Exploring causal link of 25-hydroxyvitamin D with eating disorder risk via bidirectional Mendelian randomization

Abstract

Eating disorders (EDs) are severe psychiatric diseases characterized by disordered dietary behavior and malnutrition. Previous observational studies found an ambiguous correlation between low levels of serum 25-hydroxyvitamin D (25OHD) and EDs, and the causality remains unclear. This study aims to investigate the causality between 25OHD and EDs with a 2-sample bidirectional Mendelian randomization (MR) analysis. A large database of genome-wide association studies was the source of the single-nucleotide polymorphisms employed in our study. The primary approach to evaluate the causal links between 25OHD and EDs was the fixed-effects inverse-variance weighted (IVW) model. The complementary MR approaches included MR-Egger, weighted median, weighted-mode, and MR Robust Adjusted Profile Score (MR-RAPS). False discovery rate (FDR) was applied to correct for multiple testing. In sensitivity analyses, Cochran Q, MR-Egger intercept and MR-PRESSO tests were used to validate the robustness of MR results. Lastly, we conducted multivariable MR (MVMR) to determine the direct causal effects of 25OHD on EDs. The results of forward MR analyses suggested that higher vitamin D intake tended to lower the risk of genetic susceptibility to anorexia nervosa (OR_IVW = 0.43, 95% CI = 0.25–0.74, P = .002, P_FDR = .009). Whereas, no causal link was identified between 25OHD and bulimia nervosa (OR_IVW = 0.84, 95% CI = 0.25–2.82, P = .780, P_FDR = .844). Similar results were found in replication datasets and MVMR analyses. The reverse MR analysis revealed no causal connections. No intergenic heterogeneity and no horizontal pleiotropy were detected. Our findings reminded us that daily vitamin D supplementation may be beneficial in preventing anorexia nervosa and the effect of 25OHD on bulimia nervosa needs to be further studied.

1. Introduction

Eating disorders (EDs) are severe psychiatric conditions characterized by binge eating or restriction of nutrient intake, disturbances in body image, and fear of gaining weight, with a high prevalence among adolescents and young adults.^[1] This disorder is multifactorial, resulting from the interaction of external triggers and mental factors, but it is also strongly hereditary.^[2] In recent years, individuals suffering from EDs have obviously elevated mortality rates, with those with anorexia nervosa (AN) having a higher mortality rate than those with bulimia nervosa (BN).^[3] According to the statistics, women have a 0.3% to 1% lifetime risk of developing AN, with a higher prevalence of BN.^[4,5] Previous studies have indicated that hypovitaminosis D is the second most common micronutrient deficiency among hospitalized patients with EDs, and nearly half of patients present with low serum levels of 25-hydroxyvitamin D (25OHD) concentrations.^[6,7] Clinically, EDs coexisting with hypovitaminosis D can be a serious impediment to growth and development in the period of adolescence and young adulthood. Moreover, preliminary evidence has shown that higher impulsivity is correlated with a lower level of 25OHD in patients with EDs.^[8,9] Failure to correct vitamin D insufficiency might worsen the prognosis and result in chronically unrelieved conditions in disordered eating patients.

Vitamin D is a group of lipid-soluble compounds with neurosteroidal characteristics, which is distributed into muscle tissue and fat.^[10] Food intake and solar exposure can both produce vitamin D, whose status is measured by 25OHD levels in the body. Vitamin D deficiency was defined as 25OHD concentrations <20 ng/mL, and sufficiency was defined as 25OHD concentrations >20 ng/mL. In addition to being crucial for bone metabolism, the broad expression of the vitamin D receptor and enzymes necessary for its synthesis in the central nervous system raises the possibility that diseases related to neurodegenerative or mental disorders may be affected by decreases in the production of this hormone.^[11] So far, epidemiological studies have revealed intriguing but conflicting clinical associations between vitamin D deficiency and EDs. Most studies demonstrated that disordered eating patients had a high prevalence of vitamin D deficiency and insufficiency.^[7,12–14] On the contrary, a clinical trial revealed adolescents with AN exhibited a lower prevalence of vitamin D deficiency compared to healthy controls.^[15] Additionally, a case-control study found no evidence of hypovitaminosis D in young Swedish women with AN.^[16]

Alternative approaches to improve causal inference may be more efficient given that most randomized controlled trials are costly and time-consuming, which may assist determine whether such trials are actually necessary. Mendelian randomization (MR) analysis is a method to deduce the potential causality of interesting exposures on disease outcomes by using genetic variants as instrumental variables (IVs).^[17] Genetic variants, primarily single-nucleotide polymorphisms (SNPs), are allocated at random during meiosis and fertilization, making them mostly independent of self-selected behaviors and establishing themselves well prior to the onset of the disease. This minimizes problems with confounding and reverse causality.^[18] In the current study, we implemented a 2-sample bidirectional MR analysis to evaluate the potential causal link between 25OHD and EDs (including AN and BN).

2. Materials and methods

2.1. Study design

A univariable (UVMR) MR investigation was first conducted to examine the causal links between 25OHD and EDs (including AN and BN) and then performed it in the reverse direction. Next, multivariable MR (MVMR) was employed further to assess the direct effect of 25OHD on EDs. All SNPs chosen as IVs for exposures met the following criteria: IVs were strongly linked to the exposures concerned; IVs were unbiased by any potential confounding factors; IVs must only have an impact on outcomes via exposures (Fig. 1). The flowchart of the MR analysis is presented in Figure 2.

Figure 1.

The principles and validity of MR analysis are based on 3 strict assumptions. Notes: (I) IVs were strongly linked to the exposures concerned (II) IVs were unbiased by any potential confounding factors; (III) IVs must only have an impact on outcomes via exposures. MR = Mendelian randomization, IVs = instrumental variables.

Figure 2.

Flow chart on how to conduct MR analysis step by step. 25OHD = 25-hydroxyvitamin D, BMI = body mass index, cBMI = childhood Body Mass Index, EDs = eating disorders, HC = hip circumference, MR = Mendelian randomization, MR-PRESSO = MR-Pleiotropy Sum and Outlier method, SNPs = single-nucleotide polymorphisms, WC = waist circumference, WHR = waist-hip ratio.

2.2. Data resources

The variable genetic information of 25OHD levels on 2538,249 SNPs was obtained from the MRC IEU Open genome-wide association studies (GWAS) database (https://gwas.mrcieu.ac.uk/). This meta-analysis was derived from the SUNLIGHT (Study of Underlying Genetic Determinants of Vitamin D and Highly Related Traits) Consortium, which included 79,366 individuals from 31 epidemiologic studies of European descent.^[19] The SUNLIGHT GWAS featured the benefit of avoiding cohort studies that are shared by both outcome samples, and the genetic variants related to 25OHD have been verified in multiple other samples.

For EDs, 14,477 individuals were engaged in a large-scale GWAS meta-analysis of AN, including 3495 patients with AN and 10,982 healthy controls. All participants recruited for this study were of European ancestry and were from the EDs Work Group of the Psychiatric Genomics Consortium (PGC-ED).^[20] Gene-outcome associations for BN were obtained from the fifth release of the FinnGen consortium (https://finngen.gitbook.io/documentation/), in which BN was defined by 3075B in ICD-9 and F503/F502 in ICD-10, resulting in a total of 547 cases and 213,826 controls.

In replication group, 2907 patients with AN and 14,860 controls were drawn from Genetic Consortium for AN (GCAN).^[21] Due to the relative paucity of data on BN, we selected another BN-related data used in a previous MR analysis.^[22] This dataset incorporated a European female population of 151 subjects with BN and 2291 healthy controls.^[23]

Detailed information on the GWAS datasets was exhibited in Table S1, Supplemental Digital Content, https://links.lww.com/MD/P511. All summarized data utilized in this study were obtained from publicly accessible sources, and no extra ethical approval or informed consent was required for this research.

2.3. Selection of the instrumental variables

SNPs associated with 25OHD were selected using the genome-wide significance threshold (P < 5 × 10⁻⁸) at the genetic level. In the reverse direction, we modified the criteria (P < 1 × 10⁻⁵) to increase IVs related to EDs (including AN and BN). Rigorous selection criteria (r² = 0.01, window size = 10,000) were used to extract SNPs with independent inheritance and without any linkage disequilibrium based on the 1000 Genomes European Reference Panel. After harmonization, the orientation of the alleles was corrected, and some ineligible SNPs were removed if corresponding proxy SNPs (r² > 0.8) were not available. Palindromic variants and outlier pleiotropic SNPs identified by MR-PRESSO analysis were also excluded. To avoid reverse causal IVs, we applied the MR Steiger filtering test to check the direction of causality for the extracted SNPs on exposure and outcome.^[24] SNPs explaining more variation in outcome than in exposure were ruled out. Finally, the F-statistic was utilized to evaluate the influence of weak IVs bias on the current MR analyses, the calculation formula of which is F = β² exposure/ SE² exposure. The MR standard of F-statistic > 10 was employed to indicate instrument strength.

2.4. UVMR analysis

In this MR analysis, the fixed-effect inverse-variance weighted (IVW) model was employed for the primary analysis.^[25] This approach divides the SNP-outcome correlation by the SNP-exposure correlation to generate the Wald ratio estimates for each SNP, which are then combined into causal estimates for each risk factor.^[26] Additionally, due to the inclusion of 2 groups of EDs populations, a pooled causal meta-analysis. As supplementary analyses, MR-Egger regression,^[27] weighted median,^[28] weighted-mode,^[29] and MR Robust Adjusted Profile Score (MR-RAPS)^[30] methods for MR analyses were applied to lend more robust estimations in the presence of pleiotropic instruments. Among the 4 methods, weighted median continues to yield reliable results even if more than half of the IVs are invalid. Additionally, if horizontal pleiotropy exists, it is beneficial to lower type I error and can accurately assess causal links.^[28] The validity of IVs exerts no impact on MR-Egger regression, which has the ability to regulate the base of the directional pleiotropic effect.^[27] MR-RAPS can account for systematic and idiosyncratic pleiotropy with the functions of overdispersion and robust loss and provide a robust inference for MR analysis involving plenty of weak IVs.^[30] The weighted-mode method could provide an unbiased estimate if the SNPs contributing to the largest weights are valid.^[29]

All results were presented as odds ratios (OR) with 95% confidence intervals (CI) for EDs per genetically determined unit reduction in log-transformed 25OHD levels. OR > 1 suggests a link between higher vitamin D intake and an increased risk of disordered eating, OR < 1 indicates a link between increased vitamin D intake and a lower risk of disordered eating. Scatter plots were generated to give a comparative visual evaluation of the effect estimates yielded from each MR approach.

2.5. Sensitivity analysis

To examine the heterogeneity between chosen SNPs, we applied the MR-Egger regression and IVW methods. Cochrane Q statistic was implemented to estimate the impact of heterogeneity. In terms of horizontal pleiotropy, MR-PRESSO outlier analysis and MR-Egger intercept estimates were adopted to investigate and correct any potential outliers.^[28] Additionally, the leave-one-out analysis eliminated 1 SNP at a time from the analysis and then conducted a 2-sample MR analysis utilizing the remaining SNPs as IVs to determine whether the overall causal effect is being disproportionately driven by a single SNP.^[31] Funnel plots were created to visually examine potential directional pleiotropy through symmetry.

2.6. MVMR analysis

Most studies have demonstrated that disordered eating is linked to overweight or obesity.^[32–34] Meanwhile, PhenoScanner (version 2.0)^[35] searching revealed some SNPs (25OHD) linked to other phenotypes, including trunk fat percentage, trunk fat mass, hip circumference, and body fat percentage. Considering these relationships, we conducted a MVMR to simultaneously evaluate the direct effect of 25OHD on disordered eating adjusted for obesity-related phenotypes.^[36] We combined the genetic IVs from the relevant GWASs (25OHD levels from the SUNLIGHT Consortium, childhood body mass index (cBMI) from a large GWAS meta-analysis including 61,111 European children aged 2 to 10 years,^[37] body mass index (BMI),^[38] waist-hip ratio (WHR) adjusted for BMI,^[39] waist circumference (WC) adjusted for BMI,^[39] and hip circumference (HC) adjusted for BMI^[39] from the Genetic Investigation of ANthropometric Traits (GIANT) consortium) and then clumped by linkage disequilibrium (r² < 0.01 within a window of 10,000 kb) to guarantee that instruments were independent. SNP effects and corresponding standard errors were further harmonized with the outcome EDs GWAS information. Next, MR-PRESSO test was implemented to detect potentially pleiotropic outliers. The IVW approach with multiplicative random effects was applied as the main approach. To adjust for both measured and unmeasured pleiotropy, we employed the multivariable MR extension of the IVW^[36] approach and the MR-Egger approach.^[40]

2.7. Statistical analysis

The significance threshold for the links between 25OHD and EDs (including AN and BN) is set as P < .05. To mitigate the challenge of multiple testing in a gentle manner, the P-value was modified by applying the false discovery rate, also known as q value, in the primary IVW MR analysis, setting the threshold for significant causal inference at < 0.05.^[41] Filtering, harmonization, and MR analysis were conducted in the R software (version 4.3.1) with the “TwoSampleMR (version 0.5.6),”^[42] “MRPRESSO (version 1.0),”^[43] “MendelianRandomization (version 0.5.1)” and “Mr. raps” packages.

3. Results

3.1. Genetic instrumental variants

After a series of rigorous screening processes, 10 genetic instruments were eventually retained for 25OHD in forward MR analyses. For the reverse direction, 12 AN SNPs and 5 BN SNPs were available to evaluate the causal influence of AN and BN on 25OHD. The corresponding F-statistic of each selected SNP is > 10, indicating that the results on causality allow us to exclude any potential impact from weak IVs. Detailed information on strictly chosen genetic tools for further MR analysis is presented in Table S2 to S7 (Supplemental Digital Content, https://links.lww.com/MD/P512).

3.2. UVMR analyses for the causal effects of 25OHD on EDs

The results of univariable analysis identified that 25OHD was negatively associated with the risk of AN. The OR of AN per genetically-predicted 1-SD increase of serum 25OHD concentration was 0.43 (OR_IVW = 0.43, 95% CI = 0.25–0.74, P = .002, P_FDR = 0.009; Fig. 3; Table S8, Supplemental Digital Content, https://links.lww.com/MD/P513). The similar estimates were proved in replication group (odds ratio per 1-SD increase, 0.50, 95% CI = 0.28–0.88, P = .017, P_FDR = 0.085; Table S8, Supplemental Digital Content, https://links.lww.com/MD/P513) and complementary analyses (Fig. 3; Table S8, Supplemental Digital Content, https://links.lww.com/MD/P513), supporting a protective effect of vitamin D intake against AN risk. The CIs of MR-Egger were wider than those of other methods, which can be interpreted as the power of the MR-Egger method is smaller than that of the IVW. To obtain the pooled effect size, we conducted a meta-analysis using the fixed-effect model. In the meta-analysis of estimates from 2 data sources, the odds ratio of AN was 0.46 (95% CI = 0.31–0.68, P<0.001; Fig. 4) for a 1-SD increase in genetically-predicted 25OHD. Whereas, no causal links between 25OHD and BN were detected in both test group (OR_IVW = 0.84, 95% CI = 0.25–2.82, P = .780, P_FDR = 0.844; Table S8, Supplemental Digital Content, https://links.lww.com/MD/P513) and replication group (OR_IVW = 0.92, 95% CI = 0.80–1.05, P = .225, P_FDR = 0.844; Table S8, Supplemental Digital Content, https://links.lww.com/MD/P513). The combined effects yielded no statistical significance (OR_IVW = 0.92, 95% CI = 0.80–1.05, P = .216; Fig. 4).

Figure 3.

The results of univariable and multivariable MR analysis illustrated by forest plot. The causal relationship between 25OHD and AN by univariable MR study. ^# adjusted for BMI. 25OHD = 25-Hydroxyvitamin D, AN = anorexia nervosa, BMI = body mass index, cBMI = childhood body mass index, CI = confidence Interval, MR = Mendelian randomization, OR = odds ratio.

Figure 4.

Forest plots of the combined causal estimates for the effects of 25OHD on AN and BN using fix-effects IVW method. 25OHD = 25-hydroxyvitamin D, AN = anorexia nervosa, BN = bulimia nervosa, Finn = FinnGen consortium, GCAN = genetic consortium for anorexia nervosa, IVW = inverse-variance weighted, PGC-ED = Eating Disorders Work Group of the Psychiatric Genomics Consortium.

In the sensitivity analysis, there was no evidence of heterogeneity in the Cochran Q test (P>0.05; Table S9, Supplemental Digital Content, https://links.lww.com/MD/P513). The MR-PRESSO global test (P>.05; Table S9, Supplemental Digital Content, https://links.lww.com/MD/P513) and the intercept of the MR-Egger regression (P>.05; Table S9, Supplemental Digital Content, https://links.lww.com/MD/P513) demonstrate the inexistence of any pleiotropy. The estimated effects of each exposure SNP (25OHD) on the EDs are exhibited both in forests (Fig. S1, Supplemental Digital Content, https://links.lww.com/MD/P510) and scatter plots (Fig. 5). Funnel plots were used to measure directional horizontal pleiotropy, as illustrated in Figure S2 (Supplemental Digital Content, https://links.lww.com/MD/P510). In the leave-one-out sensitivity tests, no potentially influential SNP in the selected IVs was observed in the estimated causal effect between 25OHD and BN when we sequentially eliminated each SNP and ran the MR analysis in turn (Fig. S3, Supplemental Digital Content, https://links.lww.com/MD/P510). Nevertheless, the leave-one-out plot for 25OHD on AN did not show a robust result (Fig. S3, Supplemental Digital Content, https://links.lww.com/MD/P510), but the mentioned sensitivity analysis suggested that the causal relationships were generally robust and consistent.

Figure 5.

Scatter plots of causal relationship between 25OHD and EDs, including AN and BN. (A) 25OHD on AN, (B) 25OHD on BN, (C) AN on 25OHD, (D) BN on 25OHD. Scatter plots showed the causal estimations of 5 MR analysis approaches with different colored lines. 25OHD = 25-hydroxyvitamin D, AN = anorexia nervosa, BN = bulimia nervosa, EDs = eating disorders.

3.3. The reverse analyses for the causal effects of EDs on 25OHD

To investigate whether EDs exerts an inverse effect on 25OHD, a reverse MR analysis was conducted with IVW as the major approach. There was no indication of causal effects of AN (OR_IVW = 1.00, 95% CI = 0.99–1.01, P = .729, P_FDR = 0.729; Table S8, Supplemental Digital Content, https://links.lww.com/MD/P513) and BN (OR_IVW = 1.00, 95% CI = 0.99–1.01, P = .860, P_FDR = 0.986; Table S8, Supplemental Digital Content, https://links.lww.com/MD/P513) on 25OHD. The Cochran Q test showed no signs of heterogeneity (P>.05; Table S9, Supplemental Digital Content, https://links.lww.com/MD/P513). The absence of any pleiotropy is verified by the MR-PRESSO global test (P>.05; Table S9, Supplemental Digital Content, https://links.lww.com/MD/P513) and the MR-Egger regression intercept (P>.05; Table S9, Supplemental Digital Content, https://links.lww.com/MD/P513).

3.4. MVMR analyses for assessing the direct effects of 25OHD on EDs

Considering that disordered eating was correlated with obesity or overweight, we performed a MVMR to simultaneously investigate the direct effect of 25OHD on EDs after conditioning on obesity-related phenotypes. The effect estimated for 25OHD on AN was consistent with the univariable IVW estimate after adjusting for cBMI (OR = 0.43, 95% CI = 0.24–0.77, P = .004; Fig. 3; Table S10, Supplemental Digital Content, https://links.lww.com/MD/P513), BMI (OR = 0.44, 95% CI = 0.24–0.81, P = .009; Fig. 3; Table S10, Supplemental Digital Content, https://links.lww.com/MD/P513), WHR adjusted for BMI (OR = 0.44, 95% CI = 0.26–0.75, P = .003; Fig. 3; Table S10, Supplemental Digital Content, https://links.lww.com/MD/P513), WC adjusted for BMI (OR = 0.43, 95% CI = 0.25–0.74, P = .002; Fig. 3; Table S10, Supplemental Digital Content, https://links.lww.com/MD/P513), and HC adjusted for BMI (OR = 0.42, 95% CI = 0.24–0.71, P = .001; Fig. 3; Table S10, Supplemental Digital Content, https://links.lww.com/MD/P513), respectively. Likewise, the multivariable MR estimates for 25OHD on BN were still not significant (Table S10, Supplemental Digital Content, https://links.lww.com/MD/P513). There was insufficient evidence of heterogeneity (P>.05) and horizontal pleiotropy (P>.05) according to the Cochrane Q statistic and multivariable MR-Egger intercept analysis (Table S10, Supplemental Digital Content, https://links.lww.com/MD/P513).

4. Discussion

Overall, this bidirectional MR analysis aimed to disentangle relationships between 25OHD and EDs by utilizing genetic variations as unconfounded proxies. We observed relatively robust proof indicating a protective causal effect of 25OHD on AN risk. Whereas, no significant causality between 25OHD and BN was identified in this exploration. In the reverse MR analyses, there was no causal evidence supporting the idea that EDs (including AN and BN) could affect serum 25OHD levels. Our findings were generally consistent with replication group and different MR approaches, which adds to the substantial research foundation that has already established the link between 25OHD and AN and may provide a new reference for further studies detecting the relationship between 25OHD and BN.

Results of current research investigating the association between 25OHD and EDs remain controversial.^{[7,12,13,16,44]} A pilot observational study confirmed the hypothesis that vitamin D may be involved in the modulation of impulsive symptoms in EDs patients, while this statistical significance is lost when obesity is considered as a covariate.^[9] Moreover, a recent 2-sample MR analysis suggested that there was no bidirectional causal association between 25OHD levels and AN.^[44] Hence, we conducted an univariable and multivariable MR analysis to further explore their genetic correlation and the direct effect of 25OHD on EDs after adjusting for obesity-related phenotypes. In contrast, by utilizing sample data from different sources, we produced more trustworthy results than previous research, with consistency observed via various approaches and under a series of sensitivity analyses.

One of the main findings of our study regarding individuals with EDs is a significant protective effect of vitamin D intake on the risk of AN. To date, vitamin D has been studied for its insufficiency, which could result in inadequate bone development and mineralization in ED patients, whereas emerging research highlights its important role in their psychopathology.^[8] To the best of our knowledge, vitamin D features various neuroprotective properties. The potential protective causal effect of 25OHD on the risk of AN may involve multiple biological and physiological mechanisms. Firstly, Vitamin D is involved in the regulation of serotonin synthesis and action.^[45] Emerging data indicates a malfunction in serotonin pathways within cortical and limbic areas, potentially associated with anxiety, behavioral restraint, and body image misperceptions. Alterations of these neural networks may influence mood, impulse management, along with the drive and pleasure-related facets of feeding conduct.^[46] 25OHD may help maintain adequate serotonin levels, thereby potentially reducing the risk of mood disorders and disordered eating behaviors. Secondly, the presence of vitamin D receptors has been observed in dopamine neuron-dense brain areas like the ventral tegmental area and nucleus accumbens. Dopamine is involved in reward processing and motivation, which are dysregulated in AN.^[47] Adequate vitamin D levels and 25OHD may support healthy dopamine signaling,^[48] reducing the likelihood of reward-related eating disturbances. Thirdly, AN is related to low-grade systemic inflammation, which may contribute to neuropsychiatric symptoms. For instance, relevant research on several clinical populations, including those with attention deficit hyperactivity disorder, mood disorders, and alcoholism, has suggested that vitamin D deficiency may contribute to impulsivity by increasing proinflammatory cytokines in the brain as a result.^[49] In this situation, vitamin D supplementation could ameliorate impulsive-related cognitive function.^[50,51] Likewise, solar exposure in nature obviously reduces impulsive behaviors.^[52] By mitigating neuroinflammation, 25OHD may protect against the development of AN-related neuropsychiatric disturbances.^[53] Fourthly, Vitamin D deficiency is linked to gut dysbiosis and increased intestinal permeability,^[54] which can trigger systemic inflammation. The gut-brain axis is implicated in AN,^[55] and vitamin D contributes to gut barrier homeostasis, potentially lowering AN risk through modulation of neuroimmune crosstalk.^[56] Considering all the aforementioned evidence, supplementation with vitamin D for patients suffering from AN may help to avert further medical and psychological complications. To validate our findings, more studies are required.

Another interesting result based on our analyses showed no evidence for the causality between 25OHD and BN in forward and reverse MR analyses, which contradicts most existing literature connecting inadequate vitamin D with the risk of BN.^[7,12,57] BN is characterized by recurring episodes of binge eating in which enormous amounts of food are ingested in a short period of time, which might lead to an imbalance of macronutrients in the body. The reasons for the distinctions between epidemiological studies and MR analysis can be explained as follows: Although several clinical studies have reported low vitamin D levels in patients with BN, the statistical analysis was limited due to the small sample sizes included in the research. Owing to the presence of selection bias, some patients had been in the healthcare system for some time prior to inclusion in the study. As a result, previously started treatment programs including dietary management, additional supplements, and psychiatric and pharmacological therapies, all of which may be relevant, were unavailable. In many observational studies, other confounding factors, such as laxative intake, amount of sun exposure, nutritional status and body composition that could affect vitamin D levels were not evaluated. In terms of MR analysis, there was a lack of power in these analyses because 2 BN datasets reported only a small number of BN cases and few genome-wide significant SNPs linked to 25OHD. Thus, the findings are still preliminary. More randomized controlled trials and GWAS data with larger sample sizes are warranted to further infer the causative link between serum levels of vitamin D and BN.

Our findings hold substantial clinical implications, especially within the realms of psychiatry and nutrition, where the understanding of the interplay between 25OHD and EDs is of utmost importance. Clinicians may ponder integrating vitamin D supplementation into an all-encompassing preventive plan, especially for those at elevated risk of AN onset, like individuals with a family history of the disorder, those with body image concerns, or those engaged in excessive dieting behaviors. Moreover, for patients already diagnosed with AN, monitoring and correcting vitamin D deficiency could be an important aspect of their overall treatment plan. Vitamin D plays a crucial role in immune function,^[58] bone health^[59] and mental well-being,^[45] all of which are frequently impaired in patients with AN. By addressing vitamin D status, we may not only improve the physical health of AN patients but also potentially have a positive impact on their psychological symptoms and recovery process. The lack of a causal relationship in our MR analysis indicates that the association between 25OHD and BN may be more complex than previously thought. It could suggest that other factors, such as environmental or psychological variables, play a more dominant role in the development and progression of BN. This finding encourages further research to explore these other potential factors and to better understand the etiology of BN.

Our 2-sample MR approach has revealed a novel negative causal relationship between 25OHD and AN, while concurrently demonstrating an absence of impact on BN. Previous MR studies have indeed established a link between 25OHD and AN,^[60] However, the strength of this study lies in the implementation of more rigorous statistical measures to ensure the validity and reliability of our results. Firstly, we utilized a rigorous IVs filtering criteria to eliminate pleiotropic, palindromic, and reverse causal SNPs, thereby selecting IVs that demonstrate a strong association with the exposure. Secondly, to tackle the challenge of bias arising from potentially weak IVs, we adopted the MR-RAPS method, which provided robust and dependable inferences within our MR analysis. Thirdly, Hochberg sequential P-value approach could effectively address the problem of multiple hypothesis testing. Furthermore, to validate our findings, we employed 2 distinct datasets related to EDs. Recognizing the potential for divergence in causal estimates across different datasets, we conducted a meta-analysis to consolidate our observations on the link between 25OHD levels and the risk of EDs. This approach allowed us to establish a more reliable and consistent effect size, reinforcing our conclusions on this association. Finally, we applied MVMR analysis to adjust for potential confounders, thereby investigating the direct causal effect of 25OHD on EDs. Regarding the evaluation of horizontal pleiotropy and heterogeneity, we implemented MR-Egger regression to detect average pleiotropic bias. Additionally, MR-PRESSO and Cochrane Q tests were used to identify biases deriving from outlier SNPs. Through the aforementioned sensitivity analysis approaches, no heterogeneity and potential pleiotropy affecting the results between 25OHD and EDs were detected, which effectively avoids false-negative interferences and consolidates the accuracy of our results.

This study has some limitations and deficiencies. First, the inability to generalize across ethnicities poses a major challenge for applying our results to diverse populations. EDs can be influenced by a multitude of factors, including genetic predisposition, environmental triggers, socio-cultural pressures, and psychological vulnerabilities.^[61,62] For instance, cultural norms and societal expectations regarding body image and weight vary drastically across populations, which can impact the prevalence and types of EDs observed.^[63,64] Additionally, 25OHD levels are affected by skin pigmentation, diet, and sun exposure, varying across populations. Notably, darker skin pigmentation reduces the skin’s ability to synthesize 25OHD from sunlight,^[65] potentially altering the relationship between 25OHD levels and EDs in non-European populations. Furthermore, genetic polymorphisms associated with vitamin D metabolism and receptor function may also vary across ethnic groups.^[66] These genetic differences could affect an individual’s response to vitamin D supplementation or the overall effectiveness of 25OHD in mitigating the risk of EDs. Diverse racial group studies are required for validating experimental results, paving the way for publication. Second, EDs can alter nutritional intake and absorption, affecting 25OHD levels. Without adjustment, our study might incorrectly link 25OHD to the disorder. Therefore, we performed a reverse MR analysis to investigate the impact of EDs on 25OHD levels. Nevertheless, due to the insufficient number of SNPs based on EDs, a looser instrument variable selection threshold (P < 1 × 10⁻⁵) was used in reverse MR analyses, resulting in the possibility of incorporating weak IVs. To avoid the influence of weak IVs on our MR study, F statistics were calculated and significantly more than 10, indicating a negligible possibility of bias induced by weak IVs. Further research must uncover more IVs strongly linked to EDs to firmly validate the findings of this study. Third, inaccuracies in 25OHD measurement can distort data, potentially leading to false positive or false-negative associations between 25OHD levels and EDs. These can arise from testing flaws, sample handling inconsistencies, and individual factors like the timing of blood collection relative to 25OHD consumption. Multiple measures are warranted to ensure the accuracy of our 25OHD measurements, including the use of a highly sensitive and specific tools, adherence to standardized protocols, and rigorous quality control measures. Fourth, epidemiological studies show high disordered eating in young women, but our summary statistics lacked age/gender stratification. Larger samples and detailed GWAS data are needed for accurate analysis. Fifth, our research applied a comprehensive GWAS to identify IVs specifically related to 25OHD metabolism. Among 10 selected SNPs, the half have strong associations with 25OHD according to previous research: rs10745742 (in AMDHD1), rs10741657 (in CYP2R1), rs12785878 (in NADSYN1/DHCR7), rs17216707 (in CYP24A1) and rs3755967 (in GC).^[67] Certainly, the other half awaits further proof of its strong correlation with 25OHD. However, the potential for pleiotropic effects intensifies as the number of SNPs in genes with undefined roles in the vitamin D pathway rises, potentially undermining the reliability of MR findings. More IVs strongly related to 25OHD should be included to maintain the reliability and robustness of the results in further research. Sixth, we acknowledge the narrow scope of our study, which primarily aims to investigate the causal link between 25OHD and EDs through univariate and multivariate MR analyses. Future research could potentially broaden its focus to explore the causal relationship between multivitamins and their respective bodily forms, as well as EDs. Lastly, the unstable leave-one-out plots for 25OHD on AN in our analysis may be explained by a limited number of SNPs.

5. Conclusion

In summary, this 2-sample MR analysis revealed that increasing vitamin D intake could be effective in preventing AN but not in patients with BN. The findings are quite imperative because clinicians’ awareness of the possible risk of developing AN in patients with vitamin D insufficiency will aid in early detection and individualized treatment. Further studies with detailed GWAS data and larger sample sizes are warranted to provide more powerful statistical evidence to support the causal relationships. Finally, the studies of mechanisms were needed to further enhance our comprehension of this link.

Acknowledgments

We acknowledge the generosity of the SUNLIGHT, PGC-ED, GCAN, FinnGen and the GIANT Consortiums in making their data publicly available and the donation of tissue samples from all patients.

Author contributions

Conceptualization: Xiaoying Huang, Huayang Zhang, Jiahao Cheng, Baishu Zheng.

Data curation: Xiaoying Huang, Huayang Zhang, Jiahao Cheng.

Investigation: Xiaoying Huang, Huayang Zhang, Jiahao Cheng.

Methodology: Huayang Zhang, Baishu Zheng.

Project administration: Xiaoying Huang.

Validation: Huayang Zhang.

Visualization: Xiaoying Huang, Huayang Zhang, Jiahao Cheng, Baishu Zheng.

Writing – original draft: Huayang Zhang.

Writing – review & editing: Xiaoying Huang.

Supplementary Material

medi-104-e43539-s001.docx ^{(20.4KB, docx)}

medi-104-e43539-s002.xlsx ^{(20.5KB, xlsx)}

medi-104-e43539-s003.docx ^{(25KB, docx)}

medi-104-e43539-s004.docx ^{(383KB, docx)}