of plasma protein concentrations was found the be partially heritable on the basis of the 28 million genomic variants studied. In this respect, Sun et al had previously reported a far lower heritability (8%).⁸

It is important to note that the authors validated the proteomic profiling by the SOMAScan expanded platform in 2 other cohorts, in which ≈90% of hits were consistent in directionality. They also validated the profiling using another proteomic platform (Olink Explore), in which 86% of hits were in similar direction, with 51% of all associations confirmed at a Bonferroni-adjusted P value. Both the SOMAScan and the Olink platform rely on binders for relative protein quantitation: the SOMAScan assay uses aptamers as a single binder. The Olink Explore platform uses dual antibody-based proximity extension assays. For studying the effects of genomic variants on protein levels, it is essential to exclude that the genomic variants induce platform-specific binding effects as illustrated by some highly discordant results. By replicating parts of the results of the SOMAScan using the Olink platform, the concordant results can be considered more reliable. However, even concordant results between 2 platforms may be misleading if aptamer and antibody binding occur in the same protein region that is affected by the genetic variant. This possibility should not be readily discounted and could be more reliably addressed by proteomics profiling techniques that do not rely on binders. Mass spectrometry measures peptides directly, and assays can be designed for the variant and the wild-type protein using authentic reference peptides for accurate quantitation.⁹

It is important to note that a quarter of associations were novel and missed by investigations in other ethnic groups. Several of the newly identified associations represent biologically interesting candidates.

From a functional perspective, the findings include a novel association linking the APOE gene locus with ZAP70 (zeta chain of T cell receptor–associated protein kinase 70) and MMP-3 (matrix metalloproteinase-3) protein levels in plasma, as well as a novel pleiotropic locus at the HPX (hemopexin) gene, which is associated with 9 proteins. Hemopexin may be of particular interest in Black individuals with sickle-cell disease because this gene has been found to have heme-scavenging properties leading to reduced inflammation in mouse models of sickle-cell disease.¹⁰

Of particular clinical interest are potential novel insights on the genetic susceptibility to chronic kidney disease and cardiac amyloid deposition through variants at the APOL1 and ATTR loci that revealed African-specific associations with plasma proteins. Amyloidosis may result from the V122I mutation in the TTR gene leading to misfolding of the transthyretin tetramer, ultimately resulting in abnormal protein deposition in myocardium and nerve tissue. This variant, occurring in 3% to 4% of Black individuals, was found to be a robust protein quantitative trait locus for RBP4 (retinol-binding protein 4), a binding partner of TTR, a finding that may allude to novel mechanisms involved in the pathogenesis of cardiomyopathy or neuropathy.

A variant at the APOL1 locus, rs73885319, was found to have a minor allele frequency of 23% in the Black population, whereas the variant is not present in individuals of European ancestry. In addition to being associated with levels of APOL1, Katz et al⁷ observed the sentinel single nucleotide polymorphism at this locus also determines plasma levels of CKAP2 (cytoskeleton-associated protein 2), which has been linked to tumor formation and renal tubular necrosis.⁶

All these associations between genomic variants and plasma proteins in people of African descent can only be a starting point for further mechanistic explorations. In this respect, it will be important to make these findings accessible for the broader research community, as outlined by the authors.⁷ Moreover, functional studies should follow to corroborate the findings and characterize the pathways that link genetic variations and protein levels. Also, it has to be taken into consideration that genetic variants tend to explain only a small proportion of the dynamic range of protein concentrations in plasma. Other factors, such as plasma protein synthesis, proteolysis, and renal clearance, are stronger determinants of plasma protein abundance and may lead to spurious associations.¹¹ Thus, follow-up experiments are needed to address biological plausibility, pleiotropic effects, and possible pathways to disease to enable clinical translation.

Article Information

Footnotes