Threshold effects of serum 25‐hydroxyvitamin D levels on major depressive symptoms

To the Editor: Depression is a major contributor to the global burden of disease and disability, posing a critical public health challenge. Meta-analyses have demonstrated that vitamin D supplementation significantly reduces major depressive symptoms (MDS).1 However, recent large-scale randomised controlled trials (RCTs)2-4 have yielded conflicting evidence, indicating that vitamin D supplementation may not necessarily reduce MDS risk. This discrepancy may be attributed to variations in baseline serum 25-hydroxyvitamin D (25(OH)D) levels, insufficiently characterised dose–response relationships and uncertain optimal 25(OH)D thresholds. Further studies are urgently needed to elucidate the dose–response relationships between serum 25(OH)D levels and depression risk, which will inform evidence-based intervention strategies for optimising vitamin D supplementation protocols. Therefore, our study aimed to determine the optimal serum 25(OH)D thresholds through a comprehensive dose–response analysis with MDS and to elucidate the distinct physiological roles of 25(OH)D2 and 25(OH)D3. We investigated the dose–response relationships between serum 25(OH)D metabolites and MDS using data from seven cycles (2007–2018, 2021–2023) of the National Health and Nutrition Examination Survey (NHANES). Notably, 25(OH)D biomarker data for 2019–2020 have not been released. The study was approved by the National Center for Health Statistics Research Ethics Review Board in accordance with the principles of the Declaration of Helsinki. All participants provided written informed consent. Exclusion criteria were as follows: (1) missing 9-item Patient Health Questionnaire (PHQ-9) data; (2) age < 18 years; (3) missing 25(OH)D, 25(OH)D3 and 25(OH)D2 measurements; (4) pregnant women and (5) missing important covariates (education, family poverty index, smoking, alcohol consumption, physical activity, sleep duration, body mass index [BMI], total cholesterol [TC] and high-density cholesterol [HDL], total energy intake, dietary vitamin D intake, calcium intake, hypertension, liver disease and cancer). Finally, 24 003 participants were included in the analysis (figure S1). MDS were assessed using the PHQ-9 with total scores of the 9 items ranging from 0 to 27. We referred to a PHQ-9 total score ≥ 10 as MDS. Sensitivity analyses using alternative thresholds (≥ 5, ≥ 15, ≥ 20) were conducted to verify the robustness of our findings. Serum 25(OH)D is the established biomarker of vitamin D nutritional status. Serum concentrations of 25(OH)D3 and 25(OH)D2 were quantified using ultra-high-performance liquid chromatography-tandem mass spectrometry (https://wwwn.cdc.gov/nchs/nhanes/vitamind/analyticalnote.aspx). Total serum 25(OH)D levels were calculated as the sum of 25(OH)D3 and 25(OH)D2 metabolites. Statistical analyses were performed using R version 4.3.0 with the ‘survey’ package, accounting for NHANES's complex sampling design by incorporating sample weights, clustering and stratification variables. Baseline characteristics were compared between participants with and without MDS. For categorical variables, weighted Chi-square tests accounting for complex survey design were performed, with results reported as frequencies (weighted percentages) and Chi-square statistics. For continuous variables, weighted Mann–Whitney U tests were used to compare group differences, with results presented as means (standard deviations) and Z statistics. Weighted logistic regression models were used to examine the independent associations between serum vitamin D metabolites (total 25(OH)D, 25(OH)D3, 25(OH)D2) and MDS. Restricted cubic spline (RCS) models with four knots (5th, 35th, 65th and 95th percentiles) were used to assess potential nonlinear relationships. Models were adjusted for age, sex, education level, family poverty index, enrolment periods, 6-month time period, BMI, smoking, alcohol consumption, physical activity level, sedentary time, sleep duration, hypertension, liver disease, cancer, TC, HDL, total energy intake, dietary vitamin D intake and calcium intake. We also conducted subgroup analyses stratified by key demographic characteristics (age, sex, ethnicity and education level). A two-tailed p-value < 0.05 was considered statistically significant. Additionally, to evaluate the robustness of our results to missing data, we performed sensitivity analyses using multiple imputation by chained equations, creating five imputed datasets that included all participants with missing covariates. Among 24 003 included adult participants (table S1), 2210 (weighted prevalence 8.2%) had MDS. The mean concentration of 25(OH)D2 was significantly higher in the MDS group than in the control group (4.85 vs. 4.02 nmol/L, Mann–Whitney Z = −3.21, p = 0.001). In contrast, mean levels of both 25(OH)D (62.66 vs. 65.45 nmol/L, Z = 3.36, p < 0.001) and 25(OH)D3 (57.83 vs. 61.46 nmol/L, Z = 4.70, p < 0.001) were significantly lower in participants with MDS. As illustrated in figure 1 and table S2, both 25(OH)D and 25(OH)D3 levels exhibited significant nonlinear inverse correlations with MDS risk, characterised by a distinctive three-phase pattern. Within the deficient range of 25(OH)D (< 37.0 nmol/L, figure 1A, B), MDS risk was significantly elevated. The association plateaued at 37.0–67.0 nmol/L, where MDS risk fluctuated around population average levels without further meaningful decline. Notably, clinically meaningful protective effects emerged at concentrations ≥ 67.0 nmol/L. Specifically, compared with participants with deficient 25(OH)D levels (< 37.0 nmol/L), those with concentrations of 37.0–67.0 nmol/L showed an 11% reduction in MDS risk, though this did not reach statistical significance (adjusted odds ratio [OR]: 0.89, 95% confidence interval [CI] 0.78–1.01, p = 0.075; table 1). In contrast, individuals with optimal concentrations (≥ 67.0 nmol/L) demonstrated an 18% lower risk of MDS (adjusted OR: 0.82, 95% CI 0.71–0.94, p = 0.004). Dose–response relationships between serum vitamin D metabolites and risk of major depressive symptoms. The curve of major depressive symptoms (MDS) prevalence spline curves showing associations between (A, B) total 25(OH)D, (C, D) 25(OH)D2 and (E, F) 25(OH)D3 with major depressive symptoms. 25(OH)D, 25-hydroxyvitamin D. Similar protective associations were observed for 25(OH)D3 (figure 1C, D). Compared to those with deficient levels (< 37.5 nmol/L), participants with intermediate concentrations (37.5–63.0 nmol/L) had a 15% lower MDS risk (adjusted OR: 0.85, 95% CI 0.75–0.96, p = 0.008), while those with optimal levels (≥ 63.0 nmol/L) demonstrated a 24% reduction (adjusted OR: 0.76, 95% CI 0.67–0.86, p < 0.001; table 1). Unexpectedly, elevated 25(OH)D2 levels were associated with worse mental health outcomes (figure 1E, F). Participants with 25(OH)D2 concentrations ≥ 12.9 nmol/L exhibited a 34% increased risk of MDS compared to those below this threshold (adjusted OR: 1.34, 95% CI 1.07–1.68, p = 0.010; table 1). The RCS curves (figure S2) confirmed significant nonlinear dose–response relationships for all three vitamin D metabolites (all pnon-linear < 0.001). Threshold-based categorisation demonstrated robust consistency across survey cycles, maintaining inverse trends for 25(OH)D and 25(OH)D3 but positive associations for 25(OH)D2. Notably, MDS risk was significantly elevated during the post-pandemic period (2021−2023) compared to the pre-pandemic years (figure S3). Subgroup analyses (table S3) revealed that the associations of 25(OH)D, 25(OH)D2 and 25(OH)D3 levels with MDS risk remained generally consistent across age, sex, race/ethnicity and education level strata. However, the protective effects were somewhat attenuated in older adults and became non-significant in the multiracial subgroup. Sensitivity analyses demonstrated consistent dose–response patterns and stable threshold estimates across a range of alternative PHQ-9 cutoff points (≥ 5, ≥ 15, ≥ 20) and stratified severity analyses (tables S4−S6). Multiple imputation sensitivity analyses including participants with missing covariates (n = 34 593) (table S7) showed results closely aligned with those of the complete-case analysis (table S8), indicating minimal selection bias. These findings offer novel mechanistic insights into the differential and potentially opposing roles of 25(OH)D metabolites in mental health pathophysiology. Previous studies have largely focused on total 25(OH)D levels without differentiating between 25(OH)D3 and 25(OH)D2 components. The distinct dose–response patterns observed for 25(OH)D2 versus 25(OH)D3 suggest complex, metabolite-specific biological interactions that may underlie their contrasting associations with MDS risk. Importantly, our findings on the opposing associations of 25(OH)D2 and 25(OH)D3 with depression risk are corroborated by recent independent analyses of NHANES 2021−2023 data, which similarly reported protective effects of higher 25(OH)D3 levels and detrimental associations with elevated 25(OH)D2.5, 6 This independent replication strengthens confidence that the metabolite-specific effects represent true biological phenomena rather than statistical artefacts. The divergent biological effects of 25(OH)D2 and 25(OH)D3 suggest that vitamin D3 supplementation should be preferentially recommended over D2 formulations, particularly considering the widespread availability of ergocalciferol (vitamin D2) supplements in many healthcare settings. The threshold effects observed could plausibly operate through multiple interrelated biological mechanisms. First, vitamin D is known to modulate the hypothalamic–pituitary–adrenal axis via vitamin D receptors (VDR) located in mood-regulating brain regions,7 which may potentially attenuate stress-induced cortisol dysregulation associated with depression. A second potential pathway involves the parathyroid hormone (PTH)-calcium axis, whereby vitamin D deficiency can trigger secondary hyperparathyroidism and neuronal calcium dysregulation, thereby impairing neurotransmitter function.8 Furthermore, anti-inflammatory effects are also considered critical—vitamin D has been shown to suppress pro-inflammatory cytokines (e.g., interleukin-6, tumor necrosis factor-α) which can activate the kynurenine pathway and deplete serotonin precursors.9 Consequently, the optimal threshold (≥ 67.0 nmol/L) identified in our study may represent the concentration required for sufficient VDR activation, PTH suppression and meaningful anti-inflammatory activity. Lastly, the differential effects observed between 25(OH)D2 and 25(OH)D3 could be attributed to their distinct VDR binding affinities and biological potencies, with elevated 25(OH)D2 potentially acting as a competitive inhibitor of the more active vitamin D3 signalling.10 Our identification of a risk plateau at moderate 25(OH)D concentrations (37.0-67.0 nmol/L) provides a potential explanation for the null findings of vitamin D supplementation in many large-scale RCTs. For instance, in the landmark VITAL trial,4 participants had a mean baseline 25(OH)D level of 61.6 nmol/L squarely within our identified plateau range (37.0-67.0 nmol/L), where additional supplementation would provide minimal incremental benefits. Our findings establish evidence-based thresholds to guide personalised vitamin D supplementation: maintaining total 25(OH)D > 67.0 nmol/L, restricting 25(OH)D2 to < 12.9 nmol/L and targeting 25(OH)D3 > 63.0 nmol/L. These results advocate for a precision medicine approach to depression prevention, where baseline vitamin D status informs tailored supplementation rather than universal dosing. Individuals with severe deficiency (< 37.0 nmol/L) may require aggressive repletion to achieve therapeutic levels, while those in the suboptimal range (37.0-67.0 nmol/L) might benefit from higher-dose regimens to achieve optimal levels above 67.0 nmol/L. These findings demonstrate that the effects of vitamin D supplementation are most pronounced in individuals with baseline deficiency or depressive symptoms, supporting the precision medicine paradigm where baseline vitamin D assessment guides individualised supplementation strategies. Several important limitations must be acknowledged. The cross-sectional design of NHANES precludes causal inference between vitamin D status and MDS; prospective longitudinal studies are required to establish temporal relationships. Additionally, analyses of 25(OH)D2 were limited by its undetectable levels in 76.6% of participants, confining our evaluation to the remaining 23.4% (over 5900 participants). Although this substantial subsample provides adequate statistical power, the high undetectable rate reflects the biological reality that most individuals obtain vitamin D primarily through cutaneous synthesis and dietary vitamin D3 rather than from ergocalciferol supplementation. Despite its limitation as a screening instrument that cannot replace formal clinical interviews, PHQ-9 demonstrated consistent associations in our sensitivity analyses across multiple thresholds (≥ 5, ≥ 10, ≥ 15, ≥ 20), supporting the robustness of our threshold estimates. Sample selection bias may have arisen from the exclusion of participants with missing data, though multiple imputation sensitivity analyses demonstrated consistent results. The absence of comparable large-scale datasets with metabolite-specific measurements precludes external validation, though NHANES represents one of the most comprehensive resources for this research question. Although our analysis adjusted for key demographic, clinical factors, major comorbidities and temporal variables (examination month, survey cycle), it could not fully capture several other factors influencing vitamin D metabolism. These unmeasured variables include detailed environmental measures (e.g., residential latitude, regional ultraviolet B intensity), individual sun exposure behaviours, specific gastrointestinal disorders and medications that interfere with vitamin D metabolism (e.g., anticonvulsants orlistat). While their absence may introduce residual confounding and affect the precision of our threshold estimates, the consistency of our findings across multiple sensitivity analyses that progressively incorporated a wider set of covariates suggests that unmeasured confounding is unlikely to substantially alter the observed associations. In conclusion, this large-scale analysis reveals complex nonlinear dose–response relationships between vitamin D metabolites and depression risk, demonstrating clinically important divergent effects of 25(OH)D2 and 25(OH)D3. Our identification of specific vitamin D thresholds (total 25(OH)D > 67.0 nmol/L, 25(OH)D3 > 63.0 nmol/L, 25(OH)D2 < 12.9 nmol/L) provides an evidence-based framework for precision supplementation in mental health. These findings suggest that personalised vitamin D3 supplementation, guided by baseline 25(OH)D levels and targeting optimal thresholds, may offer a promising, cost-effective strategy for depression prevention. This precision nutrition approach could potentially reconcile the longstanding discrepancy between observational studies and previous RCTs. Conceptualisation and formal analysis: Qinghua Fan and Rongtian Xu. Funding acquisition: Haibo Wang. Methodology: Rongtian Xu, Yi Guo and Zhongjian Liu. Software and visualisation: Zhongjian Liu and Yi Guo. Writing—original draft: Rongtian Xu and Qinghua Fan. Writing—review and editing: Haibo Wang. Critical revision and final approval: all authors. This study was supported by the Brain Science and Brain-like Intelligence Technology-National Science and Technology Major Project (2021ZD0200600). The authors declare no conflicts of interest. Qinghua Fan is a research assistant at the Clinical Research Institute, Institute of Advanced Clinical Medicine, Peking University in China (2024–present). He obtained a master's degree in preventive medicine from Guangxi Medical University in China (2021–2024). He has been involved in several clinical cohorts of depression during his work. His main research interests include neuroimaging and biomarker studies in patients with major depressive disorder, with particular emphasis on understanding the neural mechanisms underlying depression and developing precision medicine approaches for treatment. He is also interested in investigating the relationship between brain connectivity patterns and treatment response in depression, aiming to identify biomarkers that can predict therapeutic outcomes and guide personalised treatment strategies. All articles used in this study are publicly available data, with the relevant studies cited. This paper analyses existing, publicly available data. Any additional information required to reanalyse the data reported in this report is available from the lead contact upon request. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.