{"paper_id":"0dbd1ff6-8296-4ba0-b784-56305b6dd70f","body_text":"1\nTitle: Interplay of Human Metabolome and Gut Microbiome in Major \nDepression\nNajaf Amin1*, Jun Liu1*, Bruno Bonnechere1,2, Siamak MehmoudianDehkordi3, Matthias \nArnold3,4, Richa Batra5, Yu-Jie Chiou,1,6 Marco Fernandes7, M. Arfan Ikram8, Robert Kraaij9, \nJan Krumsiek5, Danielle Newby7, Kwangsik Nho10, Djawad Radjabzadeh9, Andrew J Saykin10, \nLiu Shi7, William Sproviero7, Laura Winchester7, Yang Yang,1,11, Alejo J Nevado-Holgado7, \nGabi Kastenmüller4, Rima F Kaddurah-Daouk3*, Cornelia M van Duijn1*\n* Contributed equally \n1. Nuffield Department of Population Health, University of Oxford, Oxford, UK \n2. REVAL Rehabilitation Research Center, Faculty of Rehabilitation Sciences, UHasselt, \nBelgium \n3. Department of Psychiatry and Behavioral Sciences, Duke University, Durham, NC, USA. \n4. Institute of Computational Biology, Helmholtz Zentrum München - German Research \nCenter for Environmental Health, Neuherberg, Germany \n5. Department of Physiology and Biophysics, Institute for Computational Biomedicine, \nEnglander Institute for Precision Medicine, Weill Cornell Medicine, New York, NY 10021, \nUSA \n6. Department of Psychiatry, Kaohsiung Chang Gung Memorial Hospital, Chang Gung  \nUniversity College of Medicine, Kaohsiung, Taiwan \n7. Department of Psychiatry, University of Oxford, Oxford, UK \n8. Department of Epidemiology, Erasmus University Medical Center, Rotterdam, the \nNetherlands \n9. Department of Internal Medicine, Erasmus University Medical Center, Rotterdam, the \nNetherlands \n10. Center for Neuroimaging, Department of Radiology and Imaging Sciences and Indiana \nAlzheimer’s Disease Research Center, Indiana University School of Medicine, Indianapolis, \nIN, USA \n11. Department of Computer Science and Engineering, Shanghai Jiao Tong University, \nShanghai, China \nKeywords: Major depression, metabolome, Krebs/TCA cycle, gut microbiome, lipids\n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprintthis version posted June 22, 2022. ; https://doi.org/10.1101/2022.06.21.22276700doi: medRxiv preprint \nNOTE: This preprint reports new research that has not been certified by peer review and should not be used to guide clinical practice.\n\n2\nAbstract:\nThe pathogenesis of depression is complex involving the interplay of genetic and \nenvironmental risk factors including diet, lifestyle and the gut microbiome. Metabolomics \nstudies may shed light on the interplay of these factors.  We study over 63,000 individuals \nincluding 8462 cases with a lifetime major depression and 5403 cases with recurrent major \ndepression from the UK Biobank profiled for nuclear magnetic resonance (NMR) \nspectroscopy based metabolites with the Nightingale platform. We identify 124 metabolites \nthat are associated with major depressive disorder (MDD), including 49 novel associations. \nNo differences were seen between the metabolic profiles of lifetime and recurrent MDD.  \nWe find that metabolites involved in the tricarboxylic acid (TCA) cycle are significantly \naltered in patients with MDD. Integrating the metabolic signatures of major depression and \nthe gut microbiome, we find that the gut microbiome might play an important role in the \nrelationship between these metabolites, lipoproteins in particular, and MDD. The order \nClostridiales, and the phyla Proteobacteria and Bacteroidetes were the most important taxa, \nwhich link the lipoprotein particles to MDD. Our study shows that at the molecular level \nenergy metabolism is disturbed in patients with MDD and that the interplay between the \ngut microbiome and blood metabolome may play a key role in the pathogenesis of MDD.\n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprintthis version posted June 22, 2022. ; https://doi.org/10.1101/2022.06.21.22276700doi: medRxiv preprint \n\n3\nIntroduction \nMajor depression is an important determinant of population health, affecting people across \nthe life span from adolescence to old age. The disease is associated with a plethora of \ndebilitating symptoms beyond emotional dysregulation1, spanning from cognition, motoric \nfunction, neurovegetative symptoms to inflammation and disturbances of the immune \nsystem. MDD is further linked to increased risks of cardiometabolic disorders and mortality2. \nCurrently, most antidepressant therapies modulate the monoamine pathway, but evidence \nis increasing for a more complex interplay of multiple pathways involving a wide range of \nmetabolic alterations spanning energy metabolism3 lipid metabolism4 and inflammation5.  \nThere is increased interest in energy metabolism and mitochondrial dysregulation as a \ncontributor to the pathogenesis of major depression6,7. The brain has high aerobic activity \nrequiring 20 times more energy than the rest of the body8 and is vulnerable to impaired \nenergy production6. A decreased brain energy production has been found in depressed \npatients 9. An excess of structural mutations in the mitochondrial DNA have been reported \nin individuals diagnosed with depression compared to controls6. There is a high prevalence \n(54%) of depression in patients with mitochondrial diseases10. Symptoms of depression such \nas loss of energy and appetite, tiredness, weakness, cognitive impairments, and sleep \ndisturbance are also frequent in mitochondrial diseases9,11 Mitochondria are cell organelles \nthat provide energy in the form of adenosine triphosphate via oxidative phosphorylation. \nSimultaneously, mitochondria are also responsible for generating reactive oxygen species \n(ROS) and anti-oxidants such as creatine, coenzyme Q10, niconitamide and glutathione, \nwhich protect cells from various deleterious effects of ROS12. The imbalance in the \ngeneration of ROS and antioxidants leads to oxidative stress, damages to lipids and proteins, \ninflammation and apoptosis 13,14.   \nDisturbed plasma lipid concentrations have been implicated in the development of \nmitochondrial dysfunction13, with high density lipoprotein (HDL) inversely correlating with \nmitochondrial DNA damage15. A recent study using Nightingale’s proton nuclear magnetic \nresonance (NMR) metabolomics platform in nine Dutch cohorts (5,283 patients with \ndepression and 10,145 controls) showed a shift towards decreased levels of high density \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprintthis version posted June 22, 2022. ; https://doi.org/10.1101/2022.06.21.22276700doi: medRxiv preprint \n\n4\nlipoprotein (HDL)-cholesterol and increased levels of very low density lipoprotein (VLDL)-\ncholesterol and tri- and diglycerides particles in patients with depression 16. The relevance \nand molecular mechanisms underlying this shift are not understood.  Gut microbiome has \nbeen shown to be a major determinant of the circulating lipids, specifically triglycerides and \nHDL17 and is also known to regulate mitochondrial function through the production of \nmicrobial metabolites including short chain fatty acids (SCFA), lipids, vitamins and amino \nacids.18 Disruptions to the gut microbiome have been found in patients with major \ndepressive disorder19,20. Metabolic signatures of the gut microbiome can be found in feces \nand blood17,21,22. In our recent study connecting the gut microbiome with the Nightingale’s \nmetabolites21 we observed association of metabolites in plasma with 32 gut microbial \ngroups. A higher abundance of family Christensenellaceae, genera ChristensenellaceaeR7 \ngroup, Ruminococcaceae (UCG002, UCG003, UCG005, UCG010, UCG014), Coprococcus, \nRuminococcaceaeNK4A214group and Ruminiclostridium6 associated with a favourable lipid \nprofile, i.e., decreased VLDLs and increased HDLs. Interestingly, decreased abundance of the \nsame groups of gut microbiota we also find associated with higher scores on depressive \nsymptoms in our study of gut microbiome and depression20. This raises the questions 1) \nwhether the gut microbiome explains part of the shift in VLDL and HDL levels seen in \npatients with depression 16 and 2) can we use the metabolic signatures of the disease based \non Nightingale’s metabolites as a tool to infer the association between gut microbiome and \nthe disease.  \nIn this study, we harness the power of the UK biobank (UKB) to study over 63000 individuals \nincluding 8462 patients with MDD and 5403 patients with recurrent MDD, who are profiled \nfor NMR spectroscopy based metabolites with the Nightingale platform. The largest study \nconducted to date discovers dysregulation of metabolites involved in the mitochondrial \nfunctioning in patients with MDD. When integrating the data with those of the Rotterdam \nStudy to understand the interplay between the blood metabolome, gut microbiome and \nMDD, we find evidence that the altered gut microbiome in patients is a key player in the \nshift of the VLDL/HDL axis in MDD patients.\n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprintthis version posted June 22, 2022. ; https://doi.org/10.1101/2022.06.21.22276700doi: medRxiv preprint \n\n5\nMethods\nStudy design and participants\nThe study was performed in the UK Biobank dataset, which comprises of more than 500,000 \nparticipants aged from 37 to 73 years during recruitment (2006 to 2010) for whom blood \nsampling was performed 23. A random subset of 118,466 individuals was profiled for \nmetabolites using a high-throughput 1H-NMR metabolomics (Nightingale Health, Helsinki, \nFinland) platform. The participants were registered with the UK National Health service and \nfrom 22 assessment centres across England, Wales, and Scotland using standardised \nprocedures for data collection which included a wide range of questionnaires, anthropological \nmeasurement, clinical biomarkers, genotype data, etc. All participants provided electronically \nsigned informed consent. UK Biobank has approval from the Northwest Multi-centre \nResearch Ethics Committee, the Patient Information Advisory Group, and the Community \nHealth Index Advisory Group. Further detail on the rationale, study design, survey methods, \ndata collection are available elsewhere 23. The current study is a part of UK Biobank projects \n30418 and 54520.\nDefinition of traits\nFor the initial analyses we considered two phenotypes including 1) lifetime major depressive \ndisorder (MDD) and 2) recurrent major depressive disorder. Both lifetime and recurrent major \ndepressive disorder were defined using the UK Biobank field code 20126, ICD10 codes F32 \n(single episode) and F33 (recurrent) or if participants were on antidepressant therapy at the \nbaseline. Individuals who reported any other mental illnesses (e.g., bipolar disorder, \nschizophrenia, psychosis, etc.) were excluded from the study. Controls included individuals \nwho had not reported depression at the baseline.  \nDefinition of covariates\nThe covariates considered in the analysis included baseline age, sex, ethnicity, fasting time, \nassessment center, lifestyle factors including body mass index (BMI), smoking status, alcohol \nintake frequency, education and status regarding multiple medication from touchscreen or \nverbal interview and technical variables during the NMR measurement, i.e., batch and \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprintthis version posted June 22, 2022. ; https://doi.org/10.1101/2022.06.21.22276700doi: medRxiv preprint \n\n6\nspectrometer. Fasting time was defined as the time interval between the consumption of \nfood or drink and blood sampling and natural log-transformed. Ethnicity was categorized to \nWhite, Asian (excluding Chinese), Black, Chinese, mixed and others. Smoking status was \ncategorized to never, previous and current. Alcohol intake frequency was categorized to 1) \ndaily or almost daily, 2) three to four times a week, 3) once or twice a week, 4) less than once \na week. Education was categorized to 1) College or University degree, 2) A levels, advanced \nsubsidiary (AS) levels or equivalent, 3) Certificated of secondary education (CSEs) or \nequivalent, 4) National vocational qualification (NVQ) or higher national diploma (HND) or \nhigher national certificate (HNC) or equivalent, 5) O levels, general certificate of secondary \neducation (GCSEs) or equivalent, 6) Other professional qualifications, and 7) none of the \nabove based on the highest qualification. Information for those who chose “prefer not to \nanswer”, was put as missing. Medication status was based on the medication codes collected \nfrom the verbal interview which were further coded to Anatomical Therapeutic Chemical (ATC) \ncodes9. The medications considered in the covariates were selected based on our previous \npublication24, including five anti-hypertensives (C08, C09, C07, C03 and C02), anti-diabetes \n(metformin and other anti-diabetes under A10), lipid-lowering drugs (C10), digoxin (C01AA), \nanti-thrombotic (B01AC06), proton pump inhibitors (PPI, A02BC), hypnotics and sedatives \n(N05) and antidepressants (N06).  \nImputation of missing values in the covariates\nFast imputation of missing values by chained random forests was performed through the R \npackage missRanger to impute the missing values for the shared covariates, including \nsmoking status, BMI, alcohol intake frequency, education, and ethnicity. The information \nused in the imputation included baseline age, sex, smoking status, pack-years of smoking, \nalcohol intake frequency, physical activity from International Physical Activity Questionnaire \n(IPAQ) groups, ethnicity, BMI, education, blood pressure and waist-hip ratio. In brief, the large \nmatrix was imputed with maximum of ten chaining interactions and 200 trees and weighted \nby the number of non-missing values; three candidate non-missing values were selected from \nin the predictive mean matching steps. \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprintthis version posted June 22, 2022. ; https://doi.org/10.1101/2022.06.21.22276700doi: medRxiv preprint \n\n7\nMetabolite profiling\nThe metabolites were measured in plasma using the targeted high-throughput 1H-NMR \nmetabolomics platform of Nightingale (Nightingale Health Ltd; biomarker quantification \nversion 2020) 25,26 which includes 249 metabolites. They include clinical lipids, lipoprotein \nsubclass profiling with lipid concentrations within 14 subclasses, fatty acid composition, and \nvarious low-molecular weight metabolites such as amino acids, ketone bodies and glycolysis \nmetabolites quantified in molar concentration units. The technology is based a standardized \nprotocol of sample quality control and sample preparation, data storage and automated \nspectral analyses. \nThe data obtained from the baseline sampling was used 25. For the samples with repeated \nmeasurements of the metabolites, one of the values was extracted at random. The \nmetabolite values which were suggested to be technical errors in the quality control provided \nby Nightingale Health during the measurement procedure were treated as missing. A natural \nlogarithm transformation of each metabolite was performed for the analysis. The zero values \nwere replaced by the lowest value except for zero. Finally the transformed values were scaled \nto standard deviation units. \nReplication \nFor replication we considered the results from the previously published study performed in \nDutch cohorts by the BBMRI-NL consortium 16. The study included 5,283 patients with \ndepression and 10,145 controls, who were characterized using the Nightingale platform. We \nfurther looked up the association of metabolites in the Predictors of Remission in Depression \nto Individual and Combined Treatments (PReDICT) study. The design and clinical outcomes of \nPReDICT have been detailed previously 27. Briefly, the PReDICT study aimed to identify \npredictors and moderators of response to 12 weeks of randomly-assigned treatment with \nduloxetine (30-60 mg/day), escitalopram (10-20 mg/day) or cognitive behaviour therapy \n(CBT, 16 one-hour individual sessions). Eligible participants were adults aged 18-65 years with \nan active untreated major depressive episode without psychotic features. Severity of \ndepression at the randomization visit was assessed with the 17-item Hamilton Depression \nRating Scale (HRSD17) 28. Eligibility required an HRSD17 score ≥18 at the screening visit and ≥15 \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprintthis version posted June 22, 2022. ; https://doi.org/10.1101/2022.06.21.22276700doi: medRxiv preprint \n\n8\nat the randomization visit, indicative of moderate-to-severe depression. Active significant \nsuicide risk, current illicit drug use (assessed with urine drug screen) or a history of substance \nabuse in the three months prior to randomization, pregnancy, lactation, and uncontrolled \ngeneral medical conditions were all excluded. Details on metabolomics profiling and \nstatistical analysis are provided in the Supplemental text.\nStatistical analysis\nAll analyses were performed in R statistical software. Descriptive analysis was performed \nusing the ‘CBCgrps’ 29 library of R.\nMetabolome-wide association analysis \nWe used logistic regression to test the association of the metabolite levels with lifetime and \nrecurrent depression. We considered four models with increasing number of covariates in the \nsubsequent models to identify the effects of most known confounders in the regression \nanalysis. Model 1 was adjusted for age, sex, fasting time, ethnicity, assessment centre, and \ntechnical variables during the NMR measurement, i.e., batch and spectrometer; model 2 was \nadditionally adjusted for BMI; model 3 further for antidepressant use and model 4 \nadditionally adjusted for most known lifestyle factors including smoking status, alcohol intake \nfrequency, physical activity, level of education and medication use for most chronic diseases. \nThe current analysis included all 249 metabolites measured by Nightingale. False discovery \nrate (FDR) of 0.05 was used to identify significantly associated metabolites. We further \nperformed a sensitivity analysis by removing those on antidepressant therapy. Finally, we \nperformed forward regression analysis in R on model 4 adjusted residuals of depression-\nassociated metabolites to identify independent metabolites. A multivariate regression \nanalysis was performed including all metabolites selected from forward regression analysis. \nMultiple testing correction was performed using false discovery rate (FDR).\nIntegration of metabolic signatures of human gut microbiome and MDD \nTo identify patterns of correlation in the metabolic signatures of MDD and the human gut \nmicrobiome we estimated correlation coefficients from the effect estimates from linear \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprintthis version posted June 22, 2022. ; https://doi.org/10.1101/2022.06.21.22276700doi: medRxiv preprint \n\n9\nmixed regression of 1) Z-score MDD on Z-score Microbe and 2) Z-score Microbe on Z-score MDD. Z-\nscoreMDD are the results of association analysis of MDD with Nightingale metabolites from the \ncurrent study while Z-scoreMicrobe are the results of association analysis of the gut microbiome \nascertained with 16S RNA sequencing and with the Nightingale metabolites published earlier \nby Vojinovic et al.21. Metabolites were clustered to 20 groups based on the method of Li & Ji \n30. These groups were used as random effects in the linear mixed regression. This step was \nperformed to control for inflation in the statistic because of high correlation between the \nmetabolites. Correlation between the metabolic profiles of the 361 gut microbial taxa and \nMDD was estimated as the square root of the product of the effect estimates from the two \nregressions described above. Significance of the correlation was tested using the Student’s T \ntest. FDR was applied to correct for multiple testing. Next, we compared these T-Statistics \n(proxy association for MDD-Microbiome based on the MDD-metabolome) as obtained in the \npresent study with the Z-scores (direct association of MDD-Microbiome) from Djawad et al. \n20.  \nMendelian Randomisation (MR) \nTo elucidate the causal relationships between depression and the associated metabolites, \nwe performed bi-directional two-sample MR using the R package of TwoSampleMR for the \ninverse variance weighted MR, heterogeneity test and pleiotropy test from MR-Egger \nregression31. The default pipelines in the packages were used. In brief, for the analysis \nusing TwoSampleMR, the genetic score was based on the top SNPs (P-value < 10-6) with \nlinkage disequilibrium R2 < 0.001 within 10,000kbps clumping distance. The overlapping \nSNPs were used without seeking proxy SNPs as we assumed that each meaningful locus \nshould have multiple SNPs significant and overlapped. The metabolite GWAS were obtained \nfrom the MRC IEU OpenGWAS in MRbase31 using all the UK Biobank participants who were \nprofiled for Nightingale metabolites (n=118,000). For MDD we used the publicly available \nresults of the largest GWAS by Howard et al. 32.  \nResults \nBaseline characteristics of the studied samples are provided in Table 1. Cases consisted of \n8462 individuals with a lifetime major depression and 5403 patients with recurrent major \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprintthis version posted June 22, 2022. ; https://doi.org/10.1101/2022.06.21.22276700doi: medRxiv preprint \n\n10\ndepression and the controls included > 55420 participants. Patients are significantly \nyounger, are more often female, smokers, have higher education and have a higher body \nmass index, are less physically active, consume less alcohol and have a black/mixed ethnic \nbackground compared to the controls (Table 1). Of the patients with history of depression, \n1958 (23%) were using anti-depressants at the time blood was drawn for the metabolomic \ncharacterization. The patients were found to use more often medication related to gastric \ndiseases, pain and addiction (Table 1).  \nMetabolome-wide association analysis for major depression using the Nightingale platform \nResults of the analysis are shown in Figure 1 and Supplementary Figures 1-4. Adjusting for \nage, sex and technical covariates, 178 (71.0%) metabolites of the 249 metabolites tested \nwere significantly (false discovery rate (FDR) < 0.05) associated with MDD in the basic model \n1. When adjusting for BMI (model 2), 163 (65.5%) metabolites remained significantly \nassociated with MDD and further adjusting for antidepressant use (model 3) yielded 132 \n(53.0%) metabolites significantly associated (FDR<0.05). In the full model further adjusted \nfor lifestyle factors including physical activity, alcohol consumption, smoking, education and \nmedication use for cardio-vascular morbidity (model 4), a total of 124 (49.8%) metabolites \nremained significantly associated with MDD (Figure 1, Supplementary Table 1). These \ninclude 27 small to extremely large VLDL particles (chylomicrons) that were increased in \nindividuals with MDD patients, 18 medium to very large sized HDL particles all of which \nwere decreased in MDD, except the triglycerides in the small and medium HDL particles, \nand 5 intermediate density lipoprotein (IDL) particles all of which, except the triglyceride \ncontent in IDL, were decreased in MDD patients. Among fatty acids, total monounsaturated \nfatty acids (MUFA) and its ratio to total fatty acids was significantly increased in MDD while \nthe ratios of linoleic acid (LA), omega 6 and polyunsaturated fatty acid (PUFA) to total fatty \nacids were significantly decreased in MDD patients. Further, apolipoprotein A1 (ApoA1), \ncholesteryl esters, citrate and sphingomyelins were significantly decreased in MDD while \nalanine and pyruvate were significantly increased in MDD patients (Figure 1, Supplementary \nTable 1). Findings were very similar when excluding those with antidepressant use instead \nof adjusting for antidepressants (Supplementary Figure 5, Supplementary Table 2). \nComparing the results of lifetime MDD with that of the recurrent depression, the metabolic \nprofiles were found to be highly correlated (Supplementary Figures 1-4). \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprintthis version posted June 22, 2022. ; https://doi.org/10.1101/2022.06.21.22276700doi: medRxiv preprint \n\n11\nComparing our results with those of the BBMRI-NL 16, 113 of the metabolites that we find \nassociated to major depression were also studied by BBMRI-NL. Supplementary Figure 6 \nshows that the effect direction of the identified metabolites from model 4 in our study is \nconsistent with that seen in the previous study by the BBMRI-NL consortium \n(Supplementary Figure 6, Supplementary Table 3. In the present study, we identified 49 \nmetabolites that were FDR significant that were not reported in the BBMRI-NL study \n(Supplementary Table 4). These include metabolites involved in mitochondrial functioning, \nincluding alanine, citrate, pyruvate, fatty acids including PUFA%, LA% and omega6%, \nsphingomyelins, IDL subfractions in addition to some VLDL and HDL subfractions that were \nnot associated earlier. The association of depression to omega 6, PUFA, citrate and pyruvate \nwas replicated in the PReDICT study (Supplementary Table 5). We could not replicate the \nfinding on alanine and no data were available on sphingomyelins, IDL or other lipid \nfractions, although most of these show consistency in direction of effects in the BBMRI-NL \nstudy (Supplementary Figure 6). \nMendelian Randomization (MR) analysis \nResults of bidirectional MR are provided in the Figure 2 and Supplementary Table 6. No \nsignificant pleiotropy was observed (Supplementary Table 7) in the MR analysis. However, \nsignificant heterogeneity was observed for most metabolites (Supplementary Table 8) \nsuggesting violation of assumptions of MR. We therefore considered the results of the \nweighted median method as these have shown to produce consistent results in the \npresence of heterogeneity33. Significant MR results were obtained after multiple testing \ncorrection using FDR < 0.05 (Figure 2) when MDD was used as the exposure and metabolites \nas the outcome. Changes in all metabolites except ApoA1, citrate, pyruvate, alanine, total \ncholesteryl ester, sphingomyelins, medium and large HDL particles, HDL cholesterol, total \nconcentration of HDL particles, total lipids in HDL and average diameter of HDL particles,  \nand a few VLDL subfractions Figure 2) appeared to be associated to the genes that are \nassociated to depression.  \nIntegration of human gut microbiome and metabolic signatures\nWe compared the pattern of association of MDD with the metabolites (metabolic signature \nof MDD) to the pattern of association of human gut microbial taxa with the metabolites \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprintthis version posted June 22, 2022. ; https://doi.org/10.1101/2022.06.21.22276700doi: medRxiv preprint \n\n12\n(metabolic signatures of the gut microbial taxa) (Supplementary Table 9, Figures 3 & 4). The \nassociation of the gut microbiota and the Nightingale metabolites is based on an \nindependent study published earlier21. Figure 3 shows all taxa whose metabolic signatures \nshow significant positive correlation (r > 0.50 & FDR < 0.05) with that of MDD and Figure 4\nillustrates all taxa whose metabolic signatures show significant negative correlation (r < -\n0.50 & FDR < 0.05) with the metabolic signature of MDD. We henceforth refer to this \ncorrelation of the metabolic signatures of MDD and gut microbial taxa as ‘proxy association’ \nbetween MDD and gut microbiome. When comparing this proxy association with the direct \nassociation of depression with the gut microbiome from a previous independent study20, we \nfind a  highly significant correlation (r=0.58, p-value < 10-16)( Figure 5). This finding suggests \nthat the Nightingale platform can be used as a proxy measure for the microbiome and that \nthe correlation of the microbiome and MDD are mainly driven by lipoproteins, especially \nVLDL and HDL particles. Of note is that the bacteria which were associated with a healthy \nlipid profile, i.e., increased levels of HDL subfractions and decreased VLDL lipid levels, were \nfound to be decreased in MDD (Figure 3).  Conversely, bacteria which were associated with \nan unhealthy lipid profile, i.e., decreased levels of lipids in HDL particles and increased levels \nof lipids in VLDL, were found to be increased in patients with MDD (Figure 4). \nOverall, we observed 223 bacterial taxa significantly associated with MDD using proxy \nassociation (FDRcorr < 0.05) (Figure 6, Supplementary Table 9). Figure 6 shows the complete \ngut microbiome profile of MDD at hierarchical levels. Family Ruminococcaceae (r=-0.60, \nFDR=2.9*10-13) and most of its genera were significantly negatively correlated with MDD \npatients. Several other families belonging to the order Clostridiales (Clostridiaceae (r=-0.59, \nFDR=8.8*10-13), Christensenellaceae (-0.68, FDR=1.8*10-17), Peptostreptococcaceae (r=-0.50, \nFDR=2.1*10-09), Defluviitaleaceae (r=-0.45, FDR=9.4*10-08) and Peptococcaceae (r=-0.41, \nFDR=2.3*10-06)) also showed significant negative correlation with MDD.  Families \nLachnospiraceae (r=0.43, FDR= 6.3*10-07) and Eubacteriaceae (r=0.38, FDR=1.3*10-05) were \nsignificantly positively correlated with MDD, however, some genera belonging to these \nfamilies were significantly negatively correlated with MDD (Figure 6). Further, several \nfamilies belonging to the phylum Proteobacteria including Methanobacteriaceae, \nRhodospirillaceae, Desulfovibrionaceae, Pasteurellaceae, Neisseriaceae and \nOxalobacteraceae and phylum Bacteroidetes including Porphyromonadaceae, Rikenellaceae \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprintthis version posted June 22, 2022. ; https://doi.org/10.1101/2022.06.21.22276700doi: medRxiv preprint \n\n13\nand Prevotellaceae were significantly negatively correlated with MDD (Figure 6, \nSupplementary Table 9).  \nDiscussion  \nWe have identified 124 metabolites associated with major depression. No differences were \nobserved between the metabolic profiles of lifetime MDD and recurrent MDD.  The findings \nof our study corroborate well with those of the previously published large study of the \nBBMRI-NL consortium16 in that most metabolites (90%) show consistency in the direction of \nassociation. Compared to the study of the BBMRI consortium, we find 49 metabolites that \nwere not significantly associated to MDD earlier, including the amino acid alanine as well as \ncitrate and pyruvate, which are two key metabolites of the energy metabolism pathway. \nThe shift of the metabolites observed in depression was associated with bacterial taxa \nbelonging to order Clostridiales, and phyla Proteobacteria and Bacteroidetes. \nOur findings extend previous reports on the major changes in lipid metabolism seen in \npatients with MDD. These include novel associations with polyunsaturated fatty acids \n(PUFA%, LA%, omega6%), sphingomyelins and various lipoprotein subfractions. In this study \nwe replicate the shift in the VLDL / HDL axis in patients with MDD16. VLDLs are produced by \nthe liver and are rich in triglyceride. The removal of triglycerides from VLDL by muscle and \nadipose tissue results in the formation of IDL particles, which we also find significantly \nassociated with MDD in the present study (not reported earlier in BBMRI-NL). VLDL, IDL and \nLDL particles are atherogenic and are found to be increased in cardio-vascular morbidities34. \nHDL, on the other hand is synthesized in the liver and the intestine35. Apolipoprotein A1 \n(ApoA1) is the major structural protein of HDL accounting for 70% of HDL protein35. HDL \nparticles are responsible for reverse cholesterol transport and have anti-oxidant, anti-\ninflammatory, anti-thrombotic, and anti-apoptotic properties34. HDLs have been shown to \nbe protective against mitochondrial dysfunction and positively associated with \nmitochondrial oxidative function15. \nThe finding that various VLDL particles and MUFA are increased and HDL particles and \nAPOA1 are decreased in depression are consistent with those of previous studies16,36. Our \nstudy provides evidence suggesting that the microbiome is a key player in the shift in \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprintthis version posted June 22, 2022. ; https://doi.org/10.1101/2022.06.21.22276700doi: medRxiv preprint \n\n14\nVLDL/HDL axis in MDD. Integrating metabolic profiles of MDD and the gut microbiome, we \nfind that the shift in most metabolites, particularly the HDL and VLDL fractions are \nassociated to several families of gut microbiota. Those belonging to the order Clostridiales, \nand phylum Proteobacteria and Bacteroidetes are decreased in MDD patients and are \nassociated with high HDL lipid levels and low VLDL lipid levels in blood while families \nLachnospiracea and Eubacteriaceae are predicted to be increased in MDD patients and are \nassociated with low levels of HDL subfractions and high levels of VLDL subfractions.  Earlier \nstudies have shown that gut microbiome is a major determinant of the circulating lipids and \nhas a bidirectional relationship with mitochondrial function17,18. Members of microbiota \nfrom order Clostridiales are known to provide transforming growth factor β enriched \nenvironment for promotion and accumulation of regulatory T cells in the gut37. These \nregulatory T cells have been found to be reduced in mood disorders38. It is interesting to \nnote that we find several families belonging to the order Clostridiales associated with MDD \nin our proxy association. Our findings are in line with the previous study where Clostridiales\nwere found to be the predominant microbes mediating psychiatric disorders including \ndepression39. \nThis significant link between the microbiome and MDD may also shed light on the \ninterpretation of the Mendelian Randomization experiments that did not yield significant \nevidence that any of the human genes associated to the metabolites investigated here. \nHowever, we find evidence that changes in lipoproteins and their subfractions and fatty \nacids associate with genes involved in MDD. We hypothesise that the metabolic change we \nobserve is part of the disease process and that the genes that have been implicated in MDD \nexplain the shift towards increased VLDL and decreased HDL subfractions (except medium \nand large HDL particles) in MDD. This may be explained by the pleiotropic effects of the \ngenes determining lipid metabolism or by the changes in diet and physical activity that are \nconsequences of the changes in a patient’s mood. Alternatively, we hypothesise that the \nchange in these metabolites may be due to altered composition of the gut microbiome in \ndepressed individuals, which may be driven by the genes that determine MDD. Such a \ngenetically driven shift in microbiome is also seen in transgenic models for Alzheimer’s \ndisease (AD) in which the introduction of the major genes involved in AD in mice, resulted in \na shift of the microbiome40.  Although we cannot exclude that the VLDL levels in the blood \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprintthis version posted June 22, 2022. ; https://doi.org/10.1101/2022.06.21.22276700doi: medRxiv preprint \n\n15\ndrive the gut microbiome and MDD, this mechanism is not supported by the Mendelian \nRandomization. Based on the Mendelian Randomization analysis, we hypothesize that MDD \ngenes drive the gut microbiome, which determines the metabolic spectrum (VLDLs and \nsmall HDLs) in the blood17. The exact mechanism is to be determined in future experiments.  \nThe second major finding is the disruption in the mitochondrial metabolism, more \nspecifically the tricarboxylic acid (TCA)/Krebs cycle. We find significantly increased levels of \npyruvate and decreased levels of citrate, which are major components of the mitochondrial \nKrebs cycle41. To our knowledge, there is no evidence in human data for an association of \nthese metabolites to MDD. Decreased citrate levels were found in the urine of rats in \nchronic unpredictable mild stress depression model42,43. Further, increased mitochondrial \nactivity generating citrate and reduced oxidative stress parameters has been observed in \npatients with bipolar depression treated with lithium compared to untreated patients44. In \nthe brain, citrate contributes to the regulation of neuronal excitability chelating and \ncontrolling the availability of divalent ions such as Ca+2 and Mg+2 41,45. Extracellular citrate is \nalso used for neurotransmitter synthesis41,45. In animal models of induced oxidative stress, \noral administration of citrate decreased brain lipid peroxidation and inflammation, liver \ndamage, and DNA fragmentation46. 90% of the total citrate in the body is localized in \nbones47,48. Osteoblast metabolic production of citrate provides the source of citrate in bone \nin the presence of zinc, which inhibits the oxidation of citrate in mitochondria48,49.  Citrate is \nreleased in the plasma during bone resorption47. Decreased levels of blood citrate \n(hypocitricemia) causes loss of bone citrate and osteoporosis. The relationship of depression \nand osteoporosis is well established50, however, Mendelian randomisation study showed no \ncausal effect of depression on osteoporosis51. Interestingly, administration of vitamin D \nincreases the plasma and bone citrate concentrations by inhibiting mitochondrial citrate \noxidation47. This implies that, if plasma citrate deficiency has a causal influence on major \ndepression, the subgroup of individuals with MDD exhibiting low citrate levels may very well \nbe treated with zinc and/or vitamin D supplements, both of which have shown to \nameliorate symptoms of depression52,53.  \nLow citrate levels may also be a result of intestinal dysbiosis or impairment of the pyruvate \ndehydrogenase complex that catalyses the conversion of pyruvate into acetyl-CoA54.  One of \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprintthis version posted June 22, 2022. ; https://doi.org/10.1101/2022.06.21.22276700doi: medRxiv preprint \n\n16\nthe questions to answer in future studies is whether acetyl-CoA links the findings on lipids \nand amino acids. Cholesterol synthesis initiates from acetyl coenzyme A (acetyl-CoA)55. \nAcetyl-CoA is synthesized in mitochondria from the oxidative decarboxylation of pyruvate, \noxidation of fatty acids or oxidative degradation of some amino acids (e.g., phenylalanine, \ntyrosine, leucine, lysine, and tryptophan)56. Acetyl-CoA is transported out of the \nmitochondria after being converted into citrate56. Cytosolic acetyl-CoA is used in fatty acid \nmetabolism and lipid biosynthesis 57. We hypothesize that citrate, which we found to be \ndecreased in depression may be the key metabolite connecting both the lipid and energy \nmetabolism pathways via Acetyl-CoA.  \nIn conclusion, we have performed the largest and most comprehensive study investigating \nthe association of NMR metabolites with major depression. We find that metabolites \ninvolved in the tricarboxylic acid (TCA) cycle are significantly altered in patients with MDD, \nsuggesting perturbations in metabolites involved in energy metabolism in patients with \nMDD. Our finding that the interplay between the gut microbiome and the blood \nmetabolome may play a key role in MDD and suggests that the gut microbiome may be a \ntarget for novel preventive and therapeutic interventions for MDD.\n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprintthis version posted June 22, 2022. ; https://doi.org/10.1101/2022.06.21.22276700doi: medRxiv preprint \n\n17\nReferences \n1. Chesney, E., Goodwin, G.M. & Fazel, S. Risks of all-cause and suicide mortality in \nmental disorders: a meta-review. World Psychiatry 13, 153-60 (2014). \n2. Friedrich, M.J. Depression Is the Leading Cause of Disability Around the World. JAMA\n317, 1517 (2017). \n3. Rezin, G.T., Amboni, G., Zugno, A.I., Quevedo, J. & Streck, E.L. Mitochondrial \ndysfunction and psychiatric disorders. Neurochem Res 34, 1021-9 (2009). \n4. Parekh, A., Smeeth, D., Milner, Y. & Thure, S. The Role of Lipid Biomarkers in Major \nDepression. Healthcare (Basel) 5(2017). \n5. Maeng, S.H. & Hong, H. Inflammation as the Potential Basis in Depression. Int \nNeurourol J 23, S63-71 (2019). \n6. Allen, J., Romay-Tallon, R., Brymer, K.J., Caruncho, H.J. & Kalynchuk, L.E. Mitochondria \nand Mood: Mitochondrial Dysfunction as a Key Player in the Manifestation of \nDepression. Frontiers in Neuroscience 12(2018). \n7. Bansal, Y. & Kuhad, A. Mitochondrial Dysfunction in Depression. Current \nNeuropharmacology 14, 610-618 (2016). \n8. Kety, S.S. Blood flow and metabolism of the human brain in health and disease. Trans \nStud Coll Physicians Phila 18, 103-8 (1950). \n9. Manji, H.  et al.  Impaired mitochondrial function in psychiatric disorders. Nature \nReviews Neuroscience 13, 293-307 (2012). \n10. Fattal, O., Link, J., Quinn, K., Cohen, B.H. & Franco, K. Psychiatric comorbidity in 36 \nadults with mitochondrial cytopathies. CNS Spectr 12, 429-38 (2007). \n11. Gardner, A. & Boles, R.G. Beyond the serotonin hypothesis: Mitochondria, \ninflammation and neurodegeneration in major depression and affective spectrum \ndisorders. Progress in Neuro-Psychopharmacology & Biological Psychiatry35, 730-743 \n(2011). \n12. Bansal, Y. & Kuhad, A. Mitochondrial Dysfunction in Depression. Curr Neuropharmacol\n14, 610-8 (2016). \n13. Madamanchi, N.R. & Runge, M.S. Mitochondrial dysfunction in atherosclerosis. Circ \nRes 100, 460-73 (2007). \n14. Allen, J., Romay-Tallon, R., Brymer, K.J., Caruncho, H.J. & Kalynchuk, L.E. Mitochondria \nand Mood: Mitochondrial Dysfunction as a Key Player in the Manifestation of \nDepression. Front Neurosci 12, 386 (2018). \n15. Parikh, N.I.  et al.  Lipoprotein concentration, particle number, size and cholesterol \nefflux capacity are associated with mitochondrial oxidative stress and function in an \nHIV positive cohort. Atherosclerosis 239, 50-4 (2015). \n16. Bot, M.  et al. Metabolomics Profile in Depression: A Pooled Analysis of 230 Metabolic \nMarkers in 5283 Cases With Depression and 10,145 Controls. Biol Psychiatry 87, 409-\n418 (2020). \n17. Fu, J.  et al.  The Gut Microbiome Contributes to a Substantial Proportion of the \nVariation in Blood Lipids. Circ Res 117, 817-24 (2015). \n18. Franco-Obregon, A. & Gilbert, J.A. The Microbiome-Mitochondrion Connection: \nCommon Ancestries, Common Mechanisms, Common Goals. mSystems 2(2017). \n19. Simpson, C.A.  et al.  The gut microbiota in anxiety and depression - A systematic \nreview. Clin Psychol Rev 83, 101943 (2021). \n20. Djawad, R.  et al. Nature Portfolio (2021). \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprintthis version posted June 22, 2022. ; https://doi.org/10.1101/2022.06.21.22276700doi: medRxiv preprint \n\n18\n21. Vojinovic, D.  et al. Relationship between gut microbiota and circulating metabolites in \npopulation-based cohorts. Nat Commun 10, 5813 (2019). \n22. Zierer, J.  et al. The fecal metabolome as a functional readout of the gut microbiome. \nNat Genet 50, 790-795 (2018). \n23. Sudlow, C.  et al. UK biobank: an open access resource for identifying the causes of a \nwide range of complex diseases of middle and old age. PLoS Med 12, e1001779 (2015). \n24. Liu, J.  et al. Integration of epidemiologic, pharmacologic, genetic and gut microbiome \ndata in a drug-metabolite atlas. Nat Med 26, 110-117 (2020). \n25. Ritchie, S.C.  et al. Quality control and removal of technical variation of NMR metabolic \nbiomarker data in ∼120,000 UK Biobank participants. medRxiv, 2021.09.24.21264079 \n(2021). \n26. Julkunen, H., Cichonska, A., Slagboom, P.E., Wurtz, P. & Nightingale Health, U.K.B.I. \nMetabolic biomarker profiling for identification of susceptibility to severe pneumonia \nand COVID-19 in the general population. Elife 10(2021). \n27. Dunlop, B.W.  et al. Predictors of remission in depression to individual and combined \ntreatments (PReDICT): study protocol for a randomized controlled trial. Trials 13, 106 \n(2012). \n28. Hamilton, M. A rating scale for depression. J Neurol Neurosurg Psychiatry 23, 56-62 \n(1960). \n29. Zhang, Z., Gayle, A.A., Wang, J., Zhang, H. & Cardinal-Fernandez, P. Comparing \nbaseline characteristics between groups: an introduction to the CBCgrps package. Ann \nTransl Med 5, 484 (2017). \n30. Li, J. & Ji, L. Adjusting multiple testing in multilocus analyses using the eigenvalues of \na correlation matrix. Heredity (Edinb) 95, 221-7 (2005). \n31. Hemani, G.  et al. The MR-Base platform supports systematic causal inference across \nthe human phenome. Elife 7(2018). \n32. Howard, D.M.  et al.  Genome-wide meta-analysis of depression identifies 102 \nindependent variants and highlights the importance of the prefrontal brain regions. \nNat Neurosci 22, 343-352 (2019). \n33. Bowden, J., Davey Smith, G., Haycock, P.C. & Burgess, S. Consistent Estimation in \nMendelian Randomization with Some Invalid Instruments Using a Weighted Median \nEstimator. Genet Epidemiol 40, 304-14 (2016). \n34. Feingold, K.R. Introduction to Lipids and Lipoproteins. in Endotext (eds. Feingold, K.R.\net al.) (South Dartmouth (MA), 2000). \n35. Mahley, R.W., Innerarity, T.L., Rall, S.C., Jr. & Weisgraber, K.H. Plasma lipoproteins: \napolipoprotein structure and function. J Lipid Res 25, 1277-94 (1984). \n36. Segoviano-Mendoza, M.  et al. Hypocholesterolemia is an independent risk factor for \ndepression disorder and suicide attempt in Northern Mexican population. Bmc \nPsychiatry 18(2018). \n37. Imdad, S., Lim, W., Kim, J.H. & Kang, C. Intertwined Relationship of Mitochondrial \nMetabolism, Gut Microbiome and Exercise Potential. Int J Mol Sci 23(2022). \n38. Bauer, M.E. & Teixeira, A.L. Neuroinflammation in Mood Disorders: Role of Regulatory \nImmune Cells. Neuroimmunomodulation 28, 99-107 (2021). \n39. Li, J.J.  et al. Clostridiales are predominant microbes that mediate psychiatric disorders. \nJournal of Psychiatric Research 130, 48-56 (2020). \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprintthis version posted June 22, 2022. ; https://doi.org/10.1101/2022.06.21.22276700doi: medRxiv preprint \n\n19\n40. Ł uc, M. et al. Gut microbiota in dementia. Critical review of novel findings and their \npotential application. Progress in Neuro-Psychopharmacology and Biological \nPsychiatry 104, 110039 (2021). \n41. Mycielska, M.E., Milenkovic, V.M., Wetzel, C.H., Rummele, P. & Geissler, E.K. \nExtracellular Citrate in Health and Disease. Current Molecular Medicine 15, 884-891 \n(2015). \n42. Liu, X.J.  et al.  Urinary metabonomic study using a CUMS rat model of depression. \nMagnetic Resonance in Chemistry 50, 187-192 (2012). \n43. Zheng, S.N.  et al.  Urinary metabonomic study on biochemical changes in chronic \nunpredictable mild stress model of depression. Clinica Chimica Acta 411, 204-209 \n(2010). \n44. Machado-Vieira, R.  et al.  Oxidative stress parameters in unmedicated and treated \nbipolar subjects during initial manic episode: A possible role for lithium antioxidant \neffects. Neuroscience Letters 421, 33-36 (2007). \n45. Westergaard, N.  et al.  Uptake, Release, and Metabolism of Citrate in Neurons and \nAstrocytes in Primary Cultures. Journal of Neurochemistry 62, 1727-1733 (1994). \n46. Abdel-Salam, O.M.  et al.  Citric acid effects on brain and liver oxidative stress in \nlipopolysaccharide-treated mice. J Med Food 17, 588-98 (2014). \n47. Costello, L.C. & Franklin, R.B. Plasma Citrate Homeostasis: How It Is Regulated; And Its \nPhysiological and Clinical Implications. An Important, But Neglected, Relationship in \nMedicine. HSOA J Hum Endocrinol 1(2016). \n48. Costello, L.C., Franklin, R.B., Reynolds, M.A. & Chellaiah, M. The Important Role of \nOsteoblasts and Citrate Production in Bone Formation: \"Osteoblast Citration\" as a \nNew Concept for an Old Relationship. Open Bone J 4(2012). \n49. Franklin, R.B., Chellaiah, M., Zou, J., Reynolds, M.A. & Costello, L.C. Evidence that \nOsteoblasts are Specialized Citrate-producing Cells that Provide the Citrate for \nIncorporation into the Structure of Bone. Open Bone J 6, 1-7 (2014). \n50. Cizza, G., Primma, S., Coyle, M., Gourgiotis, L. & Csako, G. Depression and \nosteoporosis: a research synthesis with meta-analysis. Horm Metab Res 42, 467-82 \n(2010). \n51. He, B.  et al. Depression and Osteoporosis: A Mendelian Randomization Study. Calcif \nTissue Int 109, 675-684 (2021). \n52. Petrilli, M.A.  et al.  The Emerging Role for Zinc in Depression and Psychosis. Front \nPharmacol 8, 414 (2017). \n53. Shaffer, J.A.  et al. Vitamin D supplementation for depressive symptoms: a systematic \nreview and meta-analysis of randomized controlled trials. Psychosom Med 76, 190-6 \n(2014). \n54. Thomas, G.W., Mains, C.W., Slone, D.S., Craun, M.L. & Bar-Or, D. Potential \nDysregulation of the Pyruvate Dehydrogenase Complex by Bacterial Toxins and \nInsulin. Journal of Trauma-Injury Infection and Critical Care 67, 628-633 (2009). \n55. Lent-Schochet, D. & Jialal, I. Biochemistry, Lipoprotein Metabolism. in StatPearls\n(Treasure Island (FL), 2021). \n56. Bhagavan, N.V. & Ha, C.-E. Chapter 16 - Lipids I: Fatty Acids and Eicosanoids. in \nEssentials of Medical Biochemistry (Second Edition)  (eds. Bhagavan, N.V. & Ha, C.-E.) \n269-297 (Academic Press, San Diego, 2015). \n57. Mycielska, M.E.  et al. Citrate transport and metabolism in mammalian cells: prostate \nepithelial cells and prostate cancer. Bioessays 31, 10-20 (2009). \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprintthis version posted June 22, 2022. ; https://doi.org/10.1101/2022.06.21.22276700doi: medRxiv preprint \n\n20\nAcknowledgements\n“Metabolomics data is provided by the Alzheimer’s disease Metabolomics Consortium \n(ADMC) and funded wholly or in part by the following grants and supplements thereto: NIA \nR01AG046171, RF1AG051550, RF1AG057452, R01AG059093, RF1AG058942, U01AG061359, \nU19AG063744 and FNIH: #DAOU16AMPA awarded to Dr. Kaddurah-Daouk at Duke University \nin partnership with a large number of academic institutions. As such, the investigators within \nthe ADMC, not listed specifically in this publication’s author’s list, provided data along with \nits pre-processing and prepared it for analysis, but did not participate in analysis or writing of \nthis manuscript. A complete listing of ADMC investigators can be found \nat: https://sites.duke.edu/adnimetab/team/.” \nIn order to accurately acknowledge data gathering and pre-processing by the ADMC, data \nusers must include a laboratory specific acknowledgement statement in the methods section \nof manuscripts.  These study/dataset-specific acknowledgements can be found in the Study \nPages. Depending upon the length and focus of the article, it may be appropriate to include \nmore or less of these statements.  Draft: “The Nightingale datasets have been generated at \nNightingale Health.” \nConflicts of interest \n\"Dr. Kaddurah-Daouk in an inventor on a series of patents on use of metabolomics for the \ndiagnosis and treatment of CNS diseases and holds equity in Metabolon Inc., Chymia LLC \nand PsyProtix.\"  Alternatively, \"R.K-D. formed Chymia LLC and PsyProtix, a Duke University \nbiotechnology spinout aiming to transform the treatment of mental health disorders.\" \nAll other authors declare no conflicts of interest. \nDr. Saykin receives support from multiple NIH grants (P30 AG010133, P30 AG072976, R01 \nAG019771, R01 AG057739, U01 AG024904, R01 LM013463, R01 AG068193, T32 AG071444, \nand U01 AG068057 and U01 AG072177). He has also received support from Avid \nRadiopharmaceuticals, a subsidiary of Eli Lilly (in kind contribution of PET tracer precursor); \nBayer Oncology (Scientific Advisory Board); Eisai (Scientific Advisory Board); Siemens \nMedical Solutions USA, Inc. (Dementia Advisory Board); Springer-Nature Publishing \n(Editorial Office Support as Editor-in-Chief, Brain Imaging and Behavior). \nM.A. and G.K. received funding (through their institutions) from the National Institutes of \nHealth/National Institute on Aging through grants RF1AG058942, RF1AG059093, \nU01AG061359, U19AG063744, and R01AG069901. \nM.A. and G.K. are co-inventors (through Duke University/Helmholtz Zentrum München) on \npatents on applications of metabolomics in diseases of the central nervous system; M.A. \nand G.K. hold equity in Chymia LLC and IP in PsyProtix and Atai that are exploring the \npotential for therapeutic applications targeting mitochondrial metabolism in treatment-\nresistant depression.\n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprintthis version posted June 22, 2022. ; https://doi.org/10.1101/2022.06.21.22276700doi: medRxiv preprint \n\n21\nTable 1: Descriptive Statistics for the patients with MDD and controls and recurrent depression \nLifetime MDD Recurrent MDD \nVariables 0 (n = 55420) 1 (n = 8462) p 0 (n = 67805) 1 (n = 5403) p \nage0, Median (Q1,Q3) 59 (51, 64) 57 (49, 62) < 1e-10 58 (50, 64) 57 (49, 62) < 1e-10 \nsex, n (%) < 1e-10 < 1e-10 \nFemale 25987 (47) 5474 (65) 32638 (48) 3529 (65) \nMale 29433 (53) 2988 (35) 35167 (52) 1874 (35) \nBMI, Median (Q1,Q3) 26.67 (24.17, 29.65) 26.81 (24.11, 30.32) 1.31E-06 26.68 (24.15, 29.72) 26.93 (24.1, 30.55) 1.44E-07 \nFasting_time, Median (Q1,Q3) 1.1 (0.69, 1.39) 1.1 (1.1, 1.39) < 1e-10 1.1 (0.69, 1.39) 1.1 (1.1, 1.39) 2.30E-09 \nEthnicity, n (%) 2.00E-10 3.13E-06 \n  Asian 992 (2) 153 (2) 1420 (2) 111 (2) \n  Black 751 (1) 140 (2) 1078 (2) 102 (2) \n  Chinese 182 (0) 11 (0) 248 (0) 8 (0) \n  Mixed 247 (0) 81 (1) 312 (0) 51 (1) \n  Others 461 (1) 83 (1) 670 (1) 55 (1) \n  White 52787 (95) 7994 (94) 64077 (95) 5076 (94) \nEducation, n (%) < 1e-10 < 1e-10 \n  A_levels_AS_levels_or_equivalent 6127 (11) 1042 (12) 7413 (11) 655 (12) \n  CSEs_or_equivalent 2816 (5) 480 (6) 3636 (5) 304 (6) \n  College_or_University_degree 18209 (33) 2954 (35) 21836 (32) 1918 (35) \n  NVQ_or_HND_or_HNC_or_equivalent 4054 (7) 547 (6) 4921 (7) 346 (6) \n  None 9374 (17) 1117 (13) 11925 (18) 700 (13) \n  O_levels_GCSEs_or_equivalent 11742 (21) 1884 (22) 14414 (21) 1194 (22) \n  Other_professional_qualifications 3098 (6) 438 (5) 3660 (5) 286 (5) \nSmoking status, n (%) < 1e-10 < 1e-10 \n  current 4752 (9) 1101 (13) 6217 (9) 749 (14) \n  never 31917 (58) 4232 (50) 38827 (57) 2673 (49) \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprintthis version posted June 22, 2022. ; https://doi.org/10.1101/2022.06.21.22276700doi: medRxiv preprint \n\n22\n  previous 18751 (34) 3129 (37) 22761 (34) 1981 (37) \nalcohol_freq, n (%) < 1e-10 < 1e-10 \n  Daily_or_almost_daily 11791 (21) 1661 (20) 14071 (21) 1070 (20) \n  Less_than_once_a_week 14831 (27) 2938 (35) 19037 (28) 1917 (35) \n  Once_or_twice_a_week 14867 (27) 2038 (24) 18101 (27) 1265 (23) \n  Three_or_four_times_a_week 13931 (25) 1825 (22) 16596 (24) 1151 (21) \nphysical_activity, n (%) 8.48E-06 0.06170695 \n  high 22422 (40) 3299 (39) 27078 (40) 2100 (39) \n  low 9806 (18) 1678 (20) 12358 (18) 1051 (19) \n  moderate 23192 (42) 3485 (41) 28369 (42) 2252 (42) \nMedication \nA02BC_Proton_pump_inhibitors, n (%) 4176 (8) 1107 (13) < 1e-10 5561 (8) 735 (14) < 1e-10 \nC01AA_Digitalis_glycosides, n (%) 134 (0) 15 (0) 0.30530126 164 (0) 6 (0) 0.07576164 \nC10_LIPID_MODIFYING_AGENTS, n (%) 11083 (20) 1583 (19) 0.00578375 13527 (20) 1080 (20) 0.9589938 \nA10BA02_metformin, n (%) 1456 (3) 268 (3) 0.00482339 1840 (3) 207 (4) 2.01E-06 \nA10.excl.A10BA02_Anti_diabetes_excl.Metformin, n (%) 469 (1) 62 (1) 0.31366624 614 (1) 47 (1) 0.84782497 \nB01AC06_acetylsalicylic_acid, n (%) 7719 (14) 970 (11) 8.00E-10 9503 (14) 643 (12) 1.64E-05 \nC07_BETA_BLOCKING_AGENTS, n (%) 3617 (7) 561 (7) 0.73857387 4516 (7) 365 (7) 0.80899618 \nC08_CALCIUM_CHANNEL_BLOCKERS, n (%) 4137 (7) 583 (7) 0.06264847 5058 (7) 385 (7) 0.38234663 \nC03_DIURETICS, n (%) 4220 (8) 608 (7) 0.17061941 5133 (8) 392 (7) 0.41400884 \nC02_ANTIHYPERTENSIVES, n (%) 833 (2) 103 (1) 0.04660797 1010 (1) 69 (1) 0.2345284 \nC09_AGENTS_ACTING_ON_THE_RENIN_ANGIOTENSIN_SYSTEM, n (%) 7883 (14) 1100 (13) 0.00268185 9550 (14) 712 (13) 0.06769179 \nN02A_OPIOIDS, n (%) 1651 (3) 627 (7) < 1e-10 2475 (4) 416 (8) < 1e-10 \nN02B_OTHER_ANALGESICS_AND_ANTIPYRETICS, n (%) 14312 (26) 2770 (33) < 1e-10 18370 (27) 1839 (34) < 1e-10 \nN02C_ANTIMIGRAINE_PREPARATIONS, n (%) 429 (1) 147 (2) < 1e-10 579 (1) 101 (2) < 1e-10 \nN03A_ANTIEPILEPTICS, n (%) 545 (1) 253 (3) < 1e-10 816 (1) 134 (2) < 1e-10 \nN04A_ANTICHOLINERGIC_AGENTS, n (%) 9 (0) 11 (0) 1.15E-05 14 (0) 10 (0) 3.55E-06 \nN04B_DOPAMINERGIC_AGENTS, n (%) 110 (0) 26 (0) 0.05804231 147 (0) 12 (0) 1 \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprintthis version posted June 22, 2022. ; https://doi.org/10.1101/2022.06.21.22276700doi: medRxiv preprint \n\n23\nN05A_ANTIPSYCHOTICS, n (%) 101 (0) 212 (3) < 1e-10 127 (0) 111 (2) < 1e-10 \nN05B_ANXIOLYTICS, n (%) 54 (0) 88 (1) < 1e-10 81 (0) 64 (1) < 1e-10 \nN05C_HYPNOTICS_AND_SEDATIVES, n (%) 90 (0) 155 (2) < 1e-10 155 (0) 90 (2) < 1e-10 \nN06A_ANTIDEPRESSANTS, n (%) 46 (0) 1958 (23) < 1e-10 880 (1) 1297 (24) < 1e-10 \nN06B_PSYCHOSTIMULANTS_AGENTS, n (%) 4 (0) 3 (0) 0.05378305 5 (0) 1 (0) 0.36873461 \nN06C_PSYCHOLEPTICS_AND_PSYCHOANALEPTICS, n (%) 0 (0) 4 (0) 0.00030769 5 (0) 1 (0) 0.36873461 \nN06D_ANTI_DEMENTIA_DRUGS, n (%) 290 (1) 27 (0) 0.01609039 353 (1) 16 (0) 0.03214806 \nN07A_PARASYMPATHOMIMETICS, n (%) 14 (0) 5 (0) 0.09632632 25 (0) 3 (0) 0.45870756 \nN07B_DRUGS_USED_IN_ADDICTIVE_DISORDERS, n (%) 30 (0) 17 (0) 9.76E-06 43 (0) 9 (0) 0.01309514 \nN07C_ANTIVERTIGO_PREPARATIONS, n (%) 125 (0) 39 (0) 0.00010914 161 (0) 24 (0) 0.00556512 \nN01A_ANESTHETICS_GENERAL, n (%) 4 (0) 3 (0) 0.05378305 14 (0) 2 (0) 0.33285557 \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprintthis version posted June 22, 2022. ; https://doi.org/10.1101/2022.06.21.22276700doi: medRxiv preprint \n\n24\nFigure 1: Results of metabolome-wide association analysis \nLeft Y-axis shows effect estimates and right Y-axis shows the strength of association. Red dashed line is FDR=0.05. Points denote the -log10(FDR). \nLines denote the effect estimates from the four models. \nModel 1: MDD ~ age + sex + ethnicity + fasting time + technical covariates\nModel 2: MDD ~ age + sex + BMI + ethnicity + fasting time + technical covariates\nModel 3: MDD ~ age + sex + BMI + antidepressants + ethnicity + fasting time + technical covariates\nModel 4: MDD ~ age + sex + BMI + antidepressants + education + smoking + alcohol intake + physical activity + PPI + antidiabetics + \nantihypertensives + lipid modifying agents + sleep medication + ethnicity + fasting time + technical covariates \n−0.1 0.0 0.1 0.2\nEFFECT\nTotal_C\nnon_HDL_C\nRemnant_C\nVLDL_C\nClinical_LDL_C\nLDL_C\nHDL_C\nTotal_TG\nVLDL_TG\nLDL_TG\nHDL_TG\nTotal_PL\nVLDL_PL\nLDL_PL\nHDL_PL\nTotal_CE\nVLDL_CE\nLDL_CE\nHDL_CE\nTotal_FC\nVLDL_FC\nLDL_FC\nHDL_FC\nTotal_L\nVLDL_L\nLDL_L\nHDL_L\nTotal_P\nVLDL_P\nLDL_P\nHDL_P\nVLDL_size\nLDL_size\nHDL_size\nPhosphoglyc\nTG_by_PG\nCholines\nPhosphatidylc\nSphingomyelins\nApoB\nApoA1\nApoB_by_ApoA1\nTotal_FA\nUnsaturation\nOmega_3\nOmega_6\nPUFA\nMUFA\nSFA\nLA\nDHA\nOmega_3_pct\nOmega_6_pct\nPUFA_pct\nMUFA_pct\nSFA_pct\nLA_pct\nDHA_pct\nPUFA_by_MUFA\nOmega_6_by_Omega_3\nAla\nGln\nGly\nHis\nTotal_BCAA\nIle\nLeu\nVal\nPhe\nTyr\nGlucose\nLactate\nPyruvate\nCitrate\nbOHbutyrate\nAcetate\nAcetoacetate\nAcetone\nCreatinine\nAlbumin\nGlycA\nXXL_VLDL_P\nXXL_VLDL_L\nXXL_VLDL_PL\nXXL_VLDL_C\nXXL_VLDL_CE\nXXL_VLDL_FC\nXXL_VLDL_TG\nXL_VLDL_P\nXL_VLDL_L\nXL_VLDL_PL\nXL_VLDL_C\nXL_VLDL_CE\nXL_VLDL_FC\nXL_VLDL_TG\nL_VLDL_P\nL_VLDL_L\nL_VLDL_PL\nL_VLDL_C\nL_VLDL_CE\nL_VLDL_FC\nL_VLDL_TG\nM_VLDL_P\nM_VLDL_L\nM_VLDL_PL\nM_VLDL_C\nM_VLDL_CE\nM_VLDL_FC\nM_VLDL_TG\nS_VLDL_P\nS_VLDL_L\nS_VLDL_PL\nS_VLDL_C\nS_VLDL_CE\nS_VLDL_FC\nS_VLDL_TG\nXS_VLDL_P\nXS_VLDL_L\nXS_VLDL_PL\nXS_VLDL_C\nXS_VLDL_CE\nXS_VLDL_FC\nXS_VLDL_TG\nIDL_P\nIDL_L\nIDL_PL\nIDL_C\nIDL_CE\nIDL_FC\nIDL_TG\nL_LDL_P\nL_LDL_L\nL_LDL_PL\nL_LDL_C\nL_LDL_CE\nL_LDL_FC\nL_LDL_TG\nM_LDL_P\nM_LDL_L\nM_LDL_PL\nM_LDL_C\nM_LDL_CE\nM_LDL_FC\nM_LDL_TG\nS_LDL_P\nS_LDL_L\nS_LDL_PL\nS_LDL_C\nS_LDL_CE\nS_LDL_FC\nS_LDL_TG\nXL_HDL_P\nXL_HDL_L\nXL_HDL_PL\nXL_HDL_C\nXL_HDL_CE\nXL_HDL_FC\nXL_HDL_TG\nL_HDL_P\nL_HDL_L\nL_HDL_PL\nL_HDL_C\nL_HDL_CE\nL_HDL_FC\nL_HDL_TG\nM_HDL_P\nM_HDL_L\nM_HDL_PL\nM_HDL_C\nM_HDL_CE\nM_HDL_FC\nM_HDL_TG\nS_HDL_P\nS_HDL_L\nS_HDL_PL\nS_HDL_C\nS_HDL_CE\nS_HDL_FC\nS_HDL_TG\nXXL_VLDL_PL_pct\nXXL_VLDL_C_pct\nXXL_VLDL_CE_pct\nXXL_VLDL_FC_pct\nXXL_VLDL_TG_pct\nXL_VLDL_PL_pct\nXL_VLDL_C_pct\nXL_VLDL_CE_pct\nXL_VLDL_FC_pct\nXL_VLDL_TG_pct\nL_VLDL_PL_pct\nL_VLDL_C_pct\nL_VLDL_CE_pct\nL_VLDL_FC_pct\nL_VLDL_TG_pct\nM_VLDL_PL_pct\nM_VLDL_C_pct\nM_VLDL_CE_pct\nM_VLDL_FC_pct\nM_VLDL_TG_pct\nS_VLDL_PL_pct\nS_VLDL_C_pct\nS_VLDL_CE_pct\nS_VLDL_FC_pct\nS_VLDL_TG_pct\nXS_VLDL_PL_pct\nXS_VLDL_C_pct\nXS_VLDL_CE_pct\nXS_VLDL_FC_pct\nXS_VLDL_TG_pct\nIDL_PL_pct\nIDL_C_pct\nIDL_CE_pct\nIDL_FC_pct\nIDL_TG_pct\nL_LDL_PL_pct\nL_LDL_C_pct\nL_LDL_CE_pct\nL_LDL_FC_pct\nL_LDL_TG_pct\nM_LDL_PL_pct\nM_LDL_C_pct\nM_LDL_CE_pct\nM_LDL_FC_pct\nM_LDL_TG_pct\nS_LDL_PL_pct\nS_LDL_C_pct\nS_LDL_CE_pct\nS_LDL_FC_pct\nS_LDL_TG_pct\nXL_HDL_PL_pct\nXL_HDL_C_pct\nXL_HDL_CE_pct\nXL_HDL_FC_pct\nXL_HDL_TG_pct\nL_HDL_PL_pct\nL_HDL_C_pct\nL_HDL_CE_pct\nL_HDL_FC_pct\nL_HDL_TG_pct\nM_HDL_PL_pct\nM_HDL_C_pct\nM_HDL_CE_pct\nM_HDL_FC_pct\nM_HDL_TG_pct\nS_HDL_PL_pct\nS_HDL_C_pct\nS_HDL_CE_pct\nS_HDL_FC_pct\nS_HDL_TG_pct\n0 5 10 15 20 25\n−LOG10(FDR)\n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprintthis version posted June 22, 2022. ; https://doi.org/10.1101/2022.06.21.22276700doi: medRxiv preprint \n\n25\nFigure 2: Results of bi-directional Mendelian Randomization  \nThe first column from the left shows the direction of association of 124 significantly \nassociated metabolites with MDD (Z-scores). Blue is negative association and red is positive \nassociation. The second column shows the results of the Mendelian randomization (MR) \nanalysis when MDD is the exposure and microbiome is the outcome. The third column \ndepicts the results of MR when metabolites were used as exposure and MDD outcome. \nBlack stars represent significant ones after correcting for multiple testing using FDR.\nZ_MDD Z_MR_MDD_EXPOSURE Z_MR_METAB_EXPOSURE\nXS_VLDL_C_pct\nM_VLDL_C_pct\nS_VLDL_PL_pct\nS_VLDL_FC_pct\nPUFA_by_MUFA\nXS_VLDL_CE_pct\nM_VLDL_FC_pct\nS_VLDL_C_pct\nXL_HDL_CE\nXL_VLDL_C_pct\nXL_HDL_FC\nXL_VLDL_CE_pct\nXL_HDL_C\nXL_HDL_L\nXL_HDL_P\nL_VLDL_CE_pct\nM_VLDL_CE_pct\nXS_VLDL_FC_pct\nL_HDL_C_pct\nL_HDL_CE_pct\nIDL_FC\nIDL_C\nL_HDL_FC_pct\nPUFA_pct\nXL_VLDL_FC_pct\nOmega_6_pct\nIDL_FC_pct\nM_HDL_C_pct\nUnsaturation\nL_HDL_L\nL_HDL_PL\nL_HDL_P\nHDL_FC\nL_HDL_FC\nM_HDL_FC_pct\nL_HDL_CE\nHDL_CE\nL_HDL_C\nHDL_C\nHDL_size\nIDL_C_pct\nXL_HDL_PL\nS_HDL_C_pct\nS_HDL_FC_pct\nM_HDL_CE_pct\nIDL_CE\nIDL_CE_pct\nS_LDL_PL_pct\nIDL_L\nSphingomyelins\nTotal_CE\nHDL_L\nM_HDL_FC\nM_HDL_P\nHDL_PL\nApoA1\nHDL_P\nTotal_P\nM_HDL_CE\nM_HDL_C\nCitrate\nLA_pct\nAla\nIDL_TG\nVLDL_PL\nXL_VLDL_CE\nM_VLDL_L\nVLDL_FC\nPyruvate\nXL_HDL_FC_pct\nL_VLDL_CE\nS_VLDL_L\nS_VLDL_P\nXXL_VLDL_PL_pct\nXS_VLDL_TG\nL_VLDL_PL_pct\nM_LDL_TG\nMUFA\nL_LDL_TG_pct\nXXL_VLDL_TG\nM_LDL_TG_pct\nL_VLDL_C\nL_VLDL_FC\nM_VLDL_TG\nL_VLDL_PL\nXL_VLDL_FC\nXL_VLDL_PL\nS_LDL_TG\nM_HDL_TG\nHDL_TG\nXL_VLDL_C\nXS_VLDL_PL_pct\nVLDL_L\nM_HDL_PL_pct\nXL_HDL_TG_pct\nVLDL_size\nL_VLDL_P\nL_VLDL_L\nS_VLDL_TG\nL_VLDL_TG\nXL_VLDL_P\nTotal_TG\nL_HDL_TG_pct\nM_HDL_TG_pct\nS_HDL_TG\nVLDL_TG\nXL_VLDL_L\nXS_VLDL_TG_pct\nS_VLDL_TG_pct\nIDL_TG_pct\nM_VLDL_TG_pct\nMUFA_pct\nTG_by_PG\nXL_VLDL_TG\nS_HDL_TG_pct\nL_HDL_PL_pct\nXXL_VLDL_L\nXXL_VLDL_CE\nS_LDL_TG_pct\nXXL_VLDL_PL\nXXL_VLDL_FC\nXL_VLDL_TG_pct\nXXL_VLDL_C\nXXL_VLDL_P\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n*\n*\n*\n*\n*\n*\n* *\n* *\n*\n*\n*\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n*\n*\n*\n*\n*\n*\n*\n*\n*\n*\n*\n*\n* *\n*\n* *\n* *\n*\n*\n*\n*\n*\n*\n*\n*\n*\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n*\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n* *\n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprintthis version posted June 22, 2022. ; https://doi.org/10.1101/2022.06.21.22276700doi: medRxiv preprint \n\n26\nFigure 3: Association of metabolic profiles (z-scores) of ‘healthy’ gut microbiota and MDD \nA scatter plot showing the correlation between the metabolic profiles of microbial taxa that \nare negatively associated the metabolic profile of MDD. Each dot represents a metabolite.  \nX-axis shows the association of the metabolite to microbial taxa and Y-axis shows the \nassociation of the metabolite to MDD. Different colours of the dots represent the class of \nthe metabolite and the dots highlighted with black circles are the significantly associated \nmetabolites with MDD in model 4. \n−6 −4 −2 0 2 4\n−4 −2 0 2 4\nChristensenellaceaeR7group\nMDD\nHDLs\nVLDLs\nIDLs\nLDLs\nFattyAcids\nSmallMolecules\nSignificant MDD\n−6 −4 −2 0 2 4\n−4 −2 0 2 4\nChristensenellaceae\nMDD\nHDLs\nVLDLs\nIDLs\nLDLs\nFattyAcids\nSmallMolecules\nSignificant MDD\n−4 −2 0 2\n−4 −2 0 2 4\nRuminiclostridium6\nMDD\nHDLs\nVLDLs\nIDLs\nLDLs\nFattyAcids\nSmallMolecules\nSignificant MDD\n−4 −2 0 2\n−4 −2 0 2 4\nLachnospiraceaeNC2004group\nMDD\nHDLs\nVLDLs\nIDLs\nLDLs\nFattyAcids\nSmallMolecules\nSignificant MDD\n−4 −2 0 2 4\n−4 −2 0 2 4\nEubacteriumxylanophilumgroup\nMDD\nHDLs\nVLDLs\nIDLs\nLDLs\nFattyAcids\nSmallMolecules\nSignificant MDD\n−3 −2 −1 0 1 2 3\n−4 −2 0 2 4\nCoprococcus1\nMDD\nHDLs\nVLDLs\nIDLs\nLDLs\nFattyAcids\nSmallMolecules\nSignificant MDD\n−4 −2 0 2 4\n−4 −2 0 2 4\nRuminococcaceaeUCG014\nMDD\nHDLs\nVLDLs\nIDLs\nLDLs\nFattyAcids\nSmallMolecules\nSignificant MDD\n−3 −2 −1 0 1 2\n−4 −2 0 2 4\nLachnospiraceaeAC2044group\nMDD\nHDLs\nVLDLs\nIDLs\nLDLs\nFattyAcids\nSmallMolecules\nSignificant MDD\n−6 −4 −2 0 2 4\n−4 −2 0 2 4\nRuminococcaceaeUCG005\nMDD\nHDLs\nVLDLs\nIDLs\nLDLs\nFattyAcids\nSmallMolecules\nSignificant MDD\n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprintthis version posted June 22, 2022. ; https://doi.org/10.1101/2022.06.21.22276700doi: medRxiv preprint \n\n27\nFigure 4: Association of metabolic profiles (z-scores) of unhealthy microbiota and MDD \nA scatter plot showing the correlation between the metabolic profiles of microbial taxa that \nare positively associated the metabolic profile of MDD. Each dot represents a metabolite.  X-\naxis shows the association of the metabolite to microbial taxa and Y-axis shows the \nassociation of the metabolite to MDD. Different colours of the dots represent the class of \nthe metabolite and the dots highlighted with black circles are the significantly associated \nmetabolites with MDD in model 4. \n−2 −1 0 1 2 3 4\n−4 −2 0 2 4\nLachnoclostridium\nMDD\nHDLs\nVLDLs\nIDLs\nLDLs\nFattyAcids\nSmallMolecules\nSignificant MDD\n−4 −2 0 2 4 6\n−4 −2 0 2 4\nRuminococcusgnavusgroup\nMDD\nHDLs\nVLDLs\nIDLs\nLDLs\nFattyAcids\nSmallMolecules\nSignificant MDD\n−2 0 2 4\n−4 −2 0 2 4\nFlavonifractor\nMDD\nHDLs\nVLDLs\nIDLs\nLDLs\nFattyAcids\nSmallMolecules\nSignificant MDD\n−2 −1 0 1 2\n−4 −2 0 2 4\nChristensenella\nMDD\nHDLs\nVLDLs\nIDLs\nLDLs\nFattyAcids\nSmallMolecules\nSignificant MDD\n−4 −3 −2 −1 0 1 2 3\n−4 −2 0 2 4\nFusobacteriaceae\nMDD\nHDLs\nVLDLs\nIDLs\nLDLs\nFattyAcids\nSmallMolecules\nSignificant MDD\n−4 −2 0 2 4\n−4 −2 0 2 4\nEggerthella\nMDD\nHDLs\nVLDLs\nIDLs\nLDLs\nFattyAcids\nSmallMolecules\nSignificant MDD\n−6 −4 −2 0 2\n−4 −2 0 2 4\nMegamonas\nMDD\nHDLs\nVLDLs\nIDLs\nLDLs\nFattyAcids\nSmallMolecules\nSignificant MDD\n−2 −1 0 1 2 3\n−4 −2 0 2 4\nFaecalitalea\nMDD\nHDLs\nVLDLs\nIDLs\nLDLs\nFattyAcids\nSmallMolecules\nSignificant MDD\n−2 0 2 4\n−4 −2 0 2 4\nRuminococcustorquesgroup\nMDD\nHDLs\nVLDLs\nIDLs\nLDLs\nFattyAcids\nSmallMolecules\nSignificant MDD\n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprintthis version posted June 22, 2022. ; https://doi.org/10.1101/2022.06.21.22276700doi: medRxiv preprint \n\n28\nFigure 5: Scatter plot of direct and proxy association of microbiome with major depression. \nTaxa that have a p-value < 0.05 in both proxy and direct association are annotated. \nEach dot represents a microbial taxon. X-axis depicts the proxy association (T-scores) \nbetween microbial taxa and MDD inferred through their metabolic signatures and Y-axis \ndepicts the direct association of microbial taxa with MDD (Z-scores) from the Rotterdam \nstudy performed in Radjabzadeh et al20. Taxa that are nominally significant in both proxy \nand direct associations are annotated. \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprintthis version posted June 22, 2022. ; https://doi.org/10.1101/2022.06.21.22276700doi: medRxiv preprint \n\n29\nFigure 6:  Hierarchical illustration of all healthy and pathogenic bacteria that showed \nsignificant correlation (r > 0.3 & FDR < 0.05) with MDD metabolic profile \nRed dots represent significant negative proxy association between microbial taxa and MDD \nand green dots represent significant positive proxy association between microbial taxa and \nMDD. The outermost layer depicts the phylum followed by, class, order, family and genus. \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprintthis version posted June 22, 2022. ; https://doi.org/10.1101/2022.06.21.22276700doi: medRxiv preprint","source_license":"CC-BY-4.0","license_restricted":false}