Oxidative stress, an imbalance between the production of reactive oxygen species and the ability of the cell to scavenge those species with various antioxidants, has been implicated in the pathogenesis of many chronic diseases, including type 2 diabetes mellitus, CVD, rheumatological disorders and carcinogenesis( Reference Soory 1 ). Potential beneficial effects have recently been ascribed to naturally occurring phytochemicals known as carotenoids which may reduce oxidative stress triggered by injury that characterises the pathogenesis of those chronic diseases( Reference Soory 1 ). Although the primary dietary sources of carotenoids are fruits and vegetables, they are also found in bread, eggs, beverages (e.g. carrot and tomato juices), fats and oils( Reference Rao and Rao 2 ). Among more than forty carotenoids in the human diet, only the following five carotenoids or groups of carotenoids have been shown to be consistently measurable in human serum: α-carotene, β-carotene, β-cryptoxanthin, lycopene and lutein+zeaxanthin (often combined together)( Reference Rao and Rao 2 ).
Some observational studies have shown inverse associations between carotenoids and CVD( Reference Voutilainen, Nurmi and Mursu 3 ), type 2 diabetes( Reference Montonen, Knekt and Jarvinen 4 – Reference Reunanen, Knekt and Aaran 8 ) and the metabolic syndrome (MetS) in recent national surveys( Reference Ford, Mokdad and Giles 9 – Reference Beydoun, Canas and Beydoun 11 ). Moreover, in two recent studies, using the National Health and Nutrition Examination Survey (NHANES) and Invecchiare in Chianti (InCHIANTI) data, serum total carotenoid level has been shown to be consistently inversely related to depressive symptoms( Reference Beydoun, Beydoun and Boueiz 12 , Reference Milaneschi, Bandinelli and Penninx 13 ). However, the findings are inconsistent with those reported by other studies( Reference Kataja-Tuomola, Sundell and Mannisto 14 – Reference Wang, Liu and Manson 17 ). It is also worth noting that obesity and its related disorders have been shown to be associated with an increased level of depressive symptoms in a number of studies (e.g. Beydoun et al. ( Reference Beydoun, Kuczmarski and Mason 18 ), Kimura et al. ( Reference Kimura, Matsushita and Nanri 19 ), Akbaraly et al. ( Reference Akbaraly, Ancelin and Jaussent 20 )), suggesting co-morbidity between those conditions.
Importantly, it is unclear whether the observed inverse relationships between serum carotenoids and the MetS and/or depression are due to variations in carotenoid concentration or determined by other carotenoid-containing food constituents. To identify an unconfounded role of carotenoids in health and disease, surrogate measures such as genetic polymorphisms have been used in recent studies. In fact, genome-wide association studies (GWAS) and candidate gene studies have uncovered genetic polymorphisms in a number of genes that were significantly associated with serum carotenoid status. Genes carrying the specific SNP that have been commonly tested in the literature against serum carotenoid concentrations were either directly (e.g. β,β-carotene 15,15′-mono-oxygenase, BCMO1) or indirectly (e.g. ApoE) related to carotenoid metabolism( Reference Herron, McGrane and Waters 21 – Reference Herbeth, Gueguen and Leroy 30 ). Many of these GWAS and candidate gene studies have been conducted among the individuals of European descent.
Therefore, the overall aim of the present study was to assess whether genetic polymorphisms involved in carotenoid absorption, intracellular trafficking and plasma transport are also related to a higher burden of metabolic disturbance and depressive symptoms. The present study focused on African-American adults who were part of the Healthy Aging in Neighborhoods of Diversity across the Life Span (HANDLS) study, providing the first opportunity to examine these relationships within this racial/ethnic group. The findings could elucidate whether metabolic disturbance and/or depressive symptoms are associated with genetic polymorphisms that are in turn related to low serum carotenoid status.
Materials and methods
Database and study population
Initiated in 2004 as an ongoing prospective cohort study, the HANDLS study used area probability sampling to recruit a socio-economically diverse and representative sample of African Americans and whites (30–64 years old) living in Baltimore, MD( Reference Evans, Lepkowski and Powe 31 ). The HANDLS protocol was approved by the Institutional Review Board of the National Institute on Aging. The present study used cross-sectional data from the baseline HANDLS study cohort.
A total of 3720 selected subjects participated in the household survey at phase 1 (sample 1). Of these selected subjects, 2436 (65·4 %) had complete baseline phase 2 examinations (sample 2). However, our data used a subset with complete genetic data on a sample of African-American participants of the HANDLS study (n 1024, sample 3). Of these subjects, 873 had complete depressive symptom data (sample 4a) and 910–961 had complete data on metabolic outcomes (sample 5a–5i).
Genetic data
Blood samples were collected from the participants for DNA extraction, and genome-wide genotyping was completed for 1024 participants of the HANDLS study using Illumina 1M SNP coverage. For a further description of the methods used, see online supplementary methods.
Selection of SNP of interest for the present analysis was solely based on those detected in previous GWAS and candidate gene studies as highly significant predictors of serum carotenoid status( Reference Borel 22 , Reference Hendrickson, Hazra and Chen 28 , Reference Lietz, Oxley and Leung 32 ). These SNP were extracted from high-quality imputed genotypes. Most of these selected SNP are available in our database, with the exception of two β,β-carotene-9′,10′-oxygenase (BCDO2) SNP (W80X: bovine SNP; c.196C>T: sheep SNP), which are not human SNP, and one scavenger receptor class B member 1 (SCARB1) SNP (SR-BI: intron 5)( Reference Borel 22 ). Other SNP (n 5) that were selected from one study( Reference Hendrickson, Hazra and Chen 28 ) were dropped for various reasons, the most common of which was high linkage disequilibrium with the other selected SNP. None of the remaining SNP was in strong linkage disequilibrium with each other. Consequently, twenty-four distinctive SNP with reliable values were chosen. A detailed description of these selected SNP is presented in online supplementary Table S1.
SNP for lower carotenoid status, low specific-carotenoid risk score and low total-carotenoid risk score
Of the twenty-four distinctive SNP, combinations that would allow the assessment of the effect of an increasing genetic risk of lower carotenoid level on the binary measures of metabolic disturbance and depression were created. First, we examined the independent effects of each SNP allele dosage that was previously shown to be associated with a lower specific carotenoid or a group of carotenoids. To this end, from these twenty-four SNP, twenty-four genetic exposure variables were created and termed SNPlcar (SNP for lower carotenoid status). SNP dosage was coded as is or reverse coded (0,1,2 or 2,1,0) depending on whether the minor allele was associated with lower carotenoid status or vice versa (for details, see online supplementary Table S1).
Moreover, to assess the collective associations of SNP linked to lower levels of specific and total carotenoids with the outcomes of interest, two risk scores were created: (1) low specific-carotenoid risk score (LSCRS), by summing the SNPlcar values together that pertained to that specific carotenoid; (2) low total-carotenoid risk score (LTCRS), by summing all SNPlcar values together, reflecting low levels of all carotenoids (see online supplementary Table S2 and Fig. S1). In the computation of the former score, a SNPlcar was entered into a LSCRS, when previously shown to have the most significant association with a specific carotenoid (smallest P value), particularly when multiple carotenoids were affected by the same SNPlcar. We assumed that each SNPlcar was associated with the levels of specific carotenoids based on previous findings in whites, despite potential ancestral differences in African Americans, particularly in terms of linkage disequilibrium patterns( Reference Reich, Cargill and Bolk 33 ). Since a direct way to estimate the effect size of each SNPlcar on the levels of serum carotenoids was not available for African Americans, we did not apply SNP-specific weights from previous studies on whites to account for SNP-specific differences in the effects on carotenoid status. Thus, we simply summed the risk alleles or the combinations of risk alleles together to obtain the LSCRS and LTCRS, as was done in a previous study( Reference Grimsby, Porneala and Vassy 34 ). In each LSCRS, SNPlcar included were specific to that particular carotenoid and were not double-counted in another LSCRS.
Anthropometric indices
Body weight and standing height were measured directly. BMI (weight/(height)2, kg/m2) was calculated for each participant. Waist circumference (cm) was measured using a tape measure starting from the hip bone and wrapping around the waist at the level of the navel. Obesity was defined as BMI ≥ 30 kg/m2, while central obesity was defined as a component of the MetS (see the ‘Metabolic syndrome’ section).
Metabolic outcome variables
Systolic and diastolic blood pressure
The average of the right and left sitting blood pressure values was taken to represent each of the systolic and diastolic blood pressure levels for the present analysis. Blood pressure was measured non-invasively using the brachial artery auscultation method with an aneroid manometer, a stethoscope and an inflatable cuff.
Other metabolic risk factors
Following an overnight fast, blood samples were drawn from an antecubital vein. Total cholesterol, HDL-cholesterol (HDL-C), TAG, uric acid and glucose concentrations were assessed using a spectrophotometer (Olympus 5400). Fasting serum insulin concentration was analysed with a standard immunoassay test (DPC Immulite 2000; Siemens), and C-reactive protein (CRP) concentration was analysed with an immunoturbidimetry method (Behring Nephelometer II; Siemens). Homeostatic model assessment of insulin resistance (HOMA-IR)( Reference Wallace, Levy and Matthews 35 ) was computed with a cut-off point of 2·61, reflecting a high insulin resistance level as has been suggested elsewhere( Reference Matthews, Hosker and Rudenski 36 ). Cut-off values for hyperuricaemia were >420 μmol/l (>7 mg/dl) in men and >360 μmol/l (>6 mg/dl) in women( 37 ), while elevated CRP was defined as >2·11 mg/l( Reference Ridker, Cushman and Stampfer 38 ).
Metabolic syndrome
Central obesity was defined by waist circumference ≥ 102 cm or 40 inches for men and ≥ 88 cm or 35 inches for women( 39 ). This is one of the five components in the main definition of the MetS according to the National Cholesterol Education Program Adult Treatment Panel III (NCEP ATP III) (2005)( 40 ). Using this definition, the MetS was positive when three or more of the following criteria screened were positive: (1) waist circumference >102 cm for men and >88 cm for women; (2) systolic blood pressure/diastolic blood pressure ≥ 130/85 mmHg; (3) fasting glucose ≥ 5·5 mmol/l ( ≥ 100 mg/dl); (4) TAG ≥ 1·7 mmol/l ( ≥ 150 mg/dl); (5) HDL-C < 1·04 mmol/l ( < 40 mg/dl) for men and < 1·3 mmol/l ( < 50 mg/dl) for women.
Assessment of depressive symptoms
Extensively trained psychometricians administered, among others, a baseline battery of cognitive and neuropsychological tests( Reference Lezak, Howeison and Loring 41 ) that included baseline depressive symptoms using the Center for Epidemiologic Studies-Depression scale, a twenty-item, self-report symptom rating scale that emphasises the affective, depressed mood component( Reference Radloff 42 ). The invariant factor structure of the Center for Epidemiologic Studies-Depression scale was recently shown using confirmatory factor analysis comparing NHANES I and HANDLS data( Reference Nguyen, Kitner-Triolo and Evans 43 ). A cut-off point of 16 was used to assess elevated depressive symptoms (EDS) in all analyses.
Covariates
Covariates considered as potential confounders included sex, age, education (below high school (grades 1–8), high school (grades 9–12), above high school (13+)), poverty income ratio (below v. at or above the poverty line), smoking status (current smoker v. non-smoker), drug use (current v. past or never), and ten principal components to control for any residual effects of the population structure (see online supplementary methods).
Statistical methods
Differences in means and associations of categorical variables across ‘genetic data completeness’ were tested by t and χ2 tests, respectively, using Stata 13.0 (StataCorp)( 44 ). Then, multiple logistic regression models were conducted to test the associations of SNPlcar (entered separately in each model), five LSCRS (entered simultaneously) and one LTCRS with ten binary outcomes (obesity, MetS and five components), elevated HOMA-IR, elevated CRP, hyperuricaemia and EDS. Adjusted OR and 95 % CI were estimated. Type I error was initially set at 0·05, with regression coefficients being assessed using the Wald test. Finally, to test linear dose–response relationships, quartiles of LSCRS and LTCRS were entered into the regression models as ordinal variables, and P values for trend were computed from the Wald test. Additionally, non-linear associations were tested for each quartile compared with the lowest quartile (Q1) as the common referent category. SNPlcar analyses were corrected for multiple testing by reducing type I error to α/k (k= 24 is the number of SNP tested for each phenotype). Thus, two-sided P values were presented uncorrected, with a significance level being set at 0·05/24 = 0·002.
Results
According to Table 1, the selected participants with complete genetic data were generally older, but had a few missing data on most sociodemographic and lifestyle variables compared with those without genetic data. All LSCRS (in their continuous form) were weakly to moderately correlated (R − 0·50 for lutein+zeaxanthin v. β-cryptoxanthin to +0·044 for lutein+zeaxanthin v. lycopene). Thus, it was possible to covary these gene scores in multiple logistic regression models. For descriptive purposes, mean dietary intakes of carotenoids (μg/4184 kJ per d (μg/1000 kcal per d)) are presented in online supplementary Fig. S2, stratified by sex and poverty income ratio categories. Comparisons were made between the categories, and key findings included a higher intake of β-carotene and lutein+zeaxanthin among women. However, the LTCRS was not correlated with total carotenoid intake (μg/4184 kJ per d (μg/1000 kcal per d)) (R − 0·03, P= 0·35; see online supplementary Fig. S3).
Table 1 Characteristics of the participants of the Healthy Aging in Neighborhood of Diversity across the Life Span (HANDLS) study by genetic data completeness (Number of participants and percentages; mean values and standard deviations)
HS, high school; CES-D, Center for Epidemiologic Studies-Depression; SBP, systolic blood pressure; DBP, diastolic blood pressure; HDL-C, HDL-cholesterol; NCEP ATP III, National Cholesterol Education Program Adult Treatment Panel III; MetS, metabolic syndrome; HOMA-IR, homeostatic model assessment-insulin resistance; CRP, C-reactive protein.
* P< 0·05 for null hypothesis of no difference by genetic data completeness (t or χ2 test).
† Outliers with values of CRP >50 (n 12) were removed from this sample.
While examining each of the twenty-four SNPlcar in a separate model as a predictor for each of the outcomes of interest, controlling for key potential confounders (see Table 2 and online supplementary Table S3), a few associations emerged that were against the hypothesised direction. On the one hand, these included a putative inverse relationship between SNPlcar17(BCMO1,β-carotene) and obesity; SNPlcar2(APOB,β-carotene) and several phenotypes, indicative of inflammation (CRP), dyslipidaemia (low HDL-C) and, importantly, NCEP ATP III MetS; SNPlcar10(BCMO1,β-cryptoxanthin) and hypertension; SNPlcar19(CD36,lutein+zeaxanthin) and TAG dyslipidaemia; and SNPlcar23(LPL,α-carotene) and elevated HOMA-IR. On the other hand, a number of SNPlcar showed positive associations that were in line with the hypothesis, mainly within the BCMO1 locus. These included SNPlcar14(BCMO1,β-cryptoxanthin) and EDS; SNPlcar12(BCMO1,α-carotene) and central obesity; and SNPlcar14(BCMO1, β-carotene)/SNPlcar16(BCMO1,β-carotene) and hypertension. However, none of the key findings survived Bonferroni correction.
Table 2 Gene SNP related to low carotenoid status (SNPlcar) and their associations with selected binary metabolic outcomes (obesity, National Cholesterol Education Program Adult Treatment Panel III (NCEP ATP III) metabolic syndrome (MetS)) and elevated depressive symptoms (EDS) among African-American adults assessed by multiple logistic regression analysis* (Odds ratios and 95 % confidence intervals)
ABCG5, ATP-binding cassette, subfamily G, member 5; BCMO1, β-carotene mono-oxygenase 1; CD36, thrombospondin receptor; LIPC, hepatic lipase; FABP2, fatty acid-binding protein 2; LPL, lipoprotein lipase; SCARB1, scavenger receptor class B member 1; CES-D, Center for Epidemiologic Studies-Depression.
* Each SNPlcar was entered in a separate multiple logistic regression model as the main predictor. SNP allele dosage was coded as is or reverse coded (0,1,2 or 2,1,0) depending on whether the minor allele was associated with lower carotenoid status or vice versa (for details, see online supplementary Table S1). Covariates entered as potential confounders included sex, age, poverty income ratio ( < 125 v. ≥ 125 %), education (below high school, high school or above high school), marital status (current, former, never or missing), smoking status (current, former, never or missing), drug use (current, past, never or missing) and ten principal components to adjust for population structure.
When combining SNPlcar into gene risk scores reflecting lower levels of specific carotenoids (i.e. LSCRS) and examining their associations with multiple outcomes (Table 3), several findings emerged. First, the α-carotene LSCRS was associated with a lower odds of HDL-C dyslipidaemia (Q4 (highest quartile) v. Q1 (lowest quartile): OR 0·65, 95 % CI 0·44, 0·97; P= 0·037, P for trend = 0·045), with a similar pattern being observed for the β-cryptoxanthin LSCRS (Q4 v. Q1: OR 0·61, 95 % CI 0·38, 0·96; P= 0·033, P for trend = 0·039). In contrast, this same LSCRS was associated with a higher odds of EDS (Q4 v. Q1: OR 1·83, 95 % CI 1·07, 3·12; P= 0·026, P for trend = 0·047).
Table 3 Low specific-carotenoid risk scores (LSCRS, quartiles (Q)) and their associations with selected binary metabolic outcomes (central obesity, National Cholesterol Education Program Adult Treatment Panel III (NCEP ATP III) metabolic syndrome (MetS) and its components, elevated homeostatic model assessment-insulin resistance (HOMA-IR), elevated C-reactive protein (CRP) and hyperuricaemia) and elevated depressive symptoms (EDS) among African-American adults assessed by multiple logistic regression analysis* (Odds ratios and 95 % confidence intervals)
HDL-C, HDL-cholesterol; CES-D, Center for Epidemiologic Studies-Depression.
* All LSCRS were entered in the same multiple logistic regression model (as quartiles, with the first quartile being the reference category) as main predictors, to assess their net association with each of the metabolic outcomes and with EDS. Covariates entered as potential confounders were sex, age, poverty income ratio ( < 125 v. ≥ 125 %), education (below high school, high school or above high school), marital status (current, former, never or missing), smoking status (current, former, never or missing), drug use (current, past, never or missing) and ten principal components to adjust for population structure.
Moreover, a number of non-linear associations were also noted whereby a LSCRS was either inversely or positively associated with an outcome of interest when comparing one quartile with Q1, but not with others. For instance, a lower lutein+zeaxanthin gene risk score was associated with a lower odds of TAG dyslipidaemia only when comparing Q2 with Q1 (OR 0·51, 95 % CI 0·31, 0·83; P= 0·007). Thus, only the middle part of the distribution for lower lutein+zeaxanthin status was linked to the reduced odds of this type of dyslipidaemia, whereas the remaining part of the distribution (Q3 and Q4) showed a comparable odds of this outcome with Q1. Similarly, a lower odds of EDS was found when comparing Q2 with Q1 of the low α-carotene gene score, but not with others. In contrast, a gene score reflecting a low lycopene level was associated with a higher risk of central obesity only when comparing Q2 with Q1 (OR 2·81, 95 % CI 1·09, 7·26; P= 0·033), with the association weakening with each higher quartile comparison. Importantly, the lutein+zeaxanthin LSCRS was inversely related to the odds of having NCEP ATP III MetS, though only for Q2 and Q3 v. Q1, without a significant linear trend being observed.
As detailed in Table 4, the associations of the LTCRS with metabolic outcomes and EDS were assessed by a series of multiple logistic regression, using quartiles of the risk score as the main predictor and testing for linear trend in the association. Among the key findings, an inverse and linear association between the LTCRS and HDL-C dyslipidaemia indicated that a gene score associated with low carotenoid status was potentially protective against this outcome (Q4 v. Q1: OR 0·67, 95 % CI 0·45, 0·99; P= 0·046, P for trend = 0·046). Similarly, a non-linear association was found for elevated CRP (Q2 v. Q1: OR 0·63, 95 % CI 0·43, 0·91; P= 0·015).
Table 4 Low total-carotenoid risk scores (LTCRS, quartiles (Q)) and their associations with selected binary metabolic outcomes (central obesity, National Cholesterol Education Program Adult Treatment Panel III (NCEP ATP III) metabolic syndrome (MetS) and its components, elevated homeostatic model assessment, insulin resistance (HOMA-IR), elevated C-reactive protein (CRP) and hyperuricaemia) and elevated depressive symptoms (EDS) among African-American adults assessed by multiple logistic regression analysis* (Odds ratios and 95 % confidence intervals)
HDL-C, HDL-cholesterol; CES-D, Center for Epidemiologic Studies-Depression.
* The LTCRS was entered in the multiple logistic regression model (as quartiles, with the first quartile being the reference category) as main predictors, to assess their net association with each of the metabolic outcomes and with EDS. Covariates entered as potential confounders were sex, age, poverty income ratio ( < 125 v. ≥ 125 %), education (below high school, high school or above high school), marital status (current, former, never or missing), smoking status (current, former, never or missing), drug use (current, past, never or missing) and ten principal components to adjust for population structure.
Discussion
In the present study, we examined the associations of gene polymorphisms related to low carotenoid status with various metabolic outcomes and EDS in an urban, socio-economically diverse sample of African-American adults. None of the key findings for SNP analyses survived correction for multiple testing. However, an inverse association was found between the LTCRS and HDL-C dyslipidaemia. The β-cryptoxanthin LSCRS was associated with a lower odds of HDL-C dyslipidaemia, but a higher odds of EDS.
Previous studies that examined SNP used in our SNPlcar focused on dyslipidaemia, type 2 diabetes, obesity and the MetS. Particularly, SNPlcar1(ABCG5,rs6720173:C/G,lutein+zeaxanthin) ( Reference Herron, McGrane and Waters 21 , Reference Borel 22 ) has been found to be unrelated to HDL-C dyslipidaemia or other lipids based on a study of Puerto Rican adults, while other associations were found for various other SNP studied on that gene locus( Reference Junyent, Tucker and Smith 45 ). The main function of ABCG5 (ATP-binding cassette, subfamily G, member 5) is to translocate various hydrophobic substrates including carotenoids and cholesterol across extra- and intracellular membranes( Reference Herron, McGrane and Waters 21 ).
Moreover, only a few studies directly examined the associations of SNPlcar2(ApoB-516,β-carotene) ( Reference Borel 22 , Reference Borel, Moussa and Reboul 23 ) with lipid profile and other metabolic disturbances. In one study( Reference Wojczynski, Gao and Borecki 46 ), while another ApoB SNP (rs676210) has been reported to be associated with the lowering of TAG, SNPlcar2 (rs934197:T/C) has not been found to be associated, a finding replicated by at least one other study( Reference Perez-Martinez, Perez-Jimenez and Ordovas 47 ). However, two recent studies have detected an association of the ‘T’ allele dosage of that SNP with a higher postprandial TAG level( Reference Perez-Martinez, Perez-Jimenez and Ordovas 48 ) and increased insulin resistance( Reference Perez-Martinez, Perez-Jimenez and Ordovas 49 ). In the present study, before correction for multiple testing (see Table 2 and online supplementary Table S2), the ApoB-516 ‘C’ allele dosage (SNPlcar2(ApoB)) yielded an inverse association with the MetS (OR 0·72, 95 % CI 0·52, 1·00; P= 0·048) and elevated CRP (OR 0·70, 95 % CI 0·51, 0·95; P= 0·022), which is consistent with previous studies. However, this finding was against the hypothesised direction that genetic polymorphisms linked to lower carotenoid status would be related to a higher odds of metabolic outcomes and EDS. ApoB is essential for the assembly and secretion of chylomicra and/or VLDL in the small intestine and the liver. It is also the main apo of LDL-C, a major carrier of carotenoids and TAG-rich lipoproteins( Reference Hammoud, Gastaldi and Maillot 50 ).
The main function of ApoA-IV is lipid absorption and modifying lipoprotein size( Reference Weinberg, Gallagher and Fabritius 51 ). Although no associations were detected in the present study with SNPlcar3 (ApoA-IV, rs675:A/T), previously linked to lower serum lycopene( Reference Borel 22 , Reference Borel, Moussa and Reboul 23 ), other studies have shown that this SNP was associated with the ability of fenofibrate to lower TAG levels among non-MetS patients( Reference Feitosa, An and Ordovas 52 ).
Moreover, in that same study( Reference Feitosa, An and Ordovas 52 ), one of the ApoE SNP included in SNPlcar4 (rs429358:C/T) was associated with increased LDL-C levels after fenofibrate treatment in the MetS group. ApoE2 has been reported to have established atheroprotective properties based on previous studies (e.g. Morabia et al. ( Reference Morabia, Cayanis and Costanza 53 )). However, we did not detect significant associations between SNPlcar4(ApoE) and any of the outcomes studied.
BCMO1 and BCDO2 are involved in symmetric and asymmetric carotenoid cleavage, respectively, and convert β-carotene and apocarotenals to retinal, thus influencing the circulatory levels of carotenoids( Reference Ziouzenkova, Orasanu and Sukhova 54 ). For two of the most highly studied SNP in the BCMO1 gene (SNPlcar5: rs6564851:G/T and SNPlcar6: rs6564851:T/G), with SNPlcar5(BCMO1,lutein+zeaxanthin) and SNPlcar6(BCMO1,β-cryptoxanthin), no relationship was observed with metabolic outcomes or EDS, in accordance with a meta-analysis suggesting that the loss of BCMO1 function was unrelated to a higher risk of type 2 diabetes( Reference Perry, Ferrucci and Bandinelli 55 ). Moreover, in a recent French-Canadian study( Reference Dastani, Pajukanta and Marcil 56 ), it has been demonstrated that SNPlcar7(BCMO1,β-carotene) and SNPlcar13(BCMO1,β-carotene) ( Reference Hendrickson, Hazra and Chen 28 ) had no association with a lower HDL-C level. No other SNPlcar on the BCMO1 gene locus have previously been studied in relation to metabolic disturbance or depressive symptoms. In the present study, although none of the associations remained significant after correction for multiple testing, among the notable associations before that correction, SNPlcar14(BCMO1,β-cryptoxanthin) ( Reference Hendrickson, Hazra and Chen 28 ) was associated with a higher odds of EDS (OR 2·05, 95 % CI 1·27, 3·31; P= 0·003; Table 2). Other associations detected were either in the expected direction (SNPlcar12(BCMO1,α-carotene) and central obesity; SNPlcar14(BCMO1,β-cryptoxanthin) and SNPlcar16(BCMO1,β-carotene) with hypertension) or against the hypothesised direction (SNPlcar17(BCMO1,β-carotene) and obesity; SNPlcar10(BCMO1,β-cryptoxanthin) and hypertension). Thus, further larger studies are needed to reconcile those inconsistent findings within that gene locus.
SNPlcar19(CD36,lutein+zeaxanthin) (thrombospondin receptor gene (rs13230419)) has been shown to increase the odds of the MetS by 29–40 % in the African-American population( Reference Love-Gregory, Sherva and Sun 57 ). CD36 codes for a membrane protein that facilitates the uptake and utilisation of fatty acids in key metabolic tissues. The present study found a similar putative effect, though there was only a significant or marginally significant association with the MetS before correction for multiple testing (OR 1·41, 95 % CI 0·99, 2·00; P= 0·06) and TAG dyslipidaemia (OR 0·66, 95 % CI 0·46, 0·94; P= 0·021). In contrast, the present study did not find any significant associations with SNPlcar18. SNPlcar18(CD36,low lutein+zeaxanthin with more ‘A’ alleles) ( Reference Borel 22 , Reference Borel, de Edelenyi and Vincent-Baudry 58 ) was related to the MetS in one previous case–control study of Egyptian adults, with the ‘G’ allele being more prevalent in cases (n 100) than in controls (n 100)( Reference Bayoumy, El-Shabrawi and Hassan 59 ). A similar finding was observed in another study of 317 African-American adults, in which the ‘A’ allele of CD36 (rs1761667:A/G) was associated with greater perceived creaminess regardless of the fat content of salad dressings (P< 0·01) and a higher mean acceptance of added fats and oils (P= 0·02) without a significant association with the obesity phenotype( Reference Keller, Liang and Sakimura 60 ).
Hepatic lipase (LIPC) hydrolyses TAG and phospholipids from HDL, intermediate-density lipoproteins and LDL, transforming them into smaller and denser particles, and promoting the cellular uptake of HDL-C( Reference Borel, Moussa and Reboul 61 ). For the two LIPC gene SNPlcar (both rs1800588:T/C), SNPlcar20 (TT v. others) has been previously shown to be associated with a low α-carotene level, while SNPlcar21 (C allele) has been linked to a low β-carotene level( Reference Borel 22 , Reference Borel, Moussa and Reboul 61 ). A study conducted among a large cohort of Chinese adults (n 4194) has shown that the ‘T’ allele was linked to a higher HDL-C level than the ‘C’ allele (P< 0·0001)( Reference Liu, Zhou and Zhang 62 ). The same pattern has been found in a large cohort study of Caucasian adults (n 4662), with the ‘C’ allele being associated with HDL-C dyslipidaemia (P< 0·0001)( Reference Lu, Dolle and Imholz 63 ). The present study failed to detect an association between LIPC SNPlcar and various outcomes of interest.
Fatty acid-binding protein 2 (FABP2)-related SNPlcar22 (rs1799883:A/G, GG v. others) was previously associated with a lower serum lycopene level( Reference Borel 22 , Reference Falush, Stephens and Pritchard 29 ). In a study of 315 elderly subjects with the MetS who were of European descent, the ‘G’ allele was linked to lower TAG and higher HDL-C levels (P< 0·05), indicative of lower risk for dyslipidaemia of both types( Reference Turkovic, Pizent and Dodig 64 ). There were no notable associations between this polymorphism and any of the outcomes of interest investigated in the present study. FABP2 is an intracellular protein expressed only in the intestine, which is involved in the absorption and intracellular transport of dietary long-chain fatty acids and carotenoids to their specific metabolic targets( Reference Borel, Moussa and Reboul 61 ).
For lipoprotein lipase (LPL)-related SNPlcar23 (rs328:G/C; GG v. CC, mainly low α-carotene level( Reference Borel 22 , Reference Herbeth, Gueguen and Leroy 30 )), two previous studies conducted among Caucasian adults also indicated that the ‘C’ allele was consistently linked to HDL-C dyslipidaemia( Reference Lu, Dolle and Imholz 63 , Reference Webster, Warrington and Weedon 65 ), with one of them observing an additional link to TAG dyslipidaemia( Reference Webster, Warrington and Weedon 65 ). However, a recent meta-analysis showed only a modest relationship between rs328:G/C and both types of dyslipidaemia( Reference Sagoo, Tatt and Salanti 66 ). Before correction for multiple testing, the present study was indicative of a consistent relationship in which the ‘G’ allele was associated with a lower odds of elevated HOMA-IR (OR 0·66, 95 % CI 0·44, 0·98; P= 0·037; see online supplementary Table S3). However, this SNPlcar was not found to be associated with any type of dyslipidaemia in the present study. LPL catalyses the hydrolysis of the TAG component of circulating chylomicrons and VLDL, in tissues other than the liver, and indirectly affects the concentration of carotenoids( Reference Herbeth, Gueguen and Leroy 30 ).
SCARB1 SNPlcar24 (SR-BI exon 1, rs61932577:A/G; GG v. others), previously linked to a lower level of β-cryptoxanthin( Reference Borel 22 , Reference Borel, Moussa and Reboul 23 ), was also studied in relation to lipid profiles among adults. SRBI has been shown to play a role in the metabolism of ApoB-containing lipoproteins in animal models and human subjects. In fact, SRBI constitutes a back-up pathway to the usual LDL receptor-mediated pathways for the catabolism of these lipoproteins. This is particularly relevant to adults with high ApoB-containing lipoproteins, commonly occurring in patients with familial hypercholesterolaemia( Reference Tai, Adiconis and Ordovas 67 ). Before correction for multiple testing, the present study found that SNPlcar24(SCARB1) (i.e. higher ‘G’ allele dosage) was linked to a higher odds of obesity, but no association was found with HDL-C or TAG dyslipidaemia. The associations of SCARB1 with HDL-C and TAG dyslipidaemia were investigated previously, with a higher dosage of the ‘A’ allele being related to higher HDL-C and lower LDL-C values in men, but not in women( Reference Acton, Osgood and Donoghue 68 ). This finding was replicated in a large study of US Caucasians (Framingham study: 2463 non-diabetic and 187 diabetic), in which diabetic subjects with the less common allele (allele A) had lower lipid concentrations, particularly LDL-C( Reference Osgood, Corella and Demissie 69 ). Those two studies had a consistent pattern of association found in the present study, though with different outcomes. However, two other studies found no associations of this SNPlcar with various lipid parameters( Reference Morabia, Cayanis and Costanza 53 , Reference McCarthy, Lehner and Reeves 70 ). Inconsistent with the pattern of findings from the present study and those of others, a study of seventy-seven subjects who were heterozygous for familial hypercholesterolaemia has found that the ‘A’ allele dosage of this SNP was associated with higher TAG( Reference Tai, Adiconis and Ordovas 67 ).
To our knowledge, the present study is the first to systematically examine genetic polymorphisms previously shown to be associated with the lower levels of serum carotenoids in relation to metabolic disturbance and depressive symptoms in an urban population of African-American adults, and to construct gene scores for that purpose. Despite its strengths, some limitations include a statistical power-limiting small sample size. Moreover, most GWAS yielding our SNPlcar short list came from studies of subjects of European ancestry. Finally, data on serum carotenoid concentrations were lacking, which prevented the direct assessment of SNPlcar associations with respective carotenoids and comparisons with previous studies of European ancestry subjects. Additionally, such data availability would have allowed the use of gene score weights depending on effect sizes of each SNPlcar on various carotenoids. Finally, in few gene loci included in of gene score computations, a SNP was related to multiple carotenoids, specifically BCMO1. However, to bypass this issue, the gene scores were made mutually exclusive by including only the most significant carotenoid for each of the carotenoid-specific gene score.
In conclusion, gene polymorphisms linked to low serum carotenoid status had mixed effects on metabolic disturbance and depression. Specifically, our findings do not support that gene polymorphisms associated with low carotenoid status will necessarily lead to a poorer metabolic and depressive symptom outcome. In fact, in most cases, the opposite trend was found, with the possible exception of the β-cryptoxanthin risk score and EDS. Therefore, there is a major discrepancy between what was found in studies linking serum carotenoids to metabolic disturbance and depressive symptoms and the present study that used gene polymorphisms linked to low carotenoid status as the main exposure. It is possible that different carotenoids may interact either synergistically or antagonistically with each other to affect these outcomes. Thus, similar studies on larger African-American samples are needed to test gene–gene (epistasis) interactions between these carotenoid-related gene polymorphisms.
Supplementary material
To view supplementary material for this article, please visit http://dx.doi.org/10.1017/S0007114514001706
Acknowledgements
The authors thank Dr Lori L. Beason-Held (NIA/NIH/IRP) for internally reviewing the manuscript and Dr Toshiko Tanaka (NIA/NIH/IRP) for additional help with the revision.
The present study was fully supported by the Intramural Research Program of the NIH, National Institute on Aging.
The contributions of the authors are as follows: M. A. B. had full access to the data, completed all the statistical analyses, wrote and revised the manuscript, planned the analysis, performed the data management and statistical analysis, and had primary responsibility for the final content; M. A. N. wrote and revised parts of the manuscript, participated in literature review, participated in data acquisition, plan of the analysis and statistical analysis; J. A. C. wrote and revised parts of the manuscript, and participated in literature review and plan of the analysis; M. K. E. wrote and revised parts of the manuscript and participated in data acquisition; A. B. Z. wrote and revised parts of the manuscript, and participated in data acquisition and plan of analysis. All authors read and approved the final version of the manuscript.
None of the authors has declared any conflict of interest.
