A systematic review of Mendelian randomization studies on multiple sclerosis

Mendelian randomization (MR) is a powerful approach for assessing the causal effect of putative risk factors on an outcome, using genetic variants as instrumental variables. The methodology and application developed in the framework of MR have been dramatically improved, taking advantage of the many public genome‐wide association study (GWAS) data. The availability of summary‐level data allowed to perform numerous MR studies especially for complex diseases, pinpointing modifiable exposures causally related to increased or decreased disease risk. Multiple sclerosis (MS) is a complex multifactorial disease whose aetiology involves both genetic and non‐genetic risk factors and their interplay. Previous observational studies have revealed associations between candidate modifiable exposures and MS risk; although being prone to confounding, and reverse causation, these studies were unable to draw causal conclusions. MR analysis addresses the limitations of observational studies and allows to establish reliable and accurate causal conclusions. Here, we systematically reviewed the studies evaluating the causal effect, through MR, of genetic and non‐genetic exposures on MS risk. Among 107 papers found, only 42 were eligible for final evaluation and qualitative synthesis. We found that, above all, low vitamin D levels and high adult body mass index (BMI) appear to be uncontested risk factors for increased MS risk.


| INTRODUCTION
Multiple sclerosis (MS) is a chronic neurodegenerative multifactorial disease of the central nervous system (CNS), involving in its aetiology genetic and non-genetic factors and their complex interplay (Compston & Coles, 2008;Dobson & Giovannoni, 2019). The disease can onset at all ages of life, even before 18 years old (Alroughani & Boyko, 2018), although it is most often diagnosed between 20 and 40 years of age. MS represents the most common cause of non-traumatic neurological disability in young adults showing increasing incidence and prevalence worldwide over time (Browne et al., 2014). It is estimated to affect more than 2.8 million people, therefore also representing an important socio-economic burden, considering both direct medical costs, including those of disease-modifying therapies and rehabilitation, and indirect medical costs, including costs of productivity loss of MS patients and their caregivers (Simoens, 2022).
Within this picture lies the concrete and impelling need to find the underlying cause of the disease, in other words to find the causally associated risk factors that can ultimately lead to appropriate therapeutic choices, guide prevention strategy and lay the foundations for future precision medicine.
In the last decades, the genetic architecture of MS has been extensively investigated, identifying more than 200 variants significantly associated with its susceptibility (Baranzini et al., 2009;Fazia et al., 2018;International Multiple Sclerosis Genetics Consortium, 2013; International Multiple Sclerosis Genetics Consortium & Wellcome Trust Case Control Consortium 2, 2011) and improving the understanding of the involvement of the immune system in neuroinflammation and neurodegeneration (Cree et al., 2022). Efforts have been also made to quantify the contribution of non-genetic risk factors in supporting the multifactorial aetiology of the disease. Early-life environmental risk factors (e.g., obesity in childhood or adolescence), low vitamin D levels also related to poor sun exposure (Olsson et al., 2016;Ramagopalan et al., 2010) and Epstein-Barr virus (EBV) infection (Aloisi et al., 2023;Ascherio & Munger, 2015) have been found to be associated in different studies. Instead, no evidence of association has been found between alcohol consumption and MS risk in a recent study performed on UK Biobank data (Dreyer-Alster et al., 2022).
However, although many studies have investigated genetic and non-genetic risk factors associated with MS, it is crucial to distinguish between association and causality, two different concepts that should not be confused with each other (Altman & Krzywinski, 2015). Indeed, association between variables can arise in the presence or in the absence of a causal relationship and, in this latter case, it could be the results of reverse causation, confounding or selection bias.
Causality assessment for MS-associated risk factors represents the key for understanding disease aetiology and improving preventing and therapeutic strategies.
Mendelian randomization (MR) is a widely utilized analytical approach for causal inference when randomized controlled trials (RCTs) are not feasible especially for ethical reasons Burgess & Thompson, 2015). Under the principle of MR, genetic variants (Z), significantly associated with the exposure of interest (X), are used as instrumental variables (IVs) to assess the possible causal effect of X on the investigated outcome (Y). MR studies are accurate, reliable and robust, if three important conditions are fulfilled (see Figure 1 for the graphical representation of an MR model). Condition 1 requires Z to be associated with X, Condition 2 requires Z to be independent of unobserved and/or unknown confounding factors (U) and Condition 3 requires Z to be independent of Y, given U and X, meaning that Z cannot be connected with Y other than via X (exclusion-restriction). Condition 2 is easily defensible on grounds that the genetic variants are randomized at meiosis. This 'biological randomization' process confers MR studies similarities to RCT and guarantees robustness to confounding and reverse causation (Davey Smith & Ebrahim, 2003).
Valid IVs are thus defined based on the abovereported three key assumptions, which are plausibly met F I G U R E 1 Graphical representation of a Mendelian randomization (MR) model. Three principal MR assumptions are represented in the graph: (i) The genetic instrumental variables (Z) are strongly associated with the exposure (X); (ii) Z do not affect the outcome through the confounders (U), and this is represented by the absence of a directed arrow between Z and U; and (iii) Z do not affect the outcome (Y) directly but only indirectly via the exposure of interest, and this is represented by the absence of a direct arrow between Z and Y.
when the biological process linking Z with X is well understood. However, biological effect of genetic variants and corresponding genes is usually poorly understood. The same genetic variants can affect multiple outcomes through different biological pathways, than the one under investigation, a phenomenon known as horizontal pleiotropy (Solovieff et al., 2013). Horizontal pleiotropy represents a severe violation of the exclusion-restriction (Condition 3) because the effects of Z on Y are not exclusively mediated through X. Instead, vertical pleiotropy can occur when the genetic variant affects other traits through the risk factor of interest and is in general nonproblematic.
Notably, MR studies include multiple IVs, that is, genetic variants, and the three key assumptions must hold for each of these (Davies et al., 2018). Weak IVs that poorly predict the risk factor of interest are more likely to be pleiotropic; as a rule of thumb, these are generally identified and excluded if the F statistic obtained from the statistical model between Z and X is lower than 10. Through this IVs selection, one can be more confident about IVs validity also guaranteeing sufficient statistical power (Staiger & Stock, 1997). Violation of Conditions 2 and 3 can be also verified, other than visual inspections, through the measurements of the heterogeneity between SNPs and the identification of potential horizontal pleiotropic SNPs via, for example, the Mendelian Randomization Pleiotropy RESidual Sum and Outlier (MR-PRESSO) (Verbanck et al., 2018) and Cochran's Q statistics.
Over the last decades, the methodology and application developed in the framework of MR have been dramatically improved in the different research fields, from social to health sciences. In fact, initial applications of MR made use of a single genetic variant whose association with the exposure and with the outcome was derived from the same sample. The availability of public genomewide association study (GWAS) data has made it possible to extend the methodology to include multiple genetic variants and different non-overlapping samples, so that the genetic associations with the exposure and with the outcome can be derived from two different samples (i.e., two-sample MR). These advances have provided the benefits of (i) increasing sample size and consequently increasing statistical power; (ii) obtaining a more efficient and precise evaluation of causal effects using large-scale summary data rather than individual-level data; and (iii) avoiding winner's curse bias by reducing the impact of weak instrument bias, which, in this scenario, bias the casual estimate towards the null (Lawlor, 2016).
The aim of the present study was to systematically review and summarize the studies performed so far evaluating the causal effect of genetic and non-genetic risk factors on MS risk using a two-sample MR design.

| Included studies
We reviewed the existing literature following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines for systematic reviews and meta-analysis (Liberati et al., 2009). To retrieve the documents, we used the Scopus database and PubMed. Our search strategy, carried out on 1 February 2023, is reported in Table 1 and concerned original articles in open access that were published in English language. No limits were set as to year of publication. The search process included the removal of duplicates, the screening of titles and abstracts, and the assessment of full texts for eligibility. Two authors (TF and GNB) independently screened the full text for all the papers, as the abstracts did not allow us to obtain all the information required. Discrepancies and doubts were solved after a joint article review and discussion between authors.

| Eligibility criteria
Eligibility was determined by carefully read abstracts and full texts. Inclusion criteria were as follows: (i) having clearly specified the aim of the study, (ii) having used a two-sample MR approach to investigate the causal involvement of candidate exposures on MS risk, (iii) having used a frequentist framework and (iv) having reported a detailed description of the statistical analysis performed. Furthermore, papers focusing on paediatric MS and/or investigating DNA methylation levels as exposure have also been excluded.

| Data extraction
The following data were extracted from each eligible full text: (i) the article reference (first author's last name and publication year); (ii) exposure(s) variable(s) investigated in the MR analysis and the corresponding number of selected IVs; (iii) exposure(s) variable(s) data source(s); T A B L E 1 Searching criteria used to review the existing literature using the Scopus database.
((TITLE-ABS-KEY (mendelian AND randomization) OR TITLE-ABS-KEY (mendelian AND randomisation) AND TITLE-ABS-KEY (multiple AND sclerosis) OR TITLE-ABS-KEY (ms))) AND (LIMIT-TO (OA, 'all')) AND (LIMIT-TO (DOCTYPE, 'ar')) AND (LIMIT-TO (LANGUAGE, 'English')) AND (LIMIT-TO (SRCTYPE, 'j')) (iv) outcome data sources; (v) MR analysis technique implemented; and (vi) odds ratio (OR), 95% confidence intervals (CIs) and p value of the causal association obtained in the main MR analysis. If a paper had more than one investigated outcome, we reported only results concerning MS risk in both the tables and Section 3. For the exposures investigated in more than three studies, that is, adult body mass index (BMI) and vitamin D, a forest plot was also provided to display and compare the causal estimates obtained in the different studies and settings (i.e., when, in the same study, different numbers of IVs were selected for the MR).

| RESULTS
We initially found 107 scientific papers, and after the screening of titles and abstracts, 51 articles were excluded based on the eligibility criteria. By carefully reading each of the remaining 56 full-text articles, we found that 42 were eligible for final evaluation and qualitative synthesis (see Figure 2). Table 2 summarizes all the included 42 articles. When possible, studies were reported into specific sections/categories of risk factors (e.g., BMI and vitamin D) and were ordered by year of publication. Only summary statistics of statistically significant causal associations for the main MR analysis (e.g., inverse varianceweighted [IVW] method) were explicitly listed in the sixth column.

| Vitamin D
The increasing gradient of MS prevalence with latitude towards the poles has been historically linked to a low sunlight exposure and thus lower levels of vitamin D, which has always been considered an important MS risk factor by the scientific community (Smolders, 2011). The active form of vitamin D, 1,25-dihydroxyvitamin D 3 (1,25 (OH)2D3), acts by reducing the production of proinflammatory cytokines (e.g., IFN-γ, IL-2 and IL-17) and enhancing the secretion of anti-inflammatory cytokines (e.g., IL-4 and IL-10). As highlighted in various observational studies, high levels of vitamin D, which were partially derived from sunlight exposure, exert a protective effect versus autoimmune risk, while, on the contrary, low levels appeared to be associated with an abnormal immune response (Bivona et al., 2017 Table 3 and Section 3 for the list of genes. et al., 2018). However, although there is significant epidemiological evidence for an association between MS risk and lower vitamin D (Scazzone et al., 2020;Sintzel et al., 2018), the observational nature of epidemiological designs hinders causal inference. In fact, the disease itself could led to low sun exposure and consequently low levels of 25(OH)D (reverse causation) or, in addition, confounding factors such as obesity or others unknown could drive this association. In a recent work, Wang (2022) has investigated the causal effect of serum 25(OH)D levels on MS risk in the framework of MR. Particularly, by using UK Biobank participants, he selected 20 independent variants, not associated with potential confounders such as smoking, as IVs for the MR analysis. IVW estimate, that is, OR = .46 (95% CI: .33-.63), p < .0001, as also supported by other methods such as weighted median (WM), MR-Egger, simple mode and weighted model, suggests the causal association of increased 25(OH)D levels with decreased risk of MS.
Vandebergh, Dubois, and Goris (2022), using different sets of IVs derived from SUNLIGHT and UK Biobank data, also highlighted a causal protective effect of increasing levels of serum 25(OH)D on the risk of MS. Specifically, by including 104 independent IVs, IVW estimate was equal to .72 (95% CI: .60-.88) with p = .001. The result obtained was consistent across the sensitivity analyses performed, despite the presence of heterogeneity among variants. Further analyses excluding SNPs identified as outliers and using different exposure's data sources were also performed and slightly improved the precision of the estimate.
Two other recent papers investigated this causal effect, that is, Jiang, Ge, and Chen (2021) and Harroud, Manousaki, et al. (2021). The first, selecting different sets of IVs from a meta-analysis by Manousaki et al. (2020), has shown a statistically significant (p < .0001) WM estimate, that is, OR = .40 (95% CI: .30-.52), when using a set of 24 IVs and MS ImmunoChip as outcome data source (International Multiple Sclerosis Genetics Consortium, 2013). Results were consistent across different MR methods and in line with the hypothesis of a causal protective role of vitamin D, successfully replicating this causal relationship. The second study, by using 138 IVs retrieved from UK Biobank data, also highlighted a statistically significant (p = .0006) protective causal effect of increasing vitamin D levels, that is, OR = .72 (95% CI: .60-.87), using IVW method and with consistent results from sensitivity analyses with other methods.
Lastly, also the study from Jacobs et al. (2020) showed a statistically significant (p = .001) causal association for increasing vitamin D level in decreasing the odds of MS, OR = .57 (95% CI: .41-.81), and no evidence of significant unbalanced horizontal pleiotropy was detected. Same protective effect of vitamin D was also obtained in the study from Yuan et al. (2021), OR = .77 (95% CI: .65-.93), p = .005.
It is important to highlight that these findings, all agree in underlying the casual protective role of vitamin D on MS risk, are based on GWAS performed on adult subject sample, and in this picture, it is important to keep in mind that it is likely that the protective effect of vitamin D might be prominent during childhood. In all studies, OR is referred in 1-SD increase in natural logtransformed 25(OH)D except the work from Jacobs et al. (2020), which considered 1-unit increase in natural logtransformed 25(OH)D.
See forest plot in Figure 3 for displaying and comparing the causal estimates obtained in the different studies and settings.

| BMI
Early-life/childhood and adult BMI are other putative risk factors for MS risk, whose association has been extensively investigated throughout the scientific literature in the last years (Jacobs et al., 2021;Munger et al., 2013).
The first work performed to investigate the causal effect of adult BMI, by using a two-sample MR approach, belongs to Mokry et al. (2016) and dates back to 2016. The study, making use of 70 IVs, highlighted a significant causal risk effect of increasing BMI on MS, that is, OR = 1.41 (95% CI: 1.20-1.66) with p < .0001.
In a recent MR analysis, Harroud, Manousaki, et al. (2021) confirmed the causal contribution of increasing adult BMI levels in increasing MS risk (OR = 1.33 [95% CI: 1.09-1.63], p = .005). By testing the indirect effect of BMI on MS risk mediated by 25(OH)D levels, via a multivariable MR (MVMR) analysis, the direct effect of higher BMI on MS risk persisted (OR = 1.28 [95% CI: 1.05-1.55], p = .01). In the same study, as well as in the work from Devorak et al. (2017), no causal effect was instead detected for adiponectin, a molecule with known antiinflammatory properties in both the innate and adaptive arms of the immune system and whose levels are lower in individuals with central obesity (Arita et al., 1999).
A subsequent univariable MR analysis with 588 IVs from Vandebergh, Becelaere, et al. (2022) confirmed the hypothesis for a causal risk effect of increasing adult BMI on MS, that is, OR = 1.30 (95% CI: 1.15-1.47), p < .0001, obtaining consistent results across all MR methods; moreover, no evidence for directional pleiotropy and heterogeneity was detected, as indicated by the MR-Egger regression intercept and both Cochran Q test and I 2 statistic. The deleterious causal effect of BMI was also investigated in the study by Vandebergh and Goris (2020), OR = 1.31 (95% CI: 1.15-1.49), p < .0001, confirming the results obtained so far.
As regards to the early-life BMI, Hone et al. (2022) had examined the causal relationship between genetically determined BMI at birth, infancy, early childhood and later childhood. This study has shown the presence of a causal link between early-life BMI and MS, particularly highlighting how this effect is likely to depend on the persistence of elevated BMI through adolescence/early adulthood. In fact, IVW estimate for causal effect on MS risk resulted equal to as follows: (i) OR = 1.18 (95% CI: 1.04-1.33, p = .01) for BMI at infancy, (ii) OR = 1.31 (95% CI: 1.03-1.66, p = .03) for BMI at early childhood and (iii) OR = 1.34 (95% CI: 1.08-1.66, p = .01) for BMI at later childhood. The robustness of these results was supported by the lack of horizontal pleiotropy, as suggested by the MR-Egger intercept, as well as by the consistent MR estimates obtained using Steiger filtering (Hemani et al., 2017) and leave-one-out analysis. No evidence for an effect of BMI at birth was found, probably reflecting the fact that birthweight is majorly determined by maternal and gestational factors that have a lesser impact on BMI in later life.
Childhood BMI causal effect was also investigated in a work by ; a univariable MR was performed firstly making use of 23 IVs leading to an IVW OR = 1.26 (95% CI: 1.07-1.50), p = .006, and then on a restricted set of 15 IVs (secondary analysis) leading to an IVW OR = 1.23 (95% CI: 1.05-1.43), p = .01. The further exclusion of variants associated with potential confounders had only a slight influence on the results, that is, OR = 1.29 (95% CI: 1.12-1.49), p = .001.
In the study by , increasing childhood body size was also investigated by using 277 IVs and resulted to be statistically significant, IVW OR = 1.40 (95% CI: 1.10-1.78), p = .01. Results were robust to sensitivity analyses for pleiotropy and were consistent across all MR methods used. Interestingly, an MVMR analysis was performed to estimate the direct effects of childhood BMI adjusting for adult obesity, with a direct effect resulting weaker compared to the total effect. Also, Belbasis et al. (2020) have investigated the causal effect of childhood obesity on MS by using 15 IVs, but without finding any significant evidence for a causal effect. Jacobs et al. (2020) investigated the causal effect of both childhood and adult obesity, highlighting, for both exposures, a significant causal contribution on increasing MS risk, that is, respectively, OR = 1.24 (95% CI: 1.05-1.45), p = .01, and OR = 1.14 (95% CI: 1.01-1.309), p = .04. Being the genetic architectures of both childhood and adult BMI correlated with vitamin D levels, a further analysis was performed by correcting for 25(OH)D levels. The effect of childhood BMI on MS was slightly attenuated but remained statistically significant, OR = 1.15 (95% CI: 1.00-1.33), p = .044, while the direct effect of increasing adult BMI resulted to be strongly significant, OR = 1.19 (95% CI: 1.04-1.35), p = .009. Also, Yuan et al. (2021) had investigated the effect of childhood and adult obesity, finding for both exposures a statistically significant causal effect on MS risk, respectively, OR = 1.23 (95% CI: 1.05-1.43), p = .01, and OR = 1.27 (95% CI: 1.5-1.41), p < .0001. In all studies, OR is referred in 1-SD increase in BMI levels. See forest plot in Figure 4 for displaying and comparing the causal estimates obtained in the different studies and settings.

| Smoking
Even if it is well known that free radicals, contained in high proportion into tobacco, induce oxidative stress and are involved in many neurodegenerative disorders and autoimmune diseases (Nishanth et al., 2020), the effect of smoke on the immune system is still unclear (Pryor & Stone, 1993). Moreover, smokers are characterized by high levels of proinflammatory cytokines, C-reactive protein and other inflammatory markers, thus delineating a picture of persistent autoimmunity.
In supporting observational epidemiological studies suggesting the deleterious effect of smoking on MS risk (Degelman & Herman, 2017), Mitchell et al. (2020) investigated the causal effect of smoke on MS risk. The not statistically significant IVW estimates revealed lack of evidence for a causal effect of either smoking initiation, a measure indicating whether an individual had ever smoked regularly, or lifetime smoking, a measure indicating smoking initiation, duration, heaviness and cessation, on MS risk. This result was consistent across all MR methods employed and in line with the study from Vandebergh and Goris (2020), who, in addition with smoking initiation and lifetime smoking, also investigated the potential causal effect of smoking heaviness. For all the investigated exposures, nonsignificant causal estimates were obtained. Moreover, Vandebergh and Goris (2020) also performed an MVMR to adjust for BMI levels and to estimate the directed effect of smoking initiation on MS risk, confirming the lack of a significant causal effect.

| Physical activity
As for lifestyle risk factors, physical activity plays an important role as it improves body health and reduces MS risk (Wesnes et al., 2018). Physical exercise can also act improving mobility, fatigue and health-related quality of life in MS patients with mild or moderate disability. Nonetheless, contradictory results have also been reported (Ghadirian et al., 2001). Epidemiological studies are unfortunately biased by reverse causation given the impossibility to determine, for example, whether worsened motor symptoms in MS led to less physical activity or vice versa.  performed an MR considering, as different exposures, five questionnaires investigated different aspects of physical activity: (i) moderateto-vigorous physical activity (MVPA), (ii) vigorous physical activity (VPA), (iii) overall acceleration average (AccAve), (iv) ≥2-3 days/week doing sports (SSOE) and (v) overall activity (OA). MVPA (OR = .27 [95% CI: .14-.52], p < .001), AccAve (OR = .87 [95% CI: .80-.96], p = .003) and OA (OR = .36 [95% CI: .17-.76], p = .007) showed significant causal estimates, providing further evidence that moderate physical activity is causally involved in decreasing the risk of MS. Yuan et al. (2021), in their MR study atlas, aimed at investigating the causal effect of 65 putative risk factors on the risk of MS, using data obtained from PMID: 29899525, found a protective causal relationship, even if of low entity, for moderate to vigorous level of physical activity on MS risk with an IVW OR = .12 (95% CI: .05-.32) and a p < .0001.
A previous MR study by Yang, Chen, et al. (2021) has instead investigated the causal effect of sedentary behaviour on MS, specifically television watching, computer use and driving, but none of them showed a statistically significant causal effect on MS risk in all the MR methods employed.

| Alcohol consumption
Alcohol is considered a complex modulator to the immune system (Barr et al., 2016). Indeed, ethanol modulates innate immune cells' functions, and moderate consumption of alcohol, among all, reduces inflammatory cytokine production, thus having a beneficial role in the immune system. On the other hand, chronic alcohol consumption has deleterious effects, such as inhibiting the production of growth factors. However, the comprehension of the complex interplay among alcohol intake, F I G U R E 4 Forest plot of the studies assessing the causal effect of adult body mass index (BMI) on multiple sclerosis (MS) risk. In all studies, odds ratio (OR) is referred in 1-SD increase in BMI levels. CI, confidence interval. immune response and inflammatory processes is still unknown (Romeo et al., 2007). Jiang, Zhu, et al. (2021) investigated, using MR approach, the causal effect of alcohol consumption, but no significant results were found. This result is in contrast with observational studies highlighting a preventive effect of low and moderate alcohol consumption on the incidence of autoimmune disease. Instead, the result is in line with the recent UK Biobank study (Dreyer-Alster et al., 2022) that found no evidence for an association between alcohol consumption and MS risk (OR = 1.12 [95% CI: .61-2.08], p = .314). Moreover, the MVMR analysis adjusted for BMI and smoking showed no significant direct effect of alcohol consumption on MS risk.

| COVID-related exposures
Given that MS can be triggered by viral infections and excessive stimulation of immune responses, the investigation of the causal involvement of COVID-related exposures has been of interest, also considering that coronavirus infection has been suggested to cause demyelination in animal models (Brison et al., 2011).

| Trem1 and Trem2 levels
Trem1 belongs to the immunoglobulin Trem family, and it is considered important in inflammatory responses as it contributes to the activation of immune cells and neuronal death in pathological scenarios (Lu, Liu, Sherchan, et al., 2021). Shi et al. (2022) have investigated the causal role of Trem1 plasma levels on MS, but did not find a statistically significant causal effect on MS risk. The study from Dong et al. (2022) has instead investigated the causal role of Trem2 cerebrospinal fluid levels on MS risk. Trem2 is mainly expressed by microglia, and its levels are particularly high in individuals with neurodegenerative disease. The causal role of Trem2 has been confirmed by the MR study; in fact, its increased level, albeit with a small extent, in cerebrospinal fluid has been causally linked to a higher risk of MS (OR = 1.04 [95% CI: 1.01-1.06], p = .002).

| Bone mineral density (BMD) and fracture
Inconsistent findings in literature have linked BMD and the risk of fracture with the risk of MS (Bisson et al., 2019;Olsson et al., 2018). Different factors can contribute to reduced BMD and bone metabolism in patients with MS, including reduced physical activity and depression (Cleland et al., 2020), thus increasing the risk of fracture and osteoporosis in these patients (Bisson et al., 2019).
The study from Yao et al. (2022) on a bidirectional causal effect of BMD and fracture on MS risk highlighted a significant causal protective effect of fracture (OR = .70 [95% CI: .54-.88], p = .002) on MS risk, while no significant causal effect of BMD on the risk of MS was detected. Furthermore, as highlighted by the authors, several limitations affect this study: Firstly, many IVs selected for both exposures (i.e., BMD and fracture) were not available in the outcome data sources, thus reducing and limiting the set of IVs for the MR analysis. Secondly, they did not fully exclude the potential effects of pleiotropy; in fact, using either MR-Egger or MR-PRESSO, they have found evidence of pleiotropy in the evaluation of the causal effect of BMD on MS. However, they did not find any pleiotropy in the evaluation of the causal effects of fracture on MS. So, this finding should be further verified to elucidate the molecular mechanisms underlying it.
3.9 | Leukocyte telomere length Liao et al. (2022) and Shu et al. (2022) investigated the causal effect of leukocyte telomere length on the risk of developing MS. The length of telomeres, the final portion of the chromosome, gradually shortens with cell divisions and ageing, lastly reaching a critical length in correspondence with which the cell becomes senescent. Other than biological ageing, telomere length is also related to dysregulation of immune function and immunosenescence (Fessler & Angiari, 2021), so its role in the context of MS has started to generate interest (Guan et al., 2015;Habib et al., 2020;Miner & Graves, 2021). Specifically, both previously cited studies found a statistically significant causal effect between telomeres length and MS risk, that is, OR = 2.00 (95% CI: 1.52-2.62), p < .0001, and OR = 1.91 (95% CI: 1.48-2.47), p < .0001, respectively, indicating a deleterious effect of shorter telomere length on increasing MS risk. MR estimates were concordant across all the MR methods used, and the absence of evidence for directional pleiotropy was detected through testing the statistical significance of MR-Egger intercept. Given that leukocyte telomere length is determined by genetics, environment, lifestyles and epigenetic modifications, it should be noted that these results could only partially explain its causal effect on MS.

| Herpes simplex virus (HSV) infection
The causal involvement of HSV virus infection has been investigated by , based on the hypothesis of its involvement in the pathogenesis of MS. HSV, in fact, could invade the CNS, and its presence was found in brain demyelinating plaques of MS patients by analysing post-mortem tissue (Sanders et al., 1996). However, conflicting observational studies did not find any relationship between HSV infection and MS risk, such as Etemadifar et al. (2019) and Koros et al. (2014).
With the widespread of MR methods to establish causality between exposures and outcomes,  investigated the causal effect of HSV infection and both circulating HSV-1 IgG and HSV-2 IgG, but no causal effect was found for any of the investigated exposures. It is important to note that the identified IVs explained only a small proportion of variance for circulating IgG levels of HSV-1 and HSV-2; furthermore, although no pleiotropic effects have emerged in the sensitivity analysis performed, the possibility of pleiotropy cannot be excluded.

| Gut microbiota
Another important risk factor is represented by the alteration of gut microbiota, the ensemble of microorganisms that live in the gastrointestinal tract (Jangi et al., 2016). In the study by Xu et al. (2022), 13 genera, that is, taxonomic groups, were selected as exposure, having at least one IV for each, and a statistically significant causal effect was found for Ruminococcus, using only one IV, with an OR = 2.89 (95% CI: 1.67-5.0), p = .002, and Bifidobacterium, using six IVs, with an OR = 1.38 (95% CI: 1.13-1.70), p = .01. Particularly, Bifidobacterium represents an important gut colonizer, and the causal harmful involvement of its increasing levels could be of interest for further investigations.

| Circulating interleukins
The inflammatory process results in the activation of innate immune cells and the production of interleukin (e.g., , which promote the differentiation and expansion of T cells. Inflammation and interleukins have been widely studied in the context of autoimmune diseases, even if their potential causative role remains elusive. The study from Lu, Wu, Zhang, and Liao (2021) has investigated the causal effect of circulating IL-1Ra, IL-2Ra, IL-6, IL-16, IL-17 and IL-18, finding a causal relationship between IL-1Ra and MS, OR = .94 (95% CI: .88-.99), p = .03, and IL-2Ra and MS, OR = 1.22 (95% CI: 1.12-1.32), p < .001. It is important to underlie that the significant results obtained for IL-1Ra are only suggestive, given that after multiple testing correction, the significance did not remain. Notably, another study by Vandebergh, Becelaere, et al. (2022) investigated the causal effect of IL-6 and IL-6 signalling, using both UK Biobank and CHARGE data as sources for the exposures, and showed a statistically significant causal estimate for IL-6 signalling only: OR (UK Biobank) = 1.14 (95% CI: 1.02-1.27), p = .02, and OR (CHARGE) = 1.51 (95% CI: 1.11-2.04), p = .01.
3.13 | Immune cell counts and lymphocyte cell count He et al. (2022) have investigated the causal effect of immune and lymphocyte cell count counts, driven by the observation that some susceptibility loci lie in genes implicated in immunological pathways or that may have immune modulatory effects. They found a statistically significant causal effect between (i) leucocyte count and MS with an IVW OR = 1.24 (95% CI: 1.07-1.43), p = .004; (ii) lymphocyte count and MS with an IVW OR = 1.17 (95% CI: 1.01-1.35), p = .03; and (iii) natural killer T (NKT) cells absolute count (AC) and MS with an IVW OR = 1.24 (95% CI: 1.06-1.45), p = .008. Importantly, the authors clearly stated as limitation the heterogeneity of NKT cell population, so they were unable to distinguish between Type I and Type II NKT cells.

| Circulating levels of growth factors
Growth factors are cytokines involved in the pathways of cell proliferation, differentiation and activation, and previous observational studies have suggested a role as risk factors for MS, particularly for initiation and progression of the disease (Chesik et al., 2007;Nohara et al., 2019). Lu, Wu, Ma, et al. (2021), in the framework of MR, had investigated the causal role of circulating levels of FGF23, IGF1, IGFBP3, GDF15 and VEGF, finding a causal protective role of FGF23 only with an IVW OR = .63 (95% CI: .49-.82) and p = .0005. Interestingly, as discussed by the authors, FGF23, regulating its metabolism, leads to decreased levels of vitamin D (Shimada et al., 2004) and low vitamin D levels are recognized risk factors for MS. Furthermore, FGF23 is also secreted from neurons, and it is involved in the disruption of the integrity of the blood-brain barrier. Despite this biological evidence, Lu, Wu, Ma, et al. (2021) found that increasing levels of FGF23 are inversely associated with risk of MS, therefore assuming that this potential protective effect on MS is probably not mediated via vitamin D pathways. So, further studies are required to clarify this direction of the effect and to identify the pathway where higher FGF23 levels exert its protective effect.

| Age at menarche and pubertal time
Epidemiological studies have reported an increased risk of MS with earlier age at menarche, and given that over the past decades, a precocity of age to puberty has been observed, this may also partially explain the increased incidence of MS worldwide. To clearly understand the casual contribution of sex hormone exposure and sexual maturation on disease aetiology, MS studies have been performed.
Particularly, Belbasis et al. (2020) investigated the causal contribution of increasing age at menarche, finding a protective causal effect with an OR = .91 (95% CI: .85-.99) and a p = .02, consistent across all MR methods. This result was in line with observational studies demonstrating a decreased risk for MS in women with a later age at menarche (Nielsen et al., 2017;Ramagopalan et al., 2009). Given that age at menarche is correlated with childhood BMI, the authors also run an MVMR analysis that showed lack of statistical evidence for a direct effect of age at menarche on MS risk after adjusting for childhood BMI, OR = .96 (95% CI: .88-1.04), p = .29.
A previous MR study by Harroud et al. (2019) also analysed the causal effect of pubertal timing, in terms of age at menarche. The analysis was performed by using two different sets of IVs and highlighted for each of them a statistically significant protective causal effect of later age at menarche on risk of MS. By further performing an MVMR analysis, the direct effect of age at puberty on MS, after accounting for the contribution of adult BMI, was attenuated (OR = .96 [95% CI: .89-1.04], p = .36), and the same happens when adjusting for the effect of childhood BMI, thus showing the absence of a direct effect of pubertal timing independent from BMI. We can hypothesize a complex interplay between pubertal timing and BMI, plausible bidirectional, which needs to be further investigated.

| Genes and gene expression
To investigate the causal effect of genetically predicted gene expression on MS risk is crucial to underlie the molecular pathways playing a key role in the disease and to deeply investigate the role of the expression quantitative trait loci (eQTL), that is, genetic variants associated with the level of gene expression. These kinds of studies help to shed light on how these associated genetic variations are involved in the biological pathways leading to the pathogenesis and the progression of diseases. In this picture, the study from Fazia et al. (2020) has investigated the causal effect of the expression of five candidate genes related to the proinflammatory NF-κB signalling pathway in 10 different brain regions. They found a statistically significant causal effect of NFKB1 in the cerebellum (OR = 1.39 [95% CI: 1.14-1.71], p = .001) and of CCL2 in the medulla (OR = 1.31 [95% CI: 1.12-1.54], p = .001). These results, consistent with scientific literature (Kim et al., 2014;Yue et al., 2018), were also confirmed with sensitivity analysis and bootstrap procedure.
One year later, Prince et al. (2021) investigated the causal effect of genetically predicted gene expression of 10,104 genes on autoimmune diseases, including MS, obtaining eQTL data from 31,684 blood and peripheral blood mononuclear cell samples from the eQTLGen consortium. They found 31 statistically significant causal association (see Table 3

| DISCUSSION
MR is a widely used powerful method to assess causality, allowing to positively address many of the problems affecting observational studies such as reverse causation and confounding bias. Conceptually, it adopts genetic variants as instruments to evaluate, under certain assumptions, which can only be partially verified, the causal effect of a modifiable risk factor/exposure on an outcome of interest (Emdin et al., 2017;Lawlor, 2016).
In this systematic review, we screened all the studies that, using a two-sample MR study design, have investigated the causal effect of candidate genetic and nongenetic risk factors for MS.
Several studies have highlighted a causal effect of low level of vitamin D and high BMI (both childhood and adult) on MS risk, in agreement with results from observational studies (Vandebergh, Degryse, et al., 2022). The effect of BMI has also emerged as a mediator for the causal risk effect of pubertal timing in terms of age at menarche.
As regards to other well-known lifestyle risk factors such as smoking and alcohol, their null causal effect on MS risk appears to be in contrast with the results obtained in observational studies. Indeed, smoking resulted to be not causally associated with MS risk, despite its involvement in neurodegenerative disorders and in inducing oxidative stress, as highlighted in observational epidemiological studies. The same null causal effect was found for alcohol consumption despite observational studies revealing an involvement of it in MS risk, although with controversial findings. An association between moderate alcohol consumption and a lower odds of having MS was indeed found in some studies (Andersen et al., 2019;Hedström et al., 2014;Kleerekooper et al., 2022;Weiland et al., 2014), but an increased risk has been reported as well (Abdollahpour et al., 2018;Pakpoor et al., 2014). No evidence of association between alcohol consumption and MS risk has instead been found in a recent study (Dreyer-Alster et al., 2022). Moreover, MVPA was consistently identified as a protective lifestyle factor, causally linked to decreased MS risk.
T A B L E 3 Statistically significant causal associations for the main MR analysis obtained in the study from Prince et al. (2021).

Gene
No .  Other risk factors were causally investigated through an MR approach, and some of them resulted to be causally associated with increased or decreased MS risk, and results were reported in Section 3, so that readers can have a complete summarized overall picture of all the causal exposures for MS risk.
Regarding the quality and the robustness of the results obtained, the majority of the 42 papers identified in our review as eligible for qualitative analysis attempted to assess the validity of the IVs to respect the MR assumptions, which may be of particular concern for MR studies, and performed several different sensitivity analyses with different MR methods, testing the degree of heterogeneity and pleiotropy, thus strengthening the results obtained. In addition, reverse causation was performed in many of these studies to assess the causal role of MS in varying the levels of the exposure. MVMR was also performed, in certain circumstances, to investigate the impact of direct and indirect effects. It should be noted that when investigating the causal role of many potential risk factors, the associations found need to be considered with caution if not corrected for multiple testing.
Currently available MR studies have therefore highlighted several exposures likely exerting a causal effect on MS risk, but, importantly, their clinical meaning with respect to MS aetiology needs to be still clarified. Thus, findings from MR studies require further additional interventional investigation such as RCTs to elucidate, for example, the role of vitamin D supplementation on MS risk. On the other hand, some works did not reveal potential causal associations of the investigated exposures; it is the case, among the others, of the works concerning the causal role of major depressive disorders and serum uric acid level. These results, albeit not directly informative for MS aetiology, can in part serve as a guide for subsequent scientific research. In addition, it is important to consider that although MR represents a valid approach to the study of causal relationships, it evaluates an overall casual estimation; it is likely that several distinct causal mechanisms underlie the exposure-outcome relationship, so that a risk factor can influence an outcome through different pathways with different magnitudes of causal effect.
Combining and complementing MR studies with the other research methods (e.g., observational research, RCT and lab experiments) represents the best choice to elucidate the causal role of a certain exposure on the pathogenesis of the disease under investigation.
As future direction, with the continuous development of GWASs, we should be able to successfully identify an increasing number of exposure-associated SNPs to serve as IVs for more powerful MR analysis, thus allowing to investigate additional risk factors and to replicate the results obtained so far; additionally, the development of new statistical methods for causal inference could allow to perform an MR analysis less prone to the sensitivity of its assumptions and therefore more accurate and reliable.

| CONCLUSIONS
MR studies are powerful instruments to assess causality between candidate risk factors and outcome of interest. The potential of this kind of investigation resides in the possibility to lead to personalized medicine for patients suffering from a disease. In the context of complex disease, such as MS, this potential is immense and could also help to explain the meaning and the role of the associated emerging genetic variants. However, to be really useful, results from MR studies have to be precise, reliable and robust; such a high methodological standard must be applied to prevent false positive and false negative results.