Vitamin D receptor and megalin gene polymorphisms are associated with central adiposity status and changes among US adults

We examined longitudinal associations of vitamin D receptor (VDR) and megalin (LRP2; LDL receptor-related protein-2) gene polymorphisms with central adiposity. We used data from the Baltimore Longitudinal Study of Aging (BLSA), an ongoing prospective open cohort study. Study participants consisted of non-Hispanic white adults residing in Baltimore city, with one or more visits at age ≥50 years, and complete data (n 609–617). Repeated assessments on waist circumference (WC) and waist:hip ratio (WHR) were available. Multiple linear mixed models were used to estimate mid-follow-up age central adiposity level and annual rate of change with cut-points set at the sex-specific 80th percentile. The four binary outcomes were: ‘elevated central adiposity’ (ECA-WC and ECA-WHR) and ‘significant increase in central adiposity’ (SICA-WC and SICA-WHR). SNP for VDR (four SNP: (1) rs11568820 (CdX-2:T/C); (2) rs1544410 (BsmI:G/A); (3) rs7975232 (ApaI:A/C); (4) rs731236 (TaqI:G/A)) and Megalin (three SNP: (1) rs3755166:G/A; (2) rs2075252:C/T; (3) rs4668123:C/T) genes were selected. SNP latent classes (SNPLC) and SNP haplotypes (SNPHAP) were created. Multiple logistic regression analyses indicated that, in men, higher ECA-WHR odds were associated with SNPLC Megalin2:rs3755166[–]/rs2075252[TT]/rs4668123[T–] (v. Megalin1:rs3755166[–]/rs2075252[CC]/rs4668123[–]) (OR 2·87; 95 % CI 1·15, 7·12; P = 0·023) and that SNPLC Megalin3:rs3755166[–]/rs2075252[CT]/rs4668123[–] (v. Megalin1) was linked to lower SICA-WC odds (OR 0·48; 95 % CI 0·26, 0·88; P = 0·019) (P > 0·05 for sex × SNPLC). In women, VDR3 SNPHAP (GAA:bAT) was related to lower odds of ECA-WC (OR 0·37; 95 % CI 0·16, 0·87; P = 0·023) (P < 0·05 for sex × SNPHAP), VDR1 SNPHAP (GCA:baT) was associated with greater odds and VDR3 SNPHAP (GAA:bAT) with lower odds of SICA-WC (P > 0·05 for sex × SNPHAP). Vitamin D-related gene polymorphisms were associated with central adiposity status and change. Future mechanistic studies are needed to confirm those polymorphisms' biological significance to central adiposity.

Human adiposity is heritable and polygenic (1) , with genes contributing 16-85 % for BMI (2) and 37-81 % for waist circumference (WC) (for example, Hunt et al. (3) ). Gene-environment interactions may largely determine adiposity phenotypes. Moreover, serum 25-hydroxyvitamin D (25(OH)D) concentration correlated inversely with adiposity and related metabolic disorders (4) . The lower bioavailability of the fat-soluble vitamin D through its sequestration into excessive fat tissues was a suggested mechanism (5) , implicating obesity in the aetiology of vitamin D deficiency. Conversely, vitamin D may play a causal role in obesity by modulating homeostasis of intracellular Ca which correlates inversely with dairy product consumption. Ultimately, a higher intracellular Ca triggers lipogenesis and suppresses lipolysis (6) .
Another endocytic vitamin D-binding receptor, known as megalin (or LDL receptor-related protein-2; LRP2), is expressed in many epithelial cells, belongs to the LDL receptor family and its expression is directly regulated by both vitamin D and vitamin A (18) . Vitamin D enters cells via megalin receptor bound to vitamin D-binding protein (19) . Megalin influences obesity possibly by mediating the transport of the appetite-regulating adipokine leptin through the blood-brain barrier and modulates leptin signaling (20) . Leptin, in turn, was linked to vitamin D metabolism by attenuating gene expression of renal enzyme 25-hydroxyvitamin D 3 -1α-hydroxylase in mice (21) . Megalin also facilitates transcytosis of precursor hormone thyroglobulin (22) . Leptin and thyroid hormone collectively affect adiposity through their regulation of energy metabolism, thermogenesis, glucose and lipid metabolism, appetite and food intake, and the oxidation of fatty acids (23) . Megalin is also a receptor for sex hormone-binding globulin. In fact, a cross-effect modification of oestrogen and vitamin D interventions was observed for colorectal cancer incidence in the Women's Health Initiative trial (24) , suggesting an interplay of oestrogen and vitamin D via megalin and a possible differential effect of megalin polymorphisms between sexes. To our knowledge, aside from recent genome-wide association studies (for example, Heid et al. (25) ), no study using a candidate gene approach has thus far examined megalin gene polymorphism in relation to adiposity phenotypes, particularly longitudinal changes in central adiposity, though these variations were tested for outcomes such as cognition and dementia (26) , a phenotype shown to be associated with obesity in a recent meta-analysis (27) .
In our present study, we hypothesise that selected polymorphisms in VDR and megalin genes previously shown to affect various metabolic and cardiovascular health outcomes, mainly in cross-sectional studies, are also associated with central adiposity status and longitudinal changes in a sample of non-Hispanic white US adults.

Database and study subjects
Data from the Baltimore Longitudinal Study of Aging (BLSA) were used, with methods summarised elsewhere (28) . Eligible participants for our present study had at least one visit at or beyond age 50 years (n 2 2321 of 3005), and were restricted further to non-Hispanic whites (n 3 1917), given possible differential associations of vitamin D status with adiposity in different ethnic groups. Complete genetic data among those non-Hispanic white participants eligible for analysis were available for n 4 702 BLSA participants, of whom n 4 609-617 had complete central adiposity and covariate measurements. The present study was conducted according to the guidelines laid down in the Declaration of Helsinki and all procedures involving human subjects were approved by the the Institutional Review Board (IRB) of Medstar Health Research Institute. Written informed consent was obtained from all subjects. In addition, genetic and other variables were de-identified for the purpose of statistical analysis.

Data collection and key measurements
Classification of elevated central adiposity and significant increase in central adiposity. BLSA staff clinicians assessed WC with a tape measure kept parallel to the floor, from the hipbone and wrapping around the waist at navel level while participants were holding their breath. Hip circumference was similarly measured and waist:hip ratio (WHR) was computed accordingly. Multiple assessments were available (WC, n 14 852 visits, n 2886 subjects; WHR, n 14 832 visits, n 2886 subjects) and mean and range of individual assessments (or visits) were: WC, mean 5·1 (range 1-25); WHR, mean 5·1 (range 1-25, with more than 90 % of participants having at least two visits. We conducted linear mixed models to predict individual WC and WHR at mean follow-up age and estimate annual rate of change between age 50 years and mean follow-up age (see online Supplementary Material S2), an approach previously used to predict cognitive performance and annual rate of change (26) . Using sex-specific quintiles, binary outcomes 'elevated central adiposity' (ECA) and 'significant increase in central adiposity' (SICA) were defined as the uppermost quintile (value = 1) for central adiposity level and annual rate of change, respectively, and compared with all other quintiles combined (value = 0). ECA and SICA were defined for WC and WHR, and thus four binary outcomes were obtained (ECA-WC, ECA-WHR, SICA-WC and SICA-WHR). The choice of the binary outcome (as opposed to a continuous one) was driven by the potential clinical significance of the effects as well as the ease of interpretation and replication in future studies that would use similar cut-points in independent samples.
Using latent class analysis (LCA) with sex and first-visit age as covariates and selected SNP entered into that model (one gene per model) as a three-level categorical variable, VDR and megalin SNP latent classes (SNPLC) were obtained (PROC LCA in SAS version 9.1; SAS Institute Inc.) (30) . Akaike information and Bayesian information criteria for model fit determined the appropriate number of latent classes. Using the Bayes theorem, posterior probabilities were estimated and were identical for all individuals with a particular SNP pattern per gene. Each individual belonged to a specific SNPLC when the posterior probability for this class was >0·50. For most individuals, the expected posterior probability is >0·90 for a specific latent class (30) (online Supplementary material S1, Fig. S1(a) and Fig. S1(b)).
Additionally, using Haploview version 4.2 (31) , SNP haplotypes (SNPHAP) per gene were also created. For the VDR gene, three SNP (BsmI, ApaI and TaqI) with moderately strong linkage disequilibrium (LD) were combined into SNPHAP, as was done in previous studies (for example, Beydoun et al. (26) ). Consequently, three SNPHAP were prevalent in our population, combining BsmI, ApaI and TaqI as follows: VDR 1 , GCA (baT), VDR 2 , AAG (BAt) or VDR 3 , GAA (bAT) for one or two alleles. We coded participants as: 0 = having no VDR x haplotype; 1 = having one allele carrying the VDR x haplotype; 2 = having two alleles carrying the VDR x haplotype. Using a similar approach, eight megalin haplotypes were uncovered. However, only three megalin SNPHAP were selected for our analysis (prevalence with one or two copies was >10 %). Both SNPLC and SNPHAP have been created and used in a similar fashion in a previous study (26) .
Other main covariates. To adjust for potential confounding in the main associations of interest, three sets of covariates were considered: (1) sociodemographic factors, namely first-visit age and mean follow-up ages (per individual and outcome), sex, educational attainment (years of schooling), and one lifestyle-related factor, namely smoking status (never, former or current smoker); (2) self-reported history of type 2 diabetes, hypertension, CVD (stroke, congestive heart failure, non-fatal myocardial infarction or atrial fibrillation) and dyslipidaemia at first visit. Only covariates (1) were considered as potential confounders in multiple regression models, whereas (2) were only used for descriptive purposes. In addition, EIGENSTRAT analysis (implemented as part of the EIGENSOFT package) was conducted and two top principal components were added in multiple regression models to control for any residual effects of population structure as described in a previous study (26) . The EIGENSTRAT method is a program that conducts principal components analysis to correct for stratification in genome-wide association studies.

Statistical analysis
For each selected SNP, we assessed Hardy-Weinberg equilibrium with an exact test, using Haploview version 4.2 (31) , and calculated pair-wise LD. Online Supplementary Material S3 shows the LD map for all available VDR and megalin SNP. We present means and proportions of study sample characteristics, gene SNP, SNPLC and SNPHAP distributions.
Further, we conducted multiple logistic regression analyses to examine associations of VDR and megalin SNP, SNPLC and SNPHAP with four binary central adiposity: A two-stage Heckman selection model was constructed to account for selection bias due to non-random participant selection for genetic analysis, as was done in several other studies (for example, Beydoun et al. (26) ). At a first stage, a probit model produced an inverse Mills ratio (IMR), directly derived from the predicted probability of being selected, conditional on model covariates. To adjust for this selection bias, the IMR entered the main multiple logistic regression models as a covariate in the second stage. Stratified analyses by sex were conducted. Effect modification by sex was tested by adding interaction terms of sex with SNP, SNPLC and SNPHAP. In particular, sex differences in the association between megalin gene polymorphism and various phenotypes including adiposity were hypothesised a priori (24) .
With the null hypothesis being no association between SNP (or SNPHAP, SNPLC) and the four main outcomes of interest ((1) ECA-WC; (2) ECA-WHR; (3) SICA-WC; (4) SICA-WHR), type I errors were generally set at 0·05, with main effect P values between 0·05 and 0·10 labelled as marginally significant, whereas a P value below 0·10 was considered significant for interaction terms, as was done in other studies (for example, Beydoun et al. (26) ) before correction for multiple testing. Multiple testing correction was done using a familywise Bonferroni procedure, whereby a family was defined by adiposity outcome type (i.e. status (ECA) v. change (SICA)) (32) . Within each outcome, two alternate measures were used, namely WC and WHR. The corrected statistical significance criterion for main effect P values was reduced to P = 0·05/2 = 0·025 (marginal significance: P = 0·10/2 = 0·050). Because of their lower statistical power compared with main effects, interaction terms' critical P values were reduced to 0·05 (33) . All analyses (except for LCA) were performed using Stata version 11.0 (StataCorp LP) (34) .

Study sample characteristics and gene SNP distribution
Study sample characteristics are summarised in Table 1. All examined SNP were in Hardy-Weinberg equilibrium (P > 0·05).
Vitamin D receptor and Megalin SNP latent classes' associations with central adiposity Using LCA, three SNPLC per gene were created. One key finding emerged for SNPLC related to central adiposity in the total population ( Fig. 1(a) and Fig. 1(b)). Comparing each minor SNPLC with the most dominant one, we found that Megalin 2 v. Megalin 1 was associated with significantly increased odds of ECA-WHR in the total population (OR 2·34; 95 % CI 1·18, 4·64; P = 0·015), which remained significant after correction for multiple testing ( Fig. 1(b)).

Vitamin D receptor and Megalin SNP halotypes' associations with central adiposity
Using haplotype analysis, three SNPHAP per gene were created. Each of those six SNPHAP was entered separately in the main multiple logistic regression models as an ordinal variable. In both sexes combined, VDR 1 SNPHAP was associated with an increased odds of SICA-WHR (OR 1·31; 95 % CI 1·01, 1·69; P = 0·038), though significance was not retained after correction for multiple testing (Fig. 2(a) and Fig. 2(b)).
Vitamin D receptor and Megalin SNP halotypes' associations with central adiposity: sex-stratified findings Our sex-stratified analysis with SNPHAP (Table 4) uncovered many important findings, particularly for VDR SNPHAP. Among those key findings, VDR 3 SNPHAP was associated ECA, elevated central adiposity; WC, waist circumference; WHR, waist:hip ratio; SICA, significant increase in central adiposity. * Reported any of the following conditions at first visit: stroke, congestive heart failure, non-fatal myocardial infarction, or atrial fibrillation. † WC and WHR were predicted at mean age at follow-up using a multivariate linear mixed model controlling for sex, race/ethnicity, education (years), and smoking status, with age added among the fixed-effect variables to allow for quadratic non-linear change. The slope or annual rate of change was predicted from these models at the mean age at follow-up (i.e. between age 50 years and individual mean age of follow-up for each central adiposity measure) (see online Supplementary Material S2 for more details).

Discussion
The present study examined longitudinal associations of VDR and megalin gene polymorphisms with central adiposity, using extensive data from the BLSA, an ongoing prospective open cohort study. Study participants consisted of non-Hispanic white adults residing in Baltimore city, with one or more visits at age ≥50 years, and complete data (n 609-617). Available repeated assessments on WC and WHR were used to form four binary outcomes, which were defined by multiple linear mixed models, mid-follow-up age estimators for central adiposity level and annual rate of change with cut-points set at  (v. Megalin 1 ) was linked to lower SICA-WC odds (OR 0·48; 95 % CI 0·26, 0·88; P = 0·019) (P > 0·05 for sex × SNPLC). In women, VDR 3 SNPHAP (GAA:bAT) was related to lower odds of ECA-WC (OR 0·37; 95 % CI 0·16, 0·87; P = 0·023) (P < 0·05 for sex × SNPHAP), VDR 1 SNPHAP (GCA:baT) was associated with greater odds and VDR 3 SNPHAP (GAA:bAT) with lower odds of SICA-WC (P > 0·05 for sex × SNPHAP).
Several recent cross-sectional and case-control studies have examined VDR genetic polymorphisms as potential risk markers for central adiposity and related metabolic disorders (9)(10)(11)16,17,36,37) . When testing VDR SNP associations with adiposity, a recent cross-sectional study (176 randomly selected men aged 25-65 years) found that homozygous BsmI (AA v. GG) was associated with higher BMI (29·0 v. 26·8 kg/m 2 ; P = 0·024) and higher WC (101·8 v. 96·2 cm; P = 0·014) (11) . A similar finding was observed in another cross-sectional study of 175 women with body weight and fat mass as two outcomes and the VDR SNP BsmI being of interest (10) . In a more recent study with a larger sample size (n 1773) of women, the association of fourteen VDR SNP with three adiposity measures was examined, including WC. Results suggested that the homozygous rare variant of rs3782905 found in the 3′ VDR region (LD between rs3782905 and BsmI in Caucasian HapMap is about 0·42) was associated on average with 4·4 cm larger WC compared with the homozygous common variant (Bonferroni-adjusted P = 0·004) (9) . These consistent findings for BsmI and obesity with the risk increasing allele being 'A' were confirmed when the phenotypes of interest were type 2 diabetes, fasting glucose level and CHD risk in recent studies (13,14) . In contrast, among 351 postmenopausal healthy women, VDR BsmI polymorphism ('A' risk allele) was not associated with obesity or insulin resistance but was connected with an unfavourable lipid profile (36) . Additionally, a case-control study with 309 unrelated French subjects with type 2 diabetes, and among those with early onset in particular, the TaqI SNP ('A' allele) was associated with a higher BMI and an increased prevalence of obesity, compared with the controls (16) . The Cdx-2 SNP was related only to BMI, fat mass and percentage fat mass in one study of 1215 subjects from 400 Chinese nuclear families (17) . In Table 3. Vitamin D receptor (VDR) and Megalin gene SNP latent class (SNPLC) associations with predicted central adiposity outcomes, stratified by sex: multiple logistic regression analysis (Baltimore Longitudinal Study of Aging) ‡ (Odds ratios and 95 % confidence intervals) ECA, elevated central adiposity; WC, waist circumference; WHR, waist:hip ratio; SICA, significant increase in central adiposity.
Significance for null hypothesis that Log e (OR) = 0: * P < 0·05, ** P < 0·025. † P < 0·10 for null hypothesis that Log e (OR) = 0. ‡ WC and WHR (ECA and SICA) were predicted using a linear mixed model controlling for sex, race/ethnicity, education (years), and smoking status, with age added among the fixed-effect variables to allow for quadratic non-linear change (see online Supplementary Material S2 for more details). § Based on multiple logistic regression models with outcome being ECA or SICA for WC or WHR and main exposures being VDR and megalin SNPLC entered simultaneously into the model for each outcome, stratifying by sex. The model controlled for first-visit age, mean age at follow-up, education, first-visit smoking status, first-visit self-reported type 2 diabetes, hypertension, CVD and the two principal component analysis factor scores.
the present study, only the ApaI SNP ('C' allele dosage) appeared to significantly increase the odds of SICA-WHR (P for trend = 0·024). Few previous studies have examined VDR SNPHAP (in addition to SNP) as predictors of adiposity, and none found significant associations (for example, Gu et al. (17) ). However, other studies examined VDR SNPHAP in relation to diabetes (types 1 and 2), insulin resistance and cardiovascular outcomes. For instance, in a population-based study of men and women aged 55-80 years, each copy of the baT haplotype was associated with a 20 % increased likelihood of electrocardiogram-confirmed myocardial infarction, after adjustment for established CVD risk factors (38) . The latter finding suggests that baT, which we found to increase the risk of longitudinal increase in WC among women, may also be a risk factor for cardiovascular events.
Although no prior research had tested the association between VDR SNPLC and megalin (SNP, SNPHAP or SNPLC) and adiposity, we found that they might be important risk factors for central adiposity (ECA and SICA), in some cases only in one sex. Thus, our findings need to be replicated in larger samples of adult men and women with repeated measures on WC and WHR before efforts to uncover potential reasons for sex differences can be made.
Although the exact mechanism is unknown, evidence supports the role of vitamin D on adiposity. In vitro studies have shown that vitamin D stabilises VDR and suppresses adipocyte differentiation through C/EBPα (cytidine-cytidineadenosine-adenosine-thymidine (CCAAT)/enhancer binding protein α) inhibition and PPARγ expression and activity (39) . VDR overexpression inhibited adipocyte differentiation independently of vitamin D, suggesting that the VDR plays a crucial role in adipocyte maturation. If the polymorphisms in VDR and megalin are functional or are tagging SNP that alter the availability or activity of vitamin D, it is conceivable that these SNP influence adiposity traits through the regulation of adipocyte differentiation. The Cdx-2 polymorphism in VDR has been shown to have an effect on VDR activity while the TaqI-ApaI-BsmI SNP in the 3′ untranslated region of the gene is thought to be a tag SNP (40) . In the present study, the 3′ region SNP were more strongly associated with longitudinal adiposity trajectory; therefore fine mapping of this region to identify the functional SNP may provide insight into the mechanism by which VDR SNP regulate adiposity traits. The two non-synonymous SNP in megalin (rs2075252, rs3668123) were driving the association between megalin and adiposity traits (41) . Whether these coding SNP influence megalin gene function is unknown. However, if  Tables 2 and 3. See Methods section for description of the four outcomes and the SNP haplotypes. Values are odds ratios, with 95 % CI represented by vertical bars. * P < 0·05 for null hypothesis that Log e (OR) = 0. † P < 0·10 for null hypothesis that Log e (OR) = 0.
functional, these SNP could alter vitamin D availability in cells and thereby regulate adipocyte differentiation.
The present study has several strengths. It included a large number of consecutive visits per participant, and made use of advanced statistical techniques by combining linear mixed models with multiple logistic regression analyses (42) to examine associations between gene SNP, SNPLC (defined using LCA) and SNPHAP (defined using haplotype analysis) and four measures of central adiposity status and change.
However, in light of some limitations, findings of the present study must be interpreted with caution. Indeed, the BLSA is an open cohort study of a convenience sample of participants, experiencing continuous recruitment and dropout. Moreover, genetic data were available only for a subset of the initial cohort, yielding a smaller sample size and a younger mean age. To reduce selection biases resulting from this sampling scheme, we used a number of statistical techniques, including a two-stage Heckman selection model. Further, even though observation frequency for central adiposity was high (mean about five visits), the data structure was largely unbalanced given that first-visit age and duration between visits varied across participants. Consequently, we used mixed models to predict the continuous version of the four adiposity outcomes at mean follow-up age. Our main statistical models also controlled for mean follow-up and first-visit age. Moreover, no data were readily available on potential confounders including serum vitamin D, dietary intakes of Ca and vitamin D; physical activity particularly outdoors exercise which is a main determinant of vitamin D status and is also one of the main protective lifestyle factors related to obesity; alcohol or drug use and use of medications. Finally, we cannot rule out chance, residual confounding or selection bias for positive findings, particularly for the sex-specific analyses, and lack of power for negative findings. Thus, until those findings are further replicated in another independent sample, they should be interpreted with caution.
In conclusion, the key findings of the present study point to a relationship between VDR and megalin gene polymorphisms Significance for null hypothesis that Log e (OR) = 0: *P < 0·05, **P < 0·025. † P < 0·10 for null hypothesis that Log e (OR) = 0. ‡P < 0·05 for null hypothesis that sex × SNPHAP interaction term = 0 in a model where main effect of sex was added. § WC and WHR were predicted at mean age at follow-up using a multivariate linear mixed model controlling for sex, race/ethnicity, education (years), and smoking status, with age added among the fixed-effect variables to allow for quadratic non-linear change. The slope or annual rate of change was predicted from these models at the mean age at follow-up (i.e. between age 50 years and individual mean age of follow-up for each central adiposity measure) (see online Supplementary Material S2 for more details). ║ Based on multiple logistic regression models with outcome being ECA or SICA for WC or WHR and main exposures being VDR and megalin SNPHAP entered simultaneously into the model for each outcome, stratifying by sex. The model controlled for first-visit age, mean age at follow-up, education, first-visit smoking status, first-visit self-reported type 2 diabetes, hypertension, CVD and the two principal component analysis factor scores. and central adiposity. To our knowledge this is the first study to examine longitudinal change in central adiposity in relation to polymorphisms in those two genes. However, further research is needed to replicate those findings in different populations, including populations of other racial and ethnic groups, in order to confirm the biological significance of those polymorphisms in relation to central adiposity phenotypes.