Neurological disorders have one of the highest morbidity rates globally and are leading cause of disability-adjusted life-years (DALYs), with Alzheimer’s disease (AD), other types of dementia, Parkinson’s disease (PD), and epilepsy each accounting for over 10 million DALYs in 2019 [1]. This burden continues to increase in many countries, in part due to population growth, ageing, and lifestyle risk factors [2,3], as well as rapid urbanisation without planned infrastructure, particularly in less developed countries [4–6]. The global prevalence of mental disorders has also been increasing steadily since the 1990s, resulting in them ranking as the seventh leading cause of DALYs in 2019 [7]. The DALYs from neurological and mental disorders vary substantially by country [7] and sex [2,8], although differential diagnosis rates and life-expectancy may contribute to these variations.
Studies have identified functional overlaps between neurological and psychiatric disorders [9]. For example, mild cognitive impairment, AD, and major depressive disorder share similarities in global brain functional interactions, while epilepsy, attention deficit hyperactivity disorder, and schizophrenia bear similarities in their functional connectivity in the brain [9]. Furthermore, psychiatric symptoms such as hallucinations, delusion, anxiety, and depressive state are commonly experienced by individuals with neurological disorders, such as PD or epilepsy [10–12]. A recent report on the genetic overlaps between these two groups of disorders – that is, shared genomic components underlying both neurological and psychiatric disorders – also supports an integrated approach to these outcomes [13]. Of note are potentially bidirectional links between neuropsychiatric disorders and other chronic conditions, such as cardiovascular diseases, cancer, and respiratory conditions, whereby individuals with neuropsychiatric conditions often suffer physical comorbidities and those with chronic physical conditions suffer from mental, behavioural, or neurological conditions [14–18].
The relationship between folate and neuropsychiatric disorders has been studied based on the former’s role in cellular methylation metabolism, and in turn, in the development of the central nervous system [19] and the synthesis of neurotransmitters [20–22]. Individuals with depression [23,24] and schizophrenia [25,26] have been reported to have lower plasma folate concentrations than healthy controls. While meta-analyses of randomised controlled trials (RCTs) showed that adjunctive folic acid (FA) or methylfolate treatment may be effective in improvement of symptoms in major depressive disorder [27–30], the results for schizophrenia remain inconclusive [27,31].
The evidence on the relationship between folate status and neuropsychiatric disorders is fragmented in terms of the populations studied, folate exposure assessment methods, and the specific neuropsychiatric outcomes considered. For example, there is variation in how folate exposures have been measured and/or outcomes operationalised in specific subgroups. The fact that the studies have been predominantly of cross-sectional or case-control design also make the overall evidence difficult to interpret. Moreover, policy discussions around folate (e.g. introduction of mandatory FA fortification) have been focused on a small number of outcomes, e.g. balancing the potential benefits of reduction in the prevalence at birth of neural tube defects against the potential harms of masking of vitamin B12 deficiency, and increased risk of colorectal neoplasia. It is important that a broader range of outcomes are considered in policy decision-making, considering the biological potency of FA and the impact on population nutritional exposure. The previous umbrella review on this topic was limited to systematic reviews with meta-analyses in an adult population [32].
In this sense, a comprehensive synthesis of the entirety of the existing evidence – systematic reviews with and without meta-analyses on all outcomes across age groups – with critical appraisal of the quality and credibility of the reported findings could inform researchers and knowledge users. For this purpose, we designed a series of umbrella reviews that use similar methods [33], aiming to provide a balanced understanding of potential benefits and harms associated with folate status in six different categories of health outcomes. Umbrella reviews are a useful approach to identify scientific investigations reported during a specific time period and to assess the volume and quality of the evidence in the form of systematic reviews and meta-analyses [34,35]. They also critically assess the level of confidence in the methodological quality of the evidence to potential interventions at a population level and allow for triangulation of evidence arising from different measurements, populations, and experimental designs.
METHODS
We conducted an umbrella review of systematic reviews, where we searched MEDLINE and MEDLINE in Process via Ovid (1946 to 13 February 2024), Embase Classic + Embase via Ovid (1947 to 13 February 2024), CINAHL via EBSCOHost (1981 to 13 February 2024), the Cochrane Database of Systematic Reviews (2005 to 28 December 2021), and the Database of Abstracts of Reviews of Effects via Ovid (1994 to first quarter of 2016) for systematic reviews, with or without meta-analyses, that investigated associations between folate intake or status (measured as dietary intake, supplementation, or blood concentrations) and any neuropsychiatric outcome. Syntheses examining homocysteine as a marker of folate exposure or those examining multivitamins or multiple nutrients without separate quantification of folate intake were excluded. We did not impose any restrictions on the study population or design of component studies.
Two reviewers (SY, AM, or NJ) independently screened the articles in two stages (title/abstract and full-text) and extracted data using a standardised extraction template comprising six broad sections (Table S2 in the Online Supplementary Document): study information (first author, year of publication, year of search, exposure measure, outcome measure, risk of bias assessment); study population (eligibility criteria, countries represented in the review, participants’ age, sex, other sociodemographic features, if any); exposure details (type of exposure measure, method of measurement, time of measurement); outcome details (definition of outcome, measurement tool/scale used), reported qualitative synthesis, and information about components of meta-analysis, if a meta-analysis was reported (number of primary studies, number of total participants, number of cases, reported summary effect, heterogeneity measure, measure of small study effects, methodological quality assessment, dose-response effects, subgroup analyses, if available). All discrepancies were resolved by consensus. Two reviewers (AM, NJ) then independently assessed the included syntheses for methodological quality using the ROBIS tool [36].
We categorised all evidence by type of exposure measure, outcome, and setting (population subgroups or geographical regions) to identify unique associations (unique exposure – unique outcome – unique setting). For each category of unique associations, we examined the evidence for consistencies in direction, magnitude, and statistical significance of the summary effects. If concordant, we identified the evidence with the largest sample size. If discordant, the evidence was selected based on the largest total sample size, the largest number of cases (for binary outcomes), recency of publication, and the highest methodological quality as assessed by ROBIS. Design of the component studies were not part of the identification criteria; however, in cases where the evidence selected for a unique association consisted entirely or predominantly of retrospective studies, we examined evidence in the same category comprising entirely or predominantly prospective investigations (intervention trials, prospective cohort studies, nested case-control studies, or case cohort studies) and compared the findings to see if retrospective design of component studies biased the pooled estimates in any direction.
Lastly, we evaluated the credibility of the selected evidence using predefined criteria (Table 1). For unique associations that were assessed to be of a ‘convincing’ level of credibility, we sought, based on the data reported, to re-calculate the summary effects and 95% confidence intervals (CIs); predictive intervals to understand the dispersion of effect sizes [37]; heterogeneity between the studies using I2 and P-value; small study effects using Egger’s test of symmetry [38] with a significance threshold P < 0.10; and excess significance [39] with a threshold P < 0.10. For all other levels of credibility, we reported the statistics as calculated by the review authors. Directional associations were graded as convincing, highly suggestive, suggestive, or weak; and null associations were graded as suggestive or weak. More details of the methodologies used are provided in the first paper in this umbrella review series [33].
Table 1. Criteria for credibility assessment
| Category | Associations |
|---|---|
| Directional associations | |
| Convincing | |
| With statistical significance of P < 10−6 | |
| Based on ˃1000 cases (or ˃20,000 participants for continuous outcomes) | |
| For which largest component study reports a statistically significant result (P < 0.05) and has a 95% prediction interval that excludes the null | |
| Which do not have large heterogeneity (I2<50%) | |
| Show no evidence of small study effects (P ˃ 0.10) or of excess significance bias (P ˃ 0.10) | |
| Highly suggestive | With statistical significance of P < 10−6 |
| Based on ˃1000 cases (or ˃20,000 participants for continuous outcomes) | |
| For which largest component study reports a statistically significant result (P < 0.05) | |
| Suggestive | With statistical significance of P < 0.01 |
| Based on ˃1000 cases (or ˃20,000 participants for continuous outcomes) | |
| Weak | With statistical significance of P < 0.05 |
| Null associations | |
| Suggestive | Based on ˃1000 cases (or ˃20,000 participants for continuous outcomes) |
| Which do not have large heterogeneity (I2<50%) | |
| With statistical significance of P > 0.10 | |
| Weak | With statistical significance of 0.05<P < 0.01 |
RESULTS
Overview of search results
Of the 287 systematic reviews identified for this umbrella review series [33] (Figure S1 in the Online Supplementary Document), 76 reviews (33 with meta-analyses) [40–112] examined relationships between folate exposure and the risk of neuropsychiatric disorders (Table S1 in the Online Supplementary Document). By category of outcomes, these reviews reported on neurological disorders (epilepsy, amyotrophic lateral sclerosis, migraine, AD, PD, dementia), cognitive function, and psychiatric disorders (depression, bipolar disorder, autism spectrum disorder (ASD), attention deficit hyperactivity disorder, schizophrenia, psychosis, and obsessive-compulsive disorder). Multiple sclerosis was classified as an autoimmune condition and related findings were reported in the previous paper in this series [113].
Most of the included reviews pooled from cross-sectional or case-control studies. Of the 44 unique associations identified from the meta-analyses, only 13 (30%) were based entirely on prospective investigations (RCTs or prospective cohorts) and 11 (25%) combined findings from prospective cohort studies with those from case-control or cross-sectional studies. Plasma/serum folate concentrations were examined in 26 (59%) associations and FA supplementation in 12 (27%) associations. A total of 18 unique associations were reported only in syntheses without meta-analysis, primarily related to cognitive function and depressive disorder. Most of these syntheses included a small number (<5) of component studies (Table S2 in the Online Supplementary Document).
Risk of bias assessment
More than half of the reviews (58%) included in this synthesis had a high overall risk of bias (Figure 1; Table S3 in the Online Supplementary Document). Specifically, the risk of bias was generally low among the studies in the domains of defining eligibility criteria and executing search strategies (93%), but approximately half of the reviews had a high risk of bias in the domains of identifying and selecting studies (40%), collecting data and assessment of the included studies (41%), and synthesising the findings (37%).
Figure 1. Risk of bias assessment of the included syntheses examining the relationship between folate intake/status and neuropsychiatric disorders.
Epilepsy
One meta-analysis [40] examined the relationship between FA supplementation and seizure frequency among individuals with epilepsy. The authors pooled two RCTs comparing FA dose of 15 mg/d with placebo for a duration of 3–6.5 months and reported no effect (75 participants, 18 cases; odds ratio (OR) = 0.98; 95% CI = 0.32, 2.98; I2 = 0%). The same authors qualitatively synthesised the effect of FA supplementation (5–15 mg/d for 3–6 months) compared to placebo on intelligence and behaviour in individuals with epilepsy, examining RCTs, blinded crossover studies, or blinded controlled studies. No significant difference between FA treatment and placebo groups in intelligence (three trials, 172 participants) or behaviour (three trials, 128 participants) was apparent.
Amyotrophic lateral sclerosis
One meta-analysis [41] pooled three case-control studies, comparing plasma folate concentration in individuals with amyotrophic lateral sclerosis and in healthy controls. No significant difference was observed between the two groups (338 participants, 169 cases; mean difference (MD) = −0.52 ng/mL; 95% CI = −1.89, 0.84); I2 = 83%).
Migraine
One meta-analysis [42] synthesised case-control studies reporting on a relationship between serum folate concentration and migraine. Individuals with migraine had significantly lower serum folate concentration (1406 participants, 744 cases; standardised mean difference (SMD) = −0.36; 95% CI = −0.68, −0.05; I2 = 87%) compared to healthy controls. The magnitude of the association was attenuated among adults (1,215 participants, 643 cases; SMD = −0.11; 95% CI = −0.25, 0.01; I2 = 6%). Serum folate concentration was not significantly different when comparing individuals with migraine with aura with healthy controls (405 participants, 144 cases; SMD = −0.17; 95% CI = −0.44, 0.10; I2 = 47%) or with individuals with migraine without aura (292 participants, 144 cases; SMD = −0.16; 95% CI = −0.35, 0.04; I2 = 44%).
Dementia
A higher intake of dietary folate in older individuals was associated with a reduced risk of dementia (13,473 participants; hazard ratio = 0.61; 95% CI = 0.47, 0.78; I2 = 39%) [43] examined in prospective cohorts. A lower plasma/serum folate concentration in older individuals was also associated with an increased risk of dementia (6654 participants; OR = 1.76; 95% CI = 1.24, 2.50; I2 = 81%) [43]. Older individuals with vascular dementia had significantly lower plasma folate concentrations compared to healthy controls (5,209 participants, 308 cases; SMD = −0.80; 95% CI = −1.20, −0.39; I2 = 82.6%) [44].
A narrative synthesis examined three prospective cohorts of individuals with mild cognitive impairment and if serum folate concentration was associated with their progression to all-cause dementia [45]. Two of the cohorts (n = 246) reported that a higher serum folate concentration predicted a lower risk of progression, while one (n = 55) showed a non-significant trend.
AD
A higher total folate intake was significantly associated with a reduced risk of AD in older individuals examined in prospective cohorts (4,325 participants; risk ratio (RR) = 0.50; 95% CI = 0.25, 0.76; I2 = 32.9%) [46]. This inverse association was consistent in individuals with a higher intake (≥400 μg/d) (608 participants; RR = 0.44; 95% CI = 0.18, 0.71; I2 = 35.3%; however, there was no association in those with a lower intake (<400 μg/d) (426 participants; RR = 1.15; 95% CI = 0.28, 2.02; I2 = 0%) [46].
Six meta-analyses [44,46–50] reported on the relationship between plasma/serum or cerebrospinal fluid folate concentration and AD, primarily synthesising case-control and cross-sectional studies. Based on cross-sectional studies in the general population, plasma/serum folate concentration was reported to be significantly lower in individuals with AD compared to healthy controls (7,814 participants, 3,496 cases; SMD = −0.60; 95% CI = −0.65, −0.55; I2 = 97.2%) [46]. Lower folate concentration was not associated with a higher risk of AD (401 participants, 238 cases; OR = 0.86; 95% CI = 0.46, 1.26; I2 = 0%) in individuals classified as without folate deficiency (≥13.5 nmol/L); however, it was associated with a higher AD risk in those with potential folate deficiency (<13.5 nmol/L) [46].
Older individuals with AD had significantly lower plasma/serum folate concentration than healthy controls (7833 participants, 2932 cases; SMD = −0.50; 95% CI = −0.64, −0.36; I2 = 85.5%) [44] based on prospective cohorts and case-control studies. Meta-analysis of five prospective cohorts of older adults with potential folate deficiency (<13.5 nmol/L) also indicated that lower plasma folate concentration was associated with a higher risk of AD (1,835 participants; RR = 1.88; 95% CI = 1.20, 2.57; I2 = 0%) [46]. One meta-analysis of case-control studies [50] reported that individuals with AD had significantly lower cerebrospinal fluid folate concentration than healthy controls (845 participants, 307 cases; MD = −0.56; 95% CI = −0.73, −0.38).
PD
One meta-analysis of case-control studies reported a null association (OR = 1.01; 95% CI = 0.68, 1.34; I2 = 0%) between a higher intake of dietary folate and the risk of PD [51]. The same authors reported no difference in the plasma/serum folate concentration between individuals with PD and healthy controls (1,659 participants, 735 cases; SMD = −0.12 nmol/L; 95% CI = −0.28, 0.04; I2 = 49.2%) [51]. The findings were similar in individuals treated with Levodopa (SMD = −0.14 nmol/L; 95% CI = −0.33, 0.04; total n = 1,406; case n = 692; I2 = 58.5%) and in those not treated with Levodopa (253 participants, 43 cases; SMD = 0.02 nmol/L; 95% CI = −0.31, 0.35; I2 = 0%). In non-overlapping case-control studies conducted in China, individuals with PD had a significantly lower plasma folate concentration compared to healthy controls (2,808 participants, 1,705 cases; SMD = −0.31; 95% CI = −0.47, −0.15; I2 = 74.3%) [52].
Cognitive function
A meta-analysis of four RCTs conducted in older individuals without dementia at baseline reported that FA supplementation (0.75–15 mg/d for 1–36 months), compared to placebo, did not have an effect on cognitive function (1083 participants; SMD = 0.10; 95% CI = −0.06, 0.25) [53]. The authors aggregated and standardised findings from different cognitive tests.
A lower serum folate concentration was associated with an increased risk of cognitive impairment in older individuals (9,747 participants; OR = 1.66; 95% CI = 1.40, 1.96) [54]. The association remained significant in males (2,300 participants; OR = 1.95; 95% CI = 1.55, 2.46) and females (3,115 participants; OR = 1.90; 95% CI = 1.55, 2.32). A sensitivity analysis limited to prospective cohorts showed a comparable relationship with a reduced magnitude (3103 participants; OR = 1.40; 95% CI = 1.06, 1.84).
Several narrative syntheses [55–58] examined the relationship between dietary, total folate intake, or FA supplementation and cognitive function in various populations. The evidence on the association between dietary folate or total folate intake and risk of cognitive decline was inconclusive [55].
In individuals with cognitive impairment or mild/moderate dementia, FA supplementation significantly improved memory, information processing speed, or sensory-motor speed in three RCTs (n = 1,059) and one prospective cohort (sample size not reported) but showed no difference in another RCT (n = 7) [56].
In two RCTs with children with Down syndrome, the effect of FA supplementation on cognitive performance appeared to be dependent on the dose: improvement on the cognitive function was not significant in a study with a small dose (0.1 mg/d) but significant in a study with a larger dose (1 mg/kg/d) [57]. In Chinese older individuals with mild cognitive impairment or AD, FA supplementation (0.4–1.25 mg/d for 6–24 months) was effective in improving cognitive scores compared to placebo or conventional treatments [58].
Depression or depressive symptoms
In individuals diagnosed with depression and taking antidepressants, consuming FA supplements (0.5–10 mg/d for 6–12 weeks), compared to taking antidepressants alone, was not effective in improving depressive symptoms (671 participants; SMD = 0.49; 95% CI = −0.31, 1.29; I2 = 93%) based on three RCTs and one open-label trial [59]. The authors noted differences in medications used, study durations, nutrient dosage, and participant characteristics as potential sources of high heterogeneity. In a meta-analysis combining prospective cohorts and cross-sectional studies examining pregnant women, taking FA supplementation, compared to none, was association with a reduced risk of perinatal depression (12,881 participants, 2,378 cases; OR = 0.71; 95% CI = 0.59, 0.83; I2 = 74%) [60]. Serum/red blood cell folate concentration was also significantly lower in women with perinatal depression (6,442 participants, 1,742 cases; SMD = −0.13; 95% CI = −0.18, −0.07; I2 = 61%) compared to healthy controls [60] examined in prospective cohorts.
Eight qualitative syntheses reported 11 unique associations. The relationship between dietary folate intake and depressive symptoms in the general population was inconclusive for male and null for female [61] based on prospective cohorts and cross-sectional studies. Pooling two prospective cohorts in Europe examining individuals without depression at baseline, Sanhueza et al. [62] reported that a higher intake of dietary folate was associated with a reduced risk of depressive episodes, but noted that having a history of depression may confound the pooled risk estimate. In pregnant women, the relationship between dietary folate intake and a risk of postpartum depression was null or weak based on three prospective cohorts and one cross-sectional study [63]. The association between plasma folate concentration and the risk of antenatal or postpartum depression in pregnant women was inconclusive [64].
Methylfolate supplementation was effective in reducing depressive symptoms in older individuals with depression based on two RCTs and one pre-post intervention trial [65]. In outpatients with major depressive disorder receiving antidepressants, FA supplementation did not produce difference in the treatment response based on two RCTs [66].
Among individuals with major depressive disorder taking antidepressants, those with low plasma folate concentration showed higher rates of non-response or resistance to treatments, relapses, or recurrence of depressive symptoms based on RCTs and open-label intervention trials [67].
Bipolar disorder
One meta-analysis of case-control studies reported that individuals with bipolar disorder had significantly lower serum folate concentration (1,241 participants, 481 cases; g = −0.21; 95% CI = −0.39, −0.03; I2 = 42%) [68]. In one narrative synthesis examining individuals bipolar-I disorder with current acute mania and initiating valproate or individuals who had been taking lithium for >12 months, the two RCTs included in the synthesis showed different results regarding the effect of FA supplementation on improving symptom score [69].
ASD
Based on cross-sectional and case-control studies, three meta-analyses [70–72] reported on the relationship between plasma/serum folate concentration and the risk of ASD. The plasma folate concentrations were not different between children with ASD and healthy controls in a meta-analysis of cross-sectional studies (1,665 participants, 710 cases; MD = 0.05; 95% CI = −1.28, 1.38; I2 = 98.3%) [72]. The finding of null association was comparable across the three meta-analyses.
Four meta-analyses examined the association between reported maternal FA supplement use and the risk of ASD in offspring [73–76]. Maternal FA use was associated with a significantly reduced risk of ASD in children aged 1.5–15 years (739,226 participants, 6,396 cases; OR = 0.58; 95% CI = 0.46, 0.75; I2 = 87%) [74], based on prospective cohort and case-control studies. The magnitude of the association was attenuated in another meta-analysis limited to prospective cohort studies that partially overlap with the analysis by Iglesias et al. (616,911 participants, 2,794 cases; RR = 0.90; 95% CI = 0.79, 0.998; I2 = 42.6%) [73]. This inverse association was consistently observed in subgroups of studies conducted in Asia (19,027 participants, 15,781 cases; RR = 0.67; 95% CI = 0.46, 0.97; I2 = 64.9%), Europe (596,925 participants, 2,258 cases; RR = 0.84; 95% CI = 0.68, 0.99; I2 = 53.5%), and the USA (15,953 participants, 675 cases; RR = 0.41; 95% CI = 0.17, 0.99; I2 = 68%) [73]. In younger children aged between 11 months and 6.5 years, no difference was observed between maternal FA supplementation ≥400 μg/d and <400 μg/d in terms of offspring motor development (8,805 participants; SMD = −0.02; 95% CI = −0.08, 0.04; I2 = 20%) or mental development (11,302 participants; SMD = −0.06; 95% CI = −0.11, 0.00; I2 = 35%) [74].
Two unique associations were additionally reported by narrative syntheses. The evidence on the relationship between dietary folate intake and the risk of ASD was inconclusive [77]. The evidence on the association between maternal plasma folate concentration and the risk of ASD in offspring was also inconclusive [78].
Attention deficit hyperactivity disorder
One meta-analysis of cross-sectional studies reported that children with attention deficit hyperactivity disorder (mean age ranging from 7.7 to 9.2 years) had higher plasma/serum folate concentration compared to healthy controls (1,121 participants, 537 cases; MD = 0.92; 95% CI = 0.01, 1.76; I2 = 99%) [72].
Psychosis
A meta-analysis of six cross-sectional studies reported that the serum folate concentration was significantly lower in individuals with first-episode psychosis compared to healthy controls (827 participants, 421 cases; g = −0.62; 95% CI = −1.18, −0.07; I2 = 92%) [79].
Schizophrenia
Based on case-control studies, serum folate concentration was significantly lower in individuals with schizophrenia than in healthy controls (2,739 participants, 1,463 cases; SMD = −0.57; 95% CI = −0.76, −0.37; I2 = 83%) [80]. In individuals diagnosed with schizophrenia and being treated with antipsychotics, FA supplementation (2–500 mg/d for 12 − 16 weeks) did not effectively improve total symptom score (121 participants; SMD = −0.25; 95% CI = −0.62, 0.11; I2 = 0%) or negative symptom score (104 participants; SMD = −0.20; 95% CI = −0.59, 0.18; I2 = 0%) in RCTs [81].
Obsessive-compulsive disorder
Two meta-analyses compared the plasma/serum folate concentrations between individuals with obsessive-compulsive disorder and healthy controls and reported a null association (309 participants, 172 cases; SMD = −0.09; 95% CI = −0.75, 0.58; I2 = 88%) [82].
Credibility assessment
We conducted a credibility assessment for all identified unique associations (Table 2). All but two associations were assessed as having weak credibility, primarily owing to small sample sizes. Seven associations were downgraded from a potentially suggestive to a weak level due to unavailable data. Two associations were assessed to be at a suggestive level of credibility: use of FA supplement, compared to none, was associated with a reduced risk of perinatal depression in pregnant women (OR = 0.71; 95% CI = 0.59, 0.83) [60]; and maternal FA supplementation, compared to none, was associated with a reduced risk of ASD in offspring (OR = 0.58; 95% CI = 0.46, 0.75) [74].
Table 2. Identified unique associations between folate intake/status and neuropsychiatric outcomes
| Author (year) | Outcome, subgroup | Primary study design | Exposure | Total number (number of cases) | Metric | Summary estimate (95% CI) | Estimated P-value | I2 (%) | Credibility |
|---|---|---|---|---|---|---|---|---|---|
| Ranganathan et al. (2009) [40] | Seizure frequency, individuals with epilepsy | RCT | Supplement | 75 (18) | OR | 0.98 (0.32, 2.98) | 0.971 | 0 | Weak |
| Hu et al. (2023) [41] | ALS, general population | CC | Plasma | 338 (169) | MD | −0.52 ng/mL (−1.89, 0.84) | 0.455 | 83 | Weak |
| Liampas et al. (2020) [42] | Migraine, general population | CC | Serum | 1406 (744) | SMD | −0.36 (−0.68, −0.05) | 0.025 | 87 | Weak |
| Liampas et al. (2020) [42] | Migraine, adults | CC | Serum | 1215 (643) | SMD | −0.11 (−0.25, 0.01) | 0.097 | 6 | Weak |
| Liampas et al. (2020) [42] | Migraine with aura, general population | CC | Serum | 405 (144) | SMD | −0.17 (−0.44, 0.10) | 0.217 | 47 | Weak |
| Liampas et al. (2020) [42] | Migraine with aura vs without aura | CC | Serum | 292 (144) | SMD | −0.16 (−0.35, 0.04) | 0.107 | 44 | Weak |
| Wang et al. (2022) [43] | Incident dementia, general population | PC | Dietary intake | 13,473 (NR) | HR | 0.61 (0.47, 0.78) | <0.001 | 39 | Weak (insufficient data) |
| Wang et al. (2021) [44] | Vascular dementia, general population | PC, CC | Plasma | 5209 (308) | SMD | −0.80 (−1.20, −0.39) | <0.001 | 82.6 | Weak |
| Wang et al. (2022) [43] | Dementia, general population | PC, CS | Plasma | 6654 (NR) | OR | 1.76 (1.24, 2.50) | 0.001 | 81 | Weak (insufficient data) |
| Zhang et al. (2021) [46] | AD, general population | CC | Plasma/serum | 7814 (3,496) | SMD | −0.60 (−0.64, −0.55) | <0.001 | 97.2 | Weak |
| Zhang et al. (2021) [46] | AD, folate non-deficient population | CC | Plasma/serum | 401 (238) | OR | 0.86 (0.46, 1.26) | 0.557 | 0 | Weak |
| Zhang et al. (2021) [46] | AD, folate-deficient population | CC | Plasma/serum | 885 (379) | OR | 1.94 (1.02, 2.86) | 0.011 | 0 | Weak |
| Wilde et al. (2017) [50] | AD, general population | CC | CSF | 845 (307) | MD | −0.56 (−0.73, −0.38) | <0.001 | NR | Weak |
| Zhang et al. (2021) [46] | AD, general population | PC | Total intake | 4325 (NR) | RR | 0.50 (0.25, 0.76) | 0.014 | 32.9 | Weak (insufficient data) |
| Zhang et al. (2021) [46] | AD, general population, <400 ug/d | PC | Total intake | 426 (NR) | RR | 1.15 (0.28, 2.02) | 0.781 | 0 | Weak |
| Zhang et al. (2021) [46] | AD, general population, ≥400 ug/d | PC | Total intake | 608 (NR) | RR | 0.44 (0.18, 0.71) | 0.019 | 35.3 | Weak |
| Shen et al. (2015) [51] | PD, general population | CC | Dietary intake | NR (NR) | OR | 1.01 (0.68, 1.34) | 0.954 | 0 | Weak (insufficient data) |
| Shen et al. (2015) [51] | PD, general population | CC | Plasma/serum | 1659 (735) | SMD | −0.12 nmol/L (−0.28, 0.04) | 0.141 | 49.2 | Weak |
| Shen et al. (2015) [51] | PD, levodopa treated | CC | Plasma/serum | 1406 (692) | SMD | −0.14 nmol/L (−0.33, 0.04) | 0.138 | 58.5 | Weak |
| Shen et al. (2015) [51] | PD, levodopa untreated | CC | Plasma/serum | 253 (43) | SMD | 0.02 nmol/L (−0.31, 0.35) | 0.905 | 0 | Weak |
| Dong et al. (2020) [52] | PD, Chinese population | CC | Plasma | 2808 (1,705) | SMD | −0.31 (−0.47, −0.15) | <0.001 | 74.3 | Weak |
| Wald et al. (2010) [53] | Cognitive function, general population | RCT | Supplement | 1083 (NA) | SMD | 0.10 (−0.06, 0.25) | 0.206 | NR | Weak |
| Michelakos et al. (2013) [54] | Cognitive impairment, older adults | PC, CS | Serum | 9747 (NR) | OR | 1.66 (1.40, 1.96) | <0.001 | NR | Weak (insufficient data) |
| Michelakos et al. (2013) [54] | Cognitive impairment, older male | PC, CS | Serum | 2300 (NR) | OR | 1.90 (1.55, 2.46) | <0.001 | NR | Weak (insufficient data) |
| Michelakos et al. (2013) [54] | Cognitive impairment, older female | PC, CS | Serum | 3115 (NR) | OR | 1.90 (1.55, 2.32) | <0.001 | NR | Weak (insufficient data) |
| Sarris et al. (2016) [59] | Depression, general population | RCT, trial | Supplement | 671 (NA) | SMD | 0.49 (−0.31, 1.29) | 0.229 | 93 | Weak |
| Jin et al. (2022) [60] | Depression, pregnant women | PC, CS | Supplement | 12,881 (2,378) | OR | 0.71 (0.59, 0.83) | <0.001 | 74.4 | Suggestive |
| Jin et al. (2022) [60] | Depression, pregnant women | PC | Serum/RBC | 6442 (1,742) | SMD | −0.13 (−0.18, −0.07) | <0.001 | 60.9 | Weak |
| Hsieh et al. (2019) [68] | Bipolar disorder, general population | CC | Serum | 1241 (481) | g | −0.21 (−0.39, −0.03) | 0.022 | 42.2 | Weak |
| Prades et al. (2023) [72] | ASD, general population | CS | Plasma | 1665 (710) | MD | 0.05 (−1.28, 1.38) | 0.941 | 98.3 | Weak |
| Vazquez et al. (2019) [74] | ASD, mother-offspring pairs | PC, CC | Maternal supplement | 739,226 (6,396) | OR | 0.58 (0.46, 0.75) | <0.001 | 87 | Suggestive |
| Wang et al. (2017) [73] | ASD, Asia | PC, CC | Maternal supplement | 19,027 (1,581) | RR | 0.67 (0.46, 0.97) | 0.035 | 64.9 | Weak |
| Wang et al. (2017) [73] | ASD, Europe | PC, CC | Maternal supplement | 596,925 (2,258) | RR | 0.84 (0.68, 0.99) | 0.068 | 53.5 | Weak |
| Wang et al. (2017) [73] | ASD, US | PC, CC | Maternal supplement | 15,953 (675) | RR | 0.41 (0.17, 0.99) | 0.047 | 68 | Weak |
| Vazquez et al. (2019) [74] | Motor development, mother-offspring pairs | PC | Maternal supplement | 8805 (NA) | SMD | −0.02 (−0.08, 0.04) | 0.513 | 20 | Weak |
| Vazquez et al. (2019) [74] | Mental development, mother-offspring pairs | RCT, PC | Maternal supplement | 11,302 (NA) | SMD | −0.06 (−0.11, 0.00) | 0.032 | 35 | Weak |
| Prades et al. (2023) [72] | ADHD, general population | CS | Plasma | 1121 (537) | MD | 0.92 (0.01, 1.76) | 0.039 | 98.9 | Weak |
| Firth et al. (2018) [79] | Psychosis, general population | CS | Serum | 827 (421) | g | −0.63 (−1.18, −0.07) | 0.026 | 92.4 | Weak |
| Cao et al. (2016) [80] | Schizophrenia, general population | CC | Serum | 2739 (1,463) | SMD | −0.57 (−0.76, −0.37) | <0.001 | 82.8 | Weak |
| Sakuma et al. (2018) [81] | Schizophrenia total symptom score, individuals with schizophrenia | RCT | Supplement | 121 (NA) | SMD | −0.25 (−0.62, 0.11) | 0.179 | 0 | Weak |
| Sakuma et al. (2018) [81] | Schizophrenia negative symptom score, individuals with schizophrenia | RCT | Supplement | 104 (NA) | SMD | −0.20 (−0.59, 0.18) | 0.308 | 0 | Weak |
| Yan et al. (2022) [82] | OCD, general population | CC | Plasma | 309 (172) | SMD | −0.09 (−0.75, 0.58) | 0.790 | 87.6 | Weak |
AD – Alzheimer disease, ALS – amyotrophic lateral sclerosis, AS – Asperger’s syndrome, ASD – autism spectrum disorder, CC – case-control study, CR – case report, CS – cross-sectional study, CSF – cerebrospinal fluid, FA – folic acid, MA – meta-analysis, MD – mean difference, PC – prospective cohort, PD – Parkinson disease, PDD-NOS – pervasive developmental disorder – not otherwise specified, RBC – red blood cell, RC – retrospective cohort, RCT – randomised controlled trial, SMD – standardised mean difference, SR – systematic review, VD – vascular dementia
DISCUSSION
Summary of findings
The evidence landscape around folate and neuropsychiatric disorders is complex: while its volume appears to be growing, topic distribution remains rather skewed and largely concentrated on a few outcomes, such as AD, depression, and ASD. Variation in FA fortification and supplement practices across settings may also contribute to the heterogeneity we observed. In addition, most of the reviews synthesised observational studies of various designs, which may limit direct comparisons. We also found more narrative syntheses than meta-analyses, indicating a high level of heterogeneity among primary studies in scope and definition of outcome measures.
From a total of 74 systematic reviews, we identified 44 unique associations that were pooled quantitatively and 18 that were synthesised narratively. AD and dementia accounted for 13 associations (21%), depressive disorder for 12 associations (19%), and cognitive function and ASD for 9 associations (15%) each. Most of the identified reviews had a high overall risk of bias, particularly in the domains of data collection, assessment, and synthesis. Due to the small samples pooled, all but two unique associations were assessed to be at a weak level of credibility. Two associations at a suggestive level of credibility were inverse relationships between FA supplementation and perinatal depression and between maternal FA supplementation and ASD in offspring.
Equity and global health in the evidence on neuropsychiatric outcomes
A large number of the reviews (40%) did not report the countries or regions in which the primary investigations were conducted. The remaining reviews were evenly divided into those that included studies from low- and middle-income countries (LMICs) (31%) and those that were limited to studies from high-income countries (HICs) (29%). We observed a temporal trend, where reviews incorporating studies conducted in LMICs started around 2015; the syntheses published earlier did not report the geographical distribution of the primary studies pooled or were limited to studies in HICs. This may reflect a growing awareness and research interest in neuropsychiatric health in LMICs in the past decade. A remaining concern, however, is that only a few LMICs contribute to the body of evidence, notably China, Turkey, and India. More research in other lower-income countries will enrich the evidence landscape and enhance our understanding on the complexity of the relationship between folate intake and various neuropsychiatric conditions.
Indeed, over 80% of the mental health conditions occur in 153 countries classified as LMICs [114,115], with depression projected to be one of the top three causes of death in LMICs by 2030 [115]. Factors related to the high prevalence of mental health problems in LMICs may include rapid urbanisation, economic distress, lack of social infrastructure and resources, and younger population demographics who report higher lifetime prevalence of mental disorders and higher number of lifetime disorders [6,116,117]. The burden of neurological disorders also appears to be the greatest in LMICs [118]. This disproportionate burden of neuropsychiatric conditions in LMICs emphasises the need for more research conducted in these countries or consideration of socioeconomic moderators.
Biological plausibility of associations between folate and depression and ASD
5-methyltetrahydrofolate is the primary form of folate in circulation and is concentrated across the blood-brain barrier into cerebrospinal fluid and synaptic regions. It donates its methyl group to homocysteine to form methionine, which is converted to S-adenosylmethionine (SAM), the universal methyl donor for cellular methylation reactions [119,120], except in the methylation 2′-deoxyuridine-5′-monophosphate to 2′-deoxythymidine-5′-monophosphate (dTMP), where the carbon used for the methylation is derived from methylenetetrahydrofolate. Decreased methylation of neuronal membrane phospholipids may affect neurotransmitter function [119].
The effect of low or deficient folate on the nervous system has been well demonstrated in the prevention of primary and recurrent neural tube defects by FA supplementation [121,122]; the neuropsychiatric complications of antifolate drugs [123–125]; and the disorders of peripheral nerve and spinal cord functions [126,127]. With regard to folate in depression, two hypotheses of aetiological pathways have been proposed: cellular methylation deficit and hydroxylation of tyrosine and tryptophan [120].
First, methylfolate seems to be effective in folate treatment of cases with major depressive disorder [65], while FA is not [67]. Furthermore, an association with the ineffectiveness of antidepressant treatments was observed in patients with low blood folate levels [67]. The ability to restore optimal blood folate levels is different between methylfolate and FA [128] where the former is much more effective than the latter. The effective bioavailability of folate from FA treatment is, in fact, severely limited by the low saturation level of the dihydrofolate reductase enzyme, which is necessary for the conversion of FA to tetrahydrofolate and then to methylfolate. Methylfolate supplementation, however, does not suffer from this limitation, as methylfolate is directly bioavailable. It might be useful to highlight this difference in clinical practice using these two supplements.
There are also differences between methylfolate and SAM treatments, with the latter, in addition to being a methyl group donor, also acting as a potent inhibitor of the methylenetetrahydrofolate reductase (MTHFR) enzyme, which is responsible for methylfolate synthesis. A negative feedback loop is therefore generated between MTHFR and SAM. Treatment with SAM, on the one hand, temporarily increases the availability of methyl groups for methylation reactions, but on the other, reduces the methylfolate synthesis reaction performed by MTHFR. This causes the MTHFR substrate, methylene- tetrahydrofolate, to be diverted to the synthesis of 2′-deoxythymidine-5′-monophosphate performed by thymidylate synthase (TS). The effect of SAM treatment could, at least in part, be due to improved genomic stability.
Low folate status may also be related to depression through hydroxylation of the amino acids tyrosine and tryptophan in the synthesis of catecholamines and 5-hydroxytryptamine [120]. Dihydrofolate reductase, along with dihydropteridine reductase, is involved in the regeneration of tetrahydropterin, which are essential for the hydroxylation in the brain [129]. Pteridine is required for tryptophan hydroxylase, the rate-limiting enzyme in the synthesis of serotonin, and for tyrosine hydroxylase in the synthesis of dopamine [120].
Cerebral folate deficiency has been suggested as a potential risk associated with ASD [130,131], as well as various neurological deficits [132] and neuropsychiatric conditions [133,134]. Intake of dietary folate appears to vary substantially among individuals with ASD, with more studies reporting inadequate intake often accompanied by high food selectivity and eating problems [135,136]. In addition, folate receptor alpha antibodies, which disrupt the transport of 5-methyltetrahydrofolate across the blood-brain barrier and consequently contributing to cerebral folate deficiency [137], have been observed in individuals with ASD [138]. A recent systematic review [138] reported 38% pooled prevalence of cerebral folate deficiency in individuals with ASD. Some genetic factors, such as MTHFR polymorphisms, also appears to play an important role on the risk of ASD [83,139].
Maternal folate status during pregnancy has been widely examined in relation to the risk of offspring ASD because periconceptional period is a critical window for foetal neurogenesis [140], for which high concentrations of methyl donors are required [141,142]. Most studies, however, are limited by incomplete or heterogenous measurements of folate intake/status during the perinatal period, e.g. folate intake was self-reported, or biomarkers were measured at different trimesters [23], as well as insufficient examination of other potential risk factors, e.g. maternal obesity and pre-gestational or gestational diabetes [31,143].
Limitations
This review has several limitations. First, most of the systematic reviews in our syntheses pooled primarily from observational studies, particularly cross-sectional studies and case-control studies, which cannot inform a causal relationship between folate exposure and specific outcomes. Most of the unique associations examined the differences in plasma folate concentrations between cases and controls. Second, sex-stratified analyses were not available except for one unique association (serum folate concentration and cognitive impairment among older adults). Sex differences in the prevalence, age of onset, and clinical manifestations of neuropsychiatric disorders have been well-established [144–147]. Some reports also point to sex differences in cognitive function, such as learning and memory [2,148,149]. Third, study population was not adequately described in many reviews. Underlying morbidities, duration of illness, and use of medications may be important moderators in the relationship between folate exposure and neuropsychiatric disorders. Lastly, interactions with other nutrients involved in one-carbon metabolism, e.g. vitamins B12 and B2, and choline, were not examined in this review. Future research on the role of the interactions among these nutrients will provide further insights on the folate – neuropsychiatric health relationship in a more comprehensive context.
CONCLUSIONS
In our review of the available evidence on the relationship between folate exposure and neuropsychiatric disorders, we identified 62 unique associations (44 from meta-analyses) from 74 systematic reviews across 14 outcomes. The majority of the included reviews included only observational studies and had high risk of bias. We identified two associations assessed to be at a suggestive level of credibility: FA supplementation was associated with a reduced risk of perinatal depression and maternal FA supplementation was associated with a reduced risk of ASD in offspring. These findings, along with most of the findings we reported here, should be further examined in well-designed RCTs to adequately inform policy decisions or population-level interventions. Sex-stratified analyses, which are important considerations in neuropsychiatric research, were mostly unavailable. More prospective intervention studies designed for dose-response analyses as well as dose-response analyses at the meta-analysis level using individual patient data will add to the current knowledge on folate and neuropsychiatric disorders and better inform clinical and policy decision makers.