Genetic Background of Metabolically Healthy and Unhealthy Obesity Phenotypes in Hungarian Adult Sample Population

A specific phenotypic variant of obesity is metabolically healthy (MHO), which is characterized by normal blood pressure and lipid and glucose profiles, in contrast to the metabolically unhealthy variant (MUO). The genetic causes underlying the differences between these phenotypes are not yet clear. This study aims to explore the differences between MHO and MUO and the contribution of genetic factors (single nucleotide polymorphisms—SNPs) in 398 Hungarian adults (81 MHO and 317 MUO). For this investigation, an optimized genetic risk score (oGRS) was calculated using 67 SNPs (related to obesity and to lipid and glucose metabolism). Nineteen SNPs were identified whose combined effect was strongly associated with an increased risk of MUO (OR = 1.77, p < 0.001). Four of them (rs10838687 in MADD, rs693 in APOB, rs1111875 in HHEX, and rs2000813 in LIPG) significantly increased the risk of MUO (OR = 1.76, p < 0.001). Genetic risk groups based on oGRS were significantly associated with the risk of developing MUO at a younger age. We have identified a cluster of SNPs that contribute to the development of the metabolically unhealthy phenotype among Hungarian adults suffering from obesity. Our findings emphasize the significance of considering the combined effect(s) of multiple genes and SNPs in ascertaining cardiometabolic risk in obesity in future genetic screening programs.


Introduction
Obesity is one of the alarmingly increasing global health challenges which presently affects more than a billion people worldwide [1], and it is disproportionately prevalent among vulnerable socioeconomic groups and ethnic minorities [2][3][4]. Its direct and indirect influence on non-communicable diseases (NCDs) are well-documented, with a special emphasis on the risk of type two diabetes (T2D), cardiovascular diseases (CVDs), and several types of cancer attributed to increased adiposity [5][6][7][8]. In addition to the fact that obesity increases the risk of certain diseases, the World Obesity Federation has also identified obesity itself as a progressive, chronic, relapsing disease [9]. This finding is supported by the fact that the pathophysiology of obesity is influenced by the interaction of environmental/lifestyle factors (such as an energy-dense diet and sedentary lifestyle) and genetic predisposition, which leads to increased body weight [9] and a positive energy balance.
Data from independent studies show that in a subgroup of obese individuals, no positive association between body mass index (BMI) and cardiometabolic risk was observed,

Characteristics of the Obese Study Samples
After the exclusion of subjects with incomplete geno-and/or phenotype data, a total of 398 obese individuals (317 MUO and 81 MHO) remained in the database for the current analyses (Table 1). Apart from expected differences in biomarkers' values and the prevalence of medication use, MHO participants were younger compared to MUO, with no sex and educationspecific differences. The statistical analyses were corrected for the age of the participants, thus avoiding any effects due to age differences.

The Best-Fitting Genetic Models by SNPs
Twenty-three SNPs showed the strongest association with MUO for the recessive, 12 for the codominant, and 32 for the dominant inheritance model (Supplementary Table S1).

Optimization of Genetic Risk Score and the Association of Optimized Genetic Risk Score with MUO and Related Parameters
The optimization process of genetic risk score (GRS) was performed based on SNPs that were shown to strengthen the association between GRS and MUO by logistic regression analysis, starting with the SNP with the strongest association (rs10838687: odds ratio (OR) = 1.92, 95% confidence intervals (95% CI): 1.26-2.93; p = 0.002) and moving step by step in decreasing order to the weakest (rs659366: OR = 1.03, 95% CI: 0.79-1.34; p = 0.844). During the process, 19 SNPs (for more details, see Supplementary Table S1) were selected, i.e., included in the present study. 4 of 16 The mean value of optimized GRS (oGRS) was 21.4 (95% CI: 20.9-21.9) in the MHO group and 23.7 (95% CI: 23.5-23.9) in the MUO group. The distribution of the oGRS showed a significant difference (p < 0.001) between the two groups, and significantly higher oGRS values were observed in the MUO group compared to the MHO group ( Figure 1).

24, x FOR PEER REVIEW 4 of 17
The mean value of optimized GRS (oGRS) was 21.4 (95% CI: 20.9-21.9) in the MHO group and 23.7 (95% CI: 23.5-23.9) in the MUO group. The distribution of the oGRS showed a significant difference (p < 0.001) between the two groups, and significantly higher oGRS values were observed in the MUO group compared to the MHO group (Figure 1).

Figure 1.
Comparison of the distribution of optimized genetic risk scores (oGRS) in metabolically healthy and unhealthy obese individuals. The genetic risk score was optimized for metabolically unhealthy obesity based on 19 single nucleotide polymorphisms (selected in the GRS optimization process).
The 19 identified SNPs are located in 15 genes, mainly in three main clusters (first cluster: ADIPOQ, APOB, CETP, LIPC, LIPG and LPL; second cluster: PPARG; and third cluster: C2CD4B, CDKN2B, GIPR, HHEX, SLC2A2 and SLC30A8). The genes in the first cluster, with the exception of ADIPOQ, show a strong association with lipid metabolism, while genes in the second and third clusters are associated with type two and gestational diabetes. KCTD10 and MADD genes could not be classified in either cluster ( Figure 2).

Figure 2.
Gene-gene interaction and cluster analysis results based on genes containing SNPs selected during the genetic risk score optimization process. Note: genes associated with lipid metabolism are highlighted in blue, with diabetes in general in red, and with gestational diabetes in pink. The thickness of the lines connecting genes indicates the strength of the association between them. The 19 identified SNPs are located in 15 genes, mainly in three main clusters (first cluster: ADIPOQ, APOB, CETP, LIPC, LIPG and LPL; second cluster: PPARG; and third cluster: C2CD4B, CDKN2B, GIPR, HHEX, SLC2A2 and SLC30A8). The genes in the first cluster, with the exception of ADIPOQ, show a strong association with lipid metabolism, while genes in the second and third clusters are associated with type two and gestational diabetes. KCTD10 and MADD genes could not be classified in either cluster ( Figure 2).

The Discriminatory Power of MUO-Associated Genetic and Non-Genetic Risk Factors Based on ROC Curve Analyses
Age showed the highest discriminatory power (area under the receiver operating characteristic (ROC) curve (AUC) age = 0.71, 95% CI: 0.63-0.78) among the conventional risk factors not considered among the MHO/MHO differential diagnostic criteria by Wildman et al. [26] and Meigs et al. [27] (sex, age, BMI, and education).
The 19 identified SNPs are located in 15 genes, mainly in three main clusters (first cluster: ADIPOQ, APOB, CETP, LIPC, LIPG and LPL; second cluster: PPARG; and third cluster: C2CD4B, CDKN2B, GIPR, HHEX, SLC2A2 and SLC30A8). The genes in the first cluster, with the exception of ADIPOQ, show a strong association with lipid metabolism, while genes in the second and third clusters are associated with type two and gestational diabetes. KCTD10 and MADD genes could not be classified in either cluster ( Figure 2).  Among the physical and laboratory parameters used by them to define MUO (systolic and diastolic blood pressure, WC, fasting TAG, HDL-C, glucose, C-reactive protein (CRP), and HOMA-IR), TAG level showed the highest discriminatory power (AUC TAG = 0.77, 95% CI: 0.72-0.82).
For oGRS, the power of discrimination was found to be AUC oGRS = 0.77, 95% CI: 0.71-0.83. Based on a statistical comparison of the AUC curves, the discriminatory power of the oGRS was calculated by using the four most highly MUO-related SNPs  Figure 4) and a significant correlation with MUO risk according to adjusted logistic regression analysis (OR = 1.76, 95% CI: 1.43-2.17; p = 8.77 × 10 −8 ).
In the absence of knowledge of the exact time of MUO onset, Cox regression analyses were performed using the individuals' age at the time that the questionnaire was recorded. Cox regression analysis showed that oGRS as a continuous variable was signifi- In the absence of knowledge of the exact time of MUO onset, Cox regression analyses were performed using the individuals' age at the time that the questionnaire was recorded. Cox regression analysis showed that oGRS as a continuous variable was significantly associated with an increased risk of developing MUO earlier (hazard ratio (HR) = 1.10, 95% CI: 1.05-1.15; p < 0.001). Among the genetic risk categories based on the oGRS, both in the medium (HR = 1.62, 95% CI: 1.18-2.21; p = 0.002) and the high (HR = 1.83, 95% CI: 1.26-2.65; p = 0.001) risk groups the risk of developing MHO at a younger age was significantly higher compared to the low-risk group ( Figure 5).  Figure 5. Cumulative risk of metabolically unhealthy obesity in relation to age in low (blue), m dium (red) and high (green) genetic risk groups based on Cox regression proportional hazards model analysis.

Results of Trend and Multivariate Logistic Regression Analyses on the Association of oG with the Metabolic Status
The results of the trend analyses show a significant increasing tendency in the av age oGRS and oGRS4 values of the BMI subgroups in the metabolically unhealthy in viduals (p < 0.001). Among the metabolically healthy individuals, only the average va of oGRS showed a significant result after p-value adjustment ( Table 2). The results of multivariate logistic regression analyses showed that oGRS, both s arately (OR = 1.10, p = 0.00257) and in combination with BMI (OR = 1.07, p < 0.001), sig icantly increased the risk of metabolically unhealthy status in the total population (ob and non-obese together). oGRS4 separately did not show a significant association (O 1.15, p = 0.012) with metabolically unhealthy status, only in combination with BMI (O 1.01, p < 0.001) ( Table 3).

Results of Trend and Multivariate Logistic Regression Analyses on the Association of oGRSs with the Metabolic Status
The results of the trend analyses show a significant increasing tendency in the average oGRS and oGRS 4 values of the BMI subgroups in the metabolically unhealthy individuals (p < 0.001). Among the metabolically healthy individuals, only the average value of oGRS showed a significant result after p-value adjustment ( Table 2). The results of multivariate logistic regression analyses showed that oGRS, both separately (OR = 1.10, p = 0.00257) and in combination with BMI (OR = 1.07, p < 0.001), significantly increased the risk of metabolically unhealthy status in the total population (obese and non-obese together). oGRS 4 separately did not show a significant association (OR = 1.15, p = 0.012) with metabolically unhealthy status, only in combination with BMI (OR = 1.01, p < 0.001) ( Table 3).

Discussion
Our study is the first one to assess the genetic background variations that differentiate the MHO phenotype from the MUO one in a sample of Hungarian adults, based on genetic risk models involving SNPs associated with glucose homeostasis, lipid metabolism, and adiposity.
In the present study, we found a very high prevalence of MUO in the Hungarian obese sample population (79.6%). It is known that genetic factors, which affect metabolic pathways involved in adipogenesis, fat distribution, insulin signaling, and insulin resistance, can modulate the predisposition of developing obesity-related complications and lead to MUO [28]. In our oGRS models, we involved SNPs associated exactly with these CM traits to elucidate the links between the genetic background and the transition of MHO to MUO among individuals suffering from obesity.
The combined use of 19 SNPs (in 15 genes) that we examined showed a strong association of genetic factors with the MUO phenotype. In a GWAS-based study involving nearly 50,000 Koreans, it was found that polymorphisms in the LPL, APOA5, and CETP genes are associated with a higher risk of the metabolically unhealthy phenotype in the obese [29]. Furthermore, it was also shown that polymorphisms in the CDKN2B gene are also associated with the metabolically unhealthy phenotype even in normal-weight subjects. These results are in harmony with our findings.
Based on the gene-gene interaction analysis, the fifteen genes identified can be grouped into three clusters. The first cluster contains five genes related to lipid metabolism (CETP, LIPG, APOB, LPL, and LIPC) and one gene related to glucose metabolism (ADIPOQ). The LPL and APIPOQ genes of this cluster are associated with obesity induced by the consumption of high-fat foods [30] and show a strong association with the PPARG gene, which forms the second cluster. The results of an experiment in rodents suggest that the expression pattern of the PPARG gene is associated with high fat intake, adipocyte development, and insulin resistance [31]. The direct effects of the ADIPOQ and PPARG genes on plasma lipid profile and adiponectin concentration, as well as their interaction with diet, have been demonstrated in humans [32]. Dietary habits influence the association of six genes (C2CD4B, CDKN2B, GIPR, HHEX, SLC2A2, and SLC30A8) forming the third cluster with diabetes, adipogenesis, and cardiovascular risk [33][34][35][36].
Based on these, it is possible to conclude that the direct effects on bio-mechanisms of the gene clusters we identified are likely to be influenced by dietary factors as well. This assumption is further supported by the results of our multivariable logistic regression analyses, which demonstrate that genetic risk (defined as oGRS or oGRS 4 ) in combination with an increase in BMI strongly contributes to the development of metabolically unhealthy status.
In the present study, we successfully identified a combination of four (rs10838687 in MADD, rs693 in APOB, rs1111875 in HHEX, and rs2000813 in LIPG) out of the 19 SNPs that significantly influence the risk of MUO developing. These sets of SNPs have been shown in previous studies to have significant effects on lipid and carbohydrate metabolisms. The rs10838687 in the MADD gene was found to be associated with a defect in the enzymatic conversion of proinsulin to insulin, resulting in increased fasting glucose levels, and with the development of T2D [37]. Based on our previous results, rs7944584 in linkage disequilibrium with rs10838687 is strongly associated with the early onset of insulin resistance in the Hungarian general and Roma populations [38]. The rs693 in the APOB gene increases cardiovascular risk [39,40] by raising the levels of APOB, TAG, TC, and LDL-C and reducing HDL-C [41] levels. The rs1111875 is located in the HHEX gene, which could be identified as a candidate gene for T2D using a genome-wide association approach. The association between HHEX and T2D has been reported in different ethnic groups [42]. The rs2000813 in the LIPG gene was found to be associated with lipid parameters and cardiovascular risk [43].
Based on our results, the CM markers most strongly associated with the risk of being MUO were BP, fasting glucose, HOMA-IR, and TAG, some of which are components of the metabolic syndrome (MetS). This is in line with a recent study, involving ten different cohorts from seven countries (n = 163,517 participants), which showed that BP, fasting glucose, and TAG were among the most frequent MetS components seen among MUO participants [44]. In this study, the most frequent MetS component among Finnish subjects suffering from obesity was elevated BP. In two other studies (i.e., one in Iran and one in Spain), dyslipidemia was found as the most frequent MetS component among obese individuals [45,46]. These parameters may be the most important indicators to predict the risk of metabolic deterioration to the MUO phenotype or the preservation of MHO status in the course of time.
A meta-analysis including eight longitudinal studies showed that MHO individuals are at increased risk for all-cause mortality in the long term (≥10 years), which indicates that MHO might be an intermediate stage of MUO [47] and people with MHO tend to develop metabolic dysregulation over time and have increased long-term CVD risk [48,49]. A pan-European cohort study (EPIC-CVD) showed that obese individuals without metabolic syndrome were at a higher risk of coronary heart disease than metabolically healthy individuals of normal weight (risk ratio (RR) = 1.28, 95% CI: 1.03-1.58, p = 0.001). Individuals with MHO have also a substantially higher risk for T2D than metabolically healthy individuals with normal weight (RR = 4.03, 95% CI: 2.66-6.09, p < 0.001).
The age difference of nearly 10 years between the MHO and MUO groups in our present study also supports the theory that the MHO is a dynamic condition and can transform into MUO over time [50][51][52], within 5.5 to 10.3 years of follow-up [51,52]. Our study model showed that a moderate to high genetic risk category was significantly associated with a lower mean age of participants with the MUO phenotype, strongly suggesting a link between genetic susceptibility to excessive fat adiposity and elevated CM disease risks in the early onset of obesity. Therefore, defining the variables that may predict the transition from metabolically healthy to unhealthy obesity in a specific population can help identify those who can benefit from it the most. Findings highlight the utility of potential interventions among MHO subjects with a higher susceptibility to MUO as a valid interim target, particularly in Hungary, one of the most obese countries in Europe [53].
In addition to the genetic background, other factors that may contribute to the transition of MHO to MUO have been studied in other populations. A prospective study conducted in a Spanish cohort (n = 3,052) found that any increase in BMI, waist size, or waist-to-hip ratio contributed to the transition from MHO to MUO, whereas adhering to a healthy dietary pattern, high levels of physical activity, and not smoking contributed to preventing this transition [50]. Future studies in Hungary should focus on refining and ascertaining specific factors that influence susceptibility to the transition to MUO, beyond our model.
All individuals suffering from obesity should aim for metabolic health and normal weight. Given our findings, it may be reasonable to consider genetic-based screening for obese or susceptible individuals to slow or even prevent the development of MUO. Early detection can help to avoid or at least mitigate the development of subsequent obesityrelated complications (such as diabetes and cardiovascular disease). This approach is supported by the results of a study that examined the efficacy and safety of weight-loss drugs to prevent the development of T2D [54]. Participants were classified into three CM risk groups (low, medium, and high) based on their Cardiometabolic Disease Staging Score and it was found that although the preventive phentermine/topiramate medication reduced the risk of developing diabetes in all groups compared to the placebo (lifestyle intervention only) group, the reduction was significantly greater in the high-risk group compared to the medium and low groups. Therefore, targeting patients at high risk might improve the cost-to-benefit ratio of interventions.
It is important for professionals in the field of public health, healthcare research, and clinical practice, as well as patients, to acknowledge the complexity of factors and their interactions that contribute to the manifestation of obesity. This includes not only genetic (monogenic and polygenic), epigenetic, and developmental influences but also a multitude of interactions [55][56][57][58]. Currently, there is a renewed interest in defining models to explain the origins and development of obesity, leading to renewed debate. One proposed model, known as the Energy Balance Model (EBM), views overeating (consuming more calories than expended) as the primary cause of obesity. This model places emphasis on the role of unconscious signaling by the endocrine, metabolic, and nervous systems that control food intake [59], while highlighting the contribution of inexpensive, convenient, high in fat and sugar, "ultra-processed" (go through multiple processes) foods to the development of obesity. On the other hand, the Carbohydrate-Insulin Model (CIM) suggests that the hormonal response to highly processed carbohydrates plays a role in the partitioning of energy in the body, leading to increased deposition of fat in adipose tissue and reducing the calories available for the body's metabolic needs [60]. This, in turn, can result in overeating to compensate for the sequestered calories. There have been efforts to reconcile these two models and create an integrated "push-pull" model of obesity pathogenesis [61]. Although the debate continues, public health action does not need to wait for a resolution, as both models identify major drivers of obesity and reflect the interactions of genes and the obesogenic environment.
The study has some limitations, which should be considered when interpreting our findings. Our results were not replicated and since the current study was performed in a European population, findings may not apply to non-European populations. Replication studies, including other populations, are necessary to confirm our findings and determine their applicability to ethnically diverse groups. Finally, although genetics do not change over time and it is possible to use this information prospectively, our study design remains cross-sectional, requiring future prospective studies to replicate and validate our results. Despite these stated limitations, we believe our study provides valuable information on the genetic characteristics of MUO and MHO phenotypes at a population level.
In conclusion, this study provides the first assessment on the genetic background of MHO and MUO phenotypes in a sample of Hungarian adults. Findings support the notion of early identification of individuals at high metabolic risk in populations suffering from obesity and show that in addition to environmental and lifestyle factors, one's genetic background also has an important role in the development of MUO. Further, prospective study designs are warranted aiming at using genetic risk models not only to stratify the risk of impaired metabolic health among people suffering from obesity but also in normal-weight and overweight people. In summary, our study shows that obesity varies in its impact on metabolic health and renders unfavorable effects, offering a window of opportunity for early targeted public health interventions.

Sample Population and Relevant Parameters Defined
The sample was derived from a population-based disease-monitoring program, the General Practitioners' Morbidity Sentinel Stations Program (GPMSSP) in Hungary [62]. Detailed methods of sampling and the data collection process are thoroughly described in the Hungarian Metabolic Syndrome Survey (HMSS) [63]. In brief, in the present study, 59 GPs from eight counties representing diverse socio-economic regions within Hungary were invited to participate. Medical and socio-demographic characteristics were recorded and physical examinations (weight, height, waist circumference, and blood pressure measurements) were carried out. Blood samples were collected for laboratory tests (including routine diagnostic tests for fasting glucose, insulin, C-reactive protein, HDL-cholesterol, and triacylglycerols) and DNA isolation. HOMA-IR was calculated according to the following formula: fasting insulin (microU/L) x fasting glucose (nM)/22.5. Medications for hypertension, diabetes, and lipid disturbances were also recorded.
Initially, data from 1819 participants representing 91% of invited individuals were collected. In the present study, those with complete geno-/phenotype data (n = 1282) were included. The total population was divided into three subgroups based on BMI: normal weight (BMI < 25; n = 440), overweight (BMI: 25-< 30; n = 444), and obese (BMI ≥ 30; n = 398) ( Figure 6).  59 GPs from eight counties representing diverse socio-economic regions within Hungary were invited to participate. Medical and socio-demographic characteristics were recorded and physical examinations (weight, height, waist circumference, and blood pressure measurements) were carried out. Blood samples were collected for laboratory tests (including routine diagnostic tests for fasting glucose, insulin, C-reactive protein, HDL-cholesterol, and triacylglycerols) and DNA isolation. HOMA-IR was calculated according to the following formula: fasting insulin (microU/L) x fasting glucose (nM)/22.5. Medications for hypertension, diabetes, and lipid disturbances were also recorded. Initially, data from 1819 participants representing 91% of invited individuals were collected. In the present study, those with complete geno-/phenotype data (n = 1282) were included. The total population was divided into three subgroups based on BMI: normal weight (BMI < 25; n = 440), overweight (BMI: 25-< 30; n = 444), and obese (BMI ≥ 30; n = 398) ( Figure 6).

Defining Metabolically Healthy and Unhealthy Obesity
There is no universally accepted definition for the MHO and MUO phenotypes; therefore, MHO and MUO subjects in this analysis were identified by using a combination of classifying criteria established by Wildman et al. [26] and Meigs et al. [27] (Table 4). This was achieved by considering (a) the robustness of these criteria and (b) the availability of variables in our database.

Defining Metabolically Healthy and Unhealthy Obesity
There is no universally accepted definition for the MHO and MUO phenotypes; therefore, MHO and MUO subjects in this analysis were identified by using a combination of classifying criteria established by Wildman et al. [26] and Meigs et al. [27] (Table 4). This was achieved by considering (a) the robustness of these criteria and (b) the availability of variables in our database.

DNA Isolation, SNP Selection and Genotyping
DNA was isolated using a MagNA Pure LC system (Roche Diagnostics, Basel, Switzerland) with a MagNA Pure LC DNA Isolation Kit-Large Volume according to prespecified instructions of the manufacturer. Extracted DNA was eluted in 200 µL MagNA Pure LC DNA Isolation Kit-Large Volume elution buffer.
SNPs strongly associated with obesity, lipid metabolism, and glucose homeostasis were identified by screening PubMed, HuGE Navigator, and Ensembl databases. As a result, a total of 67 SNPs (in 44 genes) were considered (Supplementary Table S1), of which (a) 23 were most strongly associated with obesity [64], (b) 22 with lipid metabolism [65], and (c) 22 with glucose homeostasis [66]. Concerning the effects of SNPs regarding the metabolic traits, overlaps are possible. Genotyping was performed using the MassARRAY platform (Sequenom Inc., San Diego, CA, USA) with iPLEX Gold chemistry by the Mutation Analysis Core Facility at the Karolinska University Hospital, Sweden. Validation, concordance analysis, and quality control were conducted by the Facility according to their protocols.

Identification and Coding of the Genetic Model Best Associated with HOMA-IR by SNPs
For each SNP, three widely used genetic inheritance models (i.e., codominant, dominant, and recessive) were examined to determine which model had the strongest association with MUO as a binary outcome (i.e., MUO vs. MHO). Multivariable logistic analysis (controlled for age, sex, and education) was conducted to test each SNP's association with MUO. Cox and Snell R 2 (the higher the better) and p-values (the lower the better) guided the selection process of the best-fitted heritability model [67]. We considered the most suitable heritability model associated with MUO for each SNP used in the optimized genetic risk score (oGRS).
Coding for each SNP was based on the following genetic model of inheritance criteria: a) Codominant genetic model: homozygote genotype with risk allele was labelled as 2, whereas heterozygote gene labelled as 1 and 0 was coded for no risk allele. b) Dominant genetic model: 2 was coded for the presence of one or two risk alleles and 0 was coded for the absence of a risk allele. c) Recessive genetic model: 2 was counted for the presence of two risk alleles, while 0 was counted for the homozygote gene with the absence of a risk allele and for the heterozygote gene.

Calculation and Optimization of the Genetic Risk Score
The oGRS was calculated using the following equation: where G i is the risk score according to the chosen heritability model (see the previous subsection). The genetic risk model optimization procedure selected SNPs with the strongest association with MUO (as a binary outcome variable). GRS optimization was performed using multivariable logistic analyses. The SNPs were tested in ascending order of p-value and each SNP was inserted into the model successively, starting from the SNP with the strongest association (lowest p-value), and the association between oGRS and MUO was examined after each succession. SNPs were selected and used for the final optimized GRS only if they increased the strength of association of oGRS with MUO. SNPs that did not affect or weakened the association were excluded from further analyses. Based on the oGRS, individuals were classified into three genetic risk groups based on tertiles. Individuals in the lowest tertile were assigned to the low-risk group, the individuals in the second tertile were assigned to the moderate-risk group, and the individuals in the third tertile were assigned to the high-risk group.

Statistical Analysis
The statistical procedures used to develop the genetic risk were tested and developed on the obesity sample population (n = 398), while its interaction with and separately from BMI was tested on the total population (obese and non-obese subpopulation, n = 1282). Chi-square (χ 2 ) was used to test the Hardy-Weinberg equilibrium (HWE) of genotyped SNPs and compare differences between non-quantitative variables within the study population. The Shapiro-Wilk test was used to examine whether the quantitative variables are normally distributed, and, if necessary, Templeton's two-step method [68] was considered to transform the non-normal variables into normal ones. The Mann-Whitney U test was used to assess the distribution of non-normally distributed data between the study groups. Multivariable logistic analyses were used to determine the association between individual SNPs, the aggregate of them (oGRS), and MUO. Cox regression analysis was used to examine the association of oGRS with age at the onset of MUO. In these analyses, the age of the individual at the time the questionnaire was collected was used as the outcome variable. All regression analyses were carried out under an adjusted model. The online software Search Tool for the Retrieval of Interacting genes (STRING-version 11.5; https://string-db.org/; accessed on 10 January 2023) was used for interaction and cluster analysis and visualization of genes and proteins [69]. A minimum interaction score of 0.400 was used and Markov Cluster Algorithm was applied for determining clusters.
The association of risk groups (low, medium, and high) based on oGRS scores with age-at-event (age at identifying MUO in the survey) was examined using multivariable logistic regression (i.e., adjusted for age, sex, BMI, and education). A statistically significant trend between the proportion of individuals with MUO and oGRS risk categories was tested with the Jonckheere-Terpstra test [70]. Receiver operating characteristic (ROC) analysis was employed to evaluate the discriminatory ability of the oGRS and the area under the curve (AUC) was used as an indicator of diagnostic accuracy. In addition, the minimum number of SNPs for which the discrimination accuracy is not significantly different compared to the oGRS was also determined on the bases of the ROC curves' analyses and the effect of them was also examined.
Multivariate logistic regression analysis was used to investigate the association between the genetic risk (defined by oGRS or oGRS 4 ) and its interaction with BMI and the risk of developing metabolically unhealthy conditions for the total study population. In the statistical analysis of interaction variables, adjustments were made for age, sex, education, BMI, and oGRSs.
Statistical tests were carried out using IBM SPSS version 26 statistics for Windows (Armonk, NY, United States). Bonferroni-corrected p-value was established for the case where several dependent or independent statistical tests were performed simultaneously on a single data set.