Multiomics Analyses Reveal Microbiome–Gut–Brain Crosstalk Centered on Aberrant Gamma-Aminobutyric Acid and Tryptophan Metabolism in Drug-Naïve Patients with First-Episode Schizophrenia

Abstract

Background and Hypothesis

Schizophrenia (SCZ) is associated with complex crosstalk between the gut microbiota and host metabolism, but the underlying mechanism remains elusive. Investigating the aberrant neurotransmitter processes reflected by alterations identified using multiomics analysis is valuable to fully explain the pathogenesis of SCZ.

Study Design

We conducted an integrative analysis of multiomics data, including the serum metabolome, fecal metagenome, single nucleotide polymorphism data, and neuroimaging data obtained from a cohort of 127 drug-naïve, first-episode SCZ patients and 92 healthy controls to characterize the microbiome–gut–brain axis in SCZ patients. We used pathway-based polygenic risk score (PRS) analyses to determine the biological pathways contributing to genetic risk and mediation effect analyses to determine the important neuroimaging features. Additionally, a random forest model was generated for effective SCZ diagnosis.

Study Results

We found that the altered metabolome and dysregulated microbiome were associated with neuroactive metabolites, including gamma-aminobutyric acid (GABA), tryptophan, and short-chain fatty acids. Further structural and functional magnetic resonance imaging analyses highlighted that gray matter volume and functional connectivity disturbances mediate the relationships between Ruminococcus_torgues and Collinsella_aerofaciens and symptom severity and the relationships between species Lactobacillus_ruminis and differential metabolites l-2,4-diaminobutyric acid and N-acetylserotonin and cognitive function. Moreover, analyses of the Polygenic Risk Score (PRS) support that alterations in GABA and tryptophan neurotransmitter pathways are associated with SCZ risk, and GABA might be a more dominant contributor.

Conclusions

This study provides new insights into systematic relationships among genes, metabolism, and the gut microbiota that affect brain functional connectivity, thereby affecting SCZ pathogenesis.

Introduction

Schizophrenia (SCZ) is a chronic, severe mental disorder. Many hypotheses have been proposed regarding the pathogenesis of SCZ, such as the neurodevelopmental hypothesis¹; the neurotransmitter hypothesis involving dopamine (DA), glutamate (Glu), gamma-aminobutyric acid (GABA), 5-hydroxytryptamine (5-HT), and other neurotransmitters²; and the immune hypothesis,³ but these hypotheses do not fully explain the pathogenesis of SCZ.

The microbiome–gut–brain axis has emerged as a hotspot in human neuroscience research. The gut microbiota synthesizes neuroactive metabolites, such as tryptophan (TRP), short-chain fatty acids (SCFAs), and indole and its derivatives, and regulates the secretion of neurotransmitters (such as DA, 5-HT, Glu, and GABA) to affect neurodevelopment and brain health.⁴ Several studies have reported alterations in the microbial composition and metabolomics in individuals with SCZ.^5–7 Based on a literature search (supplementary method 1), 16 bacterial genera have been found to be altered in individuals with SCZ by at least 2 studies (supplementary table 1), suggesting microbial dysbiosis in SCZ. Bloodstream metabolites facilitate crosstalk between the brain and the gut microbiota. Several studies have shown that alterations in plasma metabolome from SCZ patients are related to metabolic pathways of amino acids (TRP, arginine, sarcosine, etc.) and neurotransmitters (Glu, GABA, etc.).^7,8 However, the metabolites that are associated with gut microbiota and affect the brain network have not been completely elucidated. Additionally, most of the studies, which have small sample sizes, have mainly focused on 1 or 2 dimensions and account for only a small fraction of the disease. Thus, a more valuable approach may be to explore the pathological mechanisms of SCZ from a systematic perspective using multiomics analysis.

Here, we conducted a multiomics study including the serum metabolome, fecal metagenome, brain functional network, and genotype information to explore the crosstalk between the gut microbiota and host metabolism and identify multidimensional biological markers of SCZ.

Methods

Participant Recruitment and Clinical Measurements

The present study was approved by the Human Ethics Committee of the First Affiliated Hospital of Zhengzhou University, China (Approval No. 2016-LW-17). All individuals were recruited between October 2017 and January 2020. The inclusion criteria included (1) diagnosis of first-episode SCZ based on the Diagnostic and Statistical Manual of Mental Disorders fourth version (DSM-IV) criteria and confirmed using the Structured Clinical Interview for DSM-IV (SCID),⁹ (2) no use of prescription drugs, and (3) a total Positive and Negative Syndrome Scale (PANSS) score ≥ 60. All participants were evaluated by a psychiatrist. The exclusion criteria included (1) a diagnosis of neurological diseases, diabetes, heart diseases, digestive system disease, autoimmune diseases, blood diseases, or psychiatric diseases other than SCZ; (2) pregnancy or lactation; (3) treatment with any antibiotic or anti-inflammatory agent in the past month; and (4) obesity (body mass index [BMI] > 28 kg/m²).

Healthy controls (HCs) and SCZ patients were recruited from the same local geographic areas in Henan Province, China, during the same study time period. Further, enrolled SCZ patients and HCs were born and grew up in the same local geographic areas. All HCs underwent a comprehensive clinical assessment to rule out possible medical conditions and history of any psychiatric conditions. HCs must fit the same inclusion and exclusion criteria as SCZ patients except the SCZ diagnosis. All participants provided written informed consent.

Samples Collection

Venous blood (5 mL) was collected between 06:30 and 07:00 AM to determine serum levels of fasting blood high sensitivity C-reactive protein (hs-CRP), superoxide dismutase (SOD,) and metabolome profiles. Fecal samples were obtained in the morning after breakfast (8:00 to 10:00), aseptically placed in 5 mL sterile boxes (LF005, BIORISE, Shanghai, China), and stored at −80°C, until use.

Serum Metabolome Profile

Untargeted liquid chromatography—mass spectrometry (LC–MS) was used to measure serum metabolite levels. Detailed information on LC–MS are shown in supplementary method 2.

Fecal Metagenome Sequencing and Analysis

Detailed information on library construction and sequencing were provided in supplementary method 3. MetaPhlAn (v3.0.13) was used to quantitatively profile the taxonomic composition (species level) of the microbial communities in all samples subjected to metagenomic analysis and used the linear discriminant analysis (LDA) effect size (LEfSe) to detect differentially abundant taxa with cutoffs (LDA score > 3.0, P < .05).¹⁰ HUMAnN (HMP Unified Metabolic Analysis Network v3.0.0)¹¹ was used to efficiently profile pathway and gene family abundances.

Metabolic Network-Based Analysis for Inferring Mechanism-Supported Relationships in Microbiome–Metabolome Datasets (MIMOSA)

MIMOSA2 was used to construct a community metabolic model by mapping Kyoto Encyclopedia of Genes and Genomes (KEGG) ortholog abundances to Assembly of Gut Organisms through Reconstruction and Analysis (AGORA) genome-scale models and then calculating community metabolic potential scores for each taxon, sample, and metabolite¹² (supplementary method 4).

Neuroimaging Feature Extraction and Analyses

The MRI data acquisition parameters and the preprocessing steps are presented in supplementary method 5. An independent t-test (2-sided) was performed to examine GMV differences between the 2 groups with TIV, age, and gender as covariates.

Independent component analysis (ICA) was conducted with the GIFT toolbox (mialab.mrn.org/software/gift/). The number of Independent components (ICs) (N = 22) was estimated automatically by the software using the minimum description length criteria. Spatial ICA is used to decompose the participant data into linear mixtures of spatially ICs that exhibit a unique temporal profile. After 2 data reduction steps using PCA, the infomax algorithm (repeated 20 times in ICASSO), and the group-ICA back reconstruction approach, participant-specific spatial maps and time courses were obtained. Functional networks were identified as several ICs with peak activations in gray matter that showed low spatial overlap with known vascular, ventricular, motion, and susceptibility artifacts and exhibited primarily low-frequency power. Then, we calculated internetwork functional connectivity (FC) as the Pearson correlation coefficients between pairs of time courses of the functional networks and intranetwork FC via the spatial maps indexing the contribution of the time course to each voxel comprising a particular component.

Finally, to further test whether the relationships between gut microbiota, metabolites and clinical symptoms were mediated by neuroimaging features, mediation analysis was performed using the PROCESS macro (http://www.processmacro.org/). Detailed information on mediation analysis is presented in supplementary method 6.

Polygenic Risk Scores Based on Genotype Information

We calculated whole-genome SCZ polygenic risk scores (PRSs) and pathway-specific PRSs for each participant using the “standard weighted” method implemented in PRSice2 and PRSet.^13,14 Base data were built according to the variant effect sizes retrieved from the Psychiatric Genomics Consortium (PGC3), the largest Genome-Wide Association Studies (GWAS) of SCZ conducted to date, including 76 755 cases and 243 649 controls.¹⁵ Whole-genome PRSs were estimated, excluding those near the Major Histocompatibility Complex (MHC) region due to their genomic complexity and high polymorphic diversity (supplementary method 7).

Random Forest Model for Diagnosing SCZ

The cohort samples were randomly divided into a training set (70%) and a test set (30%). Tenfold cross-validation was performed ten times on a random forest model using the levels of differentially abundant metabolites, the levels of microbial species present at an altered abundance, and PRSs within the training set. We selected the optimal set of variables with the lowest cross-validation error. The predictive model was constructed using the most important variables and further applied in the receiver operating characteristic curve (ROC) analysis with the training set and test set (pROC package).

Statistical Analysis

The Shapiro–Wilk test was used to examine normality for continuous variables and the Levene’s test to evaluate variance homogeneity. The Chi-square test was used to examine independence among categorical variables. Numerical data are presented as mean and standard deviation, or median, lower, and upper quartiles. False discovery rate (FDR) was used to correct multiple testing. R software (Version 3.6.3) were used for statistical analysis.

Results

Demographic and Clinical Characteristics of Participants

We did not observe significant differences in age, sex, education level, smoking status, or BMI between the SCZ and HC groups (P > .05) (supplementary table 2). However, we observed significant impairments in several cognitive domains in SCZ compared with HCs (P < .05) (supplementary figure 1). The flowchart of the multiomics study is shown in figure 1.

Fig. 1.

Flowchart of the design of the multiomics study of patients with SCZ.

Alterations in Serum Metabolomic Profiles in SCZ, With a Focus on TRP and GABA Metabolism

We identified 15 513 metabolic features in total and 1549 features that were significantly altered in SCZ patients. Using OPLS-DA, these features could significantly discriminate the patients from the HCs (figures 2A and 2B). Of the 398 obtained metabolites, we identified 30 significantly differentially abundant metabolites between SCZ and HCs (supplementary table 3) that were mainly related to amino acid metabolism and fatty acid metabolism (figure 2C). In particular, the intensity distribution of neuroactive metabolites related to GABA metabolism, TRP metabolism, and short-chain fatty acids (SCFA) metabolism in SCZ and HCs exhibited significant differences (figure 2D). These metabolites were directly or indirectly associated with microbial metabolism. For example, Nγ-acetyldiaminobutyrate is a derivative of GABA; 2-ketobutyric acid and S-lactoylglutathione are involved in GABA and Glu metabolism; kynurenic acid (KYNA), quinolinic acid (QA), serotonin, N-acetylserotonin, and indole are involved in TRP metabolism; and isobutyric acid, l-2,4-diaminobutyrate, and ergothioneine are also related to microbial metabolism. Interestingly, neuroactive metabolites, ie, l-2,4-diaminobutyric acid and isobutyric acid, were associated with symptom severity. In summary, the microbial metabolites associated with SCZ might contribute to neuroactive function through GABA and TRP metabolism.

Fig. 2.

Differential serum levels of metabolites between the SCZ patient and HC groups. (A) Metabolic features discriminate patients with SCZ from HCs using the OPLS-DA model (VIP > 1, adj P value < .05). (B) The validation plot confirmed the validity of the OPLS-DA model. (C) Sample cluster heatmap of differentially abundant metabolites between patients with SCZ and HCs. (D) Intensity differences in key neuroactive metabolites between patients with SCZ and HCs.

Alterations in neurotransmitters (Glu and GABA) have been confirmed to be involved in the pathophysiology of SCZ.¹⁶ Evidence has shown that GABA and spermine/spermidine share the same precursor, putrescine.¹⁷ Our study showed that the levels of spermine/spermidine were significantly higher in SCZ than in HCs. This finding was supported by evidence that increased regional brain levels of spermine, spermidine, and putrescine might be associated with SCZ.¹⁷ Our study also identified an alteration in the serum sarcosine level in SCZ. Sarcosine, a Glu modulator, is a potential therapeutic drug for SCZ because it enhances the function of the glutamatergic N-methyl-d-aspartate receptor (NMDAR).^18,19 These results suggested that alterations in metabolites related to GABA/Glu were associated with SCZ.

Several studies have suggested that perturbed TRP metabolism might be involved in SCZ.^20,21 The present study also showed an alteration in the levels of TRP-related bioactive compounds. Increased KYNA and QA levels and lower levels of N-acetylserotonin in SCZ patients suggest the activation of the TRP-kynurenine (KYN) pathway. KYNA tightlxy controls glutamatergic and dopaminergic neurotransmission, and QA is capable of binding to NMDARs and inducing cellular toxicity, both of which are associated with the pathogenesis of SCZ.^22,23 Decreased levels of indole also support the hypothesis that perturbed TRP metabolism is associated with SCZ. TRP metabolism is tightly connected to GABAergic neurotransmission since QA is a Glu receptor agonist that indirectly regulates GABAergic neurotransmission²⁴; additionally, the neuroactive substance 5-HT involved in the TRP-serotonin pathway and its receptors 5-HT_2a and 5-HT_2c are widely expressed in GABAergic neurons and directly regulate GABAergic neurotransmission.²⁵

Alterations in the Microbial Community Composition in SCZ

Because various serum metabolites are derived from the gut microbiota, we conducted metagenome sequencing to investigate microbial alterations in SCZ patients. Detailed information related to the fecal metagenome assembly, microbial composition and functional annotation is summarized in supplementary table 4. Microbial diversity was significantly decreased in SCZ, characterized by decreased Firmicutes and Actinobacteria levels and increased Bacteroidetes levels (supplementary figure 2).

We identified 34 differential microbial species (figure 3A and supplementary table 5). Seven species were present at significantly elevated levels in SCZ and were mainly associated with a proinflammatory state (Bacteroides_ovatus, Bacteroides_stercoris, Veillonella_parvula, Lactobacillus_ruminis, etc.). The increased Lactobacillus level was negatively correlated with the levels of SOD (supplementary table 6), which may indirectly reflect the proinflammatory effect of Lactobacillus on the clinical phenotype. Twenty-seven species had significantly decreased abundance in SCZ patients; the levels of most of these species were negatively correlated with symptom severity measured by the total PANSS scores (figure 3B). They were associated with SCFA production and an anti-inflammatory status (Dorea_longicatena, Anaerostipes_hadrus, Ruminococcus_torques, Agathobaculum_butyriciproducens, etc.) and neuroactive substance production, eg, GABA (Blautia_obeum, Bifidobacterium_longum, Klebsiella_aerogenes) and 5-HT (Agathobaculum_butyriciproducens). The levels of those species were correlated with cognition (supplementary figure 3). Furthermore, we found that the decreased Bifidobacterium level negatively correlated with the hs-CRP, the decreased Butyricicoccus level negatively correlated with SOD, which indicated the anti-inflammatory effect of these bacteria.

Figure 3.

The metagenome analysis revealed a significantly altered microbiota and functional pathways. (A) Microbes present at significantly increased and decreased levels in the patients with SCZ were identified using LEfSe analysis. (B) Associations between altered species and PANSS scores. (C) HumanN2 revealed enriched KOs of metagenome-related differentially abundant genes between the SCZ patient and HC groups. (D) The gut–brain modules present in SCZ-associated microbial species. The dot indicates a statistically significant association between gut–brain modules present in SCZ-associated bacterial species and a metabolite.

Functional analysis by HUMAnN3 revealed 25 increased KO modules related to GABA shunt, aminobutanoate degradation, and biotin biosynthesis II, while 45 decreased KO modules related to butanoate fermentation, glycogen biosynthesis I, l-glutamine biosynthesis III, and polyamine biosynthesis II (FDR-adjusted P < .05) (figure 3C and supplementary table 7). In summary, microbe-derived functions concomitantly changed with alterations in serum metabolite levels.

We also determined whether the abundance of 56 previously reported gut–brain modules (GBMs) present in each microbial species varied significantly between SCZ and HCs. We found SCZ-associated GBMs included GABA synthesis, Glu synthesis, TRP metabolism, and acetate synthesis (figure 3D), which are consistent with alterations in serum GABA and TRP-related metabolite levels in SCZ. Klebsiella spp. was the most abundant microbe in most GBMs. The decreased TRP levels may be associated with a lower abundance of Klebsiella in SCZ patients, since it has been reported that TRP promotes the efficient growth of Klebsiella.^26,27

Crosstalk Between the Gut Microbiota and Host Metabolites

We identified significant associations between differential metabolites and intestinal microbial species (BH-adjusted P < .05) (supplementary table 8). In the HC network, Bacteroides_ovatus and Bacteroides_stercoris showed significant correlations with most metabolite partners, all of which are amino acid metabolites, particularly those related to TRP and GABA metabolism. Bifidobacterium_longum was also a hub microbe node, suggesting that Bifidobacterium_longum might be a driving species in SCZ since gadB/gadC genes involved in GABA production are widely distributed in Bifidobacterium spp. The increased abundance of Veillonella parvula was negatively correlated with decreased levels of the GABA transporter inhibitor l-2,4-diaminobutyric acid, while Streptococcus salivarius abundance was positively correlated with the levels of N-acetylserotonin, which is involved in TRP metabolism. Most significant correlations in the HC group were not observed in SCZ, suggesting a disconnection between the intestinal microbiota and metabolites in SCZ patients (supplementary figure 4). In the SCZ patient network, elevated KYNA levels were negatively correlated with a decreased abundance of Klebsiella_quasipneumoniae and positively correlated with an increased abundance of Lactobacillus_ruminis; the increased abundance of Bacteroides_stercoris was negatively correlated with decreased 2-oxoarginine levels. The correlations between the levels of microbial species and GABA- or TRP-related metabolites were disturbed in SCZ patients, suggesting that the effect of the gut microbiota in SCZ might be mediated by the metabolism of the neuroactive substances GABA and TRP.

We further explored the relationships between microbes and metabolites based on the metabolic network approach MIMOSA2. We identified 15 microbiome-governed metabolites, including l-glutamine, spermidine, acetoacetate, and spermine (supplementary table 9), among which spermidine was an important differentially abundant metabolite between SCZ and HCs. Increased spermidine/spermidine levels and Glu/GABA ratios in the frontal cortex, hippocampus, and cerebellum have been reported to be involved in the pathogenesis of SCZ.²⁸ The abundance of the SCFA (ie, butyrate)-producing strains Roseburia_faecis and Ruminococcus_torques were decreased in SCZ, which mainly contributed to spermidine synthesis (supplementary figure 5). Spermidine competes with GABA because they share the same precursor, putrescine. Thus, the elevated spermidine level modulated by microbial markers may also be associated with a decrease in GABA levels, which is involved in SCZ pathophysiology.

Brain Functional Networks Assessed Using fMRI Mediate the Relationships Between Metabolites/Microbes and SCZ

We observed a significantly lower GMV in SCZ patients than in HCs (supplementary figure 6A, supplementary table 10). Decreased volumes of the right parahippocampal gyrus (PHG.R), right middle frontal gyrus (MFG.R) and left middle frontal gyrus (MFG.L) were negatively correlated with PANSS scores. Decreased volumes of the right frontal superior orbital area (ORBmid.R) were negatively correlated with positive and total PANSS scores. Those regions with significant reductions in GMV were positively correlated with several cognitive domains. In addition, ICA results showed that dysregulation of functional connectivity within the brain network occurred mainly in the visual cortical network and the associative cortical network (supplementary figures 6B and 7). In SCZ patients, intranetwork FC in sensory–motor systems, such as the medial visual network (mVN), posterior visual network (pVN), and sensorimotor network (SEN), was significantly decreased. The decrease in anterior default mode network (aDMN)–mVN connectivity was negatively correlated with PANSS scores; aDMN–salience network connectivity was positively correlated with PANSS scores. The correlations between brain imaging features and the microbiota, metabolites, and clinical evaluations are shown in supplementary figure 8.

Mediation analyses showed that mVN–aDMN connectivity mediated the relationship of Ruminococcus_torgues with both negative and total PANSS scores. The GMV of the left MFG mediated the relationship between total PANSS scores and Collinsella_aerofaciens, which mainly produces SCFAs and was present at decreased levels in SCZ (figure 4A). This result verified the previously mentioned effects of Ruminococcus_torgues and Collinsella_aerofaciens on PANSS scores at the neuroimaging level. Lateral visual network (lVN)–executive control network connectivity mediated the relationship between the species Lactobacillus_ruminis and the metabolite l-2,4-diaminobutyric acid and reasoning and problem solving (RPS) (figure 4B). Strikingly, the increase in Lactobacillus_ruminis abundance and decrease in l-2,4-diaminobutyric acid levels might contribute to the alterations in the GABA metabolic process in SCZ. AUN–aDMN connectivity mediated the relationship between N-acetylserotonin and RPS. l-2,4-diaminobutyric acid, N-acetylserotonin, and SCFA are important differentially abundant metabolites between SCZ and HCs, and they reflect cognitive impairments based on brain function network observations.

Figure 4.

Significant associations between metabolites or microbial species and PANSS (A) or cognitive testing (B) scores mediated by GMV and functional connectivity.

Pathway-Based Polygenic Risk Score (PRS) of Aberrant Pathways From the Metabolome and Metagenome

The genome-wide PRS results revealed that the global positive risk loci might account for the differences observed in SCZ in our study (supplementary figure 9A). We further calculated pathway-based PRSs using the significantly altered metabolic pathways and important neurotransmitter pathways.¹³ We identified 19 pathways that were significantly associated with SCZ, among which 9 pathways were related to GABA (top 6 rank). Other contributing pathways were related to biotin metabolism and amino acid metabolism, particularly TRP metabolism (supplementary figure 9B). Thus, pathway-PRS analyses were leveraged to support the hypothesis that GABA and TRP neurotransmitter pathways were significantly associated with the risk of SCZ, and GABA might be a more dominant contributor.

The present study identified 84 SNP loci for SCZ (supplementary table 11). The corresponding variant genes were mainly enriched in GABA receptor activity or GABA receptor complex. The top 2 genes with variants, GABRA4 and GABRR1, have been reported to downregulate GABA receptor subunits in the superior temporal gyrus of SCZ.^29,30

We combined multiomics features, including 30 differential metabolites, 34 microbial species, and PRSs, to build a diagnostic model for SCZ. The area under receiver operating characteristic curve (AUC) in the test set was 0.984 (supplementary figure 9C), outperforming that of the classifier using PRS alone, which further confirmed that the pathology of SCZ is affected by both genetics and environmental factors. The top 10 significant features identified by the RF diagnostic model are shown in supplementary table 12.

Discussion

This study identified significant alterations in microbe-derived metabolites and dysregulated microbiome associated with GABA/TRP metabolism, and regional brain GMV and functional connectivity disturbances mediated the relationship between GABA/TRP-related species or metabolites and symptoms or cognitive impairments. An effective diagnostic model revealed microbial and metabolite biomarkers for SCZ.

We proposed a mechanism underlying SCZ (figure 5). The differentially abundant host-derived or microbial-derived metabolites associated with SCZ were mainly involved in amino acid (GABA, TRP, arginine, sarcosine, etc.) and fatty acid (stearidonic acid, SCFA, and stearic acid) metabolism, which have also been reported to be altered in large-scale meta-analysis studies.^31,32 Excitatory neurotoxic effects induced by elevated levels of the excitatory neurotransmitter Glu, a decrease in the level of the inhibitory neurotransmitter GABA, or an imbalance of Glu/GABA levels have been reported to be associated with the pathogenesis of SCZ.^33,34 We found that l-2,4-diaminobutyric acid (an inhibitor of GABA uptake) and N(g)-acetyldiaminobutyrate (a derivative of GABA) levels were decreased in SCZ, indirectly suggesting lower levels of GABA. Interestingly, GABA might be a dominant contributor, since several metabolites, such as Nγ-acetyldiaminobutyrate, 2-ketobutyric acid, S-lactoylglutathione, sarcosine, TRP-related metabolites and SCFAs, were associated with GABA metabolism. This hypothesis is supported by the abnormal behavior/cognitive impairment observed in mice receiving a fecal transplant from SCZ; the abnormal metabolism of neurotransmitters, especially Glu/GABA^5,35; and dysregulated TRP/KYN metabolism,³⁶ all of which suggest that alterations in microbe-derived metabolites, such as GABA/TRP metabolites, play critical roles in SCZ. Several of these SCZ-related metabolites, ie, SCFA and TRP metabolites, affect neurotransmitters, ie, GABA, Glu, and 5-HT, and those metabolites are involved in different metabolic pathways.

Figure 5.

Schematic of the proposed mechanisms underlying SCZ. Species present at elevated levels in patients with SCZ have proinflammatory roles, which activate the key enzyme IDO1 in the TRP/KYN pathway, leading to an increase in the level of the neurotoxin QA and an imbalance of QA/KYNA. QA is an agonist of the neurotransmitter Glu capable of binding to NMDARs, increasing glutamatergic signaling, and regulating the neurotransmitter GABA. 5-HT, an important metabolite of TRP, regulates many SCZ-related neurotransmitters, ie, GABA and Glu, since its receptors, 5-HT2aR and 5-HT2cR, are widely expressed in GABAergic and glutamatergic neurons. Alterations in GABA, Glu, and GABA/Glu may cause decreased GMV and disrupt brain functional connectivity, which might be associated with positive and negative symptoms and cognitive impairments. SCFA: short-chain fatty acid, TRP: tryptophan, IDO1: indole-2,3-dioxygenase, KYN: kynurenine, KYNA: kynurenic acid, QA: quinolinic acid, serotonin: 5-HT, TPH2: tryptophan hydroxylase 2,5-HTP: 5-hydroxytryptophan, NMDAR: N-methyl-d-aspartate receptors, GABA: gamma-aminobutyric acid, Glu: glutamate, BBB: blood–brain barrier.

Metagenome analyses revealed elevated levels of proinflammatory species, such as Bacteroides_ovatus, Bacteroides_stercoris, Veillonella_parvula, and Lactobacillus_ruminis,³⁷ and decreased SCFA production and levels of anti-inflammatory species, such as Bifidobacterium_longum, Roseburia_faecis, and Ruminococcus_torques, in SCZ.^38–41 The alterations in the abundance of these species may activate TRP/KYN pathway and inhibit TRP/5-HT pathway; thus, levels of the neurotoxin QA increased, and levels of the neuroprotectant N-acetylserotonin decreased in SCZ. Our results are consistent with previous studies showing dysbiosis of TRP metabolism.⁴² The GABAergic neuron receptor α7 nicotinic acetylcholine receptor (nAChR) is antagonized by increased KYNA levels, which impacts cognitive function by altering the levels of neurotransmitters such as glutamate and DA.³⁴ We further examined the crosstalk between the gut microbiota and serum metabolites and found that the elevated spermidine level was modulated by Roseburia_faecis and Ruminococcus_torques, which may be the dominant factor inducing the decrease in GABA levels to affect SCZ pathophysiology since spermidine competes with GABA on the same precursor, putrescine.¹⁷ Moreover, both Roseburia_faecis and Ruminococcus_torques are SCFA-producing species with an anti-inflammatory role. Their decrease may also activate the TRP/KYN pathway, thus exerting an effect on GABA metabolism and neuronal excitability and toxicity.^38,39

Further analysis of brain fMRI data complemented our investigation of the effects of the microbiome–gut–brain axis on SCZ symptoms. We found that a decreased MFG volume was associated with more severe symptoms and that a lower paracingulate gyri (DCG) volume was associated with worse cognitive function. The mediation analyses may be the first to reveal that GMV and functional connectivity disturbances mediate the relationships between Ruminococcus_torgues and Collinsella_aerofaciens and symptom severity, and the relationships between the levels of Lactobacillus_ruminis and metabolites l-2,4-diaminobutyric acid and N-acetylserotonin and cognition. According to a previous study, aberrant face processing and passive effects observed in SCZ were associated with altered GABAergic function in the visual cortex,⁴³ which is consistent with our findings in the visual network. Our study may provide the first empirical evidence that the gut microbiota and serum metabolites modulate gray matter and large-scale inter and intranetwork functional connectivity (especially the sensorimotor system) in SCZ.

Another novel area of our study is that we conducted pathway-specific PRS analyses using the genetic data in combination with the latest large-scale GWAS dataset from PGC3¹⁵ to evaluate the genetic effects of pathways on SCZ. We showed that the altered GABA/TRP pathways identified in the metabolome–microbiome results were also significantly associated with SCZ risk, and GABA might be a more dominant contributor since the top-ranked pathways that explained a high proportion of SCZ heritability were GABA-related pathways, and the top 2 genes with variations, GABRA4 and GABRR1, contributing to the decreased GABA receptor activity detected in the current patient group have also been reported as global risk loci in SCZ.²⁹ The latest GWAS performed by PGC revealed that multiple prioritized gene variants encode metabotropic receptors (Glu and GABA) and the ligand-gated NMDAR subunit.¹⁵ Therefore, multiple lines of evidence, including genetic variation, suggest that alterations in GABA/TRP metabolism and related microbial species may be the driving force underlying SCZ pathophysiology.

The present study has several strengths and limitations. In addition to the large sample size, we enrolled only first-episode and drug-naïve SCZ patients to identify multidimensional biological markers of SCZ. Our study not only generated a high-quality multiomics data resource that may benefit basic research but also provided new biological insights underlying the clinical features of SCZ pathophysiology. One limitation was that we did not match diet between the 2 groups. However, all SCZ and HCs included in the study were born and raised in the same geographic areas and had similar dietary habits. Additionally, the Gene X Environmental interaction must be deeply explored by considering trauma and stigma as pathological aspects of SCZ.^44,45 Future research with longitudinal follow-up is warranted to advance our understanding of the role of the crosstalk between gut microbiota and host metabolism and explore novel intervention approaches.

Data Availability

Data have been deposited into the CNGB Sequence Archive (CNSA; https://db.cngb.org/cnsa/) of the China National GeneBank DataBase (CNGBdb) with accession number CNP0003461. The datasets generated by this study are available from the corresponding authors upon request.