Genetics: The next step in Mendelian randomization
Understanding how variations in our genome influence our susceptibility to diseases is one of the most compelling research topics in the life sciences. Researchers have used genome-wide association studies – experiments that analyze the DNA sequences of multiple individuals – to identify statistical relationships between genetic variants and specific human traits, such as susceptibility to a disease or various body parameters.
Despite the success of this approach, major challenges persist. First, associations between variants that are located close to each other within the genome can make it difficult to determine which of these genetic changes are responsible for the phenotype of interest (a problem called linkage disequilibrium). Second, even if specific variants can be identified, it is often not straightforward to determine the molecular mechanism by which they impact the trait (Tam et al., 2019).
To overcome these difficulties, studies often include information about other modalities such as transcriptomes, proteins and metabolites (Emilsson et al., 2008; Fraser and Xie, 2009; Nicolae et al., 2010; Wainberg et al., 2019; Schadt, 2009; Suhre et al., 2011). Some ‘multi-omic’ studies use one modality, or ‘layer’, to confirm changes to another, such as confirming changes in levels of mRNA by measuring the respective protein product. However, there is a shortage of examples of mechanistic links between the different layers (Buccitelli and Selbach, 2020; Wörheide et al., 2021). Now, in eLife, Zoltán Kutalik, Eleonora Porcu and colleagues from the Swiss Institute of Bioinformatics and the University of Lausanne – including Chiara Auwerx as first author – report a new approach that uses a technique called Mendelian randomization to reveal a chain of molecular connections between the transcriptome, metabolome, and high-level physiological traits such as biomarkers associated with kidney health (Auwerx et al., 2023).
Mendelian randomization is considered to be an ‘experiment of nature’, as it uses variations already present in the genetic code to determine if exposure to certain conditions (such as the amount of cholesterol in the blood, or the expression level of a gene) affects a specific trait (for instance, increased susceptibility to heart disease). The genetic variants act as a proxy, or ‘instrument’, for exposures that are difficult or impossible to manipulate in the population being studied. Mediation analysis can then be applied to ask if the exposure is responsible for the effects of the instrumental variable on the trait of interest. However, it is necessary to proceed carefully (Sanderson et al., 2022): for example, the instrumental variable being used should not affect the trait of interest through any other mediator.
The computational framework presented by Auwerx et al. integrates results from genome-wide association studies with data on genetic variants that affect the level of transcripts or the composition of metabolites. These variants are typically referred to as eQTL (short for expression quantitative trait loci) and mQTL (metabolite QTL), and can be derived from separate population cohorts, allowing researchers to tap into the vast resources of information that are already available.
First, causal links between transcripts and metabolites are established using overlapping mQTL and eQTL as instrumental variables. Causal effects of metabolites on traits of interest are then determined in the same manner using mQTL and genetic variants identified in genome-wide association studies. The next step in the framework is purely based on this established causality: transcripts that causally affect trait-modifying metabolites have to be causally linked to the same trait, resulting in transcript-metabolite-trait triplets (Figure 1). A statistical calculation, known as multivariate Mendelian randomization, is then performed on these triplets using the metabolite-associated variants as the instrumental variable. This determines what proportion of change in the outcome is a result of the transcript directly (or via unknown mediators) impacting the trait, and what proportion is the result of changes in the level of the metabolite mediating the relationship between them.

Mendelian randomization with multiple variables.
In the first step, Mendelian randomization calculations establish causal links between: (i) transcripts (T; pink chains) and metabolites (M; green hexagons) using eQTL and mQTL as instrumental variables (IV; first row); (ii) metabolites and various phenotypes (Y, such as height), using mQTL and the genetic variants associated with the traits as instrumental variables (second row). These causal links are then overlapped to establish causal triplets (third row). These causal triplets are subsequently analyzed in another Mendelian randomization-based calculation, which evaluates the effect of the respective mQTL on the levels of the transcripts, metabolites and traits of the triplet (fourth row). From this multivariate Mendelian randomization (MWMR), the proportion of transcript changes that directly effect a trait, and the proportion that cause an effect via metabolites, can be inferred. eQTL: expression quantitative trait loci; mQTL: metabolite quantitative trait loci.
Image credit: Figure created using BioRender (CC BY 4.0).
Auwerx et al. highlight an intriguing example of genetic variants affecting the transcription of a citrate-exporting protein encoded by a gene called ANKH that has been implicated in mineralization disorders. The resulting change to the export of citrate seems to affect the level of calcium present in the serum of individuals – a connection that was not detected when only transcript levels were correlated with the calcium trait.
By extending the Mendelian randomization approach to include two modalities (transcripts and metabolites), this new framework can detect causal relationships that could not be identified by comparing the genome wide association data to a single modality only. It also provides new insights into how the transcript impacts the phenotype through metabolic changes. With multi-omics studies increasing further in size, it is highly probable that even more advanced statistical approaches may become feasible in the future.
References
-
MRNAs, proteins and the emerging principles of gene expression controlNature Reviews Genetics 21:630–644.https://doi.org/10.1038/s41576-020-0258-4
-
Common polymorphic transcript variation in human diseaseGenome Research 19:567–575.https://doi.org/10.1101/gr.083477.108
-
Mendelian randomizationNature Reviews Methods Primers 2:e00092-5.https://doi.org/10.1038/s43586-021-00092-5
-
Benefits and limitations of genome-wide association studiesNature Reviews Genetics 20:467–484.https://doi.org/10.1038/s41576-019-0127-1
-
Multi-omics integration in biomedical research - a metabolomics-centric reviewAnalytica Chimica Acta 1141:144–162.https://doi.org/10.1016/j.aca.2020.10.038
Article and author information
Author details
Publication history
- Version of Record published: March 9, 2023 (version 1)
Copyright
© 2023, Weith and Beyer
This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.
Metrics
-
- 722
- Page views
-
- 84
- Downloads
-
- 0
- Citations
Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.
Download links
Downloads (link to download the article as PDF)
Open citations (links to open the citations from this article in various online reference manager services)
Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)
Further reading
-
- Epidemiology and Global Health
- Genetics and Genomics
Background: To evaluate the utility of polygenic risk scores (PRS) in identifying high-risk individuals, different publicly available PRS for breast (n=85), prostate (n=37), colorectal (n=22) and lung cancers (n=11) were examined in a prospective study of 21,694 Chinese adults.
Methods: We constructed PRS using weights curated in the online PGS Catalog. PRS performance was evaluated by distribution, discrimination, predictive ability, and calibration. Hazard ratios (HR) and corresponding confidence intervals [CI] of the common cancers after 20 years of follow-up were estimated using Cox proportional hazard models for different levels of PRS.
Results: A total of 495 breast, 308 prostate, 332 female-colorectal, 409 male-colorectal, 181 female-lung and 381 male-lung incident cancers were identified. The area under receiver operating characteristic curve for the best performing site-specific PRS were 0.61 (PGS000873, breast), 0.70 (PGS00662, prostate), 0.65 (PGS000055, female-colorectal), 0.60 (PGS000734, male-colorectal) and 0.56 (PGS000721, female-lung), and 0.58 (PGS000070, male-lung), respectively. Compared to the middle quintile, individuals in the highest cancer-specific PRS quintile were 64% more likely to develop cancers of the breast, prostate, and colorectal. For lung cancer, the lowest cancer-specific PRS quintile was associated with 28-34% decreased risk compared to the middle quintile. In contrast, the hazard ratios observed for quintiles 4 (female-lung: 0.95 [0.61-1.47]; male-lung: 1.14 [0.82-1.57]) and 5 (female-lung: 0.95 [0.61-1.47]) were not significantly different from that for the middle quintile.
Conclusions: Site-specific PRSs can stratify the risk of developing breast, prostate, and colorectal cancers in this East Asian population. Appropriate correction factors may be required to improve calibration.
Funding This work is supported by the National Research Foundation Singapore (NRF-NRFF2017-02), PRECISION Health Research, Singapore (PRECISE) and the Agency for Science, Technology and Research (A*STAR). WP Koh was supported by National Medical Research Council, Singapore (NMRC/CSA/0055/2013). CC Khor was supported by National Research Foundation Singapore (NRF-NRFI2018-01). Rajkumar Dorajoo received a grant from the Agency for Science, Technology and Research Career Development Award (A*STAR CDA - 202D8090), and from Ministry of Health Healthy Longevity Catalyst Award (HLCA20Jan-0022). The Singapore Chinese Health Study was supported by grants from the National Medical Research Council, Singapore (NMRC/CIRG/1456/2016) and the U.S. National Institutes of Health [NIH] (R01 CA144034 and UM1 CA182876).
-
- Developmental Biology
- Genetics and Genomics
The development of tools to manipulate the mouse genome, including knockout and transgenic technology, has revolutionized our ability to explore gene function in mammals. Moreover, for genes that are expressed in multiple tissues or at multiple stages of development, the use of tissue-specific expression of the Cre recombinase allows gene function to be perturbed in specific cell types and/or at specific times. However, it is well known that putative tissue-specific promoters often drive unanticipated 'off target' expression. In our efforts to explore the biology of the male reproductive tract, we unexpectedly found that expression of Cre in the central nervous system resulted in recombination in the epididymis, a tissue where sperm mature for ~1-2 weeks following the completion of testicular development. Remarkably, we not only observed reporter expression in the epididymis when Cre expression was driven from neuron-specific transgenes, but also when Cre expression in the brain was induced from an AAV vector carrying a Cre expression construct. A surprisingly wide range of Cre drivers - including six different neuronal promoters as well as the adipose-specific Adipoq Cre promoter - exhibited off target recombination in the epididymis, with a subset of drivers also exhibiting unexpected activity in other tissues such as the reproductive accessory glands. Using a combination of parabiosis and serum transfer experiments, we find evidence supporting the hypothesis that Cre may be trafficked from its cell of origin to the epididymis through the circulatory system. Together, our findings should motivate extreme caution when interpreting conditional alleles, and suggest the exciting possibility of inter-tissue RNA or protein trafficking in modulation of reproductive biology.