Genetic associations with obesity Answers

Question 1

What association is depicted in panel A of the Figure? What association is depicted in panel B? What is the most likely explanation for the less significant P-values in panel B? What do the combined findings in panels A and B suggest to you?

Answer 1

In panel A, several SNPs in the region 52.3 – 52.4 are strongly associated (P-values between 10-5 and 10-8) with type 2 diabetes. In panel B, several SNPs in the same region are associated with adult BMI in persons with type 2 diabetes (P-values between 10-4 and 10-5). There were only 1924 people with type 2 diabetes in the study, compared with 2938 population controls; therefore, smaller sample size could explain the less significant P-values in panel B. These findings together suggest that the effect of FTO on type 2 diabetes could be mediated through its effect on BMI, a major risk factor.

horz_line_shrt.gif

Question 2

In panels A and B, the Y-axis represents ‒log10 of the P-value for each association. What does the Y-axis in panel C represent? How would you interpret the pattern in panel C?

Answer 2

The Y-axis in panel C describes linkage disequilibrium (r2) with rs9939609 for other SNPs in this region that were measured in HapMap for Caucasian European samples. These may or may not be the same SNPs measured in panels A and B; however, the pattern suggests that rs9939609 is a good marker for the observed association and that the observed clusters of nearby, highly associated SNPs in both panels reflect underlying linkage disequilibrium.

horz_line_shrt.gif

Question 3

In this study, SNP rs9939609 in the FTO gene region was strongly associated with type 2 diabetes, both in the original study population (OR = 1.27, 95% CI = 1.16 to 1.37, P = 5 x 10-8) and in a replication set of 3757 type 2 diabetes cases and 5346 controls (OR = 1.15, 95% CI = 1.09 to 1.23, P = 9 x 10-6). Analysis of the replication set was repeated with adjustment for body mass index (BMI) (OR = 1.03; 95% CI = 0.96 to 1.10; P = 0.44). How did adjusting for BMI affect the association of FTO rs9939609 with type 2 diabetes? What can you infer from this finding?

Answer 3

The replication study found a smaller association that was still highly significant; adjusting for BMI, however, abolished the association. This supports the hypothesis that FTO is associated with diabetes only through its effect on obesity.

horz_line_shrt.gif

Question 4

Two GWAS replicated the FTO-obesity association within the first year of its discovery in 2007. Since then, more than 200 additional studies related to this association have been published, including 15 GWAS and 18 meta-analyses. In addition to replicating the original finding, what further research questions could these studies have been designed to answer? List three types of questions that could be explored in genetic association studies of FTO and obesity.

Answer 4

Robustness of the association in populations with different geographic ancestry, associations with different phenotypes (e.g., BMI at different ages, different levels or classifications of BMI, measures of body composition such as skin-fold thickness, waist circumference, etc), and gene-environment interactions (e.g., with dietary intake, physical activity, etc). It could also be explored for fine mapping and secondary signals.

horz_line_shrt.gif

Question 5

Which of the relationships in the diagram (a, b, c, d) must be established before designing the Mendelian randomization study? Why?

Answer 5

Relationship a must have been established in previous studies to suggest the use of FTO genotype as an instrumental variable. Relationship b is assumed for both measured and unmeasured confounders; some relevant information may be available from prior studies or from the present study. Evaluating relationship c is the objective of the current study. Relationships d may or may not be measurable.

horz_line_shrt.gif

Question 6

Inspecting these results for effect size and statistical significance, what do you think of FTO genotype as a predictor of adult BMI? Of lifetime BMI? What implications does this have for the Mendelian randomization analysis?

Answer 6

Setting the threshold for statistical significance at P<0.05, significant associations were found for FTO genotype with both adult BMI and lifetime BMI. It is not easy to compare the results for adult BMI and lifetime BMI directly, since the latter is a standardized value (with population SD=1); however, the differences in BMI among genotype groups are not very large. A larger effect would make for a stronger instrumental variable.

horz_line_shrt.gif

Question 7

Note that the authors have laid out their results in a “2×4 table”—but they have calculated genotype-specific risks stratified by diet, rather than calculating ORg, ORe, and ORge relative to a single reference group. Because this was a cohort study, we can use prevalence to estimate the absolute risk of obesity in each genotype by diet stratum. Complete the table below. Use the calculator at http://statpages.org/confint.htmlExternal to calculate confidence intervals for each group. Label the strata Roo, Rgo, Roe, and Rge to indicate levels of genotype and environmental risk.

Saturated
fat intake

APOA2
Genotype

Obese

Not obese

Prevalence (Risk)

95% CI

Stratum

<22g/d TC+TT

150

544

<22g/d CC

24

111

­>22g/d TC+TT

135

393

>22g/d CC

39

63

Total

348

1111

0.238

0.217-0.261

 

Statistical interaction on the additive (risk difference) scale, Ird, is defined as follows:
Ird = Rge – Rgo – (Roe – Roo)

Answer 7

Note that when dealing with relative risks estimated by odds ratios rather than absolute risks,
ORoo = 1, producing the familiar formula ORge = ORg + ORe – 1.

Saturated fat intake

APOA2
Genotype

Obese

Not obese

Prevalence (Risk)

95% CI

Stratum

<22g/d TC+TT

150

544

0.216

0.186-0.248

Roo

<22g/d CC

24

111

0.178

0.117-0.253

Rgo

­>22g/d TC+TT

135

393

0.256

0.219-0.295

Roe

>22g/d CC

39

63

0.382

0.288-0.484

Rge

Total

348

1111

0.238

0.217-0.261

 

horz_line_shrt.gif

Question 8

In the absence of interaction on the additive scale, what is the expected value of Ird? Use your data from Question 7 to calculate Ird. Is there evidence of interaction on the additive scale?
The expected value of Ird is zero in the absence of interaction.

Answer 8

Ird = Rge – Rgo – (Roe – Roo)
Ird = 0.382 – 0.178 – (0.256 – 0.216) = 0.164 → evidence of interaction

horz_line_shrt.gif

Question 9

One explanation that is often cited to explain the increasing population prevalence of obesity is that today’s environment is mismatched with “energy-thrifty genes” that multiplied in the distant past, when food sources were unpredictable. In other words, according to the “thrifty genotype” hypothesis, the same genes that helped our ancestors survive occasional famines by depositing fat during times of abundance are now being challenged by environments in which food is plentiful year round. From what has been presented here, speculate on the qualifications of APOA2 as a candidate “thrifty gene.” Do you think this concept is relevant to public health efforts to prevent obesity?

Answer 9

The APOA2 CC genotype was associated with obesity only in the presence of high fat intake. The variant is common and so could have been selected for—16% of the population has the CC genotype—and although other genotype frequencies aren’t provided, the estimated C allele frequency is 40%. It seems somewhat paradoxical that the CC genotype would be protective against obesity in the low saturated fat intake group, if the role of a thrifty gene is to promote fat storage during “lean” times. Of course, the cutoff of 22g/d is simply the mean observed in this study and there’s no information on how that compares with other diets, modern or prehistoric. The effect of APOA2 CC genotype on risk of obesity is modest and there is no reason to think (nor have proponents of the thrifty gene hypothesis claimed) that only a single gene determines a thrifty phenotype. From a public health standpoint, the thrifty gene argument may be useful mainly as a way of making the point that people who eat the same diet may not be equally likely to become obese. Obesity is not simply the result of “bad behavior” but a reflection of genetics underlying the body’s response to environmental stimuli.

Page last reviewed: November 15, 2013 (archived document)