Investigating the causal role of circulating metabolites in major depressive disorder

Abstract

Background

Metabolic dysregulation has been implicated in major depressive disorder (MDD).

Aims

We aimed to explore the potential role of plasma metabolites in MDD.

Methods

We conducted Mendelian randomisation (MR) analysis to evaluate the causal effects of 871 circulating metabolites on MDD, using the Genome-Wide Association Studies datasets of MDD (N=1 035 760) and metabolites (N=8299). Bayesian colocalisation and druggability analyses were employed to identify genetic variants contributing to both MDD and levels of metabolites in plasma and to pinpoint metabolites with therapeutic potential, respectively.

Results

MR analysis identified 11 metabolites associated with MDD (false discovery rate<0.05). Eight metabolites, including arachidonate (20:4n6) (odds ratio (OR): 0.97), 1-arachidonoyl-GPC (20:4n6) (OR: 0.98), 1-(1-enyl-palmitoyl)−2-palmitoleoyl-GPC (P-16:0/16:1) (OR: 0.97), succinoyltaurine (OR: 0.98), 3-methoxycatechol sulphate (1) (OR: 0.98) and 11β-hydroxyandrosterone glucuronide (OR: 0.97), showed protective effects against MDD. Three metabolites were associated with increased risk, namely, butyrylglycine (OR: 1.03), 3-carboxy-4-methyl-5-propyl-2-furanpropanoate (OR: 1.02) and 1-(1-enyl-stearoyl)−2-oleoyl-GPE (p-18:0/18:1) (OR: 1.02). Colocalisation analysis supported shared genetic signals between five lipid metabolites and MDD, particularly at loci harbouring FADS and ATP9A. Notably, a majority of metabolites associated with MDD are being explored as therapeutic targets for various psychiatric disorders.

Conclusions

Genetically predicted levels of certain circulating metabolites make a causal contribution to MDD. Further investigation of their roles may provide novel pathophysiological insights and give clues for targeted therapies.

WHAT IS ALREADY KNOWN ON THIS TOPIC.

WHAT THIS STUDY ADDS

• This study demonstrates that genetically predicted levels of 11 plasma metabolites are causally associated with MDD, including eight protective and three risk-associated metabolites. Additionally, we identify shared genetic loci between five metabolites and MDD, and several of these metabolites are currently being explored as potential therapeutic targets for psychiatric disorders.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• These findings may influence future research into the pathophysiology of MDD and guide the development of personalised treatments targeting specific metabolic pathways.

Introduction

Depression is a complex mental phenomenon characterised by persistent low mood, loss of interest and a range of cognitive and physical symptoms.^{1 2} Major depressive disorder (MDD) is a common form of depression associated with high morbidity and mortality,3,5 yet its underlying biological mechanisms remain poorly understood.⁶ Current pharmacological treatments for depression primarily involve antidepressants targeting monoamine neurotransmitters (eg, serotonin). However, the efficacy of this therapeutic approach has been increasingly questioned, along with concerns regarding both serious and non-serious adverse events.7,9

Serving as substrates, intermediates and final products of cellular metabolism, small circulating molecules participate in the biological pathways throughout the entire body, including the brain. Levels of circulating metabolites change in many mental diseases, including MDD.¹⁰ Accordingly, pre-existing baselines for particular metabolites may predispose individuals to certain conditions, contributing to the initiation or maintenance of specific pathophysiological processes in the brain. Therefore, further investigation into the relationships between circulating metabolites and MDD could offer substantial pathophysiological insights. In particular, lipids, amino acids and nucleotide metabolites have attracted much attention because of their essential roles in neurotransmission and energy metabolism. For example, changes in plasma levels of ceramides are specific to bipolar disorder and help distinguish it from depression,¹¹ whereas reduced xanthine levels are common characteristics among patients with MDD.¹² Identifying metabolites that are genetically linked to the risks of psychiatric disorders may reveal new biomarkers and inform novel therapeutic strategies.

A major challenge in studying the role of metabolites in human disorders is the fluctuating nature of their circulating levels, which are ever-changing in response to confounding factors. Mendelian randomisation (MR) analysis uses genetic variations as instrumental variables (IVs) to infer the causal relationships between exposures and outcomes, alleviating the effects of confounders and reverse causation that often limit observational studies.¹³ MR analyses have helped uncover the effects of circulating molecules on several mental disorders, including that of C-reactive protein on schizophrenia,¹⁴ and that of arachidonic acid (AA) on reducing the risks of bipolar disorder.¹⁵ While the genetic factors predisposing to MDD are well established, its metabolome correlates are less understood. Advances in label-free metabolomics would allow a more comprehensive and systematic analysis of the relationships between MDD and individual metabolites.

Previous studies have examined the genetic relationships between MDD and plasma metabolite levels, with a particular emphasis on polyunsaturated fatty acids (PUFAs) and their derivatives.¹⁶ One study reported that higher levels of plasma 1-linoleyl-GPE (18:2) were associated with an increased risk of depression.¹⁰ However, these studies were limited in scope, focusing solely on lipid metabolites, and their findings were inconsistent. With the emergence of larger-scale metabonomic datasets, it is now possible to identify additional biomarkers across different categories of biomolecules.

In this study, we applied MR analysis to evaluate the causal effects of plasma metabolites on MDD using the summary data from large-scale Genome-Wide Association Studies (GWAS). To corroborate our findings, we conducted a colocalisation analysis and identified genetic variation shared between MDD and some of the metabolites in question. This led us to uncover biological mechanisms underlying the role of metabolites in MDD and enhanced the reliability of the genetic associations reported. Finally, we evaluated each metabolite for its therapeutic potential through druggability analysis (figure 1).

Figure 1. Flowchart of the study design. CLSA: Canadian Longitudinal Study on Aging cohort; COLOC: colocalisation; FDR, false discovery rate; GWAS, Genome-Wide Association Studies; LD: Linkage Disequilibrium; MDD, major depressive disorder; MR, Mendelian randomisation; SNPs, single-nucleotide polymorphisms.

Methods

Data sources

This study used the summary data from available GWAS datasets. The MDD dataset was sourced from the Psychiatric Genomics Consortium website, comprising 294 322 cases and 741 438 controls.¹⁷ The circulating levels of 1091 metabolites were obtained from the Canadian Longitudinal Study on Aging cohort (N=8299).¹⁸ Genome-wide correlation scans and high-throughput metabolomic measurements were conducted for each metabolite. After excluding 309 metabolite ratios and the unknown metabolites (labelled as “X-”), a total of 871 small molecules were included, covering key metabolic classes such as lipids, amino acids, xenobiotics, nucleotides, cofactors and vitamins.

All original studies received ethical committee approvals, and all participants were of European descent.

MR analysis

To investigate the potential causal relationships between the metabolites and MDD, MR analysis was executed using the ‘'TwoSamplesMR’ package in R.¹⁹ In the primary MR approach, causal estimates for each metabolite were calculated by combining single-nucleotide polymorphism (SNP)-specific Wald estimates by the inverse-variance weighted (IVW) method. To ensure the robustness of results, MR-Egger regression and Weighted Median (WM) methods were used to evaluate the sensitivity of the analyses, as they provide causal estimates under different pleiotropic assumptions. The WM method is known to provide consistent causal effect estimates with robust outputs as long as at least 50% of SNPs meet the validity assumption.

MR-Egger regression was used to assess potential pleiotropy by examining the intercept. A significant non-zero intercept (P_intercept<0.05) suggests the presence of horizontal pleiotropy. To enhance the reliability of our results, we employed Mendelian Randomization Pleiotropy RESidual Sum and Outlier (MR-PRESSO), which can identify pleiotropic biases and remove outliers.²⁰ Additionally, Cochran’s Q test and I² statistics were applied to evaluate heterogeneity. To assess the robustness of our findings, leave-one-out (LOO) analyses were performed, in which genetic variants were excluded one by one, followed by recalculation of the MR estimates. This approach allowed us to evaluate the influence of individual SNPs on the overall MR results.

Given the relatively small sample size for circulating metabolites, the SNPs with genome-wide significance at p<1×10⁻⁵ were selected as IVs, which allowed us to identify a larger set of potential associations. SNPs with moderate allele frequencies and palindromic SNPs were excluded, and the effect alleles were harmonised between exposure and outcome data to enhance the accuracy of causal inference. Using the European samples data from the 1000 Genomes project as the reference panel, SNPs were pruned for linkage disequilibrium within a window of 10 Mb and a threshold of r²<0.01 was applied to ensure the independence of selected SNPs and the validity of genetic instruments. The strength of the instruments was evaluated using the F-statistic, with F>10 indicating a low risk.²¹ The Benjamini-Hochberg procedure was applied to control the false discovery rate (FDR) in multiple comparisons. A threshold of FDR<0.05 was considered indicative of statistical significance.

Colocalisation analysis

Bayesian colocalisation analyses were conducted to evaluate intersections between the sets of genetic instruments contributing to the levels of metabolites in plasma and MDD.²² The type of analysis utilises GWAS summary data as an input to the ‘coloc’ package (V.5.1.1) in R (V.4.0.5), which can calculate the posterior probabilities for the five hypotheses and evaluate the strength of the evidence supporting each hypothesis separately. For each metabolite-MDD pair, we defined a genomic region by taking a ±250 kb window centred on each IV for a metabolite. A posterior probability greater than 0.80 for Hypothesis 4 (PP.H4.abf >0.80) was considered strong evidence for colocalisation, suggesting that both traits are influenced by the same underlying causal variant.

The specific hypotheses were as follows: H0, no SNP is significantly associated with either metabolite or MDD; H1, the SNP is significantly associated with the metabolite only; H2, the SNP is significantly associated with MDD only; H3, the SNP is significantly associated with both metabolite and MDD but driven by different causal variants; and H4, the SNP is significantly associated with both metabolite and MDD, driven by the same causal variants. To integrate colocalisation with MR findings, we identified metabolites with significant MR associations (FDR<0.05) and then assessed whether these metabolites showed strong colocalisation signals with MDD (PP.H4.abf >0.80).

Druggability evaluation

We used the DrugBank²³ and ChEMBL²⁴ databases to identify metabolite-related drug targets and gather relevant information, enabling evaluation of the feasibility of using particular metabolites as therapeutic targets. In DrugBank, metabolite names and synonyms were used as query terms, and only human targets of approved or investigational drugs were retained. In ChEMBL, compounds with reported bioactivity data were considered as having relevant interactions. Duplicate or ambiguous entries were manually curated and removed. DrugBank mainly provides details on the clinical applications of drugs, while ChEMBL offers information on the bioactivity of compounds and their interactions with biological targets. By systematic integration of drug data from both databases, we selected drugs most closely related to MDD and recorded their indications along with other detailed information.

Results

MR analysis

In MR analysis, the associations between 871 circulating metabolites and MDD were evaluated systematically. MR analyses identified 84 metabolites whose genetically predicted levels were nominally associated with MDD risk (p<0.05) (figure 2 and online supplemental figure 1). Among these, 52 metabolites were negatively associated with MDD, whereas 32 metabolites showed positive associations.

Figure 2. The causal effects of circulating metabolites on MDD. The forest plot (FDR<0.05). Green dots represent a negative effect, and red dots represent a positive effect. CI, confidence interval; FDR, false discovery rate; MDD, major depressive disorder; OR, odds ratio.

After applying the Benjamini-Hochberg correction for multiple tests, a total of 11 associations between the levels of particular metabolites and MDD remained significant (FDR<0.05) (figure 2). Eight metabolites were related to decreased risks of MDD, while three metabolites showed a positive association with MDD (online supplemental figure 2). Specifically, the list of protective metabolites included five lipids: arachidonate (20:4n6) (odds ratio (OR) 0.97, 95% confidence interval (CI) 0.95 to 0.98), 1-arachidonoyl-GPC (20:4n6) (OR 0.98, 95% CI 0.97 to 0.99), 1-stearoyl-2-arachidonoyl-GPC (18:0/20:4) (OR 0.98, 95% CI 0.97 to 0.99), 1-palmitoyl-2-arachidonoyl-GPC (16:0/20:4n6) (OR 0.98, 95% CI 0.97 to 0.99) and 1-(1-enyl-palmitoyl)−2-palmitoleoyl-GPC (P-16:0/16:1) (OR 0.97, 95% CI 0.96 to 0.99). Three protective metabolites of non-lipid nature were succinoyltaurine (OR 0.98, 95% CI 0.97 to 0.99), 3-methoxycatechol sulphate (1) (OR 0.98, 95% CI 0.96 to 0.99) and 11β-hydroxyandrosterone glucuronide (OR 0.97, 95% CI 0.96 to 0.99). The list of metabolites with positive causal relationships promoting MDD included butyrylglycine (OR 1.03, 95% CI 1.01 to 1.04), 3-carboxy-4-methyl-5-propyl-2-furanpropanoate (CMPF) (OR 1.02, 95% CI 1.01 to 1.03) and 1-(1-enyl-stearoyl)−2-oleoyl-GPE (p-18:0/18:1) (OR 1.02, 95% CI 1.01 to 1.04) (figure 2, table 1, online supplemental figure 3).

Table 1. The causal effects of the circulating metabolites on MDD by inverse-variance weighted methods.

Exposure	Outcome	OR (95% CI)	P value	FDR
1-arachidonoyl-GPC (20:4n6)	MDD	0.98 (0.97 to 0.99)	<0.001	0.02
3-methoxycatechol sulphate (1)	MDD	0.98 (0.96 to 0.99)	<0.001	0.02
Succinoyltaurine	MDD	0.98 (0.97 to 0.99)	<0.001	0.02
Arachidonate (20:4n6)	MDD	0.97 (0.95 to 0.98)	<0.001	0.02
1-stearoyl-2-arachidonoyl-GPC (18:0/20:4)	MDD	0.98 (0.97 to 0.99)	<0.001	0.02
1-(1-enyl-palmitoyl)−2-palmitoleoyl-GPC (P-16:0/16:1)	MDD	0.97 (0.96 to 0.99)	<0.001	0.03
11β-hydroxyandrosterone glucuronide	MDD	0.97 (0.96 to 0.99)	<0.001	0.03
1-palmitoyl-2-arachidonoyl-GPC (16:0/20:4n6)	MDD	0.98 (0.97 to 0.99)	<0.001	0.05
Butyrylglycine	MDD	1.03 (1.01 to 1.04)	<0.001	0.03
CMPF	MDD	1.02 (1.01 to 1.03)	<0.001	0.03
1-(1-enyl-stearoyl)−2-oleoyl-GPE (p-18:0/18:1)	MDD	1.02 (1.01 to 1.04)	<0.001	0.05

CI, confidence interval; CMPF, 3-carboxy-4-methyl-5-propyl-2-furanpropanoate; FDR, false discovery rate; MDD, major depressive disorder; OR, odds ratio.

For several lipid metabolites, the associations did not reach the threshold for stringent statistical significance after multiple testing correction (FDR≥0.05). As these nominal findings may still offer suggestive insights into the potential involvement of lipid metabolism in the pathophysiology of MDD, it is important to take them into account. In particular, nominal associations were observed for 1-arachidonylglycerol (20:4) (OR 0.97, FDR=0.05), 1-(1-enyl-palmitoyl)−2-arachidonoyl-GPC (p-16:0/20:4) (OR 0.98, FDR=0.06), 1-(1-enyl-palmitoyl)−2-oleoyl-GPC (p-16:0/18:1) (OR 0.96, FDR=0.06) and 1,2-dilinoleoyl-GPC (18:2/18:2) (OR 1.03, FDR=0.06) (online supplemental table 1). While these findings do not provide definitive evidence of causality, they are consistent with emerging literature suggesting a role of lipid dysregulation in MDD and warrant further validation in a larger, independent cohort.

Multiple sensitivity analyses supported the robustness of our findings. The intercept terms from MR-Egger regression were close to zero, and most tests showed no evidence of horizontal pleiotropy (P_pleiotropy>0.05), except 1-arachidonoyl-GPC (20:4n6), which exhibited marginal pleiotropy (P_pleiotropy=0.04) (online supplemental table 1). Moreover, MR-PRESSO removed detected outlier SNPs, thereby increasing the reliability of the causal estimates. While Cochran’s Q tests and I² statistics pointed to the heterogeneity of the study results, utilisation of the random-effects IVW method for causal effect estimation ensured their reliability. Additionally, sensitivity analyses using MR-Egger and WM techniques indicated consistency in the directions of causal effects (online supplemental table 2). The F-statistic values of all IVs were greater than 10, indicating a low risk of weak instrument bias (online supplemental table 3). Furthermore, the LOO analysis demonstrated that these estimates remained stable after excluding any single variant (online supplemental figure 4). However, the same analysis has highlighted two SNPs with effects on causal associations uncovered. One of them is rs174528 in fatty acid desaturase encoding the FADS1 gene, which affects the relationships between plasma levels of 1-arachidonoyl-GPC (20:4n6), 1-stearoyl-2-arachidonoyl-GPC (18:0/20:4) and 1-palmitoyl-2-arachidonoyl-GPC (16:0/20:4n6) with MDD. The other is rs2014355, located in the short-chain acyl-CoA dehydrogenase encoding gene ACADS, which impacts the relationships between butyrylglycine and MDD (online supplemental figure 5). Taken together, sensitivity analyses supported the stability and reliability of our findings.

Colocalisation analyses

In the GWAS data for the levels of 11 circulating metabolites and for MDD, five associations were further validated through colocalisation analysis. The colocalisation analysis was conducted using the coloc package with default prior probabilities (p₁ = 1×10⁻⁴, p₂ = 1×10⁻⁴, p₁₂ = 1×10⁻⁵). The posterior probabilities (PP.H4.abf) of these associations ranged from 0.84 to 0.99, providing strong evidence of colocalisation (, online supplemental table 4). Two chromosomal regions were found to harbour 36 protein-coding genes highlighted as possible players located within these regions, including PGA3, PGA4, PGA5, VWCE, DDB1, DAK, CYB561A3, TMEM138, TMEM216, CPSF7, SDHAF2, PPP1R32, LRRC10B, SYT7, DAGLA, MYRF, TMEM258, FEN1, FADS2, FADS1, FADS3, RAB3IL1, BEST1, FTH1, INCENP, SCGB1D1, SCGB2A1, SCGB1D2, SCGB2A2, SCGB1D4, ASRGL1, SCGB1A1 and AHNAK located in the chromosomal region 11q12.2–11q12.3, as well as NFATC2, ATP9A and SALL4 in 20q13.2 (table 2). The association between arachidonate (20:4n6) and MDD involved two chromosomal regions, while the other four lipid metabolites were solely linked to region 11q12.2–11q12.3. Lead MDD-associated SNPs were rs174546 and rs59849044, while the levels of lipid metabolites in plasma were linked to lead SNPs at loci rs174581, rs102275, rs174568 and rs78721504 (figure 3A–F). The genetic variation contributing to plasma levels of arachidonate (20:4n6), 1-arachidonoyl-GPC (20:4n6) and 1-(1-enyl-palmitoyl)−2-palmitoleoyl-GPC (P-16:0/16:1) was linked to rs102275 located downstream of the FADS1 gene, thus connecting these SNPs to the PUFA metabolic pathway.²⁵

Table 2. Colocalisation analysis on causal metabolites with the risk of MDD.

Metabolites	CHR	nSNPs	PP.H4.abf	Genes
Arachidonate (20:4n6)	11	1816	0.99	SDHAF2, PPP1R32, LRRC10B, SYT7, DAGLA, MYRF, TMEM258, FEN1, FADS2, FADS1, FADS3, RAB3IL1, BEST1, FTH1, INCENP, SCGB1D1
Arachidonate (20:4n6)	20	1931	0.84	NFATC2, ATP9A, SALL4
1-(1-enyl-palmitoyl)−2-palmitoleoyl-GPC (P-16:0/16:1)	11	1042	0.93	SYT7, DAGLA, MYRF, TMEM258, FEN1, FADS2, FADS1, FADS3, RAB3IL1, BEST1, FTH1
1-arachidonoyl-GPC (20:4n6)	11	2668	0.92	PGA3, PGA4, PGA5, VWCE, DDB1, DAK, CYB561A3, TMEM138, TMEM216, CPSF7, SDHAF2, PPP1R32, LRRC10B, SYT7, DAGLA, MYRF, TMEM258, FEN1, FADS2, FADS1, FADS3, RAB3IL1, BEST1, FTH1, INCENP, SCGB1D1, SCGB2A1, SCGB1D2, SCGB2A2, SCGB1D4, ASRGL1, SCGB1A1, AHNAK
1-stearoyl-2-arachidonoyl-GPC (18:0/20:4)	11	2668	0.96	As above
1−palmitoyl−2−arachidonoyl−GPC (16:0/20:4n6)	11	2668	0.88	As above

CHR, chromosome; MDD, major depressive disorder; nSNPs, number of single-nucleotide polymorphisms.

Figure 3. The colocalisation analyses and druggability evaluation of MDD and metabolites. (A–F) Colocalisation plots of the five metabolites with strong colocalisation evidence for MDD. The x-axis represents the genomic coordinates, and the y-axis represents the negative log10 transform p value for each genetic variant. (G) The druggability of metabolites with a causal effect on MDD. The red node represents the disease. The purple and orange nodes represent positive metabolites and negative metabolites, respectively. The green nodes represent drug property, and the blue nodes represent the example drug. MDD, major depressive disorder; SARI, serotonin antagonist and reuptake inhibitor; SNRI, serotonin and norepinephrine reuptake inhibitor; SSRI, selective serotonin reuptake inhibitor; TCA, tricyclic antidepressant.

Druggability evaluation

In the druggability analysis, we found that 9 out of 11 metabolites have already been targeted by pharmacological interventions, excluding succinoyltaurine and butyrylglycine (figure 3G, online supplemental table 5). Arachidonate (20:4n6) has been identified as a potential therapeutic target associated with sulfasalazine, a treatment for osteoarthritis and inflammatory bowel disease and melatonin, which is involved in sleep regulation. Other metabolites, including 3-methoxycatechol sulphate, CMPF and 11β-hydroxyandrosterone glucuronide, are known to be associated with tricyclic antidepressants like amitriptyline, selective serotonin reuptake inhibitors like citalopram and the dopamine receptor agonist rotigotine. Additionally, these and other lipid derivatives are targeted by buspirone, zopiclone and olanzapine in the treatment of anxiety, insomnia, schizophrenia and other psychiatric disorders. This druggability evaluation underscores the potential of these metabolites as biomarkers and therapeutic targets for the treatment of MDD.

Discussion

Main findings

MR-derived evidence

While prior studies have explored several metabolites linked to MDD, the findings have often been incomplete or inconsistent.^{10 12 16} In the present study, we identified a comprehensive set of 84 circulating metabolites associated with MDD. After multiple-testing corrections, 11 metabolites remained significantly associated with MDD. The majority of these were lipids, particularly AA derivatives, which exhibited protective effects against MDD, underscoring the potential role of lipid metabolism in the disorder. Among the lipids, PUFAs are engaged in multiple biological processes that modulate the brain-gut axis, a key pathway implicated in depression.²⁶ AA, a key ω−6 PUFA, incorporates into membrane phospholipids to preserve fluidity and structural integrity, while also promoting neuronal activity and neurotransmission.²⁷ The role of AA in depression remains controversial: while some MR research has suggested that genetically predicted levels of AA in plasma may increase the risks of MDD,²⁸ others reported protective effects. In particular, moderate-to-high AA intake may reduce suicidal tendencies²⁹ and alleviate depression symptoms.³⁰ These observations are consistent with our results, indicating that elevated levels of AA and related lipid metabolites may reduce the risk of MDD.

Moreover, succinoyltaurine, 3-methoxycatechol sulphate and 11β-hydroxyandrosterone glucuronide were also observed to confer protective effects. Conversely, metabolites such as 1-(1-enyl-stearoyl)−2-oleoyl-GPE, CMPF and butylglycine were associated with increased MDD risk. These metabolites likely influence MDD through mechanisms involving lipid dysregulation, mitochondrial dysfunction, oxidative stress and perturbation of the gut-brain axis.

Hypothesised mechanisms

Although MR analyses are inherently agnostic to downstream biology, the AA-related metabolites identified in our study point to several potential mechanisms, particularly involving neurobiological and immunological pathways that may contribute to the pathophysiology of MDD. AA and its derivatives are involved in inflammatory signalling and neuromodulation with both pro-inflammatory and anti-inflammatory functions.³¹ For instance, AA-derived prostaglandins (PGs) such as PGE₂³² and PGI₂ can suppress inflammation,^{33 34} whereas lipoxin A4 inhibits chemokine activity and cytokine synthesis.³⁵ PGD2, the most abundant eicosanoid in the brain, may increase as a compensatory response to stress.³¹

AA-containing monoacylglycerols, such as 1-arachidonylglycerol (20:4), can be hydrolysed to release free AA or isomerised to 2-arachidonoylglycerol (2-AG). As a principal endocannabinoid in the brain, 2-AG binds to cannabinoid receptors CB1 and CB2 to regulate mood, synaptic transmission and emotional homeostasis.^{36 37} Similarly, 1-arachidonoyl-GPC (20:4n6), a type of phosphatidylcholine derivative, supports membrane stability and inhibits T-cell migration, potentially mitigating neuroinflammation.³⁸ Pharmacological and animal studies further support these mechanisms. Mood stabilisers, including valproic acid and carbamazepine, have been shown to downregulate the expression of key enzymes in the AA pathway, such as cytosolic phospholipase A₂ and cyclooxygenase-2, thereby reducing AA turnover in brain phospholipids. Atypical antipsychotics such as olanzapine and clozapine also suppress AA metabolism by limiting the plasma availability of AA. In contrast, some antidepressants like fluoxetine and imipramine can enhance AA pathway activity.³⁹ Furthermore, geniposide, a compound with antidepressant properties, normalises elevated AA levels in depressive animal models.⁴⁰

These hypotheses highlight the potential roles of AA-associated lipid metabolism in immune signalling and endocannabinoid function. Nonetheless, these interpretations remain speculative and require further validation through functional assays, transcriptomic profiling or intervention studies. It is also important to note that associations between drugs and metabolites do not necessarily indicate direct causality or universal therapeutic benefit. These effects may vary with disease context, dosage and pharmacokinetics, underscoring the need for pharmacological intervention studies integrating metabolomics data to validate causal pathways.

Plasmalogens, such as 1-(1-enyl-stearoyl)−2-oleoyl-GPE (p-18:0/18:1), represent a unique class of phospholipids largely derived from PUFA metabolism, including docosahexaenoic and docosapentaenoic acid, which are essential for membrane stability and resistance to peroxidation.⁴¹ While this metabolite has been linked to foetal growth restriction, its precise role in depression remains unclear, warranting further investigation.⁴² CMPF is capable of mediating damage to multiple body systems by perturbing the mitochondrial dynamics and activating the cascade of oxidative stress in islet β-cells.^{43 44} Although a direct mechanistic link to MDD has not yet been established, its involvement in mitochondrial dysfunction suggests a potential contribution to neuronal injury in depression. Amino acid metabolites may also contribute to the onset and progression of depression, particularly through their roles as neurotransmitters. Elevated butyrylglycine, a product of butyrate metabolism, may reflect dysregulated gut microbial activity with pro-inflammatory consequences,^{45 46} thereby disturbing brain-gut signalling and promoting depressive risk.

Conversely, several metabolites appear to exert protective effects. Succinoyltaurine has been suggested to regulate the intestinal microbiota and attenuate neuroinflammation, effects that may collectively reduce depression risk.⁴⁷ The phenolic compound 3-methoxycatechol sulphate has antioxidant and anti-inflammatory properties, potentially reducing vulnerability to MDD.⁴⁸ 11β-hydroxyandrosterone glucuronide, an abundant brain metabolite of pregnenolone steroids, exemplifies the neuromodulatory potential of this pathway.⁴⁹ Pregnenolone and its derivatives interact with neurotransmitter receptors bound to microtubules, significantly enhancing neuronal activity and facilitating dopamine release, a mechanism with demonstrated therapeutic value in alleviating depressive symptoms.⁵⁰ Furthermore, druggability evaluation revealed that 11β-hydroxyandrosterone glucuronate is related to trimipramine, a tricyclic antidepressant, highlighting a translational bridge between metabolomics and pharmacology.

Colocalisation findings

Our colocalisation analysis further highlighted five lipid metabolites whose genetic regulation overlaps with loci associated with MDD. Notably, one key region (11q12.2-11q12.3) harbours the fatty acid desaturases (FADS) gene cluster, a critical regulator of PUFA metabolism. The FADS encoded by this cluster are ubiquitously expressed in the brain, where they catalyse essential desaturation steps in PUFA biosynthesis. Specifically, FADS2 governs the endogenous metabolism of ω−6 fatty acids, notably AA, while FADS1 is crucial for the desaturation of ω−3 fatty acids.⁵¹ Variants within this locus, such as rs174564 in FADS2, have previously been implicated in both PUFA levels and psychiatric traits, providing convergent evidence that lipid metabolism may contribute to MDD pathophysiology.^{52 53} In addition, the nearby genes MYRF and TMEM258 are interesting candidates for further investigation, as genetic variation within these loci has been shown to promote MDD.¹⁶ MYRF regulates the formation and maintenance of the myelin sheath, and its insufficient expression may disrupt nerve signal conduction and affect emotional stability,⁵⁴ while TMEM258 is crucial for endoplasmic reticulum stress response and intestinal inflammation. It is also worth noting the adjacent gene FEN1, encoding a DNA repair endonuclease, which harbours an SNP that has been associated with depression.^{55 56}

Another gene is ATP9A, mapped at the colocalisation region 20q13.2. ATP9A encodes a member of the P4-ATPase family that establishes phospholipid asymmetry⁵⁷ and facilitates the endocytosis-recycling pathway, mediating the transport from the endosome to the plasma membrane and thereby contributing to the balance of cellular material transport and energy metabolism. Notably, ATP9A is expressed in the central nervous system and has been linked to neurological and fatigue-related syndromes often comorbid with depression.^{58 59} Given the emerging evidence regarding its role in cellular homeostasis, ATP9A may be implicated in the pathogenesis of depression.

Limitations

MDD is associated with widespread physiological and behavioural alterations that can profoundly affect systemic metabolism.^{60 61} These include chronic activation of the hypothalamic-pituitary-adrenal axis, heightened systemic inflammation and behavioural changes such as reduced physical activity, altered sleep and appetite disturbances.62,64 Collectively, these factors can reshape fundamental metabolic processes, including lipid metabolism, amino acid turnover and energy homeostasis.^{10 65} For example, decreased levels of PUFAs and amino acids have been observed in patients with depression, possibly reflecting oxidative stress or enhanced catabolism.^{10 66 67} Similarly, dysregulation in the kynurenine pathway of tryptophan metabolism has been consistently linked to MDD, likely secondary to immune activation.68,70 These findings suggest that some of the metabolite changes associated with MDD may be consequences. While we did not perform reverse-direction MR analyses in this study, we acknowledge that bidirectional causal inference remains an important topic. Larger and more comprehensive metabolite GWASs, coupled with independent datasets, will help clarify potential metabolic consequences of MDD.

Several additional limitations should be acknowledged. First, our analyses were restricted to European ancestry populations, which reduces population stratification bias but may limit the generalisability of findings to other ethnic groups. Replication in diverse populations is needed to enhance trans-ancestry relevance. Second, our MR analysis revealed some degree of heterogeneity and pleiotropy. While the random-effects IVW method supports the validity of the observed directionality, the findings should be interpreted with caution. Third, our colocalisation assumed a single causal variant; applying multisignal approaches in future work will be important for complex loci such as FADS. Lastly, our study primarily focused on the genetic influences on MDD, but the association between metabolite levels and MDD may be affected by comorbid conditions, lifestyle and environmental exposures. Further clinical and mechanistic studies are needed to validate these findings and to better delineate the multifactorial determinants of MDD.

Implications

By integrating the most comprehensive GWAS summary data on circulating metabolites with the largest available MDD cohort, our study enhances the robustness of causal inference through MR. This approach not only reinforces the established role of lipid metabolism in MDD but also uncovers potential contributions from other metabolite classes, including amino acids and steroid derivatives. The exclusive focus on clinically diagnosed MDD improves the specificity and translational relevance of our findings. Additionally, the incorporation of druggability assessments highlights the therapeutic relevance of the identified metabolites, providing a foundation for future biomarker development and pharmacological exploration in the context of depression.

Conclusions

In summary, our study supports the causal role of circulating metabolites in the development of MDD. Some of these metabolites may serve as potential biomarkers or therapeutic targets for MDD.

Biography

Li Fu obtained a Bachelor of Medicine in 2017 and a Master's degree in psychiatry with a specialisation in the genetics of mental disorders in 2025 from Nanjing Medical University in China. Since graduating with her master’s degree, she has been working at the Department of Psychological Crisis Intervention, where she currently serves as a resident doctor. Her primary research interests include psychiatric genetics, with a focus on the genetic underpinnings and molecular mechanisms of mood disorders.

Data availability statement

Data are available upon reasonable request.