Does Greater Adiposity Increase Blood Pressure and Hypertension Risk?
Mendelian Randomization Using the FTO/MC4R Genotype
Elevated blood pressure increases the risk of experiencing cardiovascular events like myocardial infarction and stroke. Current observational data suggest that body mass index may have a causal role in the etiology of hypertension, but this may be influenced by confounding and reverse causation. Through the use of instrumental variable methods, we aim to estimate the strength of the unconfounded and unbiased association between body mass index/adiposity and blood pressure. We explore these issues in the Copenhagen General Population Study. We used instrumental variable methods to obtain estimates of the causal association between body mass index and blood pressure. This was performed using both rs9939609 (FTO) and rs17782313 (MC4R) genotypes as instruments for body mass index. Avoiding the epidemiological problems of confounding, bias, and reverse causation, we confirmed observational associations between body mass index and blood pressure. In analyses including those taking antihypertensive drugs, but for whom appropriate adjustment had been made, systolic blood pressure was seen to increase by 3.85 mm Hg (95% CI: 1.88 to 5.83 mm Hg) for each 10% increase in body mass index (P=0.0002), with diastolic blood pressure showing an increase of 1.79 mm Hg (95% CI: 0.68 to 2.90 mm Hg) for each 10% increase in body mass index (P=0.002). Observed associations are large and illustrate the considerable benefits in terms of reductions in blood pressure–related morbidity that could be achieved through a reduction in body mass index.
There is an approximately log-linear association between blood pressure (BP) and increased risk of cardiovascular events,1 and lowering BP in randomized, controlled trials yields a reduction in cardiovascular disease risk. Importantly, this reduction is not dependent on the nature of the antihypertensive therapy, strongly suggesting that it is BP lowering rather than other effects of therapeutic interventions that generates this benefit. This evidence strongly motivates public health approaches to reducing BP levels and the prevalence of hypertension within populations.2
Obesity and higher body mass index (BMI) are known associates of BP and hypertension and related disease risks.3–7 If causal, elevated BMI is a key target for effective intervention with respect to the reduction of BP. Evidence as to the importance of obesity in relation to hypertension risk is available from trials of weight reduction and the effect of this on BP as a clinical outcome. Available meta-analyses of the relationship between weight loss and hypertension suggest that a reduction of weight by even relatively small levels can reduce the risk of hypertension reliably,8,9 but they do not comment on the possible causality in these relationships.
If causal, increases in BMI should lead to an increase in the burden of hypertension. However, increasing prevalence in obesity and average BMI level has been accompanied by secular decreases in BP level and prevalence of hypertension,10 reviews having questioned the nature of associations between obesity and hypertension.11–13 Thus, it has been pointed out that the randomized, controlled trials of weight reduction, which generally involve changes in dietary intake and/or exercise, could influence BP through mechanisms other than weight loss itself.14 Consequently, understanding the influence of obesity and elevated weight on BP and hypertension requires additional study, using methods that provide insights into the causal nature of the observed associations.
One approach to strengthening causal inference, that of mendelian randomization,15 is based on the proposition that association between a disease and a genetic polymorphism that proxies for a directly measured risk factor is not generally susceptible to the reverse causation or confounding. This technique is analogous to a randomized trial, in which randomization to genotype (and, thus, exposure) takes place at conception.15
Recently, the application of this technique to questions regarding the role of obesity has been aided by the identification of genetic loci reliably associated with BMI/adiposity.16,17 After genomewide association studies for type 2 diabetes mellitus, replication,16 and meta-analysis of genomewide data in large collections of individuals,17 two loci with reliable associations with BMI have been identified. These loci are the fat mass and obesity–associated locus (rs9939609, FTO), and the melanocortin 4 receptor locus (rs17782313, MC4R). These provide suitable proxy markers for chronically elevated BMI and, importantly, markers that avoid both confounding and reverse causation. Together they account for ≈0.6% to 0.7% of observed variance in BMI in European populations, the greater part of the ≈1% now attributable to known, common, genetic polymorphisms.
The exact mechanisms of effect for these loci are not clear. The rs9939609 FTO locus may have a role in the hypothalamic regulation of appetite and food intake.18–20 However, the lack of complete understanding does not prevent the use of these independent loci to provide further evidence for the role of BMI in BP determination. Indeed, the FTO locus has already been used in this way in the investigation of the role of greater weight in the determination of components of the metabolic syndrome.21
Given the relationship between both FTO/MC4R and BMI, we aimed to assess the relationship between variation characterizing these associations and BP. Through the use of instrumental variable methods,22,23 we aim to estimate the strength of the unconfounded and unbiased association between BMI/adiposity and BP, together with the precision of such estimates.
Subjects and Methods
Copenhagen General Population Study
This is a cross-sectional study of the Danish general population initiated in 2003 and still recruiting and with focus on multifactorial phenotypes, including BP. At the time of genotyping for the present study, 37 027 unrelated individuals had been included (response rate: 45%). All of the participants were white (Danish) and were selected based on the national Danish Civil Registration System to reflect the adult Copenhagen general population aged 20 to ≥80 years. The study was approved by Herlev Hospital and a Danish ethical committee, it adhered to the principles of the Declaration of Helsinki, and all of the subjects gave informed consent.
The ABI PRISM 7900HT Sequence Detection System was used to genotype the FTO locus rs9939609 and the MC4R locus rs17782313 using TaqMan assays. Genotyping was verified by DNA sequencing in >30 individuals with each genotype. We performed reruns twice, and 99.96% of all of the available participants were genotyped.
BMI was calculated as weight (kilograms) divided by height squared (meters squared). This was log transformed to reduce skewness. To remove the dependence of BMI on sex, age, and height, log (BMI) was regressed on sex, age, age squared, log(height), and an age-sex interaction. The residuals from this model give the difference between an individual’s actual log(BMI) and that expected for his or her sex, age, and height. Exponentiating residuals gives an individual’s “relative BMI,” that is, the ratio between his or her actual BMI and that expected for his or her sex, age, and height. BP was measured by trained technicians using an automatic Digital Blood Pressure Monitor (Kivex) on the left arm, after 5 minutes of rest, and with the subject in the sitting position. The inflatable part of the cuff was 22×32 cm; however, if the circumference of the upper arm was >46 cm, we used a 32×45-cm cuff.
Plots of the difference between cumulative frequency distributions by genotypes at both FTO (rs9939609) and MC4R (rs17782313) were used to examine the nature of relationships between these loci and relative BMI. These simultaneously assess both the effect of genotypes on BMI at all ranges of this outcome (an assumption important for the application of instrumental variable analyses24) and also the most appropriate genetic model for the further use of genotypic data.
To adjust for the BP-lowering effect of antihypertensive medication, a constant value of 105 mm Hg was added to the systolic (diastolic) BP of those prescribed such medication.25 Any hypertension was defined as a systolic BP of >140 mm Hg, diastolic BP of >90 mm Hg, or the taking of antihypertensive drugs.26 Severe hypertension was taken as a systolic BP of >160 mm Hg, diastolic BP of >100 mm Hg, or the taking of antihypertensive drugs.
Smoking and alcohol consumption were dichotomized and defined as “ever” (ex-smoker or current smoker) versus “never” smokers and drinkers as those consuming >36 g of alcohol per week. Other possible confounding factors included education (0 to 9 years, 10 to 12 years, or >13 years) and annual income <400 000 krone, 400 000 to 600 000 krone, or >600 000 krone (100 000 krone is approximately $15 000).
We used Stata 10 (Stata Corp). P≈0 is equivalent to P≤10−200 or less. In observational and genetic analyses, continuous effects were estimated using linear regression with adjustment for age (quadratic), sex, height (logged), and an age-sex interaction (linear in age). Genotypes were used categorically. Logistic regression tested for association of the binary variable hypertension with tertiles of BMI and with FTO/MC4R genotypes, adjusted for age and sex.
For mendelian randomization analyses, instrumental variable methods were used to obtain estimates of the association between BMI and BP.27,28 This was performed using both rs9939609 (FTO) and rs17782313 (MC4R) as instruments for BMI and adjusting for age, sex, and height, as before. We used the generalized method of moments with robust standard errors to fit the instrumental variable models in the main analyses but checked results using limited-information maximum likelihood and 2-stage least squares. We compared the instrumental variable estimates with those from ordinary linear regression using the Durbin form of the Durbin-Wu-Hausman statistic. We examined F statistics from the first-stage regressions to evaluate the strength of the instruments.29–31
Table 1 shows baseline characteristics. Observationally, there was strong evidence for a linear association between BMI and systolic/diastolic BPs. Waist:hip ratio was strongly correlated with BMI (correlation coefficient: 0.5; P<0.001), and observational results reflected this. The correlation coefficient between BP and log-relative BMI was 0.20 (P<0.001) and 0.24 (<0.001) for systolic and diastolic BPs, respectively (Figure S1, available in the online data supplement at http://hyper.ahajournals.org). When arranged into deciles, the proportion of individuals found to have hypertension was also seen to increase consistently with BMI (Figure 1).
Each tertile increase of BMI showed an accompanying increase in the odds of hypertension of 1.73 (95% CI: 1.68 to 1.78; P≈0). This relationship was only slightly attenuated by adjustment for age, sex, education, smoking, and drinking (odds ratio: 1.71; 95% CI: 1.65 to 1.77; P≈0). For a more strict definition of severe hypertension, each tertile increase of BMI led to an odds ratio of 1.72 (95% CI: 1.67 to 1.77; P≈0). This was, again, slightly attenuated in the adjusted model (OR=1.68; 95% CI: 1.63 to 1.74; P≈0).
FTOrs9939609 was observed with a minor allele (A/fwd) frequency of 0.40 (SE: 0.002; counts: 13 019/18 057/5951). There was nominal evidence from a departure from Hardy-Weinberg equilibrium (P=0.02). MC4Rrs17782313 was observed with a minor allele (C/fwd) frequency of 0.25 (SE: 0.002; counts: 21 011/13 717/2299). This variant was observed to adhere to Hardy-Weinberg equilibrium (P=0.3). In contrast to the observed relationship between possibly confounding factors and BMI, there were no robust associations between potentially confounding factors and rs9939609 and rs17782313 genotypes in this cohort (Table S1).
Plots of the difference between the cumulative distribution functions of relative BMI stratified by both FTO and MC4R genotypes showed that both loci appeared to exert effects across the distribution of BMI (Figure 2). However, whereas FTOrs9939609 demonstrated differences in relative BMI cumulative frequency distribution (by genotype) that were indicative of an additive effect, MC4Rrs17782313 showed no substantial difference in the level of difference between groups defined as major homozygote versus heterozygote and major homozygote versus minor homozygote (Figure 2). This suggested that an assumption of additivity may not be appropriate in the case of MC4Rrs17782313 and that categorical analyses would be more appropriate for further analyses of both genotypes in this instance.
BMI showed an expected relationship with FTOrs9939609, with each rare allele accounting for a 1.18 (95% CI: 0.96 to 1.41) age- and sex-adjusted symmetrical percentage increase (P=2.0e-24; Table 2⇓). The corresponding effect for the MC4Rrs17782313 locus was a symmetrical percentage difference of 0.78 (95% CI: 0.53 to 1.04; P=2.2e-09) for BMI.
For systolic BP, adjusted age and sex showed an increase of 0.63 mm Hg (95% CI: 0.33 to 0.93 mm Hg) per rare allele of FTOrs9939609 (P=0.00004; Table 2⇑). With this, diastolic BP showed an increase of 0.26 mm Hg (95% CI: 0.09 to 0.43 mm Hg) per rare allele at FTOrs9939609 (P=0.003). For any and severe hypertension, the odds ratios per FTOrs9939609 allele were found to be 1.07 (95% CI: 1.03 to 1.10; P=0.0002) and 1.07 (95% CI: 1.04 to 1.11; P=0.00007), respectively, in the same sex- and age-adjusted data set (Table 2⇑).
For MC4Rrs17782313, BP did not show differences by genotype; however, it showed the direction of effect consistent with the expected relationships between MC4R and BMI. Age and sex adjustments showed an effect for systolic BP of 0.20 mm Hg (95% CI: −0.14 to 0.54 mm Hg) per rare allele of rs17782313 (P=0.3; Table 2⇑). Diastolic BP showed an effect of 0.08 mm Hg (95% CI: −0.12 to 0.27 mm Hg) per rare allele at rs17782313 (P=0.4). For any and severe hypertension, the odds ratios per MC4Rrs17782313 allele were found to be 1.02 (95% CI: 0.99 to 1.06; P=0.2) and 1.00 (95% CI: 0.96 to 1.04; P=0.96), respectively, in the same sex- and age-adjusted data set (Table 2⇑).
Relationships among FTOrs9939609, BP, and hypertension were maintained after adjustment for education, income, smoking, and drinking (Table 2⇑). In contrast to this, when the association between FTOrs9939609 and BP was adjusted for BMI, relationships were considerably attenuated, and no convincing evidence for association by genotype was found. Similar patterns were seen in the case of MC4Rrs17782313 (Table 2⇑).
There were strong relationships between tertiles of relative BMI and BP within this cohort (Table 3). In a linear regression model, systolic BP increased by 2.75 mm Hg (95% CI: 2.62 to 2.88 mm Hg) and diastolic BP by 1.75 mm Hg (95% CI: 1.68 to 1.83 mm Hg) for each 10% increase in BMI.
Adjustment of these relationships for the confounding factors education, income, drinking, and smoking led to attenuation of associations (Table 3). This can be considered in light of the associations of BMI and potential confounding factors presented in Table S1. Both education and drinking behavior showed strong patterns of association with BMI tertile, with those in higher BMI groups being less likely to be in the highest drinking group but more likely to be in the lowest educational bracket.
In instrumental variable analyses for the assessment of the continuous risk factor BMI on BP, using both FTOrs9939609 and MC4Rrs17782313 as instruments for BMI confirmed observational associations between BMI and BP. In analyses, including those of individuals taking antihypertensive drugs but for whom appropriate adjustment had been made, systolic BP was predicted to increase by 3.85 mm Hg (95% CI: 1.88 to 5.83 mm Hg) for each 10% increase in BMI (P=0.0002; Table 3). In the equivalent analysis, diastolic BP showed an estimated increase of 1.79 mm Hg (95% CI: 0.68 to 2.90 mm Hg) for each 10% increase in BMI (P=0.002).
For all of these analyses, the first-stage F statistic was >60,29,30 and there was no evidence of a departure of instrumental variable–derived estimates from those derived from observational analyses. Furthermore, effects from instrumental variable analysis were seen to have point estimates consistently greater than those derived from either basic or adjusted observational analyses. The correspondence between estimates derived from both observational and instrumental variable analyses is also shown in Figure 3.
In analyses using FTOrs9939609 and MC4Rrs17782313 as instruments for BMI separately, statistically equivalent estimates for the association between a 10% elevation in BMI and BP were attained. The SEs derived from these separate estimates were considerably larger (SEs: 1.23 and 1.77, respectively) than that derived from the use of both of these instruments simultaneously (SE: 1.01).
Observational estimates of the relationship between BMI and BP declined markedly with age. In data not shown, the effect of BP was seen to approximately halve in those >75 years old compared to those younger (P=0.007 interaction). In contrast to this, there was no evidence of an age-related change in the instrumental variable–derived BP/BMI effect. Instrumental variable analyses performed with the incorporation of drinking, smoking, education, age, and income as covariates in the instrumental variable model did not substantially alter results, nor did stratification by sex.
As expected, strong observational associations were found between BMI and hypertension. A 10% elevation in BMI was seen to be associated with a >3-mm Hg increase in systolic BP, this not only replicating previous findings but also adding to the weight of evidence as to the possibility of BP regulation through BMI control. However, we also found that BMI was robustly associated with socioeconomic factors, including educational status and alcohol consumption. As such, especially in light of alcohol consumption, which has been shown to be reproducibly associated with BP,32,33 the existence of confounding could not be ruled out in the explanation of these initial observations.
Within this population-based sample, the known relationships between BMI and the FTO and MC4R loci were replicated.16,17,34 Furthermore, these genotypes demonstrated 2 key properties that were important and informative to the formal undertaking of mendelian randomization analysis. First, both FTO and MC4R showed no consistent relationship with factors potentially confounding the BMI/BP relationship. Second, although the effect size of MC4R variation on BMI precluded a robust estimation of the direct genetic effect on BP, both genotypes showed directions of direct association with BP consistent with elevated BMI being positively and causally related to hypertension.
Previous work has noted that quantifiable increases in weight can be equated to expected increases in BP. Specifically, Neter et al8 showed, from meta-analysis of weight reduction–intervention studies, that a reduction in weight of ≈5 kg (by means of energy restriction, exercise, or both) was enough to achieve a reduction in systolic BP of ≈4 mm Hg. Variation at FTOrs9939609 has been shown to be related to an ≈2 kg difference in weight (from rare to common homozygote). Simple analysis shows an effect of just >1.0 mm Hg per rare FTO allele (systolic BP) in the Copenhagen cohort. This is roughly equivalent to a just >2.5 mm Hg increase in systolic BP per 5 kg of weight, and, although lower than previous estimates, it remains positive.
The application of instrumental variable analysis examining the relationship between BMI and BP gave an opportunity to appraise this association without the limitation of confounding, reverse causation, or measurement error. In this case, observational associations were supported by those associations found in a mendelian randomization framework. Furthermore and as expected from a lifelong exposure to elevated adiposity/BMI, there was actually evidence for greater estimates for the effects of BMI increase on BP increase when compared with those estimates derived from “one-off” cross-sectional analyses. This consistency in findings presents evidence that favors the consideration of BMI/adiposity increase as a causal factor in the etiology of hypertension. Also, unlike the naive observational associations, there was no attenuation of the instrumental variable estimates with age.
Possible limitations to the undertaking of mendelian randomization here include the possibility of population stratification, canalization, power deficiency, and an inability to detect the effect of acute changes in BMI on BP. However, because of the nature of this population cohort, both population stratification and power are unlikely to have considerable impact. Although the effect of acute BMI changes on BP are not the major focus here, the possible effects of canalization are lessened because of both the apparent lack of effect of the FTO locus on adiposity in early life and the concordance of MC4R-derived results with those of FTO as an instrument for adiposity.16
A further possible limitation of the approach taken in this work relates to the possible existence of pleiotropy concerning the genetic variation used as a proxy measure for BMI.35 Recent work concentrating on rare variation at the MC4R locus has suggested that there may be a BMI and insulin-independent effect on BP, and, as such, it may have the potential to introduce error to the inference taken from instrumental variable analysis.36 However, analyses here suggest an attenuation of the expected association between variation of the MC4R locus and BP (Table 2⇑), suggesting that MC4R variation retains a BMI-specific BP effect (as seen for FTO).
Overall, this research does not present a new hypothesis but adds to the available weight of evidence supporting a causal role for raised BMI in the etiology of hypertension. The exact mechanisms of this pathway may not be entirely clear; however, both observational and now mendelian randomization–derived data support the notion of targeting BMI directly as part of an effective therapeutic regime against hypertension.
We thank Dorthe Uldall Andersen and Anne Bank for technical assistance.
Sources of Funding
N.T. is funded by the Medical Research Council Centre for Causal Analyses in Translational Epidemiology grant RD1634. G.D.S. works within the Medical Research Council Centre for Causal Analyses in Translational Epidemiology funded by grant RD1634. R.H. is supported in part by Medical Research Council project grant G0601625, “Inferring Epidemiological Causality Using Mendelian Randomization.” B.G.N. is supported by the Copenhagen County Foundation and by Herlev Hospital.
- Received January 30, 2009.
- Revision received February 25, 2009.
- Accepted April 29, 2009.
Collaboration APCS. The impact of cardiovascular risk factors on the age-related excess risk of coronary heart disease. Int J Epidemiol. 2006; 35: 1025–1033.
Whelton PK, He J, Appel LJ, Cutler, JA, Havas S, Kotchen TA, Roccella EJ, Stout R, Vallbona C, Winston MC, Karimbakas J, for the National High Blood Pressure Education Program Coordinating Committee. Primary prevention of hypertension: clinical and public health advisory from the National High Blood Pressure Education Program. JAMA. 2002; 288: 1882–1888.
Batty GD, Shipley MJ, Jarrett RJ, Breeze E, Marmot MG, Davey Smith G. Obesity and overweight in relation to disease-specific mortality in men with and without existing coronary heart disease in London: the original Whitehall Study. Heart. 2006; 92: 886–892.
He Q, Ding ZY, Fong DY, Karlberg J. Blood pressure is associated with body mass index in both normal and obese children. Hypertension. 2000; 36: 165–170.
Graf C, Rost SV, Koch B, Heinen S, Falkowski G, Dordel S, Bjarnason-Wehrens B, Sreeram N, Brockmeier K, Christ H, Predel H. Data from the StEP TWO programme showing the effect on blood pressure and different parameters for obesity in overweight and obese primary school children. Cardiol Young. 2005; 15: 291–298.
Mamun AA, Lawlor DA, O'Callaghan MJ, Williams GM, Najman JM. Effect of body mass index changes between ages 5 and 14 on blood pressure at age 14: findings from a birth cohort study. Hypertension. 2005; 45: 1083–1087.
Neter JE, Stam BE, Kok FJ, Grobbee DE, Geleijnse JM. Influence of weight reduction on blood pressure: a meta-analysis of randomized controlled trials. Hypertension. 2003; 42: 878–884.
Ernsberger P, Haskew P. Health implications of obesity: an alternative view. J Obes Weight Regul. 1987; 6: 55–137.
Campos P, Saguy A, Ernsberger P, Oliver E, Gaesser G. The epidemiology of overweight and obesity: public health crisis or moral panic? Int J Epidemiol. 2006; 35: 55–60.
Frayling TM, Timpson NJ, Weedon MN, Zeggini E, Freathy R, Lindgren CM, Perry JRB, Elliott KS, Lango H, Rayner NW, Sheilds B, Harries LW, Barrett JC, Ellard S, Groves CJ, Knight B, Patch A, Ness AR, Ebrahim S, Lawlor DA, Ring SM, Ben-Shlomo Y, Jarelin M, Socio U, Bennett AJ, Melzer D, Ferrucci L, Loos RJF, Barroso I, Wareham NJ, Karpe F, Owen KR, Cardon LR, Walker M, Hitman GA, Palmer CNA, Doney ASF, Morris AD, Davey Smith G, Hattersley AT, McCarthy MI. A common variant in the FTO gene is associated with body mass index and predisposes to childhood and adult obesity. Science. 2007; 316: 889–894.
Loos RJF, Lindgren CM, Li S, Wheeler E, Zhao JH, Prokopenko I, Inouye M, Freathy RM, Attwood AP, Beckmann JS, Berndt SI, Bergmann S, Bennett AJ, Bingham SA, Bochud M, Brown M, Cauchi S, Connell JM, Cooper C, Davey Smith G, Day I, Dina C, De S, Dermitzakis ET, Doney ASF, Elliott KS, Elliott P, Evans DM, Sadaf FI, Froguel P, Ghori J, Groves CJ, Gwilliam R, Hadley D, Hall AS, Hattersley AT, Hebebrand J, Heid IM, Herrera B, Hinney A, Hunt SE, Jarvelin M, Johnson T, Jolley JDM, Karpe F, Keniry A, Khaw K, Luben RN, Mangino M, Marchini J, McArdle WL, McGinnis R, Meyre D, Munroe PB, Morris AD, Ness AR, Neville MJ, Nica AC, Ong KK, O'Rahilly S, Owen KR, Palmer CAN, Papadakis K, Potter S, Pouta A, Qi L, Randall JC, Rayner NW, Ring SM, Sandhu MS, Scherag A, Sims MA, Song K, Soranzo N, Speliotes EK, Syddall HE, Teichmann SA, Timpson NJ, Tobias JH, Uda M, Ganz Vogel CI, Wallace C, Waterworth DM, Weedon MN, Willer CJ, Wraight VL, Yuan X, Zeggini E, Hirschhorn JN, Strachan DP, Ouwehand WH, Caulfield MJ, Samani NJ, Frayling TM, Vollenweider P, Waeber G, Mooser V, Deloukas P, McCarthy MI, Wareham NJ, Barroso I. Common variants near MC4R are associated with fat mass, weight and risk of obesity. Nat Genet. 2008; 40: 768–775.
Gerken T, Girard CA, Tung YL, Webby CJ, Saudek V, Hewitson KS, Yeo GSH, McDonough MA, Cunliffe S, McNeill LA, Galvanovskis J, Rorsman P, Robins P, Prieur X, Coll AP, Ma M, Jonanovic Z, Farooqi S, Sedgwick B, Barroso I, Lindahl T, Ponting CP, Ashcroft FM, O'Rahilly S, Schofield CJ. The obesity-associated FTO gene encodes a 2-oxoglutarate-dependent nucleic acid demethylase. Science. 2007; 318: 1469–1472.
Timpson NJ, Emmett P, Frayling TM, Rogers I, Hattersley AT, McCarthy MI, Davey Smith G. The FTO/obesity associated locus and dietary intake in children. Am J Clin Nutr. 2008; 88: 971–978.
Stratigopoulos G, Padilla SL, LeDuc CA, Watson E, Hattersley AT, McCarthy MI, Zeltser LM, Chung WK, Leibel RL. Regulation of Fto/Ftm gene expression in mice and humans. Am J Physiol Regul Integr Comp Physiol. 2008; 294: R1185–R1196.
Freathy RM, Timpson NJ, Lawlor DA, Pouta A, Ben-Shlomo Y, Ruokonen A, Ebrahim S, Sheilds B, Zeggini E, Weedon MN, Lindgren CM, Lango H, Melzer D, Ferrucci L, Paolisso G, Neville MJ, Karpe F, Palmer CN, Morris AD, Elliott P, Jarvelin M, Davey Smith G, McCarthy MI, Hattersley AT, Frayling TM. Common variation in the FTO gene has a metabolic impact consistent with its effect on BMI. N Engl J Med. 2007; 57: 1419–1426.
Angrist JD, Graddy K, Imbens GW. The interpretation of instrumental variables estimators in simultaneous equations models with an application to the demand for fish. Rev Econ Stud. 2000; 67: 499–527.
European Society of Hypertension-European Society of Cardiology Guidelines C. 2003 European Society of Hypertension-European Society of Cardiology guidelines for the management of arterial hypertension [see comment] [published correction appears in J Hypertens. 2003;21:2203–2204]. J Hypertens. 2003; 21: 1011–1053.
Greenland S. An introduction to instrumental variables for epidemiologists. Int J Epidemiol. 2000; 29: 722–729.
Stock JH, Yogo M. Testing for weak instruments in linear IV regression. In: Andrews DWK, Stock JH, eds. Identification and Inference for Econometric Models: Essays in Honor of Thomas Rothenberg. Cambridge, United Kingdom: Cambridge University Press; 2005: 80–108.
Scuteri A, Sanna S, Chen W, Uda M, Albai G, Strait J, Najjar S, Nagaraja R, Orru M, Usala G, Dei M, Lai S, Maschio A, Busonero F, Mulas A, Ehret GB, Fink AA, Weder AB, Cooper RS, Galan P, Chakravarti A, Schlessinger D, Cao A, Lakatta E, Abecasis GR. Genome-wide association scan shows genetic variants in the FTO gene are associated with obesity-related traits. PLoS Genetics. 2007; 20: e115.