Background
Physiological traits are frequently dynamic, varying over time with changing or accumulating environmental and physiological factors. The influence of genotype on these traits may also vary over time through interaction with factors such as age, developmental stage or time-dependent environmental factors. Variation in the effects of genetic variants at different stages of life could significantly alter the trajectories of traits and potentially also the risks of certain diseases. Hence, studies that do not consider the possibility of longitudinal variation in genetic associations may lead to over-simplistic models of variant effects.
Type 2 diabetes (T2D) is a complex disease with numerous risk factors, including a growing number of known genetic susceptibility variants [
1‐
4]. The risk of T2D also rises with increasing age [
5]. Some of the intermediate traits of T2D, such as obesity and raised fasting plasma glucose, similarly have complex determinants and indeed both BMI and fasting plasma glucose also tend to increase with age [
6,
7]. These observations raise the question of whether time- or age-varying genetic effects exist that could influence such intermediate traits and thereby T2D risk.
To date, there have been few studies of the longitudinal effects of T2D- or obesity-susceptibility variants. Those that have examined the longitudinal effects of the
FTO variant rs9939609 with BMI, give mixed results [
8‐
13]. This SNP has been shown through cross-sectional studies to be robustly associated with raised BMI and, likely as a result of this effect, also linked to the risk of T2D [
3,
11,
14‐
16]. Some longitudinal studies also provide evidence of an age-dependent relationship between rs9939609 and BMI whilst others do not, and some were performed in children and some in adults [
8‐
13].
In the present study, we tested the cross-sectional and longitudinal associations of common genetic variants shown to be associated with the risk of T2D and/or raised BMI [
1‐
4,
14,
15,
17‐
20], with fasting glucose levels and BMI. The study utilised data from the Busselton Health Study (BHS), in which participants have completed an average of 3.9 phenotypic surveys each, over an average follow-up time of 21.2 years (95% CI 13.6-28.8 years). The variants selected were the potassium inwardly-rectifying channel J11 (
KCNJ11) single nucleotide polymorphism (SNP) rs5219 (E23K variant); the peroxisome proliferator-activated receptor gamma (
PPARG) SNP rs1801282 (Pro12Ala variant); the transcription factor 7-like 2 (
TCF7L2) SNP rs7903146; the insulin-like growth factor 2 mRNA binding protein 2 (
IGF2BP2) SNP rs4402960; the CDK5 regulatory subunit associated protein1-like 1 (
CDKAL1) SNP rs10946398; the solute carrier family 30 (zinc transporter), member 8 (
SLC30A8) SNP rs13266634; the hematopoietically expressed homeobox (
HHEX) SNP rs1111875; and the fat mass and obesity associated (
FTO) SNP rs9939609. These variants represent those with the strongest reported effects on type 2 diabetes risk and, for the
FTO variant, BMI.
Discussion
While the number of genetic variants confirmed to be associated with T2D is growing, the relationships between these variants and T2D intermediate phenotypes are yet to be fully characterised, particularly longitudinally. In this study, we investigated the cross-sectional and longitudinal associations of seven T2D-susceptibility variants in the KCNJ11, PPARG, TCF7L2, IGF2BP2, CDKAL1, SLC30A8 and HHEX genes with fasting glucose level, and of another in the FTO gene with BMI, in a large, population-based European-Australian cohort.
Our cross-sectional analyses showed that, of the seven SNPs analysed with the fasting glucose outcome, only the
IGF2BP2 SNP rs4402960 showed a significant result, the T allele at this SNP being nominally associated with raised fasting glucose (
p = 0.045). Though some previous studies have reported significant associations of rs4402960, rs7903146 and rs13266634 with raised fasting glucose [
33‐
36], others find no significant associations between these three SNPs and fasting glucose in their populations [
33,
34,
37‐
40]. Studies of the SNPs rs5219, rs1801282, rs10946398 and rs1111875 have also shown no significant associations with fasting glucose [
17,
33,
34,
38,
40‐
44]. Rather, the majority of evidence from previous studies suggests that these seven SNPs alter T2D susceptibility through effects on insulin sensitivity (rs1801282 [
45]) or pancreatic beta cell function (rs5219, rs7903146, rs10946398, rs4402960, rs13266634 and rs1111875 [
34,
36,
38,
46‐
52]). Indeed, we did observe associations of the risk allele at the
PPARG SNP rs1801282 with lower HOMA2-%S (
p = 0.01), and of the risk alleles at the
CDKAL1 SNP rs10946398 and the
SLC30A8 SNP rs13266634 with lower HOMA2-%B (
p = 0.01 and
p = 0.03, respectively). Fasting glucose level may reflect abnormalities in insulin sensitivity and beta cell function, and is a useful clinical indicator of diabetes. However, it is dependent on a number of other factors, e.g. the site of insulin insensitivity [
53], which may have influenced the results here. It may also be that effects on fasting glucose are more subtle than may be detected with the available power. We also did not observe significant associations of the SNPs rs5219, rs10946398, rs1111875 or rs9939609 with T2D, likely due in part to the small number of diabetic participants in our cohort.
Associations may also be more difficult to detect if SNP effects vary with factors such as other gene variants, lifestyle factors and age. If a SNP has an age-varying effect on a particular trait, such as an effect to raise fasting glucose level as of middle age for example, then cross-sectional analysis of participants of all ages may miss the association. On the other hand, longitudinal analysis may still identify the relationship, by examining the association of the combination of SNP allele and age with phenotype in aging individuals, through a SNP × age interaction term. However, in our longitudinal analyses of SNPs with fasting glucose level, we found no evidence of changes in associations either with age or over time, indicating that these SNPs are not involved in the age-related increase in fasting glucose level, and that their effects are not influenced by any environmental factors that have changed over the ~20 year time period examined. The curves of Figure
1 depict these results, generally showing no significant differences in the rates at which fasting glucose level increases with age for the three genotype groups of each SNP, although it is important to note again that we did not see strong evidence for cross-sectional differences.
Our investigation of the
FTO SNP rs9939609 showed no significant association with obesity. However, we did confirm that the A allele at rs9939609 was significantly associated with raised BMI, both in cross-sectional analyses of the whole adult study population (
p = 0.003), and in analyses of the unrelated sub-cohort performed across all surveys (
p = 0.004). The association between
FTO variants and BMI is not in doubt, and has been described both in childhood and old age [
11,
14‐
16,
54]. For the rs9939609 SNP, an effect size of about 0.4 kg/m
2 per susceptibility allele has been reported [
14]. But does this effect size alter with age? We found no significant evidence of this in our analysis of the rs9939609 SNP in the unrelated study population. Several other studies of the longitudinal effects of
FTO on BMI have been conducted. Some of these have reported different effects at different ages, but these findings were often limited to one sex or another subgroup after a primary analysis, and chance may have influenced these results [
8,
11‐
13]. The evidence that
FTO has different effects on BMI at different ages or at different times is therefore weak and further studies are needed.
Differences between the findings of this and other studies may have arisen from numerous sources. Most likely chance has played a strong role. Moreover, this study is underpowered to detect small to moderate genetic effects. Hence, the failure to observe significant associations must be interpreted in this light. Differences in factors that may interact with genotype but were not controlled for, e.g. lifestyle factors such as diet and physical activity, may also have influenced the results, as may have differences in the ethnic compositions of the cohorts studied. The risk allele frequencies of all eight SNPs in the BHS cohort studied here (Additional file
1: ESM Table
1) are comparable to those in European populations, as reported in the dbSNP database
http://www.ncbi.nlm.nih.gov/projects/SNP/.
It is clear that further studies are required to assess the validity of our longitudinal findings. These would ideally address the issues of analytic power and additional covariates, and test the generalisability of our results to populations from diverse ethnic backgrounds. In addition, though the average follow-up times of 18.7 years for fasting glucose level and 21.8 years for BMI enable an extended longitudinal analysis to be conducted, they cannot provide as complete a picture of age-related changes as would come from a study that followed participants through from youth to old age. However, to our knowledge, few longer running longitudinal surveys of the same individuals are able to address these longitudinal questions, and the BHS provides phenotypic data that is particularly comprehensive. The understanding of the function of the T2D-susceptibility SNPs studied here would also benefit from longitudinal analysis of additional metabolic traits, such as glucose tolerance test 30 minute and 2 hour glucose levels, which would provide further information on their relationship to impaired glucose tolerance.
Competing interests
The authors declare that they have no competing interests.
Authors' contributions
RW carried out the statistical analyses and drafted the manuscript. NW contributed statistical advice for the cross-sectional and longitudinal analyses. JB was involved in data collection and laboratory analysis. TF participated in the design and coordination of the study and helped to draft the manuscript. LP participated in the design and coordination of the study. All authors read and approved the final manuscript.