A genetic approach to examine the relationship between vitamin B12 status and metabolic traits in a South Asian population

Observational studies in South Asian populations have suggested an association between vitamin B12 status and metabolic traits; however, the findings have been inconclusive. Hence, the aim of the present study was to use a genetic approach to explore the relationship between metabolic traits and vitamin B12 status in a Sri Lankan population and to investigate whether these relationships were modified by dietary intake. A total of 109 Sinhalese adults (61 men and 48 women aged 25–50 years) from Colombo City underwent anthropometric and biochemical measurements, dietary intake analysis, and genetic tests. Genetic risk scores (GRS) based on 10 metabolic single nucleotide polymorphisms (SNPs) (metabolic-GRS) and 10 vitamin B12 SNPs (B12-GRS) were constructed. The B12-GRS was significantly associated with serum vitamin B12 (p = 0.008) but not with metabolic traits (p > 0.05), whereas the metabolic-GRS had no effect on metabolic traits (p > 0.05) and vitamin B12 concentrations (p > 0.05). An interaction was observed between B12-GRS and protein energy intake (%) on waist circumference (p = 0.002). Interactions were also seen between the metabolic-GRS and carbohydrate energy intake (%) on waist-to-hip ratio (p = 0.015). Our findings suggest that a genetically lowered vitamin B12 concentration may have an impact on central obesity in the presence of a dietary influence; however, our study failed to provide evidence for an impact of metabolic-GRS on lowering B12 concentrations. Given that our study has a small sample size, further large studies are required to confirm our findings.


Introduction
In recent years, the incidence of obesity in Sri Lanka has increased markedly [1]. The prevalence of being overweight or obese in Sri Lankan adults is 34.4% (25.2% and 9.2% in 2005 and 2006, respectively), with an upward trend being observed [1,2]. Obesity increases the risk for certain health conditions, such as insulin resistance, diabetes mellitus, and hypertension [3]. South Asians have been observed to exhibit increased visceral fat and waist circumference (WC), hyperinsulinemia, and insulin resistance; this has been termed the BSouth Asian phenotype^ [4]. Despite a known genetic contribution, the increase in obesity has been largely associated with changes in lifestyle habits [5,6]. It is imperative that modifiable risk factors for obesity and associated metabolic problems are identified, especially if they can be easily addressed.
Vitamin B 12 is a micronutrient that has been identified as a modifiable risk factor associated with the progression of metabolic disorders. In humans, vitamin B 12 acts as an essential coenzyme involved in DNA synthesis and cellular energy production [7]. Subclinical deficiency of vitamin B 12 has been linked to higher levels of homocysteine; this may have important consequences in the progression of chronic diseases, by inducing oxidative stress and inflammation [8]. Vitamin B 12 deficiency has also been linked to many other complications including an increased risk of obesity [9][10][11], diabetes [12][13][14], and cardiovascular disease [15]. Currently, one study has investigated the effect of genetically instrumented vitamin B 12 concentrations on body mass index (BMI) in individuals with European ancestry; however, there were no associations between the vitamin B 12 genetic risk score (GRS) and BMI [16].
Genetic studies have implicated several gene loci in the predisposition to vitamin B 12 deficiency, but no study has yet been carried out in the Sri Lankan population [17]. The mechanisms by which obesity and its comorbidities are related to vitamin B 12 deficiency are poorly understood. Hence, we conducted a gene-based approach to explore the relationship between metabolic traits and vitamin B 12 status in a Sinhalese cohort and investigated whether these relationships were modified by dietary intake in the Genetics Of Obesity and Diabetes (GOOD) study.

Study participants
The GOOD study is a cross-sectional study that was conducted in the city of Colombo, Sri Lanka, between April and August 2017. Healthy adults between the ages of 25 and 50 years were enrolled into the study. Exclusion criteria were having a previous history of type 2 diabetes, cardiovascular disease, or hypertension, having a BMI of more than 40 kg/m 2 or being classed morbidly obese by a physician, being blood related to other participants in the study, having any communicable disease, being pregnant or lactating, taking dietary or vitamin supplements, and taking medications that affect lipid metabolism or hypertension ( Fig. 1).

Anthropometric measures
Body weight was measured to the nearest 100 g using an electronic scale (Seca 815, Seca GmbH. Co. kg, Germany) and height was measured to the nearest millimeter using a stadiometer (Seca 217, Seca GmbH. Co. kg, Germany). The BMI calculation was based on the body weight (kg) divided by the square of body height (m). Waist circumference and hip circumference were measured using a metal tape (Lufkin W606PM®, Parsippany, NJ, USA). Body fat percentage was estimated using a handheld bioelectrical impedance analysis technique (Omron Body Fat Monitor BF306, Omron, Milton Keynes, UK).

Biochemical analysis
Blood samples (10 ml) were collected by a trained phlebotomist in the morning, after a 12-h overnight fast. Fasting serum insulin and vitamin B 12 levels were determined using the chemiluminescent microparticle immunoassay method on an Architect i1000 analyzer (Abbott Laboratories, IL, USA). Fasting plasma glucose concentrations were measured using the glucose hexokinase method using the Beckman Coulter AU5800 analyzer (Beckman Coulter®, California, USA). Glycated hemoglobin (HbA1c) was estimated by highperformance liquid chromatography using the BioRad D10 HPLC analyzer (BioRad, Hercules, CA, USA).

Dietary intake analysis
Dietary intakes were assessed using a previously validated and published [18] interviewer-administered food frequency questionnaire (FFQ) containing 85 food items. In brief, participants were asked to estimate the usual frequency (number of times per day, week, or month/ never) and the portion sizes of various food items. The recorded data was analyzed with the NutriSurvey 2007 database (EBISpro, Germany) to estimate energy as well as macro-and micronutrient consumption [19].
BThe Global Physical Activity Questionnaire( GPAQ), developed by the World Health Organization (WHO), was used to measure physical activity [20]. Individuals were classified as vigorously active, when they both exercised and engaged in demanding work activities, and moderately active, when the participants either exercised or carried out heavy physical work. The remaining study participants were classified into the sedentary group.
The 10 vitamin B 12 -related SNPs (methylenetetrahydrofolate reductase [MTHFR], rs1801133; carbamoyl-phosphate   [17]. Blood samples for the measurement of DNA were transported in dry ice to the UK. Genomic DNA was extracted from a 5-ml whole blood sample from each participant and genotyping was performed at LGC Genomics (http://www. lgcgroup.com/services/genotyping), which employs the competitive allele-specific PCR-KASP® assay.
The Hardy-Weinberg equilibrium (HWE) p values were computed for the following 20 SNPs. The SNP FUT2 rs602662 and calpain 10 (CAP10) rs3792267 deviated from the HWE; however, these SNPs were not excluded from analysis. The FUT2 SNP rs602662 previously departed from HWE in a GWA study conducted in India; the authors ruled out that the deviation was not due to a genotyping error and still used this SNP for analysis in their study [30]. In addition, the KASP™ genotyping technology used in our study has been independently assessed to be over 99.8% accurate. Validation of the KASP™ genotyping was conducted at LGC genomics, where the genotyping results were assessed by two project managers separately to confirm that the data was accurate, and this ruled out genotyping artifacts as possible reasons for deviation from HWE. The reasons for deviation from HWE could be due to population or racial grouping substructure (subgrouping), non-random mating, linkage disequilibrium (incomplete mixing of different ancestral population), or chance findings [31].

Statistical analysis
The SPSS statistical package (version 22; SPSS Inc., Chicago, IL, USA) was used for the statistical analysis. Allele frequencies were estimated by gene counting ( Table 1). The normality of variable distribution was verified by the Shapiro-Wilk test, and data not normally distributed were log transformed prior to analysis. We performed an independent t test to compare the means of the quantitative variables between men and women. Comparison of the means between the two groups was analyzed by the chi-square test for categorical outcomes.
A schematic representation of the study design is presented in Fig. 2. The unweighted, risk allele GRS method was calculated for each participant as the sum of risk allele counts across each SNP which predicted vitamin B 12 status or metabolic disease risk. The B12-GRS was generated from the SNPs in the genes MTHFR, CPS1, CUBN, CD320, TCN2, CLYBL, FUT2, TCN1, FUT6, and MUT, which have been shown to be associated with vitamin B 12 concentrations. Furthermore, another unweighted GRS was created using allele markers previously reported to be associated with metabolic disease traits. The metabolic-GRS was generated from the SNPs in the genes CAP10, KCNJ11, TCF7L2, FTO, and MC4R. A value of 0.1 or 2 was assigned to each SNP, which denotes the number of risk alleles on that SNP. These values were then calculated by adding the number of risk alleles across each SNP. The average number of risk alleles per person for the B12-GRS was 8.69 (SD = 1.70), which ranged from 5 to 15. The sample was stratified, by the median, into a Blow genetic risk group,^for those with a GRS ≤ 9 risk alleles (n = 79), and into a Bhigh genetic risk group,^for those with a GRS ≥ 10 risk alleles (n = 30). For the metabolic-GRS, the average number of risk alleles per person was 7.00 (SD = 2.28), which ranged from 1 to 13. The sample was stratified, into a Blow genetic risk group,^for those with a GRS ≤ 8 risk alleles (n = 88), and into a Bhigh genetic risk group,^for those with a GRS ≥ 9 risk alleles (n = 21). Linear regression was used to examine the association of the two GRS scores with the biochemical and anthropometric outcomes (glucose, insulin, HbAC1, vitamin B 12 , body fat %, BMI, WC, and waist-to-hip ratio (WHR)). The interaction between the two GRS scores and dietary factors on biochemical and anthropometric outcomes was determined by including interaction terms (GRS × diet) in the regression model. Models were adjusted for age, sex, BMI, and total energy intake, wherever appropriate. Correction for multiple testing was applied using Bonferroni correction [adjustment p value for association analysis was < 0.00313 [2 GRS × 8 biochemical and anthropometric outcomes (Fasting blood glucose, fasting insulin, glycated hemoglobin, vitamin B 12 , fat %, BMI, WC, and WHR) = 16 test)] and for interaction < 0.00078 [2 GRS × 8 biochemical and anthropometric × 4 lifestyle factors (dietary carbohydrate energy %, dietary protein energy %, dietary fat energy %, and physical activity levels)) = 64]. Given that there are no studies on GRS and no previously reported effect sizes for the South Asians, we were unable to perform a power calculation.

Characteristics of the participants
In this study, 109 participants (mean age, 38.34 ± 6.92 years; BMI, 24.58 ± 4.12 kg/m 2 ) were included. Table 2 illustrates the main characteristics of the study participants stratified according to sex. No significant difference between men and women was observed in the levels of fasting glucose, insulin, HbAC1, and plasma vitamin B 12 (p > 0.05). One-sided arrows with unbroken lines represent genetic associations and one-sided arrows with broken lines represent interactions between a lifestyle factor and GRS on serum vitamin B 12 /metabolic traits. We tested the association between the metabolic-GRS and vitamin B 12 concentrations and metabolic disease-related traits. We then tested the associations between the B12-GRS and vitamin B 12 status and metabolic disease-related traits. Lastly, we tested whether these genetic associations were modified by lifestyle factors (macronutrient intake and physical activity levels) Association between B12-GRS and obesity GRS with biochemical and anthropometric measurements A significant association between B12-GRS and serum vitamin B 12 was observed (p = 0.008) (Supplementary Table 1 and Fig. 3). However, this finding was not significant after correction for multiple testing. No associations between the B12-GRS and metabolic traits (p > 0.05) were observed (Supplementary Table 1). Furthermore, no associations between the metabolic-GRS and vitamin B 12 or metabolic traits (p > 0.05) were observed (Supplementary Table 2).

Interaction between the B12-GRS and dietary factors on biochemical and anthropometric measurements
An interaction was found between the B12-GRS and protein energy (%) on log transformed WC (p = 0.002). However, further stratification of participants based on their consumption of low, medium, and high dietary protein (energy %) did not show statistically significant associations between the GRS and the outcome in any of the tertiles, which could account for the small sample size (Supplementary Table 3).

Interaction between the metabolic-GRS and dietary factors on biochemical and anthropometric measurements
We observed a significant interaction between the metabolic-GRS and carbohydrate energy intake (%) on waist-to-hip ratio (P interaction = 0.015) ( Fig. 4 and Table 3). Individuals who carried eight or less risk alleles for metabolic disease had 7.47% lower WHR measurements (cm) in the highest tertile of carbohydrate energy intake (%) (mean ± SD = 78.00 ± 7.90%) compared to those with nine or more risk alleles (p = 0.035) ( Table 3).
Interactions were also seen between the metabolic-GRS and carbohydrate energy (%) on log fasting insulin concentrations (p = 0.011) and log WC (p = 0.031) and the metabolic-GRS and protein energy (%) on log fasting insulin levels (p = 0.032) and log WC (p = 0.011) ( Table 3  and Supplementary Table 3).

Interaction between the B12-GRS and physical activity on biochemical and anthropometric measurements
No statistically significant interactions were observed between the two GRSs (vitamin B 12 and metabolic) and physical activity on biochemical and anthropometric measurements ( Table 3 and Supplementary Table 3). After correction for multiple testing, none of these gene-diet and gene-physical activity interactions remained statistically significant.

Discussion
To our knowledge, this is the first study to use a genetic approach to explore the relationship between metabolic traits and vitamin B 12 status in a South Asian population. Our study confirmed the strength of the association between B12-GRS and B 12 concentrations and demonstrated the impact of genetically instrumented B 12 concentrations on waist circumference, an indicator of central obesity, through the influence of dietary protein intake. Furthermore, our study has also showed a significant effect of metabolic-GRS on waist-to-hip ratio through the influence of high carbohydrate intake. Given that the total daily intake of protein is low and carbohydrate is high in Sri Lankan adults [32], our findings, if replicated in future studies, might carry significant public health implications in terms of revising the food-based dietary guidelines which could prevent central obesity and the associated CVDrelated outcomes.
In this study, we constructed a GRS consisting of ten vitamin B 12 decreasing SNPs in genes involved in vitamin B 12 metabolism [17]. The B12-GRS was associated with vitamin B 12 levels, suggesting that it would be an ideal instrument for vitamin B 12 status. Given the lack of association between the B12-GRS and metabolic disease traits in our study, we were unable to provide evidence for linear decreases in vitamin B 12 concentrations having substantive effects on metabolic disease traits. However, we found a significant interaction between the B12-GRS and protein energy (%) on log WC. Interestingly, individuals who carried nine or less alleles had lower WC when consuming a high protein diet compared to those consuming a low protein diet. Although no statistically significant differences in WC were observed between the Fig. 4 Interaction between the metabolic-GRS and carbohydrate energy intake (%) on waist-to-hip ratio (cm) (P interaction = 0.015). Among those who consumed a high carbohydrate diet, individuals who carried nine or more risk alleles had significantly higher levels of waist-to-hip ratios compared to individuals carrying eight or less risk alleles (p = 0.035) Table 3 Interaction between the B12-GRS and lifestyle factors on anthropometric measurements alleles of the B12-GRS, the impact of the B12-GRS on WC was observed only under the influence of a high protein diet. Further investigations are required to confirm this finding to determine the clinical significance and potential applications as part of weight management interventions. At present, carbohydrates constitute the majority of the energy intake among South Asian countries such as Sri Lanka (~71.2%) [32]; in contrast, the consumption of carbohydrates is lower in Western countries (~45%) [33]. Furthermore, high carbohydrate intake has been associated with an increased risk of diabetes in a South Indian population [34] and an increase in WC among premenopausal (20-45 years) Sri Lankan women [35]. In the present study, we found a significant interaction between the metabolic-GRS and carbohydrate energy percentage on waist-to-hip ratio, where the individuals carrying more than nine risk alleles had a higher waist-to-hip ratio among those in the highest tertile of carbohydrate energy percentage. There are no previous reports of the risk variants used in our GRS, but Goni et al. [36] found that carbohydrates (total and complex) interacted with a GRS of 16 obesity/lipid metabolism polymorphisms to modify the effect on body fat mass in 711 individuals of Caucasian ancestry. In our study, we only observed interactions of the metabolic-GRS on WC and waist-to-hip ratio, which suggests that effects are likely to be on central obesity as opposed to common obesity.
South Asians have a higher risk of developing obesityrelated non-communicable diseases relative to white Caucasians despite lower BMI levels; this has been termed the BSouth Asian phenotype.^The distinctive features of this phenotype include a higher WC, abdominal adiposity combined with insulin resistance, and a greater predisposition to diabetes [4]. The role of vitamin B 12 in promoting this adverse phenotype has been suggested by Yajnik et al., who demonstrated that offspring born to mothers with a low vitamin B 12 and high folate status had a greater risk of developing insulin resistance during childhood [12]. According to Yajnik et al., vitamin B 12 deficiency prevents the generation of tetrahydrofolate from 5-methyltetrahydrofolate in the one-carbon metabolism cycle; as a result, homocysteine levels accumulate leading to altered lean tissue deposition and reduced protein synthesis [12]. Furthermore, vitamin B 12 is involved in the conversion of methylmalonyl-CoA to succinyl-CoA by the enzyme methylmalonyl-CoA mutase (adenosyl-B12 as a cofactor). Subsequently, vitamin B 12 deficiency results in elevated methylmalonyl-CoA, inhibiting the mitochondrial enzyme carnitine palmitoyltransferase, which may promote lipogenesis and insulin resistance [12,37].
No studies to date have investigated interactions between the two GRSs and physical activity on metabolic traits and B 12 concentrations in Asian Sri Lankans. Although 60% of Sri Lankan adults are reported to be highly physically active [38], no significant interactions were found between the two GRSs and physical activity on metabolic traits, which could be due to a small sample size and measurement bias associated with self-reported physical activity questionnaire. The strengths of our study include the use of a validated food frequency questionnaire [18] to measure macronutrient intake, the comprehensive measurements of lifestyle factors, and the use of GRSs which increased the statistical power of our study [39]. Nevertheless, some limitations need to be acknowledged. The first limitation concerns the relatively small sample size of the study; however, we were still able to identify significant gene-diet interactions. Furthermore, we used Bonferroni correction to correct for multiple testing and this can often lead to larger power, specifically where studies have a small sample size and a small number of disease-associated markers. This is also true for when studies have a large allele frequency difference due to a small sample size [40]. Secondly, information about the type of oil used for frying, the estimation of different dietary fat components (monounsaturated or saturated fatty acids), and vitamin B 12 intake was not collected. This could have limited our in-depth analysis of interactions of specific macronutrients and vitamins with the two GRSs. Furthermore, the study was limited to Sinhalese adults in Colombo, and the conclusions may not be applicable to other ethnic groups in Sri Lanka. Finally, none of the genetic associations or gene-lifestyle interactions were statistically significant after correction for multiple testing; however, given that this is the first study using a genetic approach to establish a relationship between vitamin B 12 status and metabolic disease outcomes in South Asians, we have taken into consideration the significant findings; hence, further large studies are required to replicate our findings.
In summary, our study suggests that a genetically lowered vitamin B 12 concentration may have an impact on central obesity in the presence of a dietary influence; however, our study failed to show an impact of the metabolic-GRS on lowering B 12 concentrations through a dietary influence. Our study also showed a significant effect of the metabolic-GRS on waist-to-hip ratio, another indicator of central obesity, through the influence of a high carbohydrate intake. However, after correction for multiple testing, none of these findings were statistically significant. Hence, further replication studies are highly warranted on large samples to confirm or refute our findings.
We also thank Dr. Suresh Surendran for the contributions for the preparation of the study. We would like to thank Dr. Seevali Surendran for the Sinhalese language editing.
Author contributions SS and KSV drafted the manuscript; SS performed the statistical analysis; and SS, VKS, DJA, and RL were responsible for the study conception. KW provided guidance to the research; SS and RL conducted data and sample collection; SS, RJ, and SA were involved in the dietary data analysis; SS and SaS were involved in the physical activity data analysis; KSV designed the gene-diet interaction study; and SS, VKS, JAL, and RJ critically reviewed the manuscript. All authors contributed to and approved the final version of the manuscript.
Funding We would like to thank the Farnborough College of Technology for contributing travel expenses towards the project. We also acknowledge the support for the GeNuIne Collaboration from the British Nutrition Foundation.
Data availability Data from this project will not be shared because additional results from the study are yet to be published.

Compliance with ethical standards
Ethics approval and consent to participate This study was approved by the Ethical Review Committee of the University of Colombo (EC-17-107) and the University of Reading Research Ethics Committee (17/ 25). All participants signed informed consent prior to their participation.

Consent for publication Not applicable.
Competing interests All authors declare that there is no conflict of interest associated with their contribution to this manuscript.
Open Access This article is distributed under the terms of the Creative Comm ons Attribution 4.0 International License (http:// creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.