Genetic Links Between Metabolic Syndrome and Osteoarthritis: Insights From Cross-Trait Analysis

Abstract

Context

Previous observational studies have indicated a bidirectional association between metabolic syndrome (MetS) and osteoarthritis (OA). However, it remains unclear whether these bidirectional associations reflect causal relationships or shared genetic factors, and the underlying biological mechanisms of this association are not fully understood.

Objective

We aimed to explore the genetic connection between MetS and OA using genome-wide association study (GWAS) summary data.

Methods

Leveraging summary statistics from GWAS conducted by the UK Biobank and the Glucose and Insulin-related Traits Consortium (MAGIC), we performed global genetic correlation analyses, genome-wide cross-trait meta-analyses, and a bidirectional two-sample Mendelian randomization analyses using summary statistics from GWAS to comprehensively assess the relationship of MetS and OA.

Results

We first detected an extensive genetic correlation between MetS and OA (r_g = 0.393, P = 1.52 × 10⁻¹⁸), which was consistent in 4 MetS components, including waist circumference, triglycerides, hypertension, and high-density lipoprotein cholesterol and OA with r_g ranging from −0.229 to 0.490. We then discovered 32 variants jointly associated with MetS and OA through Multi-Trait Analysis of GWAS (MTAG). Co-localization analysis found 12 genes shared between MetS and OA, with functional implications in several biological pathways. Finally, Mendelian randomization analysis suggested genetic liability to MetS significantly increased the risk of OA, but no reverse causality was found.

Conclusion

Our results illustrate a common genetic architecture, pleiotropic loci, as well as causality between MetS and OA, potentially enhancing our knowledge of high comorbidity and genetic processes that overlap between the 2 disorders.

Metabolic syndrome (MetS) and osteoarthritis (OA) are 2 prevalent and complex chronic noncommunicable diseases that significantly impact public health worldwide (1, 2). MetS is defined as a clustering of abdominal obesity, hyperglycemia, elevated blood pressure, and dyslipidemia (3). Approximately 25% of the global population is affected by MetS (4), and its prevalence continues to rise due to population aging, social development, and population growth (5). On the other hand, osteoarthritis is the most common form of arthritis, affecting approximately 500 million people globally, and like MetS, its prevalence is also increasing due to population aging (6).

Obesity significantly impacts OA as a key modifiable risk factor. It exerts mechanical pressure and triggers inflammation via adipose-derived mediators (7). Additionally, obesity is closely associated with MetS, characterized by abdominal obesity, hyperglycemia, hypertension, and dyslipidemia (8). Furthermore, glucocorticoid therapy for OA might elevate the risk of developing MetS (9), while metformin, used to treat MetS, shows positive effects on OA (10). These findings have sparked intriguing questions regarding the shared pathological mechanisms and genetic foundations of MetS and OA.

Genome-wide association studies (GWAS) have fundamentally altered our understanding of the genetic architecture of various complex diseases, including MetS and OA. Interestingly, several overlapping loci have been reported in GWAS of MetS and OA, suggesting possible shared genetic contributions (11-13). Recently, methodologies have been developed utilizing GWAS summary data to estimate the genetic correlation between 2 complex diseases or traits, identifying specific shared genetic variants between them (14). A recent study identified genetic variants associated with MetS and schizophrenia (15). However, there have been limited investigations leveraging large-scale GWAS data and the aforementioned genetic methodologies to explore the mechanistic links between MetS and OA.

In this study, we aimed to explore the genetic connection between MetS and OA using GWAS summary data and advanced statistical methods, so as to clarify shared genetic elements, potential causal links, and identify candidate genes and pathways underlying the joint genetic basis of both conditions.

Methods

Study Samples

Each instrumental variable (IV) was selected based on 3 Mendelian randomization (MR) assumptions. Firstly, we included single nucleotide polymorphisms (SNPs) that reached a genome-wide significance threshold (P < 5 × 10⁻⁸). Next, we identified variants with the most significant P values as independent instruments, considering linkage disequilibrium (LD) measured by r² (with a threshold of r² < 0.01 in the European 1000 Genome reference panel). Finally, we assessed the strength of instrumental variables using F-statistics, with a recommended threshold of F > 10 for MR analyses.

We collected GWAS summary data for MetS and OA from the most comprehensive GWAS in the UK Biobank. The case definition of OA is based on hospital records coded with ICD10 and/or ICD9, diagnosing OA at any site (16). The harmonized National Cholesterol Education Program (NCEP) criteria were utilized to define metabolic syndrome, establishing the 5 constituent components of the syndrome and the presence of prevalent metabolic syndrome (17). Additionally, we obtained GWAS summary data for the components of MetS, which include hypertension, waist circumference (WC), fasting blood glucose (FBG), serum triglycerides (TG), and serum high-density lipoprotein cholesterol (HDL-C). For WC, hypertension, TG, and HDL-C, we extracted GWAS summary data from the Medical Research Council Integrative Epidemiology Unit (MRC-IEU) UK Biobank GWAS Pipeline. For FBG, the summary statistics were available from the Glucose and Insulin-Related Traits Consortium (MAGIC). Further details on the specific extracted data are provided in Supplementary Table S1 (18).

Data Sources and Instrumental Variables Selection for MetS and OA

Firstly, we selected independent SNPs significantly associated with OA at a P value threshold of less than 5 × 10⁻⁸. As no SNPs met the threshold of less than 5 × 10⁻⁸ for OA, we extended the selection criteria to 1 × 10⁻⁵ to identify suitable IVs. A total of 26 independent genetic SNPs were identified at a genome-wide significant level (P < 1 × 10⁻⁵) and selected as IVs for OA (16) (Supplementary Table S2 (18)). In the MR analyses, we identified 79 independent loci associated with MetS in a GWAS that reached the threshold of significance (P < 5 × 10⁻⁸). These variants were chosen as IVs for the construction of the MR analysis (17) (Supplementary Table S3 (18)). Furthermore, for WC, we selected 31 variants with a P value less than 5 × 10⁻⁸, which were identified in a GWAS (19). In the case of hypertension, 12 variants associated with hypertension were identified from a GWAS involving 29 studies and a total of 203 006 participants of European descent, reaching genome-wide significance (20). For FBG, a set of 134 SNPs significantly associated with FBG (P < 5 × 10⁻⁸) were identified and used as IVs (21). Lastly, for TG and HDL-C, we utilized variants extracted from a representative GWAS involving 188 577 subjects from the Global Lipids Genetics Consortium (GLGC). In this GWAS, 28 SNPs were significantly associated with TG (P < 5 × 10⁻⁸), while 60 SNPs were significantly associated with HDL-C (P < 5 × 10⁻⁸), and these variants were used as IVs for the analysis (22).

Heritability and Genetic Correlation

To estimate SNP-based heritability and genetic correlation, we conducted linkage disequilibrium score regression (LDSC) analysis (14). These methods allowed us to utilize the summary statistics of each disease and obtain insights into the heritability of SNPs as well as the genetic correlation between the diseases. By employing these analytical approaches, we aimed to gain a better understanding of the genetic components and potential shared genetic factors underlying the diseases.

Cross-Trait Meta-Analysis

To identify shared SNPs with strong signals among multiple traits, we performed a meta-analysis using the Multi-Trait Analysis of GWAS (MTAG) methodology (23). This approach assumes equal SNP heritability for each trait and perfect genetic covariance between the traits. Additionally, we calculated the upper bound for the false discovery rate (maxFDR) to assess the assumption of equal variance-covariance of shared SNP effect sizes across the traits. By employing the MTAG framework, we aimed to uncover common genetic variants that contribute to the observed associations among the traits of interest.

Co-Localization Analysis

To identify potentially functional genes that share SNP associations between MetS and OA, we conducted co-localization analysis using summary-data-based Mendelian randomization (SMR) (24). SMR is a method that integrates summary statistics from GWASs and expression quantitative trait loci (eQTL) studies to examine the relationship between gene expression (exposure) and a target phenotype (outcome) using genome-wide significant SNPs as IVs. The SMR approach includes the HEIDI-outlier test, which helps differentiate between causality or pleiotropy and linkage. We utilized cis-eQTL data from GTEx V8, specifically from whole blood samples (25). P values obtained from the SMR analysis were corrected for false discovery rate (FDR) if P_FDR < .05, and P_HEIDI > .05 was considered statistically significant.

Construction of Protein-Protein Interaction Network and Functional Analysis

We employed the Search Tool for the Retrieval of Interacting Genes (STRING) version 11.5 (26), a comprehensive database that provides information on protein interactions, including direct physical interactions and indirect functional correlations. STRING encompasses data from over 14 000 species, more than 60 million proteins, and over 20 billion interactions. Using an interaction score threshold of > 0.4, we constructed the protein-protein interaction (PPI) network for the identified shared gene using STRING. To visualize the PPI network, we utilized Cytoscape software version 3.9.0 (27). Additionally, we conducted a gene enrichment analysis using Metascape (28). Adjusted P value < .05 was considered statistically significant.

MR Analysis

We employed 4 methods to investigate potential causal relationships between MetS and OA: inverse variance weighting (IVW) (29), MR Egger (30), weighted median (31), and weighted mode (32). Consistent evidence for causality using all MR methods was considered more reliable and noteworthy. The “TwoSampleMR” package was used for MR analysis with these methods. IVW assumes that uncorrelated pleiotropy has a mean of zero, thus introducing noise only to the regression of meta-analyzed SNP effects with multiplicative random effects. On the other hand, MR Egger allows for directional uncorrelated pleiotropy by incorporating an intercept into the IVW regression to control for confounding. The weighted median and weighted mode methods are capable of handling some correlated pleiotropy. The weighted median calculates the weighted median of the SNP ratio and can detect true causality even when up to 50% of the weights originate from invalid SNPs. Similarly, the weighted mode groups SNPs based on their estimated causal effects and evaluates evidence for causality using only the largest set of SNPs, relaxing the assumptions of MR and identifying genuine effects even when a majority of instruments are invalid SNPs.

We used SNPs associated with each risk factor as IVs in MR analyses. Bidirectional MR analyses were conducted to explore the potential causal effect between OA and MetS, and its components. As our summary statistics originate solely from European populations, the likelihood of sample overlap cannot be ruled out. To address the potential bias arising from the inability to directly compute the sample overlap rate, we utilized the MRlap to adjust the IVW results (33). The Bonferroni-corrected P value < .004 (0.05/12 = 0.004) was applied for determining statistical significance in MR analysis. All statistical analyses were performed using the R project (version 4.2.0).

Statistical Analysis

We first applied LDSC with GWAS summary statistics and LD scores to calculate the single phenotype and genetic correlation between the 2 phenotypes (14). To validate our results, genetic associations between components of MetS and OA were also analyzed. Then, we implemented a cross-trait meta-analysis of GWAS summary statistics using the MTAG to identify the shared risk variants associated with MetS and OA (23). We used SMR was used to identify possible functional genes for MetS and OA (24). Finally, we constructed PPI networks and gene co-expression networks (26). Additionally, in MR analysis, we used 4 methods to detect the putative causal relationships between MetS and OA. Approval from the local ethical committees of the different participating centers and informed written consent from all participants were obtained following the tenets of the Declaration of Helsinki.

Results

Genetic Relationship Between MetS and OA

LDSC analysis was conducted to estimate the SNP heritability on the liability scale for MetS and OA, yielding estimates of 9.3% and 7.1%, respectively. SNP heritability estimates for the 5 components of MetS were ranging from 5.1% to 24.4% (Supplementary Table S4 (18)).

Significantly, a genetic correlation was observed between MetS and OA (r_g = 0.393, P = 1.52 × 10⁻¹⁸), which was consistently observed for 4 MetS components: WC, TG, hypertension, and HDL-C, with genetic correlations ranging from −0.229 to 0.490. However, no significant genetic correlation was found between FBG and OA (r_g = −0.048, P = .325) (Table 1).

Cross-Trait GWAS Meta-Analysis

We identified 32 SNPs associated with the joint phenotype of MetS-OA, with significant associations at a meta-analysis level (P_meta < 5 × 10⁻⁸) and a suggestive level for individual traits (single trait P < .05). Notably, among these SNPs, we discovered one novel locus (rs731988) associated with the joint phenotype of MetS-OA, which has not been previously reported in GWAS studies on MetS or OA (Table 2). The maxFDR values, approximately 4.51 × 10⁻⁷, suggest that any bias resulting from the violation of MTAG assumptions is likely to be negligible (Supplementary Table S5 (18)).

Co-Localization Analysis, Gene-Based Analysis, and Functional Analysis

We identified 12 genes that are shared between MetS and OA, meeting the significance threshold (P_FDR < .05, P_HEIDI > .05) (Table 3). Among those 12 shared risk genes, 2 novel genes, INF80E and MAPK3, were shared by OA and MetS (Table 3)

To analyze the PPI network and explore the functional implications of the identified genes, we utilized STRING and Cytoscape. The PPI network constructed through STRING consisted of 21 nodes, 127 edges for INO80E, and 21 nodes, 91 edges for MAPK3. All the PPI enrichment P value was found to be lower than 5.35 × 10⁻¹⁴, indicating a significant enrichment of interactions among the identified genes (Supplementary Fig. S1 (18)). Additionally, the pathway enrichment analysis revealed that the 2 newly identified putative functional genes are enriched in multiple pathways, including positive regulation of telomere maintenance in response to DNA damage, MAPK signaling pathway, regulation of protein stability, and activation of immune response, among others (Supplementary Fig. S2 (18)).

The Causal Relationship of MetS and OA

We observed that MetS was initially associated with an increased risk of OA using the IVW method (OR = 1.131, 95% CI = 1.045-1.224, P = .002) (Fig. 1 and Supplementary Table S6 (18)). However, this association did not persist when using the weighted median (odds ratio [OR] = 1.060, 95% CI = 0.915-1.227, P = .440), weighted mode (OR = 1.072, 95% CI = 0.974-1.180, P = .153), and MR Egger (OR = 1.105, 95% CI = 1.008-1.211, P = .036) methods. Additionally, we did not observe any significant associations between the 5 components of MetS and the risk of OA using any of the 4 methods.

In the reverse-directional MR analysis, we found no evidence to suggest that OA has a causal effect on the development of MetS (IVW OR = 0.996, 95% CI = 0.932-1.065, P = .910). Furthermore, except for FBG, we found no causal association between the other 4 components of MetS and OA.

The results of the MRlap analyses are presented in Supplementary Table S7 (18). The causal effects of OA on the MetS, as determined by the MRlap correction, align with those obtained through the primary MR analyses, accounting for potential sample overlap. However, after MRlap correction, the causal relationship between WC and OA changed, and WC increased the risk of OA. This suggests that there may be potential sample overlap between WC and OA.

Discussion

In our study, we utilized large-scale GWAS summary data of MetS, OA, and its 5 components to estimate the shared genetic etiology and causal relationships between MetS and OA. We discovered a positive genetic correlation between OA and MetS and identified shared genetic loci. Our study enhances our understanding of the genetic contribution to MetS and OA, elucidates the overlapping etiology and processes of their co-occurrence, and provides an initial insight into the potential regulatory function of shared genetic factors, which warrants further investigation.

Based on our findings, the phenotypic association between MetS and OA may be attributed to the shared genetic susceptibility underlying these 2 conditions. At the whole-genome level, we employed LDSC to establish significant genetic correlations between OA and MetS. In addition, our study revealed positive correlations between OA and WC, hypertension, TG, and HDL-C. Our results align with previous epidemiological studies that observed a bidirectional association between MetS and OA (34). Earlier epidemiological research also identified bidirectional associations between OA and WC, hypertension, TG, and HDL-C (35, 36), consistent with our findings. Previous studies have also reported a bidirectional association between OA and FBG (35, 37), whereas our study found no association between OA and FBG. The possible reason for this discrepancy could be the limited sample size of the GWAS summary data we used, which may not have provided sufficient statistical power to detect small causal effects. In cases with a small sample size, even if a true causal relationship exists, it may not be statistically significant. Furthermore, this study investigates the genetic association between OA and FBG based on GWAS summary data. However, previous studies using macro-level data might have been influenced by lifestyle and environmental factors, potentially contributing to the observed association between OA and FBG. Therefore, future larger-scale GWAS studies are needed to investigate the association between OA and FBG.

In this study, MR analysis provided support for MetS increasing the risk of OA, suggesting that the observed association between these 2 diseases may not solely be driven by shared genetic factors but could also involve a causal effect. Additionally, MR analysis showed no evidence of a causal relationship between OA and several MetS components, including WC, hypertension, TG, and HDL. This is consistent with previous findings from MR analysis (38). However, it revealed a protective effect of OA on elevated FBG, in contrast to previous studies reporting an increased risk of FBG with OA (35). The discrepancy in the FBG results may be attributed to the inclusion of 30% non-European populations in the GWAS summary data used for FBG analysis. To gain more comprehensive insights into the relationship between FBG and OA, larger-scale GWAS data from European populations are needed in future investigations.

MetS and OA often co-occur in individuals. To effectively treat and prevent these conditions, it is insufficient to simply understand their commonalities on a surface level. This necessitates researchers to delve deeper into the underlying mechanisms of these diseases. Although the epidemiological and pathological characteristics shared by MetS and OA have been reported, the understanding of their co-occurrence is still limited, and the explanations are not yet fully elucidated, especially at the genetic level. As more studies unravel the genetic factors influencing these 2 diseases, investigating the genes and pleiotropy may contribute to a better understanding of how genetic variations contribute to the development and association of these diseases.

In this context, we utilized a cross-trait GWAS meta-analysis and co-localization analysis to identify several functional genes shared between MetS and OA. Among these shared genes, some have been previously reported to be associated with either MetS or OA (PABPC4, CEP68, RBM6, TMEM258, RPAP1, JMJD7, and CLEC18A). Additionally, we identified 2 potential novel putative functional genes, INO80E and MAPK3, which have not been previously reported in the context of MetS or OA. INO80E, also known as coiled-coil domain-containing protein 95, belongs to the INO80 subfamily of chromatin remodeling complexes and is involved in various crucial cellular functions, including transcription regulation, DNA replication and repair, telomere maintenance, and chromosome segregation (39). Previous research suggests that the INO80 chromatin remodeling complex is a key regulator of metabolic functions (40). INO80 modulates the activity of immune-related genes by regulating chromatin structure, thereby influencing gene transcription and expression (41). Furthermore, INO80 is also associated with transcription and DNA repair (39). Given the close relationship between DNA repair and the immune response, INO80 might indirectly impact immune function through DNA repair mechanisms (42). Another study implicated INO80 in immunoregulation (43). The INO80 complex influences histone acetylation, which is closely associated with metabolic states. Disruption of histone acetylation might lead to the onset of metabolic disorders (44). Collectively, INO80E may contribute to the development of OA and MetS through the regulation of metabolism and immune response. Furthermore, studies have shown that the INO80 chromatin remodeling complex is essential for embryonic stem cell self-renewal, reprogramming, and embryonic development (45). Interestingly, a study demonstrated that embryonic stem cell–based therapy could be utilized for the treatment of OA (46). MAPK3, also known as extracellular signal-regulated kinase 1 (ERK1), has been shown to mediate intracellular signaling, promote inflammatory response and cartilage destruction, and act as a regulator of gene expression mediated by cell growth, inflammation, chondrocyte, and osteoblast receptors (47). Additionally, the inactivation of MAPK3 may promote or inhibit the development of obesity and insulin resistance (48, 49). Liver-specific loss of ERK1 improves systemic insulin and glucose tolerance (50). In the hypothalamus, ERK1 can inhibit appetite and promote peripheral energy expenditure (51, 52). By delving deeper into the functional aspects and interactions of these genetic loci, we can enhance our comprehension of the link between OA and MetS, facilitating their application in personalized medicine. This will enable the provision of more effective, accurate, and individualized treatment strategies to patients. Moreover, the identification of shared genetic loci may present opportunities for potential therapeutic targeting, underscoring the importance of further research into the functions and regulatory mechanisms of these loci. This could open up new avenues for the development of innovative medications or interventions.

Genetic correlations have been estimated across MetS and OA, and functional analyses can be constructed that could improve the power to describe the shared biological etiology of MetS and OA. Functional enrichment analysis reveals the association of the MAPK signaling pathway with both MetS and OA. The MAPK signaling pathway consists of 4 distinct cascades, including ERK1/2, Jun amino-terminal kinases (JNK1/2/3), p38-MAPK, and ERK5. It has been reported that the MAPK signaling pathway can regulate pro-inflammatory cytokines by controlling mRNA levels (53). Inhibition of the MAPK signaling pathway has shown anti-inflammatory and antioxidant effects and improved prognosis in cancer patients (54, 55). In animal experiments, tangeretin was found to suppress OA development by downregulating the activation of NF-κB through the activation of the Nrf2/HO-1 axis and suppressing the MAPK signaling pathway (56). Furthermore, in rat models, spironolactone prevented diet-induced MetS by inhibiting the PI3-K/Akt and p38 MAPK signaling pathways (57). These findings from pathway enrichment analysis provide important clues for further investigating the pathological and physiological mechanisms of MetS and OA. Dysregulation of these pathways plays a crucial role in the onset and progression of both MetS and OA, offering potential targets for future therapeutic strategies. Further experimental validation and clinical research are needed to confirm the roles of these pathways in MetS and OA and explore their potential therapeutic value.

Our study has certain limitations. Firstly, our study focused only on individuals of European ancestry, as the current large-scale GWAS for MetS and OA is primarily applicable to European populations. Further investigations are needed to explore the genetic architecture of MetS and OA in other populations. Secondly, rare variant genetic associations could not be assessed since SNPs with minor allele frequency (MAF) less than 0.01 were automatically filtered out in the MTAG analysis. Thirdly, the GWAS summary data for FBG were derived from trans-ethnic GWAS meta-analysis. Genetic ancestry and potential population stratification may bias our results, despite the majority of GWAS samples being of European ancestry. Additionally, there might be potential sample overlap in the GWAS summary data between WC and OA. We aim to conduct further analyses if updated data becomes available.

Conclusions

Our findings suggest a common genetic architecture, pleiotropic loci, as well as causality between MetS and OA, highlighting an intrinsic link underlying the 2 complex diseases.

Acknowledgments

We would like to thank the GWAS Catalog for providing us with the OA, MetS, and FBG GWAS summary data; and the MRC-IEU OpenGWAS database for providing the WC, hypertension, TG, and HDL-C GWAS summary data.

Funding

This study was funded by grants from the National Natural Science Foundation of China (82273710, 81872687), Anhui Provincial Natural Science Foundation (2108085Y26, 2108085QH361) and Research Fund of Anhui Institute of Translational Medicine (2021zhyx-B04).

Author Contributions

H.F.P. and J.N. take responsibility for the content of the manuscript, including the conception and design of the study and final approval of the version to be submitted. J.X.H. and S.Z.X. contributed to the acquisition, analysis, and interpretation of the data and wrote the manuscript. T.T., J.W., and T.H. contributed to the analysis and interpretation of the data; S.Y.M. and L.Q.J. contributed to the statistical expertise; and J.N. contributed to the conception and design.

Disclosures

None.

Data Availability

MetS GWAS summary statistics can be downloaded from https://www.ebi.ac.uk/gwas/downloads/summary-statistics. OA GWAS summary statistics can be downloaded from https://www.ebi.ac.uk/gwas/publications/29559693. FBG GWAS summary statistics can be downloaded from https://magicinvestigators.org/. WC, hypertension, TG and HDL-C can be downloaded from https://gwas.mrcieu.ac.uk/. GTEx eQTL summary statistics can be downloaded from https://www.gtexportal.org/home/.

Code Availability

The following software packages were used for data analyses: LD score regression software: https://github.com/bulik/ldsc; MTAG: https://github.com/JonJala/mtag; Summary-data-based Mendelian Randomization: (SMR): https://yanglab.westlake.edu.cn/software/smr/#eQTLsummarydata; Mendelian Randomization (MR): https://cran.r-project.org/web/packages/MendelianRandomization/index.html; Metascape: https://metascape.org/.

References

Abbreviations

 
• eQTL

  expression quantitative trait loci

 
• ERK1

  extracellular signal-regulated kinase 1

 
• FDR

  false discovery rate

 
• GWAS

  genome-wide association study

 
• FBG

  fasting blood glucose

 
• HDL-C

  high-density lipoprotein cholesterol

 
• ICD

  International Classification of Disease

 
• IV

  instrumental variable

 
• IVW

  inverse variance weighting

 
• LDL-C

  low-density lipoprotein cholesterol

 
• LDSC

  linkage disequilibrium score

 
• MetS

  metabolic syndrome

 
• MR

  Mendelian randomization

 
• MTAG

  Multi-Trait Analysis of GWAS

 
• OA

  osteoarthritis

 
• OR

  odds ratio

 
• PPI

  protein-protein interaction

 
• SMR

  summary-data-based Mendelian randomization

 
• SNP

  single nucleotide polymorphism

 
• STRING

  Search Tool for the Retrieval of Interacting Genes

 
• TG

  triglycerides

 
• WC

  waist circumference

Author notes