We found a causal association between vitamin D status and HDL cholesterol and triglycerides when using the filaggrin genotype as an IV for vitamin D status, although not statistically significant when applying the Bonferroni adjusted significance level. Vitamin D status seemed to be associated with a more favorable lipid profile overall, strengthening causal inference between vitamin D status and lipid profile. Further, we replicated the association between filaggrin genotype and vitamin D status reported by Thyssen et al 
in the Inter99 population. The inverse association between vitamin D status and metabolic syndrome was statistically non-significant. On the other hand, we found no association between vitamin D status and blood pressure, BMI, or waist circumference. However, the confidence intervals are relatively wide and do not exclude a causal effect of vitamin D indicating that large studies are needed to exclude causal effects of vitamin D on these traits.
A recent mendelian randomization study found no significant associations between vitamin D associated SNPs and systolic blood pressure, BMI, total cholesterol, or triglycerides after adjustments 
. Our results extend these findings regarding diastolic blood pressure, waist circumference, metabolic syndrome, HDL-, LDL- and VLDL-cholesterol. Our results regarding serum lipids are in line with both the results from cross-sectional studies which have found a higher vitamin D level to be associated with a favorable lipid profile 
and prospective studies which have shown an inverse association between vitamin D status and triglycerides 
. However, the evidence from the few randomized controlled trials (RCTs) examining a possible effect of vitamin D supplementation on lipid profile is inconclusive 
. Jorde et al summarized the results from 10 placebo-controlled double-blind intervention studies with vitamin D supplementation as divergent. They found some studies showing a positive and some a negative effect of vitamin D supplementation. None of the intervention studies were, however, designed for evaluating the relation between vitamin D and lipids, and they were all underpowered 
Regarding the observed lack of association between vitamin D status and blood pressure, a meta-analysis on RCTs of vitamin D supplements and blood pressure found weak evidence to support a small effect of vitamin D supplementation on lowering the blood pressure in hypertensive patients 
whereas another meta-analysis found a non-significant reduction of the systolic blood pressure 
. As for the observed lack of association with BMI and waist circumference, it may be speculated whether the inverse association with vitamin D status seen in traditional observational studies 
can be explained by the fact that the fat soluble vitamin D is sequestered in the adipose tissue resulting in lower levels in obese individuals 
, i.e. that obesity causes low vitamin D status. Given a larger or older study sample, we might have been able to find an association for both blood pressure and anthropometrics.
While the mechanism by which vitamin D could affect the lipid profile is unclear, it may be due to suppression of parathyroid hormone (PTH) secretion by vitamin D, since PTH can reduce lipolysis 
. Alternatively, vitamin D may increase calcium level, thereby reducing hepatic triglyceride formation and secretion 
. Finally, vitamin D may have an effect on insulin secretion and sensitivity 
The estimates from the 2SLS are higher than the estimates from the OLS. As Mendelian randomization estimates are based on life course differences in the exposure –here vitamin D status–effect estimates can be larger than those derived from traditional observational estimates. Also, the approach avoids the underestimation of risk associations caused by regression dilution in traditional prospective studies 
. However, sometimes the most recent exposure is the strongest determinant as for the impact on cholesterol levels where a quick response to a raise or decline in a determinant can be expected.
The strengths of our study are the large samples of the general population; the ethnic homogeneity which enables genetic association studies; the detailed information on covariates; the objective measurements of instrument, exposure, and outcome; and the Mendelian randomization approach which has the potential to avoid some of the limitations of observational epidemiology (confounding, reverse causality, and regression dilution bias) for making causal inferences.
Unmeasured confounders could be factors such as sun exposure or dietary habits. Compared to RCTs, Mendelian randomization studies can be done in a representative sample with no required random treatment allocation. Certain limitations of RCTs, such as limited generalizability, high costs, feasibility and ethics, also make the Mendelian randomization approach attractive 
Using a genetic variant as proxy for vitamin D status is supposed to give better causal inferences for several reasons. First, unlike vitamin D status, genetic variants are generally not associated with the behavioral, social, and physiological factors that confound the association between vitamin D and cardiovascular risk factors. Second, genetic variants associated with vitamin D status will not be influenced by the onset of disease, and the estimates will therefore be less biased by reverse causation. Third, often a genetic variant will indicate long-term levels of exposure and will not suffer from the measurement error inherent in phenotypes that have high levels of variability like vitamin D status 
. The estimates from the analyses can be interpreted as the causal effect of vitamin D status on the outcome if the instrument is correlated with vitamin D status; is independent of the unmeasured confounders; and only affects the outcome through vitamin D given the unmeasured confounders 
. Regarding the dichotomous outcome metabolic syndrome, the estimate from the two-stage procedure is only an approximation of the causal OR. Palmer et al has suggested several possibilities of estimating the causal OR 
. We also did the suggested probit analysis and got comparable results (p
0.03 as compared to the reported p
The limitations of the study are the risk of selection bias if filaggrin genotypes were unequally distributed among the persons who died before the study; and the heterogeneity of vitamin D measurements. Importantly, the effects of the filaggrin genotype on serum lipids in particular were very consistent for the cohorts when analyzed separately. Although Mendelian randomization is a potentially powerful technique for strengthening causal inference, several issues could violate the IV assumptions: canalization i.e. developmental changes compensating for genetic variation; linkage disequilibrium between filaggrin genotype and other causal variants; pleiotropy which refers to a single gene having multiple biological functions 
; and epigenetic effects i.e. non-Mendelian, heritable changes in gene expression not accompanied by changes in DNA sequence 
. Given an inheritance of gene expression from one’s parents, we also need to assume a random distribution of epigenetic changes at conception to comply with the core assumptions of the Mendelian randomization methodology. The analyses are based on the assumption that filaggrin genotype only affects cardiovascular risk factors through vitamin D status.
The implication of using three mutations in the Inter99 and Health2006 studies while only using 2 in the Monica10 study needs consideration. The R2447X mutation is rare compared to the two other mutations, and the effects of all three FLG mutations are supposed to be the same 
. Further, excluding the R2447X mutation from the analyses did not change the results.
25-OH-D concentrations are different between the studies. This could be due to several factors such as the different methods for measuring vitamin D, evaporation during storage, or a real decrease in vitamin D levels in the population over the years. It is recommended to store samples for measurement of 25-OH-D at -80°C but studies have demonstrated stability of 25-OH-D in serum samples under different conditions 
. In general, measuring serum 25-OH-D is associated with methodological concerns, and variations between methods are considerable 
. Differences between methods are well known and variations among laboratories using the same method or assay are significant. We adjusted for the method of measuring/different levels of vitamin D by adjusting for study population. Results of analyses for each cohort separately were consistent with the combined analyses of the three cohorts which indicate that methodological differences between the cohorts did not influence our results substantially.
We investigate several outcomes/hypotheses, and it should be considered whether multiple testing represents a concern. Although we had some a priori evidence to support our hypotheses, we provided the Bonferroni adjusted significance level along with the traditional level of significance to decrease the risk of false-positive results (type 1 error). When p-values were adjusted for multiple testing by the Bonferroni method, none of the IV associations remained statistically significant emphasizing that our results need confirmation in other populations.
In conclusion, our results support a causal effect of higher vitamin D status on a more favorable lipid profile and possibly a beneficial effect in the development of the metabolic syndrome, although our results need to be confirmed in other studies. Further, we replicated the results from a previous study reporting a higher vitamin D status among filaggrin mutation carriers. A key issue in instrumental variable analyses is having a sufficiently strong instrument. Filaggrin genotype and other vitamin D related genetic variants only explain a small proportion of the variance in the observed vitamin D levels compared to the variance explained by strong determinants such as sun exposure and diet, and future research should focus on developing more efficient IV tools, e.g. by including more genetic determinants of vitamin D status or even a genetic risk score based on several SNPs.