Post genome-wide association study challenges for lipid traits: describing age as a modifier of gene-lipid associations in the Population Architecture using Genomics and Epidemiology (PAGE) study

Summary

Numerous common genetic variants that influence plasma high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), and triglyceride (TG) distributions have been identified via genome-wide association studies (GWAS). However, whether or not these associations are age dependent has largely been overlooked. We conducted an association study and meta-analysis in more than 22,000 European Americans between 49 previously identified GWAS variants and the three lipid traits, stratified by age (males: <50 or ≥50 years of age; females: pre- or post-menopausal). For each variant, a test of heterogeneity was performed between the two age strata and significant P_het values were used as evidence of age-specific genetic effects. We identified seven associations in females and eight in males that displayed suggestive heterogeneity by age (P_het<0.05). The association between rs174547 (FADS1) and LDL-C in males displayed the most evidence for heterogeneity between age groups (P_het=1.74E-03, I²=89.8), with a significant association in older males (P=1.39E-06) but not younger males (P=0.99). However, none of the suggestive modifying effects survived adjustment for multiple testing, highlighting the challenges of identifying modifiers of modest SNP-trait associations despite large sample sizes.

Introduction

Since 2005, 1,617 genome-wide association studies (GWAS) have been published for common human diseases and traits yielding 249 genetic variants associated at genome-wide significance (Hindorff et al., 2010). A maturing step in the GWAS-era is the identification of all common genetic variation associated with the disease or trait of interest. Arguably, this step is progressing rapidly for the lipid traits high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), and triglycerides (TG), for European-descent populations. The most recent meta-analysis of nearly 100,000 European-descent populations identified 95 loci associated with HDL-C, LDL-C, and/or TG (Teslovich et al., 2010).

Beyond identification of all common genetic variants, there is much interest in the post GWAS era in identifying modifiers of GWAS-identified variants (Ober & Vercelli, 2011). Sex, age, and environmental exposures such as diet and smoking status are known to affect lipid levels (Third Report of the National Cholesterol Education Program (NCEP) Expert Panel, 2002), but little is known about their interactions with genetic variants associated with lipid levels. The understanding of these potential modifiers of genetic effects may be important in both public health genomics and personalized medicine as, unlike one’s genetic make-up, some environmental exposures have the potential to be attenuated or amplified. Furthermore, understanding of these modifiers may also lead to a better understanding of the biology of the disease process, perhaps leading to improved preventive and/or treatment approaches (Manolio et al., 2006;Ober & Vercelli, 2011).

As mentioned above, it is well-known that lipid levels can vary considerably with age, with mean total cholesterol levels tending to increase with increasing age (Boomsma et al., 1996;Ericsson et al., 1991;Reilly et al., 1990;Snieder et al., 1997). But the change is not always gradual. Indeed, there are four specific times in the life cycle that result in significant changes in one’s lipid profile: birth and the first few following years, adolescence, middle age, and advanced age (Snieder et al., 1997). Often these age-related changes differ among the sexes. For example, total cholesterol levels tend to peak around age 50 in men (Jousilahti et al., 1996;Kronmal et al., 1993) and at menopause in women (Bonithon-Kopp et al., 1989;Hjortland et al., 1976;Matthews et al., 1989). Why lipid levels change over time is likely due to a number of factors, but whether or not the genetic effects of variants important in lipid metabolism vary with age has yet to be thoroughly explored.

While the interest in and importance of identifying modifiers of genetic effects, such as age, is recognized, so too are the challenges in designing a study to identify such modifiers. The Population Architecture using Genomics and Epidemiology (PAGE) study, a collaboration of investigators with access to large population-based cohorts and cross-sectional epidemiologic studies, began in 2008 with the aim of characterizing GWAS-identified variants, and their modifiers, for a number of different traits (Matise et al., 2011). In one report, we replicated and characterized 49 GWAS-identified lipid associations in the diverse population-based samples available within the PAGE study (Dumitrescu et al., 2011). In this report we extend our initial post-GWAS work to investigate if age modifies the association of these GWAS-identified variants for HDL-C, LDL-C, and TG within ~22,000 European-descent participants.

Methods

Study Populations and Phenotypes

Briefly, PAGE study samples were drawn from four large population-based studies or consortia: EAGLE (Epidemiologic Architecture for Genes Linked to Environment), based on three National Health and Nutrition Examination Surveys (NHANES) (Centers for Disease Control and Prevention, 2004;Centers for Disease Control and Prevention, 2010;Centers for Disease Control and Prevention (CDC), 2002), the Multiethnic Cohort (MEC) (Kolonel et al., 2004), the Women’s Health Initiative (WHI) (The Woman’s Health Initiative Study Group,1998; Anderson et al., 2003), and Causal Variants Across the Life Course (CALiCo), a consortium of several cohort studies: Atherosclerosis Risk in Communities Study (ARIC) (The ARIC Investigators, 1989), Coronary Artery Risk Development in Young Adults (CARDIA) (Friedman et al., 1988), and Cardiovascular Health Study (CHS) (Fried et al., 1991). All selected individuals are of European-descent and all studies were approved by Institutional Review Boards at their respective sites.

While the overall PAGE study consists of individuals representing most decades of the human age spectrum, it is important to note that each individual PAGE study site collected samples at different stages in the life course. Several PAGE study sites represent older cohorts (CHS, MEC, and WHI) and younger to middle aged cohorts (ARIC and CARDIA). Only one study, EAGLE accessing NHANES, collected data for all age ranges (from 12 years of age to the elderly). For studies with repeated measures, baseline age and lipid levels were used in this analysis.

Serum HDL-C, triglycerides, and total cholesterol were measured using standard enzymatic methods. LDL-C was calculated using the Friedewald equation (Arking et al., 2006), with missing values assigned for samples with triglyceride levels greater than 400 mg/dl. For PAGE study sites with longitudinal data, the baseline measurement was used for analysis.

SNP Selection and Genotyping

SNP selection, genotyping, and quality control have been previously described (Dumitrescu et al., 2011). Briefly, each PAGE study site performed genotyping independently using commercially available genotyping arrays (Affymetrix 6.0, Illumina 370CNV BeadChip), custom mid- and low-throughput assays (TaqMan, Sequenom, Illumina GoldenGate or BeadXpress), or a combination thereof (Affymetrix, Santa Clara, CA; Illumina, San Diego, CA, USA; TaqMan, Applied Biosystems, Foster City, CA; Sequenom, San Diego, CA). Quality control was implemented at each study site using common criteria. A total of 49 SNPs reported here were genotyped in two or more PAGE study sites and were previously associated with HDL-C, LDL-C, and/or triglycerides in published (as of 2008) candidate gene and genome-wide association studies. Characteristics for all 49 SNPs are listed in Table S1.

Statistical methods

All tests of association were performed by each PAGE study site using a pre-specified analysis protocol prior to meta-analysis. The study protocol was limited to fasting participants of European-descent regardless of lipid lowering medication usage given that previous analyses demonstrated that lipid lowering medication usage did not appreciably alter single-SNP association results, as treated individuals represented only a small subset of the overall study (Dumitrescu et al., 2011). Participants with >1,000 mg/dl were excluded from the TG analyses, as very high TG levels may be due to rare mutations, disease or illness, or simple human error when recording the data. Triglyceride levels were natural-log transformed.

Linear regression was performed with HDL-C, LDL-C, or ln(TG) as the dependent variable and a SNP as the independent variable, assuming an additive genetic model. Sex effects on lipid trait distributions are well-known (Reilly et al., 1990;Third Report of the National Cholesterol Education Program (NCEP) Expert Panel, 2002), and one PAGE study site (WHI) is composed of only women. Therefore, analyses were stratified by sex unless otherwise indicated. Selected PAGE study sites also included study site or site of ascertainment as a covariate in all models. Age distributions varied across the different PAGE studies (Tables S1-S3), with little age overlap across some studies. Therefore, to gain the largest sample size possible for each analysis, individuals were stratified into two age groups. Males were divided into two categories based on age (<50 years or ≥50 years of age), while females were divided into two categories based on menopausal status (pre-menopausal and post-menopausal). Participants <18 years of age and >70 years of age were excluded.

Meta-analysis, using a fixed-effects inverse-variance weighted approach, was first performed across PAGE studies within each age group using Meta Analysis Helper (METAL) (Willer et al., 2010). Then, for each variant, a 1-degree of freedom χ²test for heterogeneity was performed between the two age groups. The quantity I², a measure of the degree of inconsistency between age groups, was calculated using the equation I²=100% × (Q-degrees of freedom)/Q (Higgins et al., 2003). Genetic associations were considered modified by age if the test of heterogeneity between age strata was significant at P_het<0.05, uncorrected for multiple testing. Aggregate data from the meta-analysis as well as individual tests of association from each PAGE study site will be made available via dbGaP (Mailman et al., 2007;Matise et al., 2011).

Results

Table 1 summarizes the characteristics of the PAGE samples studied here, including the total number of individuals, mean age, and mean lipid levels for each age group in both males and females. Out of a total of 8,550 male participants and 13,772 female participants, the majority fell into the older age group (68% of males were 50+ years of age and 76% of females were post-menopausal), reflecting the study recruitment of PAGE studies. The older age group was nearly 30 years older on average than the younger age group for both sexes. For all three lipid traits and both sexes, lipid levels were slightly higher in the older age group than in the younger age group (Table 1). Mean lipid levels by each PAGE study site are shown in Tables S2-S4.

Table 1. PAGE Study Characteristics.

All values reported as mean (SD) unless otherwise indicated. All individuals are of European-descent.

Trait	Males		Females
	Age<50 yrs	Age≥50 yrs	Pre- menopausal	Post- menopausal
N	2727	5823	3335	10437
Age (yrs)	35.7 (10.3)	63.2 (9.0)	38.3 (11.8)	66.0 (8.6)
HDL-C (mg/dl)	44.3 (11.5)	46.8 (14.3)	56.1 (14.8)	57.3 (16.4)
LDL-C (mg/dl)	122.4 (35.3)	135.4 (35.6)	114.1 (33.9)	134.3 (36.7)
Triglycerides (mg/dl)	117.9 (96.1)	144.0 (83.5)	91.5 (58.1)	149.6 (81.2)

For each variant, a test of heterogeneity between age groups was used to identify associations where age may modify the effect of the SNP on lipid levels. In meta-analyses stratified by sex and menopausal status, seven tests of heterogeneity were significant in females at P_het<0.05: two for HDL-C, two for LDL-C, and three for (ln)TG (Table 2). The level of heterogeneity for these tests was high, ranging from 74.3-87.9%. The most significant test of heterogeneity in females was rs9989419 at CETP for LDL-C (P_het = 3.98E-03; I² = 87.9). This association was significant in post-menopausal women but not in pre-menopausal women and the genetic effect size (β) was in opposite directions among the two age groups (P=0.32, β=−0.84 and P=1.25E-03, β=2.36 in pre- and post-menopausal women, respectively). Indeed, a majority (4/7) of the significant tests of heterogeneity involved a SNP-lipid trait association that was significant in one age group (at P<0.05) but not the other. However, not all of the significant tests of heterogeneity were as easily interpretable: more than half (4/7) of the genetic effect sizes in pre- versus post-menopausal females trended in the same direction, two involved a SNP-lipid trait association that was significant in both age groups, and one SNP-lipid trait association was not significant in either strata. Interestingly, for SNPs with associations stronger among pre-menopausal women compared with post-menopausal women, such as rs3764261 and rs4803750 (Table 2), these data are consistent with the hypothesis that genetics plays a greater role early in the aging process compared with the environment.

Table 2. Lipid SNPs with evidence of menopausal status as a modifier in females of European-descent.

Single-SNP tests of association were performed within each age group and in the combined dataset. Beta and standard error (SE), p-value (P), and sample size (N) are given for each association. The beta estimate is per additional copy of the coded allele (CA). Tests of heterogeneity were performed between the two age strata and the resulting p-value (P_het) and I² index are also reported. Only results where P_het<0.05 are reported. Coded allele frequency (CAF).

SNP	Gene	Lipid Trait	CA (CAF)	Pre-Menopausal			Post-Menopausal			All Females			Phet	I2
				Beta (SE)	P	N	Beta (SE)	P	N	Beta (SE)	P	N
rs2197089	LPL	HDL	T (0.54)	0.29 (0.36)	0.41	3291	1.20 (0.29)	3.17E- 05	8123	0.84 (0.24)	1.73E- 04	11414	0.049	74.3
rs3764261	CETP	HDL	T (0.33)	4.48 (0.37)	1.97E- 33	3201	3.41 (0.38)	1.09E- 19	4677	3.95 (0.26)	1.58E- 50	7878	0.042	75.7
rs4803750	BCL3	LDL	A (0.93)	13.08 (2.52)	2.04E- 07	1473	5.28 (1.65)	1.36E- 03	4536	7.62 (0.38)	3.32E- 08	6009	9.42E-03	85.2
rs9989419	CETP	LDL	A (0.39)	−0.84 (0.83)	0.32	3041	2.36 (0.73)	1.25E- 03	5239	0.97 (0.55)	0.08	8280	3.98E-03	87.9
rs1883025	ABCA1	TG	T (0.26)	−0.05 (0.02)	1.73E- 03	1111	−0.01 (0.02)	0.69	1166	−0.03 (0.01)	0.02	2277	0.038	76.8
rs562338	APOB	TG	T (0.20)	0.02 (0.01)	0.27	3192	−0.03 (0.01)	0.02	7084	−0.01 (0.01)	0.24	10276	0.025	80.0
rs754523	APOB	TG	T (0.69)	0.02 (0.01)	0.13	3176	−0.02 (0.01)	0.07	4777	−0.003 (0.01)	0.67	7953	0.021	81.3

Tests of heterogeneity suggested that eight genetic associations were modified by age (<50 years versus ≥50 years) in males: three for HDL-C, three for LDL-C, and two for (ln)TG (P_het<0.05; Table 3). The level of heterogeneity for these tests was high, ranging from 74.9-89.8%. None of the significant heterogeneity tests identified in males overlapped with those identified for females (Table 3 versus Table 2). One variant, rs6102059 at MAFB, is listed twice in Table 3 as its associations with both HDL-C and ln(TG) displayed evidence of heterogeneity. The most significant test of heterogeneity in males was for LIPC rs174547 and LDL-C (P_het = 1.74E-03; I² = 89.8). As in females, this significant test of heterogeneity appears to be driven by the fact that the SNP-trait association was highly significant in one age stratum but not the other (P=0.99 and P=1.39E-06 in males <50 and ≥50 years old, respectively). Unlike in females, however, this trend was observed for all eight significant tests of heterogeneity.

Table 3. Lipid SNPs with evidence of age as a modifier in males of European-descent.

Single-SNP tests of association were performed within each age group and in the combined dataset. Beta and standard error (SE), p-value (P), and sample size (N) are given for each association. The beta estimate is per additional copy of the coded allele (CA). Tests of heterogeneity were performed between the two age strata and the resulting p-value (P_het) and I² index are also reported. Only results where P_het<0.05 are reported. Coded allele frequency (CAF).

SNP	Gene	Lipid Trait	CA (CAF)	Males <50			Males ≥ 50			All Males			Phet	I2
				Beta (SE)	P	N	Beta (SE)	P	N	Beta (SE)	P	N
rs3135506	APOA1/C3/ A4/A5	HDL	C (0.06)	−0.41 (0.80)	0.61	1677	−2.42 (0.54)	6.26E- 06	5325	−1.80 (0.45)	5.38E- 05	7002	0.037	77.1
rs4775041	LIPC	HDL	C (0.29)	0.61 (0.34)	0.07	2674	1.52 (0.28)	5.47E- 08	5438	1.15 (0.22)	9.48E- 08	8112	0.038	76.7
rs6102059	MAFB	HDL	T (0.30)	1.08 (0.44)	0.01	1695	−0.04 (0.32)	0.91	3614	0.35 (0.26)	0.17	5309	0.041	76.2
rs174547	FADS1	LDL	T (0.67)	0.01 (1.03)	0.99	2391	4.26 (0.88)	1.39E- 06	3452	2.46 (0.67)	2.41E- 04	5843	1.74E-03	89.8
rs2156552	LIPG	LDL	A (0.84)	3.11 (1.25)	0.01	2404	−0.26 (0.91)	0.77	5324	0.91 (0.73)	0.21	7728	0.029	79.0
rs1566439	CETP	LDL	T (0.60)	2.71 (1.37)	0.05	1424	−0.38 (0.72)	0.60	4832	0.30 (0.64)	0.64	6256	0.046	74.9
rs1748195	ANGPTL3	TG	C (0.67)	0.09 (0.02)	1.47E- 05	1552	0.02 (0.01)	0.11	3460	0.04 (0.02)	2.48E- 04	5012	4.89E-03	87.4
rs6102059	MAFB	TG	T (0.30)	−0.05 (0.02)	0.02	1537	0.003 (0.01)	0.82	3517	−0.01 (0.01)	0.31	5054	0.032	78.3

Given that these tests of association stratified by sex and age were greatly impacted by sample size and power, we also performed tests of association stratified by age with sexes combined (Table 4). Pre-menopausal women were analyzed together with men <50 years of age, while post-menopausal women were analyzed together with men >50 years of age. In this sex-combined analysis, eight tests of heterogeneity were significant at P_het<0.05: two for HDL-C, four for LDL-C, and two for (ln)TG. A majority (5/8) of the significant heterogeneity tests identified in all participants overlapped with those identified in females or males only (Table S5). For these overlapping associations, the magnitude of heterogeneity (I²) was, in general, slightly lower in the sex-combined analysis than the sex-stratified analysis. The most significant test of heterogeneity in all participants was rs4803750 at BCL3 for LDL-C (P_het = 5.88E-03; I² = 86.8), which was also significant in the female-only analysis. Associations that showed evidence of modification by age in the sex-combined analysis only included rs1699614 (CILP2/PBX4/NCAN) for HDL-C, rs6586891 (LPL) for HDL-C, and rs11206510 (PCSK9) for LDL-C.

Table 4. Lipid SNPs with evidence of age as a modifier in all individuals of European-descent.

Single-SNP tests of association were performed within each age group and in the combined dataset. Beta and standard error (SE), p-value (P), and sample size (N) are given for each association. The beta estimate is per additional copy of the coded allele (CA). Tests of heterogeneity were performed between the two age strata and the resulting p-value (P_het) and I² index are also reported. Only results where P_het<0.05 are reported. Coded allele frequency (CAF).

SNP	Gene	Lipid Trait	CA	Younger Males & Females			Older Males & Females			All			Phet	I2
				Beta (SE)	P	N	Beta (SE)	P	N	Beta (SE)	P	N
rs1699614	CILP2/PB X4/NCAN	HDL	T (0.12)	0.69 (0.43)	0.11	5965	−0.49 (0.34)	0.16	13509	−0.03 (0.27)	0.92	19474	0.033	77.9
rs6586891	LPL	HDL	A (0.64)	0.57 (0.25)	0.02	5987	1.24 (0.20)	8.06E- 10	13579	0.97 (0.16)	5.31E- 10	19566	0.034	77.7
rs11206510	PCSK9	LDL	T (0.79)	3.27 (0.79 )	3.44E- 05	5487	0.92 (0.61)	0.13	12349	1.81 (0.48)	1.86E- 4	17836	0.019	81.8
rs174547	FADS1	LDL	T (0.67)	1.15 (0.66)	0.08	5442	3.19 (0.69)	3.32E- 06	6338	2.13 (0.48)	7.50E- 06	11780	0.032	78.3
rs4803750	BCL3	LDL	A (0.93)	11.10 (2.03)	4.38E- 08	2393	4.74 (1.10)	1.66E- 05	9023	6.19 (0.97)	1.58E- 10	11416	5.88E-03	86.8
rs9989419	CETP	LDL	A (0.40)	−0.29 (0.63)	0.64	5408	1.40 (0.51)	5.83E- 03	10433	0.74 (0.40)	0.06	15841	0.036	77.1
rs1748195	ANGPTL3	TG	C (0.67)	0.06 (0.01)	3.85E- 06	3597	0.02 (0.01)	0.06	7926	0.03 (0.01)	1.72E- 05	11523	0.011	84.5
rs1883025	ABCA1	TG	T (0.26)	−0.04 (0.01)	8.21E- 03	2065	−0.003 (0.01)	0.81	2669	−0.01 (0.01)	0.07	4734	0.049	74.1

Discussion

Epidemiologic studies have long documented that lipid trait distributions are shaped by sex, age, and environmental exposures such as smoking or diet. As lipid levels tend to increase with age, we explored the potential modifying effect of age on previously identified lipid-SNP associations by testing for heterogeneity between single-SNP associations stratified by age or menopausal status. At a nominal P_het<0.05, seven associations in females, eight associations in males, and eight associations in the sex-combined dataset displayed significant heterogeneity by age. In total, after accounting for overlap of results among the sex-stratified and sex-combined analyses, 18 distinct associations (comprised of 17 SNPs and all three lipid traits) showed suggestive modifying effects of age.

It should be noted, however, that none of the p-values presented here were adjusted for multiple testing. Indeed, if we were to use a Bonferroni adjustment for 441 tests of heterogeneity (=49 SNPs*3 lipid traits*3 sex-defined datasets), none of our tests would remain significant at p<1.34E-04 (=0.05/441). Of course, these associations are highly correlated and correction for 441 tests is overly conservative. To further complicate interpretation, tests of heterogeneity are generally known to be underpowered when only a small number of studies is included in the meta-analysis (Hardy & Thompson, 1998). Due to potentially low power, some investigators have recommend using a non-Bonferroni-adjusted significance threshold of P_het<0.10, rather than the standard P_het<0.05 (Higgins et al., 2003). Given this limitation, our results may err on being overly conservative. Furthermore, the high I² values observed here provide additional evidence of heterogeneity of genetic effects between age groups, with up to nearly 90% of the total variability in beta estimates attributed to heterogeneity. Therefore, while replication remains the gold standard, we present our most significant findings in the interest of hypothesis generation.

As potential environmental modifiers are often ignored in genetic association studies, it has been argued that accounting for these modifiers may help with replication across studies and with generalization across racial/ethnic groups. Using the association between rs1883025 at ABCA1 and triglycerides in the sex-combined analysis (Table 4) as an example, when all ages are included the resulting p-value is suggestive of association (p=0.07) and would not meet most studies’ statistical significance threshold. However, when this association is stratified into younger and older age groups, the SNP effect is clearly present in the younger cohort (β=−0.04 and p=8.21E-03) but not the older cohort (β=−0.003 and p=0.81), even though the sample sizes are similar.

To our knowledge, only a few studies have explored whether age modifies lipid-SNP associations. For example, Shirts et al (Shirts et al., 2011) tested 17 previously identified lipid-SNP associations for interactions with age. No evidence for modification was found for the majority of associations; however, a single significant gene-age interaction (rs646776 at SORT1 and LDL-C) was identified and replicated (Shirts et al., 2011). We found no evidence of age-related modification of this association in our data. Neither did our results support those in Webster et al (Webster et al., 2009), which showed some evidence of an age effect for rs328 and HDL-C and rs3135506 and triglycerides. Again, we found no evidence that these SNPs are involved in age-related changes in lipid levels. It is important to note that, in general, these studies found little evidence that the effects of previously established SNP-lipid associations vary significantly over time, a conclusion concordant with our results.

Our study faced many of the challenges inherent to studies of gene-environment or gene-gene interactions (Hunter, 2005;Ober & Vercelli, 2011). As mentioned earlier, correction for multiple testing can be a major burden that may only be overcome by large effect sizes or large population sizes. PAGE is a large and diverse collaboration. However, our study was limited to individuals of European-descent given that, following stratification by sex and age group, our power would be greatly limited in other racial/ethnic groups with smaller sample sizes. Therefore, the results presented here are not necessarily generalizable to populations of non-European descent.

Another challenge commonly faced by genetic studies of environmental modifiers is interpretation of results. Even when a statistically significant modifier effect is identified, it is often difficult to dissect the biological etiology of the effect. This study was no exception. While age is an easy variable to collect and harmonize across studies, it is most likely a surrogate for numerous accumulating physiological changes. Indeed, there are many possibilities as to why we observe age-related heterogeneity, including survival bias, discrepant medication use between age groups (ex: hormone replacement therapy in post-menopausal women), and genetic differences between younger and older populations not examined here. Also, it is difficult to distinguish between the effects of age and sex, given that both are surrogates for age- and sex-dependent exposures. In the present analysis, we did not explicitly test for sex as a modifier of lipid trait profiles given its complex relationship with age. Despite our attempts to account for both variables in stratified analyses, the exact mechanism by which age modifies genetic effects on lipid levels is difficult to elucidate without further experimentation.

The lipid levels tested in this study were based on only one time point for each individual. Therefore, we were unable to compare inter-individual change in lipid levels and their associations with genetic variants over time. While longitudinal studies are ideal for studying any environmental modifier, they are more expensive, tend to be smaller in sample size, and require more time to collect exposure data compared to cross-sectional studies. Therefore, cross-sectional and case-control studies may be used as a first pass for identifying potential modifiers. Additionally, we must note that our analysis was limited to 49 established loci drawn from GWAS studies published in or before 2008. Therefore, as additional GWAS have been performed, many additional lipid-associated loci have been identified, which we were not able to investigate here.

The results of this study have an important implication for genetic association studies of lipid traits. While it is common practice to simply adjust for age as a confounder, this assumes that the relationship between age and lipid levels is homogenous across the different genotypes of the variant under study. Our results suggest that this assumption does not hold true for some variants. Therefore, caution must be exercised when combining or comparing genetic associations across studies with different age distributions.

In summary, while our results suggest that effects of established gene variants on lipid traits may vary with age, further studies are required to assess the validity of these findings. Due to the difficulties inherent in studying genetic modifiers, large collaborative efforts with longitudinal measures are necessary both to test for the existence of gene-environment interactions and to begin to understand the mechanisms behind them.