Treatment of Metformin Decreases the Risk of Varicose Vein: A Mendelian Randomization Study

Background Varicose vein is a common illness of the vascular system which affects life quality and social function of patients. We aimed to assess the causality between metformin and varicose vein using a two-sample Mendelian randomization(cid:0)MR(cid:0)analysis based on genome-wide association study (GWAS) summary data. Methods Twenty-ve single nucleotide polymorphisms (SNPs) were selected from the GWAS summary data from Neale Lab and MRC-IEU Consortium available on the MR-base platform. Inverse variance weighted (IVW), MR-egger method, weighted median method and weighted mode were adopted. Results were evaluated by pleiotropy test using an Egger regression method and sensitivity analysis preforming a leaving-one-out (LOO) method. Analyses were performed using R package “TwoSampleMR”. Results The result of IVW method showed that one SD increased treatment of metformin was linked with approximately 10% lower risk of varicose vein (OR,0.90; 95%CI, 0.85-0.96; P, 6.99e-04). It was similar to that measured by other methods in the aspect of effect size and direction. There is no evidence to supporting genetic pleiotropy in the MR-Egger regression method (intercept=2.5e-04, P=0.33). No single SNP was detected to be strongly driving the overall causal effect in a LOO sensitivity analysis. The genetically predicted treatment of metformin was negatively casually associated with varicose veins. Conclusions This study suggested that treatment of metformin was a casual protective factor of varicose vein. Further researches are required to conrm our ndings and explore the potential mechanisms of metformin on varicose vein.


Introduction
As one of the most illnesses of the vascular system, varicose vein is a syndrome of super cial venous hypertension and varices due to venous blood re ux, mainly caused by increased venous pressure, weak vein wall and insu ciency venous valves, usually occurs in the lower extremities [1][2]. Varicose vein is characterized by pigment, desquamation, itching and edema of ankle. If untreated, it will lead to venous ulceration and venous thrombosis, which seriously affects life quality of patients as well as aggravates heavy society and economic burdens [3]. Advanced age, female and obesity are important risk factors for varicose vein [4][5][6]. Nowadays, instead of effective medical therapy, minimally surgery is the most widely used method in clinical practice, but leaves longterm problems of reopening [3]. In the molecular level, disturbances of transforming growth factor-beta one (TGF-β1) signaling pathway may contribute to varicose vein [7]. Therefore, researches are still required to improve the treatment on varicose vein.
Metformin is a classic oral anti-diabetic drug which has been clinically used to treat the patient with type 2 diabetes for over 50 years [8]. Recently, in addition to the glucose-lowering effect, some studies indicate its all-round bene cial impact including diabetes prevention, obese treatment, polycystic ovary disease treatment, anticancer effect, antiaging effect, cardioprotective effect and so on [9][10][11]. There are reports that the mechanism of metformin action might be signi cantly lessening the level of TGF-β1 expression through AMPK activation [12][13][14]. There was a retrospective study published in 2019 showed that the use of metformin decreased the incidence of varicose vein in people with type 2 diabetes [15]. However, observational studies are easily affected by confounding factors and reverse causality, so that it is di cult to con rm whether there is causal association between metformin and varicose vein [16].
Mendelian randomization (MR) is a new method which considers genetic variants as instrumental variables for exposures in order to evaluate whether there are causal associations between exposures and disease outcomes [17]. Since genetic variants are randomly distributed in the population and preserve stability over time, MR is highly similar to the Randomized Controlled Trials (RCTs) which escape from confounding factors and reverse causality [18].
Moreover, it has the advantage of cost and time saving when compared with RCTs. In recent years, the rapid development of Genome-wide association study (GWAS) supplies great convenience to MR, making GWAS-based two-sample MR an e cient and feasible approach [19][20].
In this study,we intended to investigate a potentially causal association between metformin and varicose vein using MR and estimate the range of effects.

Study design
This study applied two-sample MR analysis to estimate the causal association between metformin and varicose vein. The institutional review board approved the MR study due to the data from summary GWAS study and without clinical intervention to the patients. The two-sample MR study, which allows the association data of genetic variants -exposures and genetic variants -disease outcomes to come from two independent samples of the same population, greatly improving the e ciency and feasibility of etiology researches, has been widely used to identify causal associations between exposures and disease outcomes [21]. It is based on three hypotheses: (1) genetic variants is robustly related to exposures; (2) genetic variants is independent of confounding factors; (3) genetic variants affects disease outcome only through exposure factors [22]. In the study, several single nucleotide polymorphisms (SNPs) were selected as instrumental variables from the GWAS of metformin and varicose vein obtained through MR-base platform. The F statistics and Q statistics were respectively calculated to ensure small bias caused by weak instruments and low heterogeneity. Four methods of Mendelian randomization, including inverse variance weighted (IVW), weighted median method, weighted mode and MR-egger method were adopted to assess the causal association between metformin and varicose vein [23][24]. Eventually, results were evaluated by pleiotropy test using an Egger regression method and sensitivity analysis preforming a leaving-one-out method.

Sources of data
We mainly used the GWAS summary data on metformin from Neale Lab Consortium in 2017 that was included in UK controls, all of whom were Europeans. The GWAS summary data on varicose vein was developed by MRC-IEU consortium in 2018 which with 18,818 cases and 444,115 controls rooted Europe ( Table 1). The above GWAS summary data was available on the MR-base platform (http://www.mrbase.org). The MRbase platform was a database which integrates GWAS data with software that automates 2-sample MR, making MR analysis more rigorous and convenient [25].  [27]. One SNP (rs649698) was gotten rid of MR analysis for strand ambiguity. The rest 25 SNPs explained 0.49% of the variations in using of metformin across individuals.
The F statistic calculated was more than 10, which was generally believed to reduce bias caused by weak instruments  Table 1).

Statistical analyses
In this study, screening out for independent SNPs was completed in MR-base platform, while two-sample MR analysis was preformed to assess potential associations between metformin and varicose vein using R (version 3.4.2) package "TwoSampleMR" (version 0.3.4) to achieve automated operations [29]. Four Mendelian randomization method incorporating IVW approach, weighted median method, weighted mode, and MR-Egger method were resorted to explore causal effects of metformin for varicose veins. Results of MR of were expressed as odds ratios (OR) and con dence interval (CI). Forest plot and scatter plot showed the results more intuitively conducted by R. Differences were considered to be statistically signi cant if P-value < 0.05.
We implemented a pleiotropy test preforming Egger regression method to test whether there was a violation of hypothesis of MR. Sensitivity analysis was adopted using a leave one out (LOO) method in which one SNP was removed in turn to estimate causal effects of the rest by IVW method for assess whether individual SNP affected MR estimates.

Results
The MR estimates of the four methods including the IVW approach, weighted median method, weighted mode, and MR-Egger method were shown as OR and 95% CI in the  Fig. 1. Causal estimates of two SNPs (rs34872471 and rs76895963) showed clear negative correlations between treatment of metformin and risk of varicose veins. Figure 2 revealed the regression curve of causal effect between using of metformin and risk of varicose vein. The casual association estimated by MR-Egger method, weighted median method and weighted mode method were consistent to that by IVW method, no matter in the aspect of effect size or direction ( Table 2, Fig. 1 and Fig. 2). In this study, there is no evidence to supporting directional horizontal pleiotropy in the MR-Egger regression method (intercept = 2.5e-04, P = 0.33). No single SNP was detected to be strongly driving the all causal effect of metformin on risk of varicose vein in a LOO sensitivity analysis. The more detail information about LOO sensitivity analysis was showed in the Fig. 3 and appendix Table 2.

Discussion
Varicose vein is a common clinical illness that limits physical and social activities of patients. Stretch socks, harden agent and surgery are the main prevention and treatment measures instead of effective medical therapy at present, but leaves the problem of long-term reopening and surgery complications [3]. Metformin is an inexpensive classical but all-round bene cial drug. In this MR study, the major nding was that metformin was casually associated with varicose vein. Our results indicated that the treatment of metformin predicted a reduction in the risk of varicose vein by 10%.
Little researches were done to seek the association between treatment of metformin and risk of varicose vein. A cohort study exhibited that metformin decreased the incidence of varicose vein in type 2 diabetes patients, which was in accordance with the result in our analysis [15]. However, the truth of causal association between metformin and varicose vein without diabetes remained unknown. Concerning the unethical approaches of taking metformin in patients without diabetes, RCT is not appropriate for this issue. Mendelian randomization analysis is a new method widely used to identify causal associations in a lot of etiology researches [17]. Due to the strength of MR, this analysis is able to overcome the limitation of methodology in retrospective study, such as selection bias, indication confounder and so on. Through MR analysis, our study revealed that treatment of metformin was linked with approximately 10% lower risk of varicose veins.
The possible mechanisms of the sheltering effect of metformin against varicose veins have not been clari ed. Impaired microvascular endothelial function plays an important role in generation and progress of varicose veins [30]. The sensation of venous hypertension and valvular incompetence lead to the recruitment of in ammatory cells and the release of in ammatory mediators, which was triggered by the activation of endothelial cells. It has been well demonstrated that metformin can improve the endothelial function and relieve the in ammatory reaction [31][32][33]. In the molecular level, a research about vascular wall pathology showed TGF-β1 signaling pathway worked in development of varicose veins [7]. It is widely accepted that the mechanism of metformin action might be signi cantly lowering down the level of TGF-β1 expression through AMPK activation [12][13][14]. Thus, the protection function from metformin on varicose vein might be associated with impaired endothelial function, in ammation and valvular incompetence while further investigations were required.
This study had several great advantages. First, this was the rst MR study to investigate the casual relationship between metformin and varicose vein. Compared with RCT, it was an enormous saving of time, money and effort using MR method. Second, MR analysis escaped from the impact of reverse causation and confounding factors, which were common in conventional researches. Finally, the large sample size (over 750000) and robust instrumental variables (F statistics > 10) leaded to more credible causal estimates. However, our study also leaved much to be desired. First, two large consortiums included in this study were both European populations. It was questionable whether our ndings could apply to other populations, such as Asian race. Second, due to the limitation from the GWAS summary data, we are not able to obtain the data in the individual level, for which our result provided a linear relationship between metformin and varicose vein while the relationship might be U shape in real world. The taking dosage and frequency of metformin for individual patient were not able to assess and thus subgroup analysis were di cult to conduct. Third, through the method of MR, we were not able to investigate the possible mechanism metformin and varicose vein.

Conclusion
In this study, we found the negatively casual association between metformin and risk of varicose vein using two-sample Mendelian randomization method based on GWAS summary data. But further researches are demanded to add to the ndings and explore the potential mechanism of metformin on varicose vein.

Declarations ETHICS APPROVAL AND CONSENT TO PARTICIPATE
Not applicable.

COMPETING INTERESTS
The authors have no con ict of interests. Scatter plot of SNPs associated with metformin and the risk of varicose vein. The plot relating the effect sizes of the SNP-metformin association (x-axis, SD) and the SNP-varicose vein associations (y-axis, log (OR)) with 95% con dence intervals. The regression slopes of the lines correspond to causal estimates using four MR methods including the IVW approach, Weighted mode, Weighted median, and MR-Egger method.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.