Non-causal Effect of Circulating Vitamin D Levels on the Risk of Rheumatoid Arthritis: A Two-sample Mendelian Randomization Study

Background: Recently, many studies have indicated the potential roles of vitamin D in rheumatoid arthritis (RA). In the present study, the published available GWAS summary data was used to perform two-sample Mendelian randomization (MR) to infer the causal effect between circulating vitamin D levels and RA risk. Methods: Single nucleotide polymorphisms (SNPs) which signicantly associated with exposure were selected as instrumental variables from larger-scale genome-wide association study (GWAS). The robust analytical methods including MR Egger, inverse variance weighted (IVW), weighted median and weighted mode were conducted to infer the causal links. Results: The IVW method suggested that there was no causal relationship of genetically predicted circulating vitamin D on RA (Asian: odds ratio (OR)=0.911, 95% condence interval (CI), 0.542-1.534, P=0.727; European: OR=0.897, 95% CI, 0.633-1.269, P=0.538). MR Egger (Asian: OR=0.937, 95% CI, 0.373-2.352, P=0.895; European: OR=0.853, 95% CI, 0.469-1.553, P=0.639), weighted median (Asian: OR=0.895, 95% CI, 0.499-1.606, P=0.711; European: OR=0.885, 95% CI, 0.620-1.264, P=0.503) and weighted mode (Asian: OR=0.904, 95% CI, 0.515-1.588, P=0.738; European: OR=0.898, 95% CI, 0.618-1.306, P=0.604) demonstrated that vitamin D was not directly related to RA as well. Conclusions: The MR analysis indicates that there is no evidence of causal effect of genetically predicted circulating vitamin D levels on RA risk.


Introduction
Rheumatoid arthritis (RA) is a systemic chronic autoimmune disease, mainly manifested by the proliferation of synovial lining cells and the erosion of cartilage and bone tissue. RA appears mostly in women and may occur at any age, with a peak incidence between 50 and 60 years old [1]. In general, progressive disability and shortened life expectancy are almost universal in RA patients [2], furthermore, physical pain also seriously affects the quality of life [3]. The exact pathogenesis of RA is not yet known [4], but it is believed that environmental, genetic, and infectious factors may interact in susceptible individuals and the heritability of RA is approximately 65% [5].
Vitamin D is a fat-soluble steroid that with the function of promoting the absorption of calcium and phosphorus, mainly including vitamin D2 and D3. It is generally believed that vitamin D is closely related to calcium homeostasis and bone formation. In addition, vitamin D has the potential to modulate immune cells, particularly T lymphocytes [6]. Vitamin D de ciency is common in autoimmune rheumatic diseases [7] which is accepted to be an established risk factor of numerous autoimmune diseases [8]. Vitamin D is involved in innate and adaptive immune responses and therefore insu cient vitamin D levels may be related to the loss of immune tolerance [9,10]. The inherent anti-in ammatory functions of vitamin D highlight its potential therapeutic value in autoimmune diseases [11][12][13]. However, no consensus is reached on whether there are causal effects between vitamin D and RA. Two-sample Mendelian randomization (MR) is a technique which integrates large-scale genome-wide association study (GWAS) summary data, and single nucleotide polymorphisms (SNPs) signi cantly associated with exposure are used as instrumental variables to infer the causal links between exposure and outcome. MR method must follow the three assumptions. First, instrumental variables are independent of confounding factors. Second, instrumental variables are signi cantly related to exposure. Third, instrumental variables affect the outcome only through exposure [14]. The advantage of MR is that the causal inference is not interfered by common confounding factors such as environment and lifestyle.
Given the current results of the association between vitamin D and RA was inconsistent, and the causal effect between vitamin D and RA had not yet been established, the two-sample MR was performed to infer whether there were causal effects of circulating vitamin D levels on RA risk.

GWAS summary data
Summary data for the circulating vitamin D levels were selected form a larger-scale GWAS analysis, including 79,366 European ancestry [15]. The GWAS summary data for RA were obtained from a larger dataset with 19,234 individuals (4,873 Asian and 14,361 European ancestry) [16]. The selection of large sample GWAS was conducive to obtaining more comprehensive and reliable results. The corresponding information of SNPs including effect allele, other allele, effect sizes, standard error and effect allele frequency (EAF) was collected.

Instrumental variables selection
The following quality control steps were used to select the optimal instrumental variables to avoid the impact of bias on the results. First, SNPs signi cantly associated with exposure were served as potential instrumental variables (P<5×10 -8 ). Second, the clumping process (R 2 <0.001, clumping distance=10,000 kb) was performed to avoid the underlying bias caused by strong linkage disequilibrium (LD). Third, the SNPs which with minor allele frequency (MAF) less than 0.01 were excluded from the present study. Fourthly, if the information of vitamin D related SNPs was not present in the outcome GWAS, the proxy SNPs with high LD (r 2 >0.80) were chosen to replace the variants of interest. Finally, to ensure that the effect of the selected SNPs on circulating vitamin D levels and the effect of the same SNPs on the outcomes corresponded to the same allele, the palindromic SNPs were excluded.
One of the assumptions of MR analysis is that the included SNPs must be highly correlated with exposure. In the present study, the F statistic (F=R 2 (n−k−1)/k(1−R 2 )) was used to explore whether there were instrumental variables that were weakly correlated with exposure.
In this equation, R 2 represents the cumulative explained variance of the selected instrumental variables on circulating vitamin D levels, k is the number of selected instrumental variables and n represents the sample size. When F>10, the correlation between instrumental variables and exposure is considered to avoid the bias caused by weak instrumental variables.

Statistical analysis
In this study, MR methods with robust testing power including inverse variance weighted (IVW), MR Egger, weighted median and weighted mode were used to infer the causal effects of vitamin D with RA. The IVW conducts a meta-analysis approach to divide the β coe cient of the SNPs of RA by the β coe cient of the SNPs of vitamin D to obtain an overall estimate of the effect of vitamin D on RA [17]. In the absence of horizontal pleiotropy, or when the horizontal pleiotropy is balanced, IVW linear regression enable to obtain unbiased causal estimates [18]. Directional pleiotropic test, causal effect test and causal effect estimation are the three components of MR Egger [19]. Under the Instrument Strength Independent of Direct Effect (InSIDE) condition, even if all instrumental variables are invalid, MR Egger can also give a consistent causal effect estimate [20]. Weighted median is suitable for the situations where up to 50% of the information comes from invalid instrumental variables. If most of the instrumental variables in the model do not meet the requirements of using MR for causal inference, the weighted mode approach is effective [21]. Weighted mode is usually used as a supplementary analysis. MR-Egger regression was conducted to test whether the instrumental variables had potential pleiotropic effects. Cochran Q statistic was performed to quantify the heterogeneity among selected SNPs. Leave-one-out sensitivity analysis was used to test whether the overall estimates were affected by single SNP.
The TwoSampleMR package in R software (version 4.0.3) was used to conduct statistical analysis.

Genetic variants selection
In the analysis of Asian population, after the SNPs with strong LD were excluded, 8 SNPs were selected as potential instrumental variables. After excluding a palindromic SNP (rs8018720), 7 variants of interest were selected as instrumental variables (Table 1). For European population, 5 SNPs were served as instrumental variables after excluding a palindromic SNP (rs8018720) ( Table 1). The remained SNPs explained 2.2% (Asian) and 1.8% (European) of the variance of circulating vitamin D levels. F statistics were greater than 10, suggesting weak instrumental variables were less likely to exist.

Two-sample MR analysis
For the Asian population, IVW result indicated that there was no causal effect between circulating vitamin D levels and the risk of RA (odds ratio (OR)=0.911, 95% con dence interval (CI), 0.542-1.534, P=0.727) (  (Fig. 1a). Leave-one-out sensitivity analysis showed that the results were reliable and stable (Fig. 2a). The results of MR-Egger regression showed that there was no horizontal pleiotropy between instrumental variables and RA (P=0.946) (Fig. 3a). Heterogeneity tests demonstrated that there was no signi cant heterogeneity exist (P=0.914) (Fig. 4a).

Discussion
In the current study, the two-sample MR was performed to infer the potential causal links between vitamin D and RA. The results demonstrated that there was no causal effect of genetically predicted circulating vitamin D on RA risk.
In addition to its inherent functions of regulating calcium and phosphorus metabolism and promoting bone development, the immunomodulatory properties of vitamin D have also been extensively studied. However, existing evidence on the association between vitamin D and RA was contradictory. A study found that the serum 25OH vitamin D level of early RA patients was signi cantly lower than that of healthy controls [22]. A study focusing on serum and synovial uid vitamin D metabolites in RA patients found no signi cant difference in concentrations of 25(OH)D3, 24,25(OH)2D3, 1,25(OH)2D3 and 25(OH)D2 when compared with healthy controls, except for 3epi-25(OH)D3 [23]. Beyer et al. suggested that most RA patients did not suffer from hypovitaminosis D and it was not related to the differences in the severity of RA as well [24]. However, a recent study showed that vitamin D de ciency was associated with RA activity and severity, and might predict the disability and imaging progression of early RA patients within 1 year [25]. There were some researches on the vitamin D-related gene polymorphisms and RA. A study demonstrated that the serum vitamin D level of RA patients might be normal, but the polymorphisms of the vitamin D receptor (VDR) gene restricted the anti-in ammatory effect of vitamin D by changing the 1,25(OH)2 D3 binding site [26]. A meta-analysis found an association between VDR gene polymorphisms and RA susceptibility [27]. However, the study pointed out that some mutations would make carriers susceptible to RA, while others would reduce the susceptibility of RA.
The reasons for the inconsistent conclusions might be that the sample size was different and the studies with small sample size had a high contingency. In addition, the circulating vitamin D levels of participants were closely related to their eating habits and lifestyles, therefore, it was unreliable to estimate the association between vitamin D and RA without adjusting these confounding factors. However, some data indicated that hypovitaminosis D status was related to the progression of RA and vitamin D supplementation might alleviate the clinical symptoms of RA [28]. A study showed that treatment with vitamin D supplementation in RA patients with secondary osteoporosis could effectively increase bone mineral density [29]. Nevertheless, a study involving 1180 individuals indicated that vitamin D supplementation did not provide other bene ts for anti-rheumatic therapy [30]. The speci c roles of vitamin D in the pathology of RA and whether vitamin D supplementation is effective remains to be con rmed in the future.
In general, there are several limitations in the present study should be noticed. First, since the absence of clinical parameters, and disease activity information in original GWAS, further subgroup analysis was impossible. Second, our study was unable to determine whether overlapping participants were involved in the exposure and outcome GWAS used in the two sample MR analyses. Nevertheless, the deviation from participants overlap could be minimized by the F statistic [31].
In summary, the two-sample MR analysis suggest that there is no causal effect between circulating vitamin D levels and RA risk. However, further studies should utilize updated data from large-scale GWAS to obtain more reliable results.