The aim of this study was to discover novel variants associated with lipid levels in children and to test if those associations were also significant in adults. We performed a GWAS of children undergoing treatment for ALL ascertained at St. Jude Children’s Research Hospital, and attempted replication in an independent population of youths and adults from NHANES III. Three of the 52 lipid-genotype associations tested in NHANES III children replicated at p<0.05, including intronic SGSM2 rs2429917 at p=0.009. However, these associations did not generalize to NHANES III adults. We also identified a genotype × age interaction with SGSM2 rs2429917 for transformed LDL-C in Blacks, supporting a genetic basis for the differences observed in lipid levels in children compared to older individuals.
Age as a modifier of genetic association studies has only recently been highlighted in the literature. In one study of the 100K data of the longitudinal Framingham Heart Study, Lasky-Su et al describe an age-dependent association between ROBO1
and obesity where the association was stronger among the pediatric cohorts compared with adult cohorts (Lasky-Su et al., 2008
). Although Lasky-Su et al do not speculate on the mechanism behind the age-dependent interaction, it is interesting to note that heritability estimates for obesity in children tend to be higher (Haworth et al., 2008
) compared with estimates in adults (Brown et al., 2003
). Somewhat consistent with these observations in obesity are the observations of heritability estimates for the lipid traits. That is, some studies have found that heritability of select lipid levels tends to decrease with age (Beekman et al., 2002
; Heller et al., 1993
). However, the review by Snieder et al (Snieder et al., 1999
) concluded that heritability estimates for HDL-C, LDL-C, and triglycerides remain relatively stable with age.
While the magnitude of the genetic influence on lipid metabolism may not change with age, the importance of different genes may. In other words, different genes may be expressed in childhood and adolescence compared to adulthood. In regards to lipid metabolism, longitudinal twin studies support this possibility (Friedlander et al., 1997
; Nance et al., 1998
; Williams and Wijesiri, 1993
), and an extended parent-twin study determined that different genes are expressed in adolescence compared to adulthood (Snieder et al., 1997
). It is also possible that the same genes function throughout life, but are expressed at different levels depending on the decade of life. Supporting this latter hypothesis is the observation that younger patients heterozygous for ABCA1
mutations that cause Tangier disease have significantly higher HDL-C levels than older patients heterozygous for ABCA1
mutations (Clee et al., 2000
). There is evidence that normal ABCA1
function increases over time (Clee et al., 2000
), which suggests pronounced HDL-C deficiency between age groups may be highlighting the heterozygous carriers’ inability to do so.
Given the proposed and observed differences between children and adults for these traits, we purposefully performed a discovery study in children as this subset may allow for discovery of novel genes associated with lipid levels compared with previously published GWAS from adults. For our work presented here, the most promising novel candidate as a result of this study is rs2429917, located in the intron of SGSM2
, or small G protein signaling modulator 2. SGSM2
is ubiquitously expressed in various tissues, including the liver, and as the name implies, acts as a modulator of G-protein signaling through its interaction with a subfamily of RAS proteins (Yang et al., 2007
). Proteins involved in G protein-mediated signal transduction are associated with a number of cellular mechanisms, including differentiation and proliferation. It is also important to note that rs2429917 is located in a fairly gene-dense region of chromosome 17, including SMG6
, and METT10D
, all within ~100 kb flanking SGSM2
. Based on their biological functions, none of these neighboring genes are compelling candidates for association with lipid metabolism. However, SMG6
is an intriguing candidate given its essential association with telomerase activity (Reichenbach et al., 2003
; Snow et al., 2003
) and, thereby, aging. Deletion of Est1p (the S. cervisiae
homolog to human SMG6
) in yeast leads to ever-shorter telomeres over time despite functional telomerase activity (Lundblad and Szostak, 1989
). Telomere shortening occurs in all mitotic tissues (excluding germline tissue) as humans age and has been shown to contribute to mortality in many age-related diseases, including heart disease (Cawthon et al., 2003
). Although these findings require further study, it is interesting to speculate that these data may point to the involvement of previously unsuspected pathways contributing to lipid metabolism.
This study had several limitations, including that the discovery GWAS was underpowered due to its small sample size. Even with our largest population (n=282 in Whites), and an allele frequency of 5%, we had 80% power to detect only large effect sizes (R2
>11%) at genome-wide significance. The majority of published lipid GWAS-identified variants have small effect sizes and explain only a small percent of the variance of lipid traits in the population (Teslovich et al., 2010
; Manolio, 2009
). However, to our knowledge, no GWAS has been performed on children with lipid levels; therefore, it is unknown whether the effect size and/or the significance of these well-known variants remain constant over a lifetime.
Another limitation is that the discovery cohort consisted of children undergoing treatment for ALL. Although medications administered for treatment of ALL are not known to affect lipid levels, side effects (such as loss of appetite or nausea) of these medications may cause nutritional differences that affect lipid levels. While this possibility was not directly measured and, therefore, cannot be completely ruled out, we observed that the children in the St. Jude cohort maintained healthy appetites (data not shown).
Despite the small discovery sample size, we were able to detect nominally significant associations, of which three replicated at p<0.05 in an independent dataset. Furthermore, examination of genetic variation known to influence lipids in European-descent populations demonstrates that true associations can be detected in spite of the low power of the study. That is, of the 26 established lipid-associated SNPs in 23 genes (including CETP, LPL, GCKR, APOB,
etc; Table S4
), we detected seven associations with p-values ≤ 0.05. As an example, rs328 is a non-synonymous SNP in LPL
and has been shown to have a reproducible effect (~19 mg/dL in one study) (Kathiresan et al., 2008
) on lowering triglycerides. In our GWAS of children, rs1741102, a proxy for rs328 (r2
=1 in HapMap CEU), was associated at p=0.02 with β=−0.17, corresponding to −21.3 mg/dL, which is consistent in both the previously reported magnitude and direction of effect.
A benefit of using NHANES III data is that it allows for genetic studies in a large, diverse population with a wide age-range. However, it is a cross-sectional study. Since our data suggest that there may be age-specific genetic influences, longitudinal data are necessary to derive further conclusions and to replicate this interaction.
Post-mortem studies on young adults and children have shown that atherosclerosis starts early in life (Expert Panel Blood Cholesterol Levels Child Adolesc, 1992
), even though clinical symptoms usually do not manifest until decades later. The potential temporal nature of factors, both genetic and non-genetic, that contribute to cardiovascular disease is important for better understanding of the etiology of the disease. While it is often assumed in genotype-phenotype association studies that genetic effects are stable over a lifetime, the possibility of important age-effects should not be ignored when studying the genetics of lipid metabolism.