We have found associations (with p values less than 0.05) for SGA with the diabetes related SNP in KCNJ11 and the obesity related SNPs in FTO, PFKP, PTER, SEC16B and BDNF. After controlling for potential confounders the association with the FTO SNP did not remain significant, whilst the other 5 SNPs were positively associated with SGA in the multivariable model.
The T allele of KCNJ11 SNP rs5219 is associated with type 2 diabetes in adults and has been shown to be associated with reduced insulin secretion (see table for references) so our study result finding that this risk allele is associated with being SGA is compatible with the fetal insulin hypothesis where genetically mediated reduced insulin secretion beginning in-utero results in reduced birthweight, and later increases the risk of developing T2 D. Two previous studies evaluated the diabetes related KCNJ11
variant with birthweight and found no association [31
]. It is possible that this variant interacts with other factors, either genetic or environmental, that exist within the ABC cohort but are not present in the other two studies.
Our study found that the high risk allele for obesity in the PTER SNP (C allele at rs10508503) was associated with being SGA. This finding would fit with the fetal insulin hypothesis only if this allele had a direct effect on increasing insulin resistance prior to manifesting as increased BMI later in life, and thus manifests as low birth weight and later leads to obesity. Alternatively, the association of this obesity gene in SGA babies may be due to some survival advantage of being a "thin-fat" baby in terms of inappropriate fat mass for body size [33
], not discerned by the simple measure of birth weight. It may be that most of these SGA babies grow into genetically predisposed obese children and adults; hence we are observing the association with post-natal obesity in our SGA cohort.
Conversely, our study found an association of SGA with the low risk alleles for obesity in the BDNF and SEC16B genes suggesting that these alleles may confer a propensity to small size beginning in-utero, since the same SNP in BDNF has been associated with thinness in women [34
]. It would be interesting to examine whether this sub-group of SGA babies go on to have improved metabolic outcomes later in life by having a lower risk of obesity.
Since we began this study associations between common variants in type 2 diabetes susceptibility genes have been tested in several large birthweight cohorts. Freathy et al
, 2009 [35
] looked at five type 2 diabetes susceptibility genes and found that the CDKAL1
loci were associated with reduced birth weight. They did not detect an association with CDKN2A/B
, and SLC30A8
. All 5 of these loci were included in our study and we did not detect an association for any of them with SGA. Zhao et al
, 2009 [36
] also observed an association between lower birth weight and the CDKAL1
locus. However, no association was found with 19 other diabetes genes examined, including KCNJ11
for which we found an association with SGA. Pulizzi et al
, 2009 [37
] investigated 9 diabetes genes, all of which are included in our study but were not found to be associated with SGA. Of the tested variants, the risk variant in HHEX
showed a trend towards a low birthweight and the risk variant in the CDKN2A/2B
locus was associated with high birthweight. The three studies described above investigated birthweight. Only TCF7L2
has been studied for association with SGA [38
]. The gene was not associated with SGA in these two cohorts or our own. However, an association has been described between TCF7L2
and birthweight, although the effect was strongest with maternal genotype and after adjustment for maternal genotype fetal TCF7L2
genotype was not associated with birth weight [40
We examined the publically available British 1958 birth cohort database http://www.b58cgene.sgul.ac.uk/
for our significant genes. SNPs in BDNF
were associated with birthweight but KCNJ11, PTER, PFKP
did not show any associations.
The failure to replicate the associations reported by Freathy et al, Pulizzi et al and Zhao et al and our reporting of significant results for different genes may be due to the different phenotype used (birthweight vs. SGA) and/or due to different study populations with different environmental and genetic influences.
To summarise, we have identified five SNPs/genes which are associated with SGA. While noting that replication in independent samples is essential, our data provides evidence that genetic variation in type 2 diabetes and obesity susceptibility genes such as KCNJ11, BDNF, PFKP, PTER and SEC16B have a possible role in SGA as well as their established roles in obesity and/or diabetes.
We recognise that the association observed with these SNPs are unlikely to survive any adjustment for multiple testing and could thus be false positives. But it is possible that we may be seeing small genetic effects here and as our sample size is small compared to the majority of genetic association studies today we have low power to detect these associations with a high level of statistical significance. Calculations of statistical power using PS 2.1.31 [41
] show that for the ABC study we have 31.61% power to detect an odds ratio of 1.2, 56.84% power to detect an odds ratio of 1.3 and 78.01% power to detect an odds ratio of 1.4 for a SNP with a minor allele frequency of 0.48 (such as rs864745 in JAZF11). For SNPs with a lower allele frequency the power would be less. For example for a SNP with a minor allele frequency of 0.12 (such as rs6602024 in PFKP) we have 17.36% power to detect an odds ratio of 1.2, 31.42% power to detect an odds ratio of 1.3 and 48.05% power to detect an odds ratio of 1.4.
It is also possible that these genes may have more subtle effects and could affect a related phenotype, rather than be directly associated with SGA. Although beyond the scope of this paper it would be interesting to look at these SNPs/genes in relation such phenotypes e.g., catch up growth. Alternatively, the associations could reflect underlying LD with other markers in these genes. Further analysis in these genes with which we demonstrate an association with SGA is therefore required. Also, further investigation of the 36 genes for which we found no association should not be ruled out. The lack of association of these genes with SGA in our sample could be explained by a lack of power and we cannot rule out that we were unable to detect smaller effects of these variants. It is also possible that these obesity and/or diabetes genes may lead to small decreases in birthweight but do not result in the more severe SGA phenotype. Alternatively, it may be possible that any direct effects of susceptibility genes resulting in an individual being born SGA (by reduced insulin secretion) may be offset by an opposing effect from the maternal genotype (through the effects of the same variants on maternal glucose levels) [42
]. Unfortunately, maternal DNA samples are not presently available from the ABC cohort and so we are unable to test this.
During revision of this manuscript Freathy and colleagues reported a meta-analysis of genome-wide association studies and followed up the top hits in 13 replication studies [43
]. They identified two loci, in ADCY5
and near CCNL1
, that are associated with birth weight and explain 0.3% and 0.1% of the variance in birth weight, respectively. Both loci were also associated with smallness for gestational age. SNPs in ADCY5
have recently been implicated in regulation of glucose levels and susceptibility to type 2 diabetes [44
], providing further evidence that the association between lower birth weight and/or SGA and subsequent type 2 diabetes does indeed have a genetic component.