Predicting anthropometric and metabolic traits with a genetic risk score for obesity in a sample of Pakistanis

Obesity is an outcome of multiple factors including environmental and genetic influences. Common obesity is a polygenic trait indicating that multiple genetic variants act synergistically to influence its expression. We constructed a genetic risk score (GRS) based on five genetic variants (MC4R rs17782313, BDNF rs6265, FTO rs1421085, TMEM18 rs7561317, and NEGR1 rs2815752) and examined its association with obesity-related traits in a sample of Pakistanis. The study involved 306 overweight/obese (OW/OB) and 300 normal-weight (NW) individuals. The age range of the study participants was 12–63 years. All anthropometric and metabolic parameters were measured for each participant via standard procedures and biochemical assays, respectively. The genetic variants were genotyped by allelic discrimination assays. The age- and gender-adjusted associations between the GRS and obesity-related anthropometric and metabolic measures were determined using linear regression analyses. The results showed that OW/OB individuals had significantly higher mean ranks of GRS than NW individuals. Moreover, a significant association of the GRS with obesity-related anthropometric traits was seen. However, the GRS did not appear to affect any obesity-related metabolic parameter. In conclusion, our findings indicate the combined effect of multiple genetic variants on the obesity-related anthropometric phenotypes in Pakistanis.

Obesity is a multifactorial and complex metabolic disorder involving a chronic imbalance of energy homeostasis that has adverse implications for health such as dyslipidemia, arterial hypertension, coronary heart disease, type 2 diabetes mellitus, ovarian polycystosis, gallbladder lithiasis, sleep apnea syndrome, arthropathy, cerebral vasculopathy, and some neoplasms 1 . The substantially increasing worldwide prevalence of obesity has made it one of the major global public health concerns 2 . It is generally ascribed to unhealthy lifestyle choices or environmental factors but it is also known to have a strong genetic component. Thus, obesity is highly heritable and inherent genetic variations can confer augmented predisposition in some people while protection in others 3,4 . In rare instances, genetic predisposition to obesity can be owing to a large-effect mutation that disrupts energy homeostasis 5 . Nevertheless, no such monogenic mutation can be indicated for the prevailing majority of severely obese people [6][7][8] . The genetic predisposition to obesity is but rather an outcome of the cumulative effects of several genetic variants with individually modest effects. This polygenic inheritance involving numerous common genetic variants has been reported to account for the majority of inherited predisposition to complex and common diseases in addition to obesity [9][10][11] . Therefore, it can be said that a single genetic variant can significantly affect the risk of disease occurrence in case of monogenic Mendelian disorders while multiple genetic variants (that individually play only a modest role in conferring disease risk) collectively make a considerable contribution in manifestation of complex traits and common disorders. In this context, the concept of genetic risk score (GRS) has emerged by which the cumulative small effect sizes of multiple genetic variants can be exploited to generate an overall score for predicting these traits and disorders. The GRS has been shown to provide reliable risk prediction for many complex genetic traits 12 . In particular, the remarkable potential of GRS for identifying individuals at a higher risk of developing obesity has been revealed 13,14 .
The GRS is based on simple counts of disease-causing alleles. Generally, the GRS of any complex genetic trait can be computed by adding up risk alleles. Calculating a GRS by uniting information for multiple genetic variants offers a tool for examining genetic contributions in the manifestation of obesity in samples much Significant differences in study variables including GRS between OW/OB and NW Individuals. First, all anthropometric and metabolic parameters were compared between OW/OB and NW individuals. Anthropometric parameters included body weight, height, body mass index (BMI), waist circumference (WC), hip circumference (HC), skinfold thicknesses (SFTs), waist-to-hip ratio (WHR), waist-to-height ratio (WHtR), body fat percentage (%BF). Similarly, metabolic parameters encompassed systolic blood pressure (SBP), diastolic blood pressure (DBP), the product of triglycerides and glucose (TyG), fasting blood glucose (FBG), fasting insulin, homeostatic model assessment-insulin resistance (HOMA-IR), total cholesterol (TC), triglyceride (TG), high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), very-low-density-lipoprotein cholesterol (VLDL-C), coronary risk index (CRI), atherogenic index (AI), triglyceride-to-HDL-C ratio (TG/HDL-C). All anthropometric and metabolic parameters except height, HDL-C, and CRI differed significantly between OW/OB and NW individuals being higher in OW/OB as compared with NW individuals (Table 1). Similarly, the mean ranks of GRS were found to differ significantly between OW/OB and NW individuals being higher in OW/OB as compared with NW individuals (Table 2).
Allele frequencies and association of individual variants with overweight/obesity. The risk allele frequencies (RAFs) of all the variants are given in Table 3. The RAFs for all the variants except rs6265 were higher in the OW/OB as compared to NW individuals; however, this difference was found to be statistically significant for the TMEM18 rs7561317 only. Consequently, the multinomial logistic regression analysis also revealed a significant association of only rs7561317 with overweight/obesity ( Table 3).
Association of the GRS with obesity-related metabolic traits. No significant association of the GRS with any of the obesity-related metabolic variables such as SBP, DBP, TyG, FBG, fasting insulin, HOMA-IR, TC, TG, HDL-C, LDL-C, VLDL-C, CRI, AI, and TG/HDL-C was seen (Table 5). Table 1. Comparison of the obesity-related anthropometric and metabolic parameters between OW/ OB and NW individuals. OW/OB overweight/obese, NW normal-weight, BMI body mass index, WC waist circumference, HC hip circumference, WHR waist-to-hip ratio, WHtR waist-to-height ratio, SFT skinfold thickness, % BF body fat percentage, SBP systolic blood pressure, DBP diastolic blood pressure, TyG the product of triglycerides and glucose, FBG fasting blood glucose, HOMA-IR homeostatic model assessmentinsulin resistance, TC total cholesterol, TG triglyceride, HDL-C high-density lipoprotein cholesterol, LDL-C low-density lipoprotein cholesterol, VLDL-C very-low-density-lipoprotein cholesterol, CRI Coronary Risk Index, AI Atherogenic Index, TG/HDL-C triglyceride-to-HDL-C ratio, SD standard deviation, CI confidence interval, q-value corrected p value. The comparison of the obesity-related anthropometric and metabolic parameters between overweight/obese and normal-weight individuals was determined by Mann-Whitney U test. The mean ± SD and mean ranks were calculated to show the difference of parameters between overweight/ obese and normal-weight individuals. The analysis was corrected for multiple comparisons via the Benjamini-Hochberg method of false discovery rate (FDR) control. A p value < 0.05 was considered statistically significant. The significant p values are indicated in bold.

Discussion
Overweight/obesity is a multifactorial metabolic disorder having a complex genetic background 38 . In this context, the estimated heritability of obesity ranges from 40 to 70% 39 . Numerous candidate gene studies and GWAS have discovered several obesity-related genetic variants 40 , however, the additive effect of these variants on obesity risk has been reported in very few studies. In particular, the studies for exploring the combined effect of multiple genetic variants on the risk of obesity and related traits regarding Pakistani population are in their infancy. As common obesity has a polygenic inheritance, the assessment of cumulative effect based on GRS is very crucial to fully comprehend the components of genetic architecture and physiopathology of obesity. The striking potential of GRS for identifying individuals at a higher risk of developing overweight/obesity has been revealed in this regard 13,14 .
The rs1421085 of the FTO gene (T > C) is anticipated as the causal variant for overweight/obesity and it has been linked with a higher total energy intake and more eating episodes per day 41,42 . The obesity-increasing effect of FTO rs1421085 has been shown to disrupt ARID5B-mediated repression of IRX3 and IRX5 expression in preadipocytes that in turn leads to the excessive accumulation of triglycerides, increased adipocytes size, reduced mitochondrial oxidative capacity, and reduced white adipocytes browning, resulting in reduced mitochondrial thermogenesis 43 . The variant rs17782313 (T > C) located 188 kb downstream of the MC4R gene is strongly linked with obesity and the disruption of the transcriptional control of MC4R has been proposed as the likely mechanism of this variant 44 . Likewise, the rs6265 variant (G > A) in the coding region of the BDNF gene that involves Valine to Methionine substitution at the 66th amino acid position (Val66Met) of the N-terminal domain of pro-BDNF has also been linked with obesity. The intracellular trafficking and depolarization-induced release of BDNF are Table 3. Comparison of RAFs of the variants between OW/OB and NW individuals along with the association of the variants with overweight/obesity. OW/OB overweight/obese, NW normal-weight, RAFs risk allele frequencies, OR odds ratio, CI confidence interval. The data represents risk allele frequencies (RAFs) in overweight/obese and normal-weight individuals along with odds ratios and confidence intervals in parenthesis. Association was tested using multinomial logistic regression assuming an additive model. The analysis was performed by adjusting age and gender. An adjusted p value < 0.05 was considered significant. The significant p value is indicated in bold.  Table 4. Association of the genetic risk score with obesity-related anthropometric variables. BMI body mass index, WC waist circumference, HC hip circumference, WHR waist-to-hip ratio, WHtR waist-to-height ratio, SFT skinfold thickness, % BF body fat percentage, SE standard error, CI confidence interval, q-value corrected p value. The association of the genetic risk score with anthropometric traits was determined by linear regression. Effect size (β) and 95% confidence intervals were computed to seek rise or fall in the selected parameter per each risk allele increase. All association analyses were adjusted for age and gender and corrected for multiple comparisons via the Benjamini-Hochberg method. A p value < 0.05 was considered statistically significant. The significant p values are indicated in bold.  45,46 . The rs7561317, located about 22 kb downstream of TMEM18 is also an obesity-associated genetic variant that may diminish the expression of the TMEM18 gene and thus play a role in the manifestation of obesity 33,47 . The rs2815752 variant (G > A) positioned 20 kb upstream of the NEGR1 gene presumably decreasing its expression is linked with increased risk of obesity 48 . The current study computes the GRS based on the above-discussed five obesity-linked key loci (FTO rs1421085, MC4R rs17782313, BDNF rs6265, TMEM18 rs7561317, and NEGR1 rs2815752), which were previously found to be individually associated with obesity and related anthropometric and metabolic measures in the same sample of Pakistani population [23][24][25][26] . In our previous studies, the aforementioned individual associations were sought using multiple genetic models and in the case of obtaining association in more than one model, the h-index was calculated to indicate the relevant mode of inheritance (such as dominant, recessive, or overdominant). Also, associations of the two variants FTO rs1421085 24 and TMEM18 rs7561317 25 were obtained without gender stratification while those of the other two namely MC4R rs17782313 23 and NEGR1 rs2815752 26 were observed after gender stratification in females only. Moreover, no association was found for BDNF rs6265 49 . However, in the current study, we used a single additive model for simultaneously encompassing all the aforementioned five different genetic variants in the analysis to compute GRS for obesity-related anthropometric and metabolic phenotypes. Moreover, the sample population was not stratified based on gender in the current study. Therefore, the use of an additive model and non-stratification of the sample population might be the possible reasons that only rs7561317 was found to be associated with the overweight/obesity, when before computing GRS, individual associations of the aforementioned variants with obesity were sought in the current study.
The current study has been undertaken to determine the susceptibility when the combined effects of the above-mentioned five variants are considered because common obesity has a polygenic inheritance as indicated before. Moreover, the cumulative effect of these five variants for the increased risk of obesity has not been studied before in the Pakistani population. In the present study, the GRS based on the aforementioned five genetic variants appeared to be significantly associated with the obesity-related anthropometric parameters such as body weight, BMI, WC, HC, WHR, WHtR, SFTs, and BF% in a sample of Pakistanis regardless of age and gender. This indicates that individuals with high GRS may be more predisposed to the development of overweight/obesity. It must be noted that four out of five genetic variants could not exhibit any significant association with overweight or obesity when associations for individual variants were sought in the current study. However, the significant association of GRS based on a count of risk-associated alleles across a panel of five aforementioned genetic variants was observed, which point to the benefit of identifying a significant association based on the additive count of multiple genetic variants that individually may or may not appear to exhibit a significant association with the risk of overweight/obesity in a sample population. This implies that the use of GRS is comparatively a better approach to estimate the genetic risk of overweight/obesity based on different multiple common risk variants rather than individual risk variants especially when the sample size is not very large.
Previously, two studies also reported a significant association of GRS based on obesity-associated genetic variants with the increased risk of being obese in Pakistanis 50, 51 . However, the GRS in these aforementioned two Table 5. Association of the genetic risk score with obesity-related metabolic traits. SBP systolic blood pressure, DBP diastolic blood pressure, TyG the product of triglycerides and glucose, FBG fasting blood glucose, HOMA-IR homeostatic model assessment-insulin resistance, TC total cholesterol, TG triglyceride, HDL-C high-density lipoprotein cholesterol, LDL-C low-density lipoprotein cholesterol, VLDL-C very-low-densitylipoprotein cholesterol, CRI Coronary Risk Index, AI Atherogenic Index, TG/HDL-C triglyceride-to-HDL-C ratio, SE standard error, CI confidence interval, q-value corrected p value. The association of the genetic risk score with metabolic traits was determined by linear regression. Effect size (β) and 95% confidence intervals were computed to seek rise or fall in the selected parameter per each risk allele increase. All association analyses were adjusted for age and gender and corrected for multiple comparisons via the Benjamini-Hochberg method. A p value < 0.05 was considered statistically significant. www.nature.com/scientificreports/ studies were based on different sets of genetic variants, which differed not only between these studies but also differed with our study. This indicated the additive power of GRS over an individual variant because common obesity has a polygenic inheritance implying that various obesity-associated genetic variants act in concert to modulate the bodyweight 52 . Our and the two aforementioned studies 50,51 can be considered to create a panel of genetic variants for estimating the risk of an individual for developing obesity. The association of the GRS with almost all anthropometric traits (body weight, BMI, WC, HC, WHR, WHtR, skinfold thicknesses, and %BF) in our study showed the additive effect of these variants on fat distribution resulting in increased body weight. It is important to note that WC and WHtR are the main determinants of abdominal obesity, cardiovascular diseases, and increased mortality risk [53][54][55][56] . Thus, the higher GRS (constructed in our study) may escalate the risk of cardiovascular disease and mortality via increasing WC and WHtR. An interesting and important aspect of the current study is that we not only computed and sought the association of the GRS with the obesity-related anthropometric measures but also with obesity-related metabolic measures. Nevertheless, the GRS in the current study showed a lack of association with all metabolic traits including the parameters related to glucose and lipid metabolism, and blood pressure. Likewise, none of the aforementioned five genetic variants in our previous studies individually showed a significant association with any of the obesity-related metabolic traits in the same sample of the Pakistani population [23][24][25][26]49 . However, it must be noted that the values of nearly all the metabolic parameters (like anthropometric variables) were found significantly higher in overweight/obese individuals as compared to normal-weight individuals. Moreover, the GRS computed in our study was based on the genetic variants mainly involved in energy homeostasis and body weight regulation via regulating appetite and energy expenditure as already mentioned in the introduction section. Thus, it can be said that these variants may indirectly influence metabolism via increasing BMI and other obesity-related anthropometric measures.
In the end, it can be concluded that the cumulative or combined effect based on GRS may be valuable for better comprehension of the genetic susceptibility to obesity and may help researchers to better understand trait biology.

Material and methods
Study design, setting, and ethics. It was an analytical observational study based on a case-control Sample population and phenotypic traits. The study involved a total of 606 participants including 306 overweight/obese (OW/OB) and 300 normal-weight (NW) individuals. The participants were recruited from the general population of Karachi, a metropolitan city of Pakistan, including universities and colleges of the city using the simple random sampling without replacement technique. However, all the participants were not permanent residents of the city. All the participants or their parents/guardians (in the case of children or adolescents) signed the written informed consent before participation in the study. The demographic information regarding the recruited participants was acquired through a questionnaire. The body mass index (BMI) reference ranges for adults were considered according to World Health Organization (WHO), whereas for children/ adolescents, BMI reference ranges were considered according to growth charts of the Center for Disease Control and Prevention (CDC). However, the individuals with a history of taking drugs such as antidepressants, phenothiazine, and steroids, etc., and having any type of endocrinological disorder were not included in the study. Blood pressure (BP) was measured from the right-upper arm according to the standard procedure by using a mercury column sphygmomanometer (Certeza medical, Germany) while the individual was sitting comfortably. The BP of each individual was measured twice to calculate the average value. The body weight was measured in kilograms (kg) with the precision of up to 0.1 kg by using a mechanical weighing balance (Seca 755, Germany). Moreover, the body height was measured in centimeters (cm) with the precision of up to 0.1 cm by using a portable stadiometer (Seca 214, Germany). Both measurements were assessed without shoes and in light clothes. To calculate BMI, height was converted into meters (m) and then the value of weight (kg) was divided by the value of height in the square (m 2 ). The waist circumference (WC) and hip circumference (HC) were measured with a non-elastic measuring tape. The skinfold thicknesses (SFTs) such as biceps, triceps, suprailiac, abdomen, thigh, and sub-scapular were measured in millimeters (mm) by using a skin-fold caliper (Slim Guide, MI, USA). All the measurements were taken thrice to compute the average value for each SFT. The waist-to-hip ratio (WHR) and waist-to-height ratio (WHtR) were calculated using the values of WC, HC, and height. The body fat percentage (%BF) was computed using the sum of SFTs by employing gender-specific formulae 57 . The fasting (8-12 h) blood samples from each participant were collected in two separate vacutainer tubes for subsequent serum and DNA isolation. Fasting blood glucose was measured using a glucose monitoring system (Abbott, UK) while fasting insulin levels were measured by ELISA (enzyme-linked immunosorbent assay) using a commercial kit (DIA source, Belgium). The values of fasting glucose and insulin were used to calculate homeostatic model assessment-insulin resistance (HOMA-IR) by applying relevant formula 58 . Moreover, lipid parameters (total cholesterol, triglycerides, HDL-C, and LDL-C) were determined on a chemistry analyzer (Hitachi 902, Japan) by consuming commercial kits (Merck, Germany) based on enzymatic endpoint assays. The values of triglycerides were divided by 5 to calculate VLDL-C. Also, coronary risk index (CRI), atherogenic index (AI), triglyceride-to- www.nature.com/scientificreports/ HDL-C ratio (TG/HDL-C) were calculated 59 . Furthermore, the product of triglyceride and glucose (TyG) index was also computed 60 .
DNA extraction and genotyping. DNA  Statistical analysis. All statistical analyses were performed utilizing the Statistical Package for Social Sciences version 21 (SPSS, IBM statistics). Hardy-Weinberg Equilibrium (HWE) test was applied to determine whether the genotypic distributions of all the variants were in HWE. The Shapiro-Wilk test was availed to check the normality of the data. The obesity-associated risk alleles of MC4R rs17782313, BDNF rs6265, FTO rs1421085, TMEM18 rs7561317, and NEGR1 rs2815752 were C, A, C, G, and A, respectively. The homozygous wild type (non-risk), heterozygous (risk), and homozygous mutant (risk) genotypes were coded as 0, 1, and 2, respectively. The GRS was constructed by summing the number of risk alleles (0, 1, and 2) of the aforesaid genetic variants.
Since we included five bi-allelic variants, an individual may have a minimum of 0 and a maximum of 10 risk alleles. However, none of the individuals was found to have either 0 or 10 risk alleles. The study individuals were found to have a minimum of 1 and a maximum of 9 risk alleles. The GRS and the other continuous variables were compared between overweight/obese and normal-weight individuals by employing Mann-Whitney U test. Allelic frequencies of all the variants were calculated by direct count. The association of each variant with overweight/obesity was sought by assuming an additive model. Multinomial logistic regression was applied and the odds ratio (OR) along with 95% confidence intervals (CI) was calculated to predict the overweight/obesity risk associated with each variant as well as with GRS. The association of GRS with obesity-related anthropometric and metabolic parameters was determined by linear regression. For conformity to the linearity assumption of the linear regression, all non-normal data were transformed into normal distribution through a rank-based inverse normal transformation. The effect sizes (β) along with 95% CI were determined for all anthropometric and metabolic parameters. All analyses were adjusted for confounders such as age and gender and also corrected for multiple comparisons by the Benjamin-Hochberg method as per requirement 61 . A p value < 0.05 was considered statistically significant for all analyses.

Data availability
All data generated or analyzed during this study are included in this published article.