Centers for Disease Control and Prevention
Centers for Disease Control and Prevention		 Centers for Disease Control and Prevention CDC Home Search CDC CDC Health Topics A-Z    
Office of Genomics and Disease Prevention  
Office of Genomics and Disease Prevention

 

 Journal Publication

This commentary was published with modifications in Lancet 2003;9388:930-931.

Commentary

Mendelian randomisation: a new spin or real progress?

by Julian Littlea,b and Muin J. Khourya

a Office of Genomics and Disease Prevention, Centers for Disease Control and Prevention, Atlanta, GA 30341, USA
b Epidemiology Group, Department of Medicine and Therapeutics, University of Aberdeen, Aberdeen, UK

Article Outline

A recent statement that "In the genetics of complex diseases, association is in danger of becoming a rather dirty word" [1] calls to mind that similar sentiments were expressed nearly 30 years ago after a series of studies were published on the associations between various diseases and blood groups and HLA variants. [2] During this period, the human genome project has been completed, and new technologies for genomic analysis have been developed. Epidemiological methods and understanding of the biases of human observational studies have also advanced.

 

Genetic association studies are undergoing a renaissance under the banner of Mendelian randomisation. George Davey Smith and Shah Ebrahim [3] recently suggested that studies of the association between diseases and gene variants of known function may share with randomised controlled trials the advantage of excluding confounding as an explanation for a relation. Thus, in a population-based study of a genotype-disease association, the random assortment of alleles at the time of gamete formation (Mendel's second law) results in a random association between loci in a population and is independent from environmental factors. In theory, this random assortment brings about a similar distribution of variants at unlinked genetic loci between individuals with and without disease. This situation is analogous to adequately sized randomised controlled trials in which the random assignment to the intervention or control results in similar distributions of confounders (both measured and unmeasured) between the trial groups. The second step to Davey Smith and Ebrahim's reasoning is that for genes known to modulate the effects of environmental exposure, genetic variants with known functional effects can be considered as markers of altered exposure to an environmental factor of putative causal importance. Therefore the investigation of gene-disease associations potentially enables the effect of environmental exposures to be determined, excluding confounding as an explanation for the association.

 

Although these developments provide an exciting promise for epidemiological studies of gene-disease associations in the Human Genome Project era, there are some caveats. First, size matters. By contrast with the theoretical promise of Mendelian randomisation, the non-replication of association studies is well known.[4] Davey Smith and Ebrahim note that the major factor accounting for non-replication of results is likely to be inadequate statistical power, coupled with publication bias. This issue is analogous to the experience with randomised controlled trials, where evidence from small trials has not been confirmed in subsequent larger trials. [5] Publication bias is one explanation, but the distribution of confounders may also have differed between groups in smaller trials. As with trials, more consistent associations will be likely to be observed as the investigation of gene-disease associations matures––ie, moves from small innovative studies to large well-designed studies in which potential biases are kept to a minimum. In a recent meta-analysis, an excess of studies replicated the initial report, which likely was not due to publication bias. [6] The scientific record needs to be as complete as possible, which will be facilitated by initiatives such as the Human Genome Epidemiology Network. [7]

 

Second, the concept of Mendelian randomisation is predicated on a lack of effect of linkage disequilibrium (alleles at nearby loci preferentially associated with the alleles of interest). Empirical studies in human beings suggest that variation in linkage disequilibrium at all distances is great and is not predictable from one region of the genome to another.[8] Studies of microsatellite polymorphisms show linkage disequilibrium between a few loci that are separated by many megabases (>1 cM). [9] In addition, uncertainty exists about whether patterns are similar between populations. Differences in patterns of linkage disequilibrium between populations may partly account for the variable results from studies of gene-disease associations. [4] Therefore care is needed when making inferences in view of the current lack of knowledge about linkage disequilibrium patterns.

 

Third, the strategy discussed by Davey Smith and Ebrahim depends on knowledge of gene function. However, the contrasting associations between the MTHFR C677T polymorphism and cardiovascular disease and cancer, and between the NAT2 polymorphism and cancers of the bladder and colon (used as examples by Davey Smith and Ebrahim), suggest a need to understand not only gene function but also disease mechanisms. Even for a well-studied gene such as MTHFR, information on the functional effects of the variants on enzyme activity is limited. For the C677T and A1298C variants, there are few papers on enzyme activity in vitro,[10, 11 and 12] and enzyme activity in compound heterozygotes is unclear. [13] Moreover, the proposed mechanism by which the homozygosity for the C677T variant has an inverse association with colorectal cancer (reduced incorporation of uracil into DNA and associated effects on DNA/chromosome breakage) is not well supported. In one study, uracil misincorporation into human lymphocyte DNA, after folate depletion in vitro, was similar for all MTHFR genotypes.[14] In another study, [15] these genotypes were not significantly associated with uracil misincorporation or DNA strand breakage in lymphocytes. Homozygosity for the C677T variant was associated with increased DNA damage in coronary artery disease. [16] Overall, a tremendous expansion in biological information has occurred, stimulated by biotechnological innovation and the application of high-throughput technology. However, there may be concerns about selection of study participants and tissues (or cells) included in the studies of gene function and gene and protein expression on which this biological information is based.

Fourth, population stratification remains a issue.[17, 18 and 19] On the basis of work on the non-Hispanic white population in the USA and in Europe, this bias does not appear to be substantial when epidemiological principles of design, conduct, and analysis are rigorously used. [18 and 20] However, this may not apply to all genes and all populations. [20]

 

Finally, gene-environment interactions are important. The lack of a gene-disease association in an epidemiological study would not exclude gene-environment interaction in a subgroup.[21] Whilst recognising the implications of misclassification associated with many current methods for assessing exposure for testing hypotheses of departure from multiplicative interaction, [22] abandoning attempts to investigate joint effects of exposure and genotype would be inappropriate. Because of the implications of gene-environment interaction, a high-risk intervention may have a greater public-health effect than a mass-prevention strategy. For example, a screen-and-treat strategy is more cost effective than universal supplementation in lowering homocysteine by folic acid and vitamin B12. [23]

Mendelian randomisation offers a potentially exciting perspective on gene-disease associations in the genomics era. The realisation of this concept will require the integration of molecular biology and rigorous epidemiological principles of study design and analysis.

References

  1. Editorial. In search of genetic precision. Lancet 2003; 361: 357.

  2. T. Strachan and A.P. Read. Human molecular genetics, BIOS, Oxford (1996).

  3. G. Davey Smith and S. Ebrahim, Mendelian randomization": can genetic epidemiology contribute to understanding environmental determinants of disease?. Int J Epidemiol 32 (2003), pp. 1–22.

  4. J.N. Hirschhorn, K. Lohmueller, E. Byrne and K. Hirschhorn, A comprehensive review of genetic association studies. Genet Med 4 (2002), pp. 45–61.

  5. R. Collins and S. MacMahon, Reliable assessment of the effects of treatment on mortality and major morbidity I: clinical trials. Lancet 357 (2001), pp. 373–380.

  6. K.E. Lohmueller, C.L. Pearce, M. Pike, E.S. Lander and J.N. Hirschhorn, Meta-analysis of genetic association studies supports a contribution of common variants to susceptibility to common disease. Nat Genet 33 (2003), pp. 177–182.

  7. J. Little, M.J. Khoury, L. Bradley, M. Clyne et al., The Human Genome Project is complete: how do we develop a handle for the pump?. Am J Epidemiol 157 (2003), pp. 667–673.

  8. K.G. Ardlie, L. Kruglyak and M. Seielstad, Patterns of linkage disequilibrium in the human genome. Nat Rev Genet 3 (2002), pp. 299–309.

  9. J.K. Pritchard and M. Przeworski, Linkage disequilibrium in humans: models and data. Am J Hum Genet 69 (2001), pp. 1–14.

  10. R. Rozen, Genetic predisposition to hyperhomocysteinemia: deficiency of methylenetetrahydrofolate reductase (MTHFR). Thromb Haemost 78 (1997), pp. 523–526.

  11. N.M.J. van der Put, F. Gabreëls, F. Stevens et al., A second common mutation in the methylenetetrahydrofolate reductase gene: an additional risk factor for neural-tube defects?. Am J Hum Genet 62 (1998), pp. 1044–1051.

  12. I. Weisberg, P. Tran, B. Christensen, S. Sibani and R. Rozen, A second genetic polymorphism in methylenetetrahydrofolate reductase (MTHFR) associated with decreased enzyme activity. Mol Genet Metab 64 (1998), pp. 169–172.

  13. K.J.A. Lievers, G.H.J. Boers, P. Verhoef et al., A second common variant in the methylenetetrahydrofolate reductase (MTHFR) gene and its relationship to MTHFR enzyme activity, homocysteine,and cardiovascular disease risk. J Molec Med 79 (2001), pp. 522–528.

  14. J.W. Crott, S.T. Mashiyama, B.N. Ames and M.F. Fenech, Methylenetetrahydrofolate reductase C677T polymorphism does not alter folic acid deficiency-induced uracil incorporation into primary human lymphocyte DNA in vitro. Carcinogenesis 22 (2001), pp. 1019–1025.

  15. S. Narayanan. The effect of folic acid and genetic polymorphisms on DNA stability and colorectal cancer [dissertation], University of Aberdeen, Aberdeen, UK (2001).

  16. M.G. Andreassi, N. Botto, F. Cocci et al., Methylenetetrahydrofolate reductase gene C677T polymorphism, homocysteine, vitamin B12, and DNA damage in coronary artery disease. Hum Genet 112 (2003), pp. 171–177.

  17. D.C. Thomas and J.S. Witte, Point: population stratification: a problem for case-control studies of candidate-gene associations?. Cancer Epidemiol Biomarkers Prev 11 (2002), pp. 505–512.   

  18. S. Wacholder, N. Rothman and N. Caporaso, Counterpoint: bias from population stratification is not a major threat to the validity of conclusions from epidemiological studies of common polymorphisms and cancer. Cancer Epidemiol Biomarkers Prev 11 (2002), pp. 513–520.

  19. L.R. Cardon and L.J. Palmer, Population stratification and spurious allelic association. Lancet 361 (2003), pp. 598–604.

  20. K.G. Ardlie, K.L. Lunetta and M. Seielstad, Testing for population subdivision and association in four case-control studies. Am J Hum Genet 71 (2002), pp. 304–311.

  21. M.J. Khoury, W. Stewart and T.H. Beaty, The effect of genetic susceptibility on causal inference in epidemiologic studies. Am J Epidemiol 126 (1987), pp. 561–567.

  22. D. Clayton and P.M. McKeigue, Epidemiological methods for studying genes and environmental factors in complex diseases. Lancet 358 (2001), pp. 1356–1360.

  23. B.K. Nallamothu, A.M. Fendrick, M. Rubenfire, S. Saint, R.R. Bandekar and G.S. Omenn, Potential clinical and economic effects of homocyst(e)ine lowering. Arch Intern Med 160 (2000), pp. 3406–3412.