Metabolic features of recurrent major depressive disorder in remission, and the risk of future recurrence

Recurrent major depressive disorder (rMDD) is a relapsing-remitting disease with high morbidity and a 5-year risk of recurrence of up to 80%. This was a prospective pilot study to examine the potential diagnostic and prognostic value of targeted plasma metabolomics in the care of patients with rMDD in remission. We used an established LC-MS/MS platform to measure 399 metabolites in 68 subjects with rMDD (n = 45 females and 23 males) in antidepressant-free remission and 59 age- and sex-matched controls (n = 40 females and 19 males). Patients were then followed prospectively for 2.5 years. Metabolomics explained up to 43% of the phenotypic variance. The strongest biomarkers were gender specific. 80% of the metabolic predictors of recurrence in both males and females belonged to 6 pathways: (1) phospholipids, (2) sphingomyelins, (3) glycosphingolipids, (4) eicosanoids, (5) microbiome, and (6) purines. These changes traced to altered mitochondrial regulation of cellular redox, signaling, energy, and lipid metabolism. Metabolomics identified a chemical endophenotype that could be used to stratify rrMDD patients at greatest risk for recurrence with an accuracy over 0.90 (95%CI = 0.69–1.0). Power calculations suggest that a validation study of at least 198 females and 198 males (99 cases and 99 controls each) will be needed to confirm these results. Although a small study, these results are the first to show the potential utility of metabolomics in assisting with the important clinical challenge of prospectively identifying the patients at greatest risk of recurrence of a depressive episode and those who are at lower risk.

Eicosanoids are a major class of pro-inflammatory and anti-inflammatory lipids synthesized from 20carbon, polyunsaturated fatty acids, including the omega-6 fatty acid, arachidonic acid (C20: 4), and 20-carbon omega-3 and omega-9 fatty acids. 13(S)-hydroxyoctadecadienoic acid  was decreased in females, but not changed in males. Similarly, 15-hydroxyeicosatetraenoic acid  was decreased in females, but not in males ( Figure 2C-G, Table S3, S4). HETE are anti-inflammatory and pro-resolving oxylipins that also have antitumor effects 11 . In contrast, males with rrMDD had increases in 3 eicosanoids: which are proinflammatory mediators made by neutrophils, eosinophils, and mast cells 12,13 . Males also had an increase in the vasodilatory and anti-inflammatory epoxyeicosatrienoic acids 8,and 11,EET, but females did not ( Figure 2CD and G). The large number of disturbances in eicosanoid metabolism made this the most statistically significant single pathway disturbance in males ( Figure   2F).

Acyl-carnitines and lactate
Associated with the decreased mitochondrial medium chain (C4-C10) fatty acid oxidation capacity in females was an increase in medium chain acyl-carnitines and L-carnitine. There was also an increase in lactate, reflecting an increase in glycolytic ATP production. While males had a defect in mitochondrial long-chain (C12-C18) fatty acid oxidation and an associated increase in long-chain acylcarnitines. Lactate was not significantly increased in males (p = 0.22; Figure 2D, Table S4). A C8 medium chain, monounsaturated acyl-carnitine (2-octenoylcarnitine) known to be associated with obesity and risk of diabetes 14 was decreased in males but not in females with rrMDD ( Figure 2G).

Sphingolipids
Sphingolipid metabolism, including ceramides and sphingomyelins, was disturbed in both males and females with a history of recurrent major depressive disorder. However, in females, the non-2'hydroxylated sphingomyelins were increased, while the 2'-hydroxy sphingomyelins were decreased (Table S3 and S5). In males, only the 2'-hydroxysphingomyelins were decreased and other sphingomyelins were unchanged ( Figure 2G, Table S4 and S6).

Folate, 1-Carbon, methylation, transsulfuration, and polyamine metabolism
Although males and females with rrMDD shared disturbances in folate-1-carbon metabolism, the metabolites used were different. Females had decreases in 5-methyltetrahydrofolic acid (mTHF), homocysteine, and hypotaurine, with an increase in the methylation inhibitor S-adenosylhomocysteine (SAH). The polyamine spermine was also increased in females but not males ( Figure 2G, Table S3-S6). Spermine synthesis depends on the decarboxylation of S-adenosylmethionine (SAM) to dcSAM and its use as a 3-carbon aminopropyl donor in polyamine synthesis.

Tryptophan and indole metabolism
In addition to a decrease in plasma serotonin, females with rrMDD also had an increase in the proinflammatory and neurotoxic metabolite of tryptophan, quinolinic acid and its downstream product, the vitamin B3 redox cofactor, niacinamide ( Figure 2G, Tables S3-S6). Males did not. Indoxyl-sulfate, a metabolite regulated by both microbiome and liver metabolism, was decreased in males but not in females ( Figure 2G).

Pyrimidines
Orotic acid was decreased in both males and females with rrMDD ( Figure 2G). Orotic acid is the product of the 4 th step of de novo pyrimidine synthesis catalyzed by the mitochondrial enzyme dihydroorotate dehydrogenase (DHOD).

Serotonin and tryptophan metabolism
Serotonin was decreased in both males and females, but its precursor tryptophan, was unchanged in this cohort of drug-free rrMDD subjects ( Figure 2G).

Phenylalanine and tyrosine metabolism
Phenylketones is the name given to molecules made largely by gut bacteria from phenylalanine and tyrosine left over after, or diverted from, catecholamine neurotransmitter and protein synthesis.
Phenyllactic acid is a phenylketone that was decreased in females. Hydroxyphenylacetic acid is a phenylketone that was decreased in males with rrMDD ( Figure 2G, Table S3, S4).

Study Limitations
A small but significant difference was found in some participant characteristics among the female subjects (Table 1). These included a small increase in waist circumference and the number of overthe-counter (OTC) supplements, vitamins, and non-CNS prescription medications. We chose not to correct for these differences statistically for two reasons. First, in the case of waist circumference, we and others have found that this and other measures of obesity such as body mass index (BMI) are intrinsically linked to disturbances in mitochondrial function leading to oxidative stress 15 , and as such is not a confounder of the metabolomic signature, but a causal factor or mediator. Statistical efforts to correct for weight can inadvertently over-correct for a mediator of the metabolic phenotype and lead to an increase in type II error. Second, in the case of non-CNS prescription medications, no more than 1-4 patients (2%-10%) were taking any particular drug. As only a minority of subjects was found to take any particular medication, we felt the chances of introducing an error by inadvertent over-correction or over-simplification by grouping different types of drugs by category, were greater than the potential benefit of correcting for a factor that was present in only a minority of the subjects. Larger studies will be required to sort out the effects of non-CNS medications on the metabolomic features of recurrent major depressive disorder.

Mechanistic Implications
Lipids dominated both the diagnostic and prognostic metabolic markers found in rrMDD. Purine abnormalities were also a consistent feature in both diagnostic and prognostic markers. The importance of lipids and purines in major depressive disorder 16 , anxiety 17 , autism spectrum disorder 18 , and other mental health disorders 19,20 has been underscored by several recent studies.
The prognostic metabolites that were found to regulate the risk of recurrence were united in serving dual functions in the cell: 1. as matter, these molecules function as building blocks and intermediates in metabolism; 2. as information, they can act to modify macromolecular targets and change their function by phosphorylation, methylation, acetylation, myristoylation, farnesylation, etc. In addition, these metabolites act as ligands for transmembrane G-protein coupled or nuclear receptors that change gene expression, or act as allosteric regulators that change the conductance of solute carriers (SLCs) and ATP-binding cassette (ABC) transporters that conduct metabolites as organic anions and cations, and ion channels that conduct inorganic cations like Na + , K + , Mg 2+ , and Ca 2+ , and anions like Cl -. In the case of purines, a third function is also well known: the role of molecules like ATP as energy. Metabolites that have both metabolic and signaling functions have been called "metabokines".
All neurotransmitters currently targeted by drug therapies for major depressive disorder are metabokines 21 , and purines are purinergic metabokines that act as ligands for ionotropic and metabotropic receptors 22,23 . A large majority of metabokines have been found to be coordinately regulated in response to cellular injury or threat. The coordinated regulation of metabokines has been placed in evolutionary context as the biological process that underlies healing and aging and called the cell danger response (CDR) 21,24,25 . The molecular aspects of the CDR include the integrated stress response (ISR) 26 . The long-distance coordination of the CDR is effected by metabokines used for remote sensing and signaling (RSS) 27 . The interconversion of these 3 functions of metabolites as matter, information, and energy requires the movement of molecules up and down gradients between membrane-bounded compartments-between organelles, between cells, the extracellular space, and between organ systems-such that information is stored in the form of dynamically-interacting structures acting as sub-systems in the larger whole-body system. The subsystems of the body are connected and coordinated by the flow of materials they exchange 28-30 , the energy they produce 31,32 , and the information they store as structural order 33,34 and epigenetic modifications 35,36 .

Potential Clinical Implications for Prognosis, Management and Treatment
Low methylcysteine was the single most predictive risk factor for recurrence found in females with recurrent major depression in remission. Low methylcysteine has been found to be a risk factor for a number of other mental health disorders including schizophrenia 37,38 . Dietary sources of methylcysteine include onions, garlic, and cruciferous vegetables like cabbage, broccoli, and kale and many legumes like peas, soy, and kidney beans 39 . Recent clinical trials 40,41 have found that dietary interventions can increase plasma methylcysteine levels, in addition to several sphingolipids and phospholipids. However, methylcysteine alone accounted for only 6% of the risk of recurrence in females and was not a predictor of recurrence in males. The broader benefits of tailored healthy dietary interventions should be considered as a potential intervention for patients with recurrent major depressive disorder. Low phospholipid and sphingolipids, which can also be increased by including a healthy diet, together added another 47% of the recurrence risk in females and 68% in males. The importance of sphingolipids in depression has been underscored by the recent demonstration that a common mechanism of action for many antidepressant drugs is to inhibit the stress-related conversion of sphingomyelin to ceramide by inhibiting acid sphingomyelinase (ASM), thereby restoring more normal sphingomyelin levels, autophagy, and organellar quality control 42 .
Abnormalities in metabolites traceable to the microbiome also accounted for 6% of the recurrence risk in both males and females. Although low methylcysteine was the top predictor of recurrence in females, a decrease in 10 sphingolipids and 5 phospholipids was responsible for a larger proportion of the overall metabolic risk. Low 15-HETE, b-carotene, and 7 sphingolipids were the top predictors in males.
If methylcysteine, phospholipids, sphingolipids (ceramides, sphingomyelins, glycosphingolipids), bcarotene, and the microbiome are addressed together, more than 80% of the metabolic risk of recurrence might be amenable to intervention. However, vitamin supplementation alone is not a solution. Accumulating evidence suggests that over-supplementation with purified vitamins can distort their natural stoichiometric proportions in vivo. Vitamin ratio distortions can produce relative deficiencies in unsupplemented vitamins because balanced proportions in vitamin cofactors are required to maintain balanced metabolic fluxes. Improvements in weekly activity and exercise can be used to facilitate the whole body integration of metabolism, and improve brain-body and interorgan communication, mitochondrial quality control by mitophagy 43 , lipid, endocrine, inflammation, and microbiome health [44][45][46] . Clinical trials of a whole food diet rich in plant-based foods containing methylcysteine, carotenoids, phospholipids, sphingolipids, and fiber for a healthier microbiome, with 23.
Burnstock, G. The therapeutic potential of purinergic signalling.  Figure S1. Principal component analysis of metabolomics in rrMDD. A. Diagnostic differences, rrMDD vs healthy controls (n = 84 females). B. Prognostic differences, prospective analysis of rrMDD subjects with recurrence and non-recurrence in 2.5 years (n = 42 females). C. Diagnostic differences, rrMDD vs healthy controls (n = 42 males). D. Prognostic differences, prospective analysis of rrMDD subjects with recurrence and non-recurrence in 2.5 years (n = 20 males). The green lines indicate the cumulative proportion of phenotypic variation explained by up to 5 principal components. The blue lines indicate the fraction of variance explained by each component.     Table S1. Raw AUC and recurrence data. Females Table S2. Raw AUC and recurrence data. Males Table S3. Biochemical pathways disturbed with recurrent major depressive disorder in remission: Females. Table S4. Biochemical pathways disturbed with recurrent major depressive disorder in remission: Males. Table S5. Statistical analysis. Females Table S6. Statistical analysis. Males Table S7. Prognostic pathways. Females Table S8. Prognostic pathways. Males Table S9. Representative diagnostic and prognostic metabolites for recurrent major depressive disorder in remission and the risk of recurrence. Females Table S10. Representative diagnostic and prognostic metabolites for recurrent major depressive disorder in remission and the risk of recurrence. Males Table S11. Log transformations. Females Table S12. Z-score transformations. Females Table S13. Log transformations. Males Table S14. Z-score transformations. Males

Figure S1
LysoPC ( 14,8,    in females with a history of recurrent major depressive disorder in drug-free remission identified by Cox proportional hazard analysis. rrMDD subjects were followed prospectively for 2.5 years: n = 42 females (24 with recurrence, 18 no recurrence). re S2 Figure S5. Kaplan-Meier style recurrence curves for the top metabolic predictors of recurrence risk in males with a history of recurrent major depressive disorder in drug-free remission identified by Cox proportional hazard analysis. rrMDD subjects were followed prospectively for 2.5 years: n = 20 males (11 with recurrence, 9 no recurrence). e S3