See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/360078408

Multi-omics analyses of serum metabolome, gut microbiome and brain

function reveal dysregulated microbiota-gut-brain axis in bipolar depression

Article in Molecular Psychiatry · April 2022

DOI: 10.1038/s41380-022-01569-9

CITATIONS READS

71 2,073

35 authors, including:

Li Zhiming Jianbo Lai

China National Genebank-Shenzhen First Affiliated Hospital, Zhejiang University School of Medicine
19 PUBLICATIONS 593 CITATIONS 98 PUBLICATIONS 11,241 CITATIONS

SEE PROFILE SEE PROFILE

Chuanfa Liu Yuzhe Sun

The Chinese University of Hong Kong Beijing Genomics Institute
13 PUBLICATIONS 215 CITATIONS 32 PUBLICATIONS 690 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Li Zhiming on 01 June 2022.

The user has requested enhancement of the downloaded file.

 Molecular Psychiatry www.nature.com/mp

ARTICLE
Multi-omics analyses of serum metabolome, gut microbiome
and brain function reveal dysregulated microbiota-gut-brain
axis in bipolar depression
Zhiming Li1,6,7,8,17, Jianbo Lai1,2,3,4,17, Peifen Zhang1,17, Jiahong Ding6,7,8,17, Jiajun Jiang1, Chuanfa Liu6,7,9, Huimin Huang1,5,
Hefu Zhen6,7,8, Caixi Xi1, Yuzhe Sun6,7,8, Lingling Wu1, Lifang Wang6,7,8, Xingle Gao1, Yan Li6,7,8, Yaoyang Fu1, Zhuye Jie6,7,
Shenghui Li 10, Danhua Zhang 1, Yiqing Chen1, Yiyi Zhu1,5, Shaojia Lu1,2,3,4, Jing Lu1,2,3,4, Dandan Wang1,2,3,4, Hetong Zhou1,2,3,4,
✉
Xiuxia Yuan11,12, Xue Li11,12, Lijuan Pang11,12, Manli Huang1,2,3,4, Huanming Yang6,7, Wenwei Zhang6,7,8, Susanne Brix 13,14 ,
✉ ✉ ✉ ✉
Karsten Kristiansen 6,14,15 , Xueqin Song 11,12 , Chao Nie 6,7,8 and Shaohua Hu 1,2,3,4,16

© The Author(s), under exclusive licence to Springer Nature Limited 2022

The intricate processes of microbiota-gut-brain communication in modulating human cognition and emotion, especially in the context
of mood disorders, have remained elusive. Here we performed faecal metagenomic, serum metabolomics and neuroimaging studies on
a cohort of 109 unmedicated patients with depressed bipolar disorder (BD) patients and 40 healthy controls (HCs) to characterise the
1234567890();,:

microbial-gut-brain axis in BD. Across over 12,000 measured metabolic features, we observed a large discrepancy (73.54%) in the serum
metabolome between BD patients and HCs, spotting differentially abundant microbial-derived neuroactive metabolites including
multiple B-vitamins, kynurenic acid, gamma-aminobutyric acid and short-chain fatty acids. These metabolites could be linked to the
abundance of gut microbiota presented with corresponding biosynthetic potentials, including Akkermansia muciniphila, Citrobacter spp.
(Citrobacter freundii and Citrobacter werkmanii), Phascolarctobacterium spp., Yersinia spp. (Yersinia frederiksenii and Yersinia aleksiciae),
Enterobacter spp. (Enterobacter cloacae and Enterobacter kobei) and Flavobacterium spp. Based on functional neuroimaging, BD-related
neuroactive microbes and metabolites were discovered as potential markers associated with BD-typical features of functional
connectivity of brain networks, hinting at aberrant cognitive function, emotion regulation, and interoception. Our study combines gut
microbiota and neuroactive metabolites with brain functional connectivity, thereby revealing potential signalling pathways from the
microbiota to the gut and the brain, which may have a role in the pathophysiology of BD.

Molecular Psychiatry; https://doi.org/10.1038/s41380-022-01569-9

INTRODUCTION Bipolar disorder (BD) is a prevalent and debilitating mental illness

The bidirectional interaction between the brain and the gut with complex aetiology. Emerging evidence has indicated BD as a
microbiota, the microbiota-gut-brain (MGB) axis, has emerged as a whole-body disease with changes in the gut microbiota and serum
research hotspot in psychiatry. Healthy gut microbiota is essential metabolome [6, 7]. Recent reviews [8, 9] have summarised evidence
for maintaining physical and mental well-being [1], while dysbiosis regarding the gut microbiota composition across major psychiatric
in the microbial community—imbalances in the composition disorders including BD, showing that higher levels of Eggerthella,
and function of gut microbes—may disrupt immunological and Lactobacillus, Enterococcus, Flavonifractor and Streptococcus, and
metabolic processes [2, 3], thus reshaping host emotion, cogni- lower levels of Coprococcus, Faecalibacterium and Ruminococcus,
tion, behaviour, and contribute to the pathophysiology of major were observed in BD patients compared to healthy controls (HCs).
psychiatric disorders [4, 5]. Identified bacterial genera were reported to be involved in the

1
Department of Psychiatry, the First Affiliated Hospital, Zhejiang University School of Medicine, Hangzhou 310003, China. 2The Key Laboratory of Mental Disorder’s Management
in Zhejiang Province, Hangzhou 310003, China. 3Brain Research Institute of Zhejiang University, Hangzhou 310003, China. 4Zhejiang Engineering Center for Mathematical Mental
Health, Hangzhou 310003, China. 5Wenzhou Medical University, Wenzhou 325035, China. 6BGI-Shenzhen, Shenzhen 518083, China. 7China National GeneBank, BGI-Shenzhen,
Shenzhen 518120, China. 8Shenzhen Key Laboratory of Neurogenomics, BGI-Shenzhen, Shenzhen 518083, China. 9College of Life Sciences, University of Chinese Academy of
Sciences, Beijing 100049, China. 10Key Laboratory of Precision Nutrition and Food Quality, Department of Nutrition and Health, China Agricultural University, Beijing 100083,
China. 11Department of Psychiatry, First Affiliated Hospital of Zhengzhou University, Zhengzhou 450052, China. 12Biological Psychiatry International Joint Laboratory of Henan,
Zhengzhou University, Zhengzhou 450052, China. 13Department of Biotechnology and Biomedicine, Technical University of Denmark, DK-2800 Kgs, Lyngby, Denmark.
14
Qingdao-Europe Advanced Institute for Life Sciences, Qingdao, Shandong 266555, China. 15Laboratory of Genomics and Molecular Biomedicine, Department of Biology,
University of Copenhagen, DK-2100 København, Denmark. 16Department of Neurobiology, NHC and CAMS Key Laboratory of Medical Neurobiology, School of Brain Science and
Brian Medicine, and MOE Frontier Science Center for Brain Science and Brain-machine Integration, Zhejiang University School of Medicine, Hangzhou 310058, China. 17These
authors contributed equally: Zhiming Li, Jianbo Lai, Peifen Zhang and Jiahong Ding. ✉email: [email protected]; [email protected]; [email protected]; [email protected];
[email protected]

Received: 12 October 2021 Revised: 6 April 2022 Accepted: 7 April 2022

 Z. Li et al.
2
production of short-chain fatty acids (SCFAs) [6, 10, 11], and a structured clinical interview. The 24-item Hamilton Depression Rating
associated with glutamate and gamma-aminobutyric acid (GABA) Scale (HAMD-24) [20] and the Montgomery–Asberg Depression Rating
metabolism [9]. However, due to the limited number of studies, Scale (MADRS) [21] were used to evaluate the severity of depression; the
modest sample sizes and inconsistent methodology, more investi- Hamilton Anxiety Rating Scale (HAMA) [22] was used to measure the
gations with updated databases and high-resolution metagenomic severity of anxiety; the Young Mania Rating Scale (YMRS) [23] was used to
assess the severity of mania. The HAMD-24 score ≥14 was defined as
sequencing technologies are essential to uncover the interrelation- 'having a current depressive episode' and was set as the threshold for
ship between altered gut microbiota and BD aetiology [10–12]. inclusion. All patients needed to meet the following inclusion criteria: (i)
Beyond the mere identification of BD-related microbial taxa, HAMD-24 score no less than 14; (ii) drug-naïve or drug-free for at least
attempts have been made to interpret the potential functionality 3 months; (iii) no other psychiatric comorbidity or apparent suicidal
of 'signature' gut microbes. One important aspect was to examine thoughts (suicidal ideation scored ≤2 in MADRS) to exclude individuals not
the gut microbiota-derived metabolites. In fact, there had been a suitable for standardised pharmacotherapy in the ISBD study. Exclusion
growing interest in investigating the role of microbial tryptophan- criteria included: (i) chronic, severe cardiovascular or cerebrovascular
kynurenine metabolism pathway in BD aetiology [13, 14]. The diseases or other organic diseases of the brain (e.g. epilepsy and tumours);
(ii) chronic or acute inflammatory, autoimmune disorders; (iii) history of
reason may be that not only is tryptophan the precursor of
substance abuse, such as alcohol and tobacco; (iv) current pregnancy or
serotonin, a neurotransmitter is well known for its entangled breastfeeding subjects; (v) consumption of antibiotics, prebiotics or
relationship with mood disorders [15]; but also altered levels of probiotics within 4 weeks before screening; (vi) having contraindications
tryptophan, kynurenine, kynurenic acid and xanthurenic acid were for MRI scanning (e.g. metal implant); (vii) failing to provide the informed
found in the metabolic profiling of BD, defined as dysregulated consent. HCs with no DSM-IV psychiatric disorders were recruited from
kynurenine pathway [13]. Notably, the bloodstream, a transporter local communities and were required to abide by the exclusion criteria.
of metabolites, acted as a bridge for the 'crosstalk' between the Laboratory examinations on the thyroid function, blood inflammatory
brain and the gut microbiota [7]. Studies had shown that the factors and T-cell subsets in patients were performed to identify potential
serum metabolome of BD patients differed from that of healthy physical illnesses. Prior to the enrolment, all the participants were inquired
individuals, reflecting changes in pathways related to the about their dietary habits to exclude individuals with special dietary
requirements, such as food allergy, food intolerance, vegetarianism, or due
metabolism of specific amino acid, lipid, the citric acid cycle and to religious or cultural reasons. Other relevant demographic and clinical
polyunsaturated fatty acids [16–18]. However, serum metabolomic profiles e.g. age, sex, height, weight, body mass index (BMI), marital status,
alterations could be disease-specific or associated with medica- drinking and smoking history, age of onset and previous history of
tions to treat BD [17–19]. Therefore, a cross-sectional design with medication of participants were collected (Supplementary Table 1).
drug-naïve or drug-free patients would be desirable to detect the Notably, there was no significant difference between the BD group and
metabolic biomarkers for BD. the HC group in relation to age, sex, and BMI, analysed by Wilcoxon’s rank-
In an attempt to fully explore the components of the MGB axis sum test and Fisher’s exact test (p > 0.05; Supplementary Table 2).
and how the gut microbiota may influence host metabolism and
neural networks, we hypothesised that an integrated analysis of the Stool and plasma sample collection
gut microbiota, the serum metabolome and resting-state functional Faecal samples from all 149 recruited participants were collected during the
connectivity (rsFC) might not only unravel their relationships but clinical examination visit. Participants were asked to deposit stool into a
also reflect the pathophysiological underpinnings of BD. collection bowl and hand the bowl over to a clinical assistant. The stool
From a multi-omics approach, we employed shotgun metage- sample was aliquoted using a scoop into a tube which was snap-frozen in dry
nomic sequencing, untargeted mass spectrometry, and resting-state ice and stored at −80 °C within half an hour after collection. A blank swab
functional magnetic resonance imaging (rs-fMRI) to investigate the was added to the faecal preservation solution as a control. The elbow vein
blood (5 ml) was collected from 80 BD patients and 38 HCs (thirty-one
multidimensional differences between 109 unmedicated, depressed individuals out of the total 149 participants refused to be sampled, and were
BD patients and 40 healthy counterparts. With the advantage of high- thus removed from this session) during 6 a.m.to 7 a.m. in a fasting state with
resolution technologies and multidimensional data, we would be in a vacutainer tubes containing heparin, and was immediately centrifuged for
better position to identify quantifiable and informative biomarkers for 10 min (3000 rpm, 4 °C). Each aliquot (1.5 ml) of the plasma samples was
BD. Moreover, plausible interactive links between the gut and the stored at –80 °C until the ultrahigh-performance liquid chromatography was
brain could be evaluated through the concomitant changes between performed with liquid chromatography-mass spectrometry (LC/MS) analysis.
different omics. The current study would help to provide a
comprehensive landscape of the gut-brain axis in the clinical context Metabolome profiling of human serum samples
of acute bipolar depression. Sample preparations. Untargeted metabolomics analysis using LC/MS was
conducted. Briefly, the supernatant of the NIST standard curve correction
solution was obtained by mixing the serum sample with 80% methanol,
METHODS vortexed, and centrifuged at 12,000 rpm at 4 °C for 10 min. After thawing all
Ethics statement samples at 4 °C, the samples for quality control and LC-MS detection were
As part of the Integrated Study of Bipolar Disorder (ISBD, ChiCTR-COC- prepared at the same centrifuge condition as previously described [24].
17011401), this study was approved by the Ethics Committee of the First
Affiliated Hospital, School of Medicine, Zhejiang University (Ref.2017-397) LC-MS-based serum metabolic profiling and identification of metabolites. The
and the BGI Review Board of Bioethics and Biosafety (BGI-IRB20153). All raw data was converted into the mzXML format using ProteoWizard
applicable institutional regulations concerning the ethical use of informa- (v3.0.8789). The identification, filtration and alignment of peaks were
tion and samples from human participants were followed during this conducted by the R-package XCMS (R-v3.1.3). After quality control, the
study. Each individual or his/her legal guardian provided the signed metabolite annotations of the LC-MS data were verified using databases
informed consent before enrolment. including LipidMaps (http://www.lipidmaps.org), massbank (http://www.
massbank.jp/), Human Metabolome Database (HMDB) (http://www.hmdb.ca),
Metlin (http://metlin.scripps.edu), mzclound (https://www.mzcloud.org), as well
Human population recruitment as the metabolome database, which was constructed by the BioNovoGene
A hundred and nine BD patients with a current depressive episode and (Suzhou, China) to avoid missing any important metabolite.
forty healthy controls from the same geographical area (Zhejiang Province,
China) were recruited from the Department of Psychiatry, the First
Affiliated Hospital, Zhejiang University School of Medicine. According to Metabolite pathway identification
the Diagnostic and Statistical Manual of Mental Disorders, 4th Edition The biological pathways of key metabolites that manifested significant
(DSM-IV) criteria for BD-I, BD-II and BD not otherwise specified (NOS), the differences between BD patients and HCs were annotated. Biological
diagnosis was confirmed by two trained psychiatrists fulfilling the Chinese pathway analysis was performed through metabolite set enrichment
version of the Mini-International Neuropsychiatric Interview (MINI) through analysis using the MetaboAnalyst tool suite [25].

Molecular Psychiatry
 Z. Li et al.
3
MRI data acquisition and processing Faecal DNA extraction and metagenomic sequencing
Acquisition. Forty-four BD patients and thirty-seven HCs (68 individuals According to the manufacturer’s instructions, DNA was extracted from
out of the total 149 participants were not able to attend the MRI session) thawed faecal samples with OMEGA‐soil DNA Kit (Omega Bio‐Tek, USA).
underwent a 20-min MRI session including structural and functional scans The extracts were treated with DNase-free RNase to eliminate RNA
on a 3.0 Tesla GE Signa HDxt scanner (GE Healthcare, Waukesha, Wisconsin, contamination. The DNA quality was examined by NanoDrop 2000 UV–vis
USA), equipped with an eight-channel phased array head coil. All subjects spectrophotometer and 1% agarose gel electrophoresis. The DNA library
were instructed to remain still and awake with their eyes open during the was constructed according to the manufacturer’s instructions (Illumina).
whole session. Cushions were used to restrict head movements and Applying the same workflow as described previously [33], we constructed
earplugs for reducing the noise. one paired-end (PE) library with an insert size of 350 bp for each sample,
Resting-state functional images using an echo-planar imaging protocol followed by a high-throughput sequencing with PE reads of length 2 × 150
were acquired with the following parameters: TR (repetition time) = 1800 bp, using NEXTFLEX Rapid DNA-Seq (BioScientific, Austin, TX, USA). PE
ms, TE (echo time) = 30 ms, flip angle = 90 degrees, voxel size = 3.75 × sequencing was performed on Illumina NovaSeq (Illumina Inc., San Diego,
3.75 × 4 mm3, field of view = 240 × 240 mm2, 28 axial slices per volume, CA, USA). Low quality or human genomic DNA reads were removed [34].
180 time points/volumes. High-resolution 3D T1-weighted magnetisation- Human genomic DNA reads were identified via SOAP2.21 [35] and were
prepared rapid acquisition with gradient echo (MPRAGE) structural images removed if they shared >95% sequence [34] with the human genome
were acquired for anatomical reference, with parameters of TR = 7.05 ms, reference sequence (hg38).
TE = 2.85 ms, flip angle = 8 degrees, voxel size = 1 mm3 isotropic, field of
view = 240 × 240 mm2.
Construction of species and KO profiles
The high-quality reads were aligned to the gut microbiome genomes
Data preprocessing and denoising. Both structural and functional images catalogue [36] by bwa (default parameters) and 91.97 ± 3.31% reads (n
were preprocessed with the CONN-20.b toolbox [26] (http://www.nitrc.org/ = 149; Supplementary Table 4) were mapped. Sequence-based contigs
projects/conn, default preprocessing pipeline), based on SPM12 [27]. abundance profiling was performed [37] by jgi_summarize_bam_con-
Functional scans were realigned to the corresponding T1 images and tig_depths (default parameters). Reads were mapped to the genomes
resampled along the phase-encoded direction to carry out susceptibility catalogue and the number of reads counted formed a mapping depth or
distortion correction for adjusting head motion and possible deformation
abundance matrix. Considering the different sequencing depths of
due to field inhomogeneities (realign and warp). Functional slices
different samples, we used the mapping depth matrix of normalisation
(interleaved and bottom-up) were time-shifted and resampled for slice- to estimate the abundances of contigs. For the species profile, we used
timing correction. Since functional images were notoriously prone to head the species assignment of each contig from the original genome
motion artefacts, a more conservative approach to detecting outlier scans catalogue and took the median of the relative abundance of contigs
due to excessive head motion was employed. Scans with composite from the same species to generate the abundance of that certain
subject motion threshold over 0.5 mm or the observed global blood-
species. The KO functional profile of each species was estimated as
oxygen-level-dependent (BOLD) signal changes above 3 standard devia-
follows: the relative abundance of a KO was calculated as the summation
tions were marked as outliers [28]. of the relative abundance of its corresponding contigs.
All anatomical and functional images were normalised to the standard
Montreal Neurological Institute (MNI) space, and then segmented
into grey matter, white matter, and cerebrospinal fluid (CSF) classes Species count, α-diversity and β-diversity
(segmentation and normalisation). Direct normalisation mapped the The calculation of species count was based on a documented method [38],
functional data (interpolated in isotropic 2 mm voxels; resolution which was to tally non-zero species in each sample. The α-diversity (within-
consistent with the MNI average mask) to the reference structural data. sample diversity) was estimated on the basis of the gene profile of each
The resulting functional data were smoothed with a Gaussian kernel of 6 sample according to the Shannon index, using R (3.5.0) vegan package
mm full-width half maximum to boost the BOLD signal-to-noise ratio. To [39]. The vegdist function in vegan package was implemented to compute
further mitigate the influence of motion-related and physiological noise, the Bray–Curtis dissimilarity and the β-diversity (between-sample diversity).
necessary denoising procedures had been applied. CONN’s anatomical
component correction strategy computed the confounding effects of
noise components from white matter and CSF, which were linearly Statistical analysis
Multivariate analysis. Multivariate statistical analyses were applied to
regressed out of the global signal using aCompCor [29]. Band-pass
discriminate between BD patients and healthy individuals. Principle
filtering [0.008 Hz, 0.09 Hz] was employed to the preprocessed functional
time series. component analysis (PCA) of subject-level RRC matrices was performed,
using the ade4 package in the R platform [40]. Distance-based redundancy
analysis (dbRDA) was carried out based on the Bray–Curtis dissimilarity on
Regions of interest (ROIs). We defined the whole brain (excluding the the serum metabolome, gut microbial composition and functional profile
cerebellum) into 136 different ROIs to conduct the ROI-based rsFC analysis. using capscale [39] (function of the vegan package in R).
Ninety-one bilateral cortical ROIs were defined from FSL Harvard-Oxford
Atlas maximum likelihood cortical atlas [30] and fifteen bilateral subcortical
ROIs were defined from FSL Harvard-Oxford Atlas maximum likelihood Co-inertia analysis (CIA) analysis. CIA was performed on gut microbial
abundance profiles and brain functional RRC matrices profiles, serum
subcortical atlas (see Supplementary Table 9 for the ROIs list). Based on
metabolite abundance profiles and functional RRC matrices of samples to
CONN’s default clustering and ordering algorithms, we also included an
additional 30 networks-based ROIs defined from CONN’s independent assess the relationship between gut microbiome, serum metabolome
component analyses of the Human Connectome Project dataset [26] into and rsFC.
the rsFC analysis.
PERMANOVA tests. PERMANOVA [41] was conducted on the species-
Neuroimaging analyses. We investigated the ROI-based rsFC at the abundance and serum metabolite profiles of the samples to assess the
subject level with the CONN-20.b toolbox [31]. Average BOLD timeseries effect of each clinical measure [42] using the Bray–Curtis dissimilarity and
of all predefined ROIs were analysed in a pairwise manner to compute the 999 permutations in R (3.5.0, vegan package [39]). Clinical measures with
adjusted p < 0.05 were considered salient to associate with species and
Fisher-transformed bivariate correlation coefficient between each pair of
serum metabolites. In addition, we applied the PERMANOVA to investigate
ROIs. Pairwise ROI-to-ROI connectivity (RRC) matrix characterised the entire
networks of connections for each subject. Subject-level RRC matrices were the possible effects of each BD-related serum metabolites and gut species
extracted from CONN to be studied subsequently with serum metabolome on the significant rsFC clusters in bipolar brains. Gut species and serum
and gut microbiome data. metabolites with adjusted p < 0.05 were considered salient to associate
We examined between-sample RRC matrices (BD vs HC group) and with the alterations of rsFC in BD.
carried out the functional network connectivity analysis using multi-
variate parametric generalised linear models (GLM) [32], where group- Association analysis. Spearman’s correlation was used to ascertain the
relevant functional connections were organised into significant network pairwise correlations between clinical measures, BD-related serum
clusters with a cluster-level false discovery rate (FDR)-corrected p < 0.05 metabolites, BD-related gut microbial function and BD-related gut
threshold. microbial composition [43].

Molecular Psychiatry
 Z. Li et al.
4

Fig. 1 Altered metabolites in serum of BD patients compared to healthy controls. a The effect size of phenotype indices contributed
significantly to the variance (R2) of the serum metabolome (all subjects, BD: 80, HC: 38). b A clear discrepancy of serum metabolomes between
BD patients and healthy controls, revealed by the dbRDA. The metabolites (squares), which were identified as the main contributors to the
discrepancy are specified. c Major metabolic pathways involved in the differentially enriched metabolites comparing BD patients and healthy
controls. d Reaction steps for steroid hormone biosynthesis, tryptophan metabolism, arginine and proline metabolism, and citrate cycle (TCA
cycle). Metabolites enriched in HCs are shown in yellow, whereas the metabolites enriched in the BD are shown in red.

Hypothesis testing and multiple testing correction. Differential abundance RESULTS

of gut microbial composition, gut microbial function and serum Serum metabolomic profiling reveals an apparent disparity
metabolites were tested by two-tailed Wilcoxon’s rank-sum test between between BD patients and HCs
HCs and BD patients. Functional connectivity strength of rs-fMRI was Serum samples were analysed by untargeted LC/MS. After quality
tested by Analysis of Variance (ANOVA) between HCs and BD patients. FDR control, data filtering, and normalisation (see Methods), we
adjustment was employed by the Benjamin-Hochberg method [44] (using
the R package p.adjust), and the local FDR was provided in the article. identified 12,127 metabolic features across 118 samples. Of these,
8918 features (73.54%) were altered in BD (FDR <0.05), indicating
extensive metabolic changes in BD patients. To investigate the
Random forest models potential function of BD-related metabolites, we first annotated
According to the previously reported method [45], the concentrations of
metabolic features by using the Kyoto Encyclopaedia of Genes
neuroactive metabolites in serum and disorder status classification were
modelled using the random forest 4.6–12 package [46] based on the
and Genomes (KEGG) database. A total of 265 annotated serum
species' compositional profiles. metabolites were obtained, including plasma metabolites with
important functions in humans. Given the symptomatologic
Variable selection. Firstly, the random forest regression model was used to complexity and heterogeneity of BD, we also collected beha-
predict the concentration of the serum metabolites, and the largest vioural data from patients, including diagnostic questionnaires
variable of IncMSE was selected as the first variable. Secondly, the first and demographic information (Supplementary Table 1). Notably,
variable and the remaining variables were combined into two variables to we found no evidence that behavioural measures were directly
predict the neuroactive metabolites, and the Q2 values of the predicted related to the serum metabolome (PERMANOVA test, p > 0.05).
results were computed and compared (the variables that maximise Q2 Nevertheless, the severity of BD symptoms (quantitative measures
were selected). More variables were added iteratively in the same way, derived from questionnaires of MADRS, HAMD and HAMA) and
until Q2 was no longer increased.
P the proportion of T-cell subsets were significantly associated with
2
ðyi  ybi Þ the serum metabolome (Fig. 1a, PERMANOVA test, p < 0.01).
Q2 ¼ 1  P
ðyi  yi Þ2 Additionally, we found that years of education and sex exerted
moderate but statistically significant effects on the serum metabo-
yi and ybi are the i-th observation value and predicted value of the serum lome of patients with BD (Fig. 1a, PERMANOVA test, p < 0.05). The
metabolites, respectively. BD disease status, which accounted for almost 12% of the variance,
was the main factor for changes in the serum metabolome,
Model training. Neuroactive metabolites were predicted by the variables distinguishing BD patients from HCs (Fig. 1a).
selected above, where Q2 was calculated, and the variables of importance
We further examined the data with a dbRDA and found that the
were obtained.
serum metabolome in BD patients was in stark contrast to that
Cross-validation. We applied the leave-one-out cross-validation method. of HCs (Fig.1b PERMANOVA test, p < 0.01), where 138 of 265
Each sample was independently assumed as a validation set, and the metabolites were significantly associated with BD (Supplementary
remaining 117 samples were assumed as training sets. Q2 was calculated Fig. 1 and Supplementary Table 3). Moreover, we observed
by the predicted value and the experimental value. Cross-validation was that levels of more than half (73.2%) of the metabolites were
used throughout all random forest modelling processes. reduced across all BD patients (FDR <0.05, |fold change | > 1.35;

Molecular Psychiatry
 Z. Li et al.
5
Supplementary Fig. 1 and Supplementary Table 3). The BD significantly associated with microbial changes. In addition, BMI
patients exhibited distinct patterns of metabolic pathway and sex were associated with the gut microbiome changes in BD
changes, where the 138 serum metabolites were involved in 64 patients (PERMANOVA test, p < 0.05, Supplementary Fig. 5a). BD
metabolic pathways, including citrate cycle, fatty acid biosynth- patients had significantly lower species counts (p < 0.01) and
esis, glutathione metabolism and arginine and proline metabolism bacterial Shannon-diversity (p = 0.062) than controls (Supplemen-
(Fig. 1c, d and Supplementary Table 3). Metabolites that were tary Fig. 5b, c), indicating that the diversity and richness of the gut
reduced in abundance in BD reflected pathways involved in microbiota in BD patients were relatively poor. We observed a
steroid hormone biosynthesis, tryptophan metabolism, phenyla- higher β diversity of the BD microbiota (p < 0.01), implying a more
lanine metabolism, pyrimidine metabolism, pantothenate and heterogeneous community structure among BD individuals than
CoA biosynthesis, and butanoate metabolism (Supplementary that in HCs (Supplementary Fig. 5e). The dbRDA showed that the
Table 3). Similarly, we found that the decrease of serum GABA, taxonomic composition and functional potential of the BD
common in BD patients, might be pertinent to the increase of microbiota differed markedly from that in HCs (Supplementary
spermine as they shared the same precursor, putrescine (Fig. 1d). Fig. 5e, f), where we identified 600 species associated with BD
The concomitant co-variation of GABA and neuroactive steroids in (Fig. 2a and Supplementary Table 5, FDR <0.05, |fold change | > 2).
BD, found in the present study, demonstrated the theoretical Specifically, 136 species were enriched in BD patients, while 464
plausibility of targeting neuroactive steroids in future BD were depleted (Fig. 2a and Supplementary Table 5); species that
treatment (Fig. 1d). were most enriched in BD patients included Streptococcaceae
Mounting evidence further suggested that dysregulation of the (nine species; Streptococcus mitis, Streptococcus oralis, Streptococ-
metabolic fate of tryptophan via the kynurenine pathway may be cus pseudopneumoniae, Streptococcus mitis and Streptococcus spp.),
implicated in a range of severe psychiatric disorders, including BD Fusobacteriaceae (three species; Fusobacterium varium and
[47, 48]. In line with previous studies [49], the levels of kynurenine Fusobacterium spp.), Tissierellaceae (one species; Urmitella timo-
and kynurenic acid were significantly lower in BD patients than nensis), Bacteroidaceae (three species; Bacteroides barnesiae,
those in HCs (Fig. 1d). Notably, we found that indolepyruvate, Bacteroides togonis and Bacteroidaceae spp.), and Actinomyceta-
enriched in BD patients, might compete with the synthesis of ceae (five species; Actinomyces graevenitzii, Actinomyces oris,
kynurenic acid and serotonin, all of which shared the same Actinomyces spp. and Varibaculum cambriense) (Fig. 2a and
precursor, tryptophan (Fig. 1d). Additionally, vitamins involved in Supplementary Table 5). Depleted species included Akkermansia-
the production of neurotransmitters were significantly correlated ceae (four species; Akkermansia muciniphila and Akkermansia spp.),
with the symptom severity of BD, including folic acid (vitamin B9), Yersiniaceae (five species; Yersinia aleksiciae, Yersinia frederiksenii
pyridoxine (vitamin B6), pantothenic acid (vitamin B5) and and Serratia spp.), Enterobacteriaceae (12 species; Citrobacter
riboflavin (vitamin B2) (Supplementary Fig. 2), suggesting the freundii, Citrobacter werkmanii, Citrobacter spp., Cronobacter
potential effects of regulating vitamin intake on BD symptoms. malonaticus, Enterobacter cancerogenus, Enterobacter cloacae,
Essential amino acids [50] and vitamins [51] cannot be Enterobacter kobei and Enterobacter mori), Acidaminococcaceae
synthesised by humans and have to be acquired from the diet. (eight species; Acidaminococcus fermentans, Acidaminococcus spp.,
In addition, tryptophan derivatives (kynurenine, kynurenic acid, Phascolarctobacterium succinatutens and Phascolarctobacterium
serotonin), tyrosine derivatives (tyramine, dopamine), and some spp.), Eubacteriaceae (13 species; Eubacterium eligens, Eubacterium
B-vitamins (such as folic acid, pantothenic acid and pyridoxine) spp.), Ruminococcaceae (43 species, Ruminococcaceae spp.,
have been reported to be produced by the gut microbiota via the Ruminococcus spp., Faecalibacterium prausnitzii, Anaerotruncus
degradation of diet-derived amino acids and purine [52, 53] colihominis and Anaeromas silibacillus), Morganellaceae (three
(Supplementary Fig. 3). species, Providencia alcalifaciens), Flavobacteriaceae (two spe-
We clustered BD-related serum metabolites (Supplementary cies, Flavobacterium spp.) (Fig. 2a, Supplementary Fig. 6 and
Table 3) and examined associations of cluster abundance with the Supplementary Table 5). With respect to the existing literature,
symptom severity in BD patients. Importantly, the clusters Streptococcaceae is of particular interest, as several Streptococ-
including “neuroactive metabolites” of gut microbiota derivatives caceae, including S. vestibularis, have been found to be
(B-vitamins, kynurenic acid, GABA, SCFA derivatives) were strongly associated with schizophrenia [54]. A. muciniphila negatively
associated with the symptom severity of BD across the entire correlated with BMI (Spearman correlation, r = −0.14, p =
cohort (Supplementary Fig. 4). Taken together, we here discovered 0.0976), which was depleted in BD patients, was also reported
apparent changes in serum metabolomics of BD; results that in to be decreased in overweight, obesity, type 2 diabetes mellitus,
part could be explained through the analysis of metabolic and hypertension [55–57]. Consistent with the previous finding
pathways. of Enterobacteriaceae (unclassified genus) related to BD [58], our
analysis here identified nine species of Enterobacteriaceae
Taxonomic and functional characterisation of the gut associated with BD (Fig. 2b).
microbiota in BD Functional analyses of single species showed that species
To investigate whether the gut microbiota-mediated metabolomic enriched in BD/HC encoded functions related to neuroactive
changes in BD, we analysed the gut microbiota using metagenome metabolites mediating central neuronal processes, such as acetyl-
shotgun sequencing across the 149 faecal samples, generating an CoA metabolism, polyamine biosynthesis, cofactor and vitamin
average of 107.3 million high-quality reads (16.1 Gb of data) per biosynthesis, and aromatic amino acid metabolism (Supplementary
sample on an Illumina HiSeq platform (Supplementary Table 4). The Fig. 7). 15.4% of the BD-enriched species and 17.3% of the HC
high-quality sequencing reads were aligned to a comprehensive enriched species displayed significant differences in these functions,
reference, Unified Human Gastrointestinal Genome (UHGG), com- possibly reflecting disturbances of acetyl-CoA, polyamine, aromatic
prising 4644 species-level genomes [36], which allowed on average amino acid, cofactor, and vitamin availability in BD. In addition, BD/
91.9 ± 0.03% of the reads to be mapped (Supplementary Table 4), HC enriched species encoded a variety of amino acid, carbohydrate,
highlighting a considerable coverage of the gut microbiome for and methane metabolic functions (Supplementary Fig. 8). More
subsequent analyses. The classification resulted in a total of 3835 importantly, these microbial-derived functions concomitantly chan-
inferred prokaryotic species, and annotation of 9106 functional ged with the corresponding BD-associated serum metabolites
categories using the KEGG database. (Fig. 3). We thus concluded that the enrichment of metabolites in
With respect to the behavioural data, we found that the BD BD patients was associated with gut microbiota-mediated AAA
disease status was the major factor contributing to the alterations biosynthesis, SCFA biosynthesis, choline-related function, cofactor
in the gut microbiome. The severity of BD symptoms was also and vitamin biosynthesis.

Molecular Psychiatry
 Z. Li et al.
6
a
Archaea b
Methanomassiliicoccaceae
BD-enriched species Healthy control-enriched species
1
Methanobacteriaceae 1 1 Euryarchaeota Enterobacteriaceae
Akkermansiaceae
Spirochaetaceae 2 Bacteria * * **
* * 1e−3
Brachyspiraceae 2 Spirochaetes * ** **
*
Akkermansiaceae Bacteria *

Relative abundance
4 1e−05 *
Erwiniaceae 1 Verrucomicrobia

Relative abundance
Succinivibrionaceae 2
1e−6
Neisseriaceae 2
Pseudomonadaceae 1 1 1e−07 *
Yersiniaceae
**
5
Helicobacteraceae 1
1e−9
Campylobacteraceae 1
1e−09
Vibrionaceae 1 Bacteria
Alcaligenaceae 1 Proteobacteria
Morganellaceae 1 3

2)

1)
7)

2)

1)
5)
)

0)

6)
6)

6)

71

3)
47

32

21
05

27

41
59

74

74
99

76
Hafniaceae

45
1

74
00

13

43
00

15

42
00

43

43
95

03

31
17
00

E1

E1
E0

E1

E1
E0

E1

E1
E0

E1

E2
Pasteurellaceae

ME
1

ME

OM

OM
OM

OM

OM
OM

OM

OM
OM

OM

OM
NO
NO

EN

EN
N
Acetobacteraceae

EN

EN
N

EN

EN
5

EN

EN

N
GE
GE

GE
GE

GE

(G

(G
(G

(G
(G

(G
(G

(G

.(
.(

.(
ri (
Desulfovibrionaceae 1 1

ii (

ila

ila
p.

p.
ae

ae
us

di i

pp
p

p
mo

an
sp

sp

sp
sp

iph

iph
un
en

ac

ac

es

rkm
Sutterellaceae 2 12

ter

sia

sia
r

clo

clo
fre

cin

cin
ter

og
cte

ea
ac

er
ac

an

an
we

mu

mu
r

iac
ter

ter
ba

cte
nc
ob
Rhodospirillaceae 2 2

ob

rm

rm
ac

ac
tro

ter

sia

sia
ca

ba

cte
ter
ter

ke

ke
ob

ob
Ci

ac
tro

an

an
En

ter
Enterobacteriaceae

ba
En

Ak

Ak
3 12

ter

ter
Bacteria

ob

rm

rm
Ci
ac

tro
En

En

ter

ke

ke
ob

Ci
Fusobacteriaceae 3 1

En

Ak

Ak
ter
Fusobacteria Streptococcaceae

En
Carnobacteriaceae 1
Tissierellaceae 1 ** * * ** * ** Yersiniaceae
Clostridiales Family XIII 2 **
* *
Sporomusaceae 1 1e−06 *
**

Relative abundance
Staphylococcaceae 1

Relative abundance
1e−08 * *
Planococcaceae 1 1e−07
Lactobacillaceae 1 2
Bacillaceae 3 1e−09
Paenibacillaceae 3 1e−08
Peptococcaceae 1
Streptococcaceae 9 1e−10
1e−09
Oscillospiraceae 3 11
Erysipelotrichaceae 3 8 Bacteria
Peptoniphilaceae 1 9 Firmicutes

7)

6)
)

)
9)

1)

6)
2)
17

84

9)
6)
5)

2)
8)
78

86
33

43

96
73

59
Christensenellaceae

34
20
67

78

19
2

40
07

07
58

31

31
76

31
43
44
17

20

32
43
E2

E2
E0

E2

E2
E1

E2
E1
E2
ME

ME

E2
E1
Catabacteriaceae 2 10

OM

OM
OM

OM

OM
OM

OM
OM
OM

OM
OM
NO

NO
EN

EN
EN

EN

EN
Enterococcaceae
EN
1 1

EN
EN
EN

EN
EN
GE

GE
(G

(G
(G

(G

(G
(G

(G
(G
(G

(G
(G
s(

e(
Acidaminococcaceae 8 p.

p.
tis

tis

tis
p.

p.
ae
ali

p.
nia

p.
nii
sp

sp
mi

sp

mi

mi

sp
sp

sp
or

ici

se
mo
us

us
Peptostreptococcaceae 1 2
us

us

us

us

ks

ae
us

us

rik

tia
cc

cc

eu
cc

cc

cc

cc

ale
cc

cc

ce
de

rra
co

co

pn
co

co

co

co
Lachnospiraceae 15 29
co

co

nia
fre

Se
nia
pto

pto
pto

pto

pto

pto
do
pto

pto

rsi
nia
rsi
re

re

eu
re

re

re

re
Eubacteriaceae 3 13
re

re

Ye
St

St

Ye

rsi
ps
St

St

St

St
St

St

Ye
us

Veillonellaceae 1 8
cc
co

Clostridiaceae 14 38 Ruminococcaceae
pto

1e−03 * *
**
re

Selenomonadaceae 1 1 * * *
St

*
Ruminococcaceae 8 43 Bacteria ** * * *
Fibrobacteraceae 1 Fibrobacteres ** *
Marinilabiliaceae 1 *
Relative abundance

*
Dysgonamonadaceae 1 1e−06
Flavobacteriaceae 2
Bacteroidaceae 3 9
Rikenellaceae 7 Bacteria
Porphyromonadaceae 3 16 Bacteroidetes
1e−09
Muribaculaceae 2 6
Odoribacteraceae 2 2
Prevotellaceae 8 31
Mycoplasmataceae
Bacteria
5
Tenericutes
)

6)

7)
29
)

)
1)
)
)

)
)
82

08

32
)

)
21
10

14
03

35
07

36
41

14

Tannerellaceae
23

5
26
58

19

71
50
00

60
00

35
93

07
23

39
08
10
)

06

13

25
67

09
00

14

19

23
13

26
E1

E1
E1

Micrococcaceae
ME

1
ME

ME

ME
63

ME
ME

ME

ME

ME
ME

ME
Bacteria
OM

OM
OM
NO
04

NO

NO

NO
NO
NO

NO

NO

NO
NO

NO
Atopobiaceae 1
EN

EN
ME

EN
GE
GE

GE

GE
GE
GE

GE

GE

GE
GE

GE
Actinobacteria
_G

_G
_G
NO

T_

Actinomycetaceae
T_

T_

T_
5 2
T_
T_

T_

T_

T_
T_

T_
UT

UT
UT
GU
GE

GU

GU

GU
GU
GU

GU

GU

GU
GU

GU
(G

(G
(G
T_

Eggerthellaceae
.(

1
.(

.(

.(
ii (
s(

.(

.(

.(
.(

.(
pp

p.

p.
p.
GU

pp

pp

pp
pp

pp

pp
itz

pp
ini

pp
sp

sp
sp
es
es

es

es

Coriobacteriaceae 1
sn

es

es

es
om

s(

es

es
m

m
us
ea
au
ea

ea

ea
illu

riu

riu

ea

ea

ea
lih

ea

ea
cc
ac

Bacteria
pr
ac

ac

ac
ac
co

Uncultured 30
cte

cte

ac

ac

ac
ac
co

ac

114
cc
cc

cc

cc
m
ilib

cc

cc

cc
cc
no

cc
us

ba

ba
co
riu
co

co

co
co

co

co
co
ss

co
mi
nc

ali

ali
no
cte
no

no

no
ino

no

no

BD-enriched species Healthy control-enriched species

no

no
Ru
tru

ec

ec
ma

mi
mi

mi

mi
ba

mi

mi
mi

mi
m
Fa

Fa
ro

Ru
ro

Ru

Ru

Ru
ali

Ru

Ru

Ru
Ru

Ru
ae

ae

ec
An

An

Fa

Fig. 2 Gut microbiota species associated with BD compared to controls. a Statistically significant results (FDR <0.05, fold change ≥2| fold
change ≤0.5) of the case-control discrepant analyses. Per microbial family, the number of increased (blue) or decreased (orange) species are
shown respectively, including 135 species in BD belonging to 33 families and 465 species in HC belonging to 59 families. b The boxplots show
the prominent species (including Yersiniaceae spp., Akkermansiaceae spp., Streptococcaceae spp., Ruminococcaceae spp. and Enterobacteriaceae
spp.) that differed significantly in abundances between BD patients and HC.

Microbiota alterations correlate with serum metabolome aleksiciae (Supplementary Fig. 3 and Supplementary Fig. 10b). To
changes in patients with BD substantiate the potential role of gut microbes in the production of
To further explore the links between the gut microbiota and neuroactive metabolites, we focused on the microbial metabolic
serum metabolome composition, we carried out an inter-omics co- pathway/genes predicted to encode enzymes, critical in the main
inertia analysis of the abundances of gut microbes and serum synthesis pathways of these compounds (Supplementary Fig. 2 and
metabolites. Strong connections were identified between gut Supplementary Table 6). We identified 1840 species, which contained
microbes and serum metabolites (Supplementary Fig. 9, RV = 0.265, the whole metabolic pathway or encoded the key synthetases for
p < 0.05), where 98.5% (261/265) of serum metabolites were related to neuroactive metabolites in our reference genomes catalogue
at least one gut microbe. In particular, neuroactive metabolites (Supplementary Table 6). In addition to pyridoxine, these genes/
[59–61], including B-vitamins (pantothenic acid, riboflavin, folic acid module abundances were significantly more abundant in BD patients
and pyridoxine), SCFA derivatives (3-methylthiopropionic acid and and in BD-enriched microbial species (Supplementary Fig. 11). Serum
2-hydroxybutyric acid), kynurenic acid and GABA, were related to concentrations of neuroactive metabolites significantly correlated to
various gut microbes (Supplementary Fig. 10a), such as A. muciniphila, the abundance of the metabolic pathway/cognate synthetase-
F. prausnitzii, E. cloacae, Ruminococcaceae spp., F. prausnitzii and Y. encoding genes in specific species. In particular, A. muciniphila,

Molecular Psychiatry
 Z. Li et al.
7

M10

M11

M12
M1

M6
M3
M9
M2
M8

M7
M4

M5
Functional modules of the gut microbiota
M00579:Acetyl−CoA => acetate
M00377:Reductive acetyl−CoA pathway
M00307: Pyruvate => acetyl−CoA
M00620:Acetyl−CoA => oxoglutarate
+
M00036:Leucine => acetoacetate + acetyl−CoA **
SCFA biosynthesis M00032:Lysine => acetoacetyl−CoA
M00088:Acetyl−CoA => acetoacetate/3−hydroxybutyrate/acetone
+
M00569:Catechol => acetyl−CoA / 4−methylcatechol => propanoyl−CoA
M00013:Propanoyl−CoA => acetyl−CoA
M00545:Trans−cinnamate => acetyl−CoA
+
Choline related function M00090:Phosphatidylcholine (PC) biosynthesis, choline => PC
M00555:Betaine biosynthesis, choline => betaine
+ * * + + *
M00133:Polyamine biosynthesis, arginine => spermidine **
Polyamine biosynthesis M00136:GABA biosynthesis, prokaryotes, putrescine => GABA
M00134:Polyamine biosynthesis, arginine => ornithine => putrescine
M00089:Triacylglycerol biosynthesis
M00093:Phosphatidylethanolamine (PE) biosynthesis, PA => PS => PE
+ +
Lipid metabolism M00098:Acylglycerol degradation +
M00100:Sphingosine degradation
M00099:Sphingosine biosynthesis
+
M00131:Inositol phosphate metabolism, Ins(1,3,4,5)P4 => myo−inositol +
M00120:Coenzyme A biosynthesis, pantothenate => CoA
M00115:NAD biosynthesis, aspartate => NAD
+
M00125:Riboflavin biosynthesis, GTP => riboflavin/FMN/FAD
M00119:Pantothenate biosynthesis, valine/L-aspartate => pantothenate
+ * + + +
M00841:Tetrahydrofolate biosynthesis, GTP => THF + +
M00126:Tetrahydrofolate biosynthesis, GTP => THF + + +
M00124:Pyridoxal biosynthesis, erythrose−4P => pyridoxal−5P
M00140:C1-unit interconversion, prokaryotes **
+ * ** + + + + ** **
M00114:Ascorbate biosynthesis
M00840:Tetrahydrofolate biosynthesis, GTP => THF + +
M00572:Pimeloyl−ACP biosynthesis + + * +
Cofactor and vitamin biosynthesis M00127:Thiamine biosynthesis, AIR => thiamine−P/thiamine−2P
M00577:Biotin biosynthesis, pimelate => pimeloyl−CoA => biotin
+ +
M00573:Biotin biosynthesis, long−chain−acyl−ACP => biotin
M00123:Biotin biosynthesis, pimeloyl−ACP/CoA => biotin
M00117:Ubiquinone biosynthesis, chorismate => ubiquinone +
M00843:L−threo−Tetrahydrobiopterin biosynthesis +
M00842:Tetrahydrobiopterin biosynthesis, GTP => BH4 +
M00116:Menaquinone biosynthesis, chorismate => menaquinone + + *

Spearman correlation coefficient

M00118:Glutathione biosynthesis, glutamate => glutathione

0.3
M00112:Tocopherol/tocotorienol biosynthesis
M00622:Nicotinate degradation, nicotinate => fumarate
M00022:Phosphoenolpyruvate + erythrose-4P => chorismate
+ *
M00044:Tyrosine degradation, tyrosine => homogentisate
M00350:Capsaicin biosynthesis, L-Phenylalanine => Capsaicin
M00039:Monolignol biosynthesis, phenylalanine / tyrosine
M00533:Homoprotocatechuate => 2-oxohept-3-enedioate
Aromatic amino acid M00037:Melatonin biosynthesis, tryptophan => serotonin => melatonin

0
degradation M00137:Flavanone biosynthesis, phenylalanine => naringenin + + + +
M00038:Tryptophan metabolism, tryptophan => kynurenine => 2-aminomuconate
M00042:Catecholamine biosynthesis, tyrosine => adrenaline
+
Aromatic amino acid
M00024:Phenylalanine biosynthesis, chorismate => phenylalanine + + +
M00025:Tyrosine biosynthesis, chorismate => tyrosine + + +
biosynthesis M00040:Tyrosine biosynthesis, prephanate => pretyrosine => tyrosine +* +

−0.3
M00023:Tryptophan biosynthesis, chorismate => tryptophan

BD-enriched Healthy control-enriched

Fig. 3 Relationships between specific gut microbial functions and the concentration of BD-related serum metabolites. The heatmap
displays the Spearman correlation coefficients between functional modules and serum metabolite clusters. Black boxes highlight the BD-
associated metabolites and their corresponding functional modules. The significance levels in the correlation tests are denoted as: +p < 0.05;
*p < 0.01; **p < 0.001. Details of metabolite clusters were shown in Supplementary Table 3.

Ruminococcaceae spp., Citrobacter spp., Eubacterium spp. and Yersinia- A classification model based on the species discussed thus far
ceae spp. were significantly associated with neuroactive metabolites provided an area under the receiver operating characteristic curve of
(Supplementary Fig. 12). 0.81, differentiating BD patients from HCs (Supplementary Fig. 14a). In
Guided by these findings, we applied random forest models to this model, A. muciniphila, C. freundii, E. cloacae and Y. frederiksenii
estimate the correlation between each neuroactive metabolite were the major contributors (Supplementary Fig. 14b). These findings
and the abundance of species that contained the metabolic suggest that microbes involved in the production of neuroactive
pathway/synthetase-encoding genes of that particular neuroac- metabolites may be potential diagnostic biomarkers of BD.
tive metabolite. Random forest models that maximised the
power of the neuroactive metabolites concentration prediction Resting-state functional connectivity patterns in BD
in serum identified 154 microbial species (Fig. 4 and Supple- To further investigate to what extent the intestinal microbiota and
mentary Table 7). The models accounted for, on average, 22% of neuroactive metabolites might influence brain activity, we
the variance of the target metabolite concentrations in serum, collected rs-fMRI of 44 BD patients and 37 HCs (Supplementary
indicating that the corresponding species largely contributed to Table 1). Based on the CONN’s data-driven hierarchical clustering
the production of neuroactive metabolites. Y. frederiksenii, Y. algorithm [26] of ROI-to-ROI spatial proximity and functional
aleksiciae, A. muciniphila, C. freundii, C. werkmanii, E. cloacae, similarity metrics [31] (see Methods for detail), 9180 pairwise
Ruminococcaceae spp. and Enterobacter kobei were the major connections of 136 ROIs were classified into 210 clusters. We first
constituents in the random forest models (Fig. 4 and Supple- performed the PCA and found that the RRC matrix of BD patients
mentary Table 7). Coherent with changes in neuroactive markedly differed from that of HCs (Fig. 5a, PERMANOVA test, p <
metabolites in BD patients, most of the species (28.2%) were 0.05). An FDR-corrected cluster-level q-value of 0.05 was applied to
more depleted in BD (Fig. 4 and Supplementary Table 7). properly control the family-wise error rates and threshold of the
Importantly, species linked to the production of neuroactive RRC statistical map by merely including significant connectivity
metabolites correlated robustly to BD symptom severity (MADRS, clusters (Supplementary Fig. 15). Sixty-nine out of 210 clusters
HAMD, HAMA and YMRS) (Supplementary Fig. 13). Based on were significant (GLM, FDR q < 0.05, Supplementary Table 8) with
these findings, we hypothesised that the intestinal microbiota 1401 significant individual connections (post-hoc t-test, FDR q <
could affect BD pathophysiology, possibly through regulating 0.05, Supplementary Table 9), which resulted in 20 'networks'
certain neuroactive metabolites. (Fig. 5b and Supplementary Table 10). All significant clusters were

Molecular Psychiatry
 Z. Li et al.
8
Flavobacterium spp.

BD-enriched species
Staphylococcus argenteus Sutterella spp. Prevotella multisaccharivorax
Healthy control-enriched species Brevibacillus massiliensis

not significant different species

Capnocytophaga spp.
Genus Kynurenic acid
Pantothenic acid Prevotella spp.
Riboflavin
Enterobacter
Enterobacter cancerogenus
Exiguobacterium spp. Fusobacterium spp.
Yersinia Muribaculaceae spp.
Citrobacter Pseudomonas aeruginosa Campylobacter concisus
Capnocytophaga spp
Enterobacter mori
Akkermansia Serratia spp.

Enterobacter cloacae Enterobacter cloacae

Yersinia frederiksenii Fusobacterium spp.
Enterobacter kobei
Phascolarctobacterium spp． Yersinia frederiksenii Enterobacter asburiae Bacteroides togonis
Enterobacter cloacae
Ruminococcaceae spp. Bacteroidaceae spp.
Yersinia aleksiciae
Providencia rustigianii Plesiomonas shigelloides
Phascolarctobacterium spp． Ruminococcus spp. Bacteroides barnesiae
Pyridoxine Akkermansia muciniphila
Folic acid
Akkermansia spp.
Ruminococcaceae spp. Akkermansia muciniphila
Citrobacter freundii
Hungatella hathewayi Citrobacter werkmanii Akkermansia spp.
Citrobacter freundii
Citrobacter spp.
Eubacterium spp.

Enterobacteriaceae spp.
Bacteroides spp.

Alistipes spp.

2-Hydroxybutyric acid
3-Methylthiopropionic acid
gamma-Aminobutyric acid Porphyromonadaceae spp Holdemania massiliensis

Providencia rettgeri Prevotella spp.

Providencia alcalifaciens Bacteroides spp.

Lachnospiraceae spp.

Fig. 4 Alterations in the gut microbial composition in patients with BD contribute to the loss in the biosynthesis of neuroactive
metabolites. Network views of neuroactive metabolites and species. Squares represent the neuroactive metabolites and the surrounding
connected circles represent the species identified in the random forest models.

a b
PERMANOVA test p ＜ 0.001 BD
PERMANOVA R2 ： 0.04
HC
PC2 (4.26%)

Subcallosal Cortex.vs. Left Thalamus

Left Superior LOC.vs. Left Inferior LOC

Right Centrall O
Opercular Cortex vs. RightThalamus
ntr
t Opercular Cortex.vs. LeftThalamus
Right Central

ft Lateral
Right IFG, pars triangularis.vs.DMN.Left L
La Parietal
Right IFG pars triangularis.vs.DMN. Right Lateral Parietal

PC1 (19.86%)

Fig. 5 Characteristics of brain functional connectivity in BD patients. a Differences in brain functional connectivity between patients with
BD and healthy controls, was revealed by PCA. Arrows indicate the individual functional connections identified as the major contributor.
b Functional connectome ring of BD-HC contrasts. CONN’s default hierarchical clustering algorithm yielded 20 networks, displayed in sagittal
brain views. All connections within- and between-networks were analyzed using GLM to establish differences in functional connectivity
between BD and HC subjects. Significant clusters (FDR q < 0.05) with significant individual connections are shown as coloured curves (reddish
scale: BD > HC contrasts; bluish scale: BD < HC contrasts). The opacity of connection curves corresponds to respective t-statistics.

Molecular Psychiatry
 Z. Li et al.
9
a b
2
co-inertia analysis (p=0.039, RV coefficient: 0.395 n = 61) co-inertia analysis (p=0.073, RV coefficient: 0.368, n = 81)
BD
HC
Sample of Gut miicrobiome
1
5.0 Sample of functional connections

2.5

−1

0.0
−2 BD
Axis2(15.08%)

Axis2(7.84%)
HC
Sample of serum metabolome
Sample of functional connections
−3
−2.5

Axis1(23.81%) Axis1(23.19%)
−1 0 1 2 −4 −2 0

FNC BD-enriched Network-3 Network-4 FNC

Network-3
Network-2 Control-enriched Network-2 Gut microbiome
Serum metabolites
Network-4 Streptococcus mitis
Network-20 Lachnoclostridium spp. Network-5
Clostridium spp.
Lachnoclostridium spp.Akkermansia
Network-20 muciniphila
Streptococcus spp.
Network-5 Akkermansia spp. Network-6
Clostridium spp.
Clostridium spp. Clostridium spp.
2-Hydroxybutyric acid
Network-19
gamma-Aminobutyric acid Network-6 Clostridium spp. Network-7
Pantothenic acid Enterobacteriaceae spp. Clostridium spp.
Enterobacter spp.
Network-7 Clostridium spp. Clostridium spp. Network-8
Methylmalonic acid Clostridium spp. Lachnoclostridium phocaeense
Network-19 Actinomyces graevenitzii
Riboflavin Streptococcus spp. Clostridium spp.
Clostridium spp.
Network-8 Clostridium spp. Lachnoclostridium spp. Citrobacter werkmanii
Network-18 Clostridium spp. Clostridium spp. Clostridium spp.
Pyridoxine Network-9
Folic acid Clostridium spp. Actinomyces spp.Lachnoclostridium
Yersinia frederiksenii Clostridium spp. Actinomyces graevenitzii
11-Dehydrocorticosterone Enterobacter cloacae
Kynurenic acid Yersinia frederiksenii Yersiniaceae spp.
4-Pyridoxic acid Citrobacter freundii Yersinia aleksiciae Streptococcus pseudopneumoniae
Network-9 Clostridium spp. Akkermansia spp. Actinomyces oris
Clostridium spp. Enterobacter asburiae
Enterobacter cloacae
Corticosterone Enterobacter mori Lachnoclostridium spp. Clostridium spp.
Network-18 Citrobacter freundii Clostridium spp. Bacteroides barnesiae
Enterobacter kobei Clostridium spp.
Network-17
Ribose 1,5-bisphosphate Network-10
Edge count Enterobacter cancerogenus
Network-10
Edge count Clostridium spp.
5 5 Network-11
10 10 Clostridium spp.
Cortisol 25 Clostridium spp.
30 Network-16 Network-11
50
Network-12 c Network-17 Network-12
60 100 Network-16 d

Fig. 6 Gut microbiota and serum metabolome of bipolar disorder correlate with brain functional connectivity. Co-variation analysis of
serum metabolome and gut microbiota versus functional connectivity matrices. a gut microbiota versus functional connectivity matrices.
b Subject-level serum/gut data and functional connectivity matrices are shown as circles and squares, respectively; connection lines reflect
the serum/gut data and functional connectivity matrix in one individual. The correlation networks between serum metabolomics and brain
functional network connectivity clusters (FNC clusters) (c), gut microbiota and FNC clusters (d) are shown. PERMANOVA test was used to
quantify the relationship between serum metabolites/gut microbes and BD-associated functional connectivity. The connection lines represent
significant correlations (p < 0.05) between serum metabolites/gut microbes and functional connectivity networks.

annotated and summarised according to CONN’s built-in network regulation in BD [62]. However, robustly strengthened FC was
clustering of atlas ROIs with regard to anatomical proximity and also identified within the subcortical regions, particularly the
functional similarity (Supplementary Table 8). limbic system ('network-18' and 'network-19'), thalamus and
Our results revealed that compared to the HC group, the BD striatum including the caudate nucleus and putamen ('network-
group featured general reductions in functional connectivity (FC) 20'), which spoke to potentially elevated neural communications
between limbic areas, including the hippocampal formation and in the emotion and reward circuits in BD (Supplementary Fig. 16a).
amygdala ('network-18' and 'network-19', Supplementary Fig. 16a In fact, previous studies found that BD patients with hypomania
and Supplementary Table 8), and multiple cortical regions. exhibited increased connectivity between the ventral striatum and
Highlighted cortical regions were the middle temporal gyrus thalamus [63], emphasising the involvement of thalamic-striatal
(MTG, classified into the default-mode network [DMN]; 'network-9' connectivity in BD.
and 'network-10'), inferior temporal gyrus (ITG temporo-occipital Another noteworthy finding was the complex role of
part, classified into the dorsal-attention network [DAN]; 'network- interoceptive-sensorimotor networks played in the FC of BD. On
12'), and inferior frontal gyrus (IFG, classified into the language the one hand, hyperconnectivity in BD was observed encompass-
network, 'network-8'). Over half (61.3%) of the significant ing auditory ('network-1'), language (superior temporal gyrus, STG,
hypoconnectivity clusters (HC > BD contrast) revolved around ['network-3' and 'network-5'] and IFG ['network-8']) and sensor-
the limbic system ('network-18' and 'network-19'), indicating imotor (precentral/postcentral gyri, 'network-16' in Fig. 5b and
possible disturbances in cognitive function and emotional Supplementary Fig. 16b) areas. On the other hand, connections

Molecular Psychiatry
 Z. Li et al.
10
between parts of the sensory regions and subcortical areas were DISCUSSION
largely attenuated in BD. The evidence seems to suggest that BD Accumulating evidence suggests that disturbed gut microbiota
patients may experience aberrant sensory information processing may contribute to the pathophysiology of bipolar disorder, yet the
and emotional appraisals of interoceptive activities [64]. Arguably, underlying mechanism remains unresolved [70]. By comparing the
the FC patterns between the auditory areas ('network-1' and part gut microbiota, serum metabolome, and rsFC patterns between
of 'network-7') and subcortical regions were less conclusive unmedicated BD patients and HCs, we found that BD was
(Supplementary Fig. 16b). Increased FC of the auditory areas with characterised by alterations in gut microbial composition, func-
the hippocampus and amygdala and reduced FC of the auditory tional potential and metabolic pathways impinging on the MGB
areas with the thalamus were both observed, which might explain axis. The altered microbial and functional modules linked the gut
the finding of some BD patients having psychosis-like experiences microbiota with dysregulation of microbiota-derived neuroactive
[65]. Converging evidence thus far has demonstrated abnormal- metabolites (pantothenic acid, riboflavin, folic acid, pyridoxine,
ities of FC in BD and stressed the importance of cognition, kynurenic acid, GABA and SCFAs). Moreover, further analyses of
emotion, and interoceptive-sensory perception related connectiv- functional connectivity in the bipolar brain complemented our
ity networks. investigation of the MGB axis, revealing disturbances in the
hippocampus, amygdala, superior temporal gyrus and sensor-
BD-related serum metabolites and gut microbes are tightly imotor gyrus. Our multi-omics study has drawn tight lines
linked to brain functional connectivity between specific microbiota-derived neuroactive metabolites
Next, we assessed the effect size of the serum metabolomics, gut and highlighted neural networks, depicting a more nuanced
microbiota and rsFC. The effect size of the serum metabolome and picture of MGB communication and how that may affect human
rsFC (Pearson r = 0.395, p = 0.039) was greater than that of the gut cognition and behaviour in the context of BD.
microbiota and rsFC (Pearson r = 0.368, p = 0.073) (Fig. 6a, b), The crosstalk between the gut and the brain may take place
possibly reflecting that neural signals can alter the sensorimotor through multifarious pathways. For instance, the gut microbiota
and secretory functions of the gut through complex neurohu- can interact intimately with the intestinal immune system and
moral pathways, and the derivatives from gut microbes can thus affect neuroimmunity; microbial products and metabolites
regulate the brain function through visceral and endocrine can signal through enteroendocrine cells and enterochromaffin
circulation afferent signals [66]. Furthermore, we analysed the cells to modulate the secretion of neuropeptides and neuro-
association between BD-related serum metabolites/BD-related gut transmitters; microbiota-regulated hormones can directly inter-
microbes and ROI-based FC. We found that 86.96% (120/138) of act with intrinsic enteric neurons and gut innervating vagal and
BD-related serum metabolites were significantly correlated spinal afferents; micronutrients of microbial products provide
(Supplementary Table 11, PERMANOVA test, p < 0.05) with at least nutrition for the brain actively transporting across the
one individual connection. In particular, folic acid, which was blood–brain barrier [66, 71, 72]. Existing evidence has supported
documented to be related to the brain development [67] and that mitochondrial dysfunction plays a crucial role in the
regulation of mood [68] and cognition [69], was correlated with aetiology of BD [73, 74]. The citrate cycle is imperative for the
most of the significant clusters (85.51%, 59/69, Fig. 6c and synthesis of mitochondrial ATP, and it was recently reported that
Supplementary Table 11), including brain regions and networks of abnormality in the citric acid cycle of the mitochondria might
the hippocampal formation and amygdala ('network-18' and contribute to the development of BD [16].
'network-19'), thalamus and striatum ('network-20'), language Our work has contributed to the current evidence base by
areas ('network-3', 'network-4' and 'network-5') and sensorimotor elaborating BD-specific neuroactive microbes and metabolites,
areas ('network-10' and 'network-16') (Fig. 6c and Supplementary involved in the abovementioned pathways of gut-brain commu-
Fig. 17). In addition, other neuroactive metabolites, such as nication. We found that BD-associated A. muciniphila, Citrobacter
kynurenic acid, pyridoxine, GABA and riboflavin were significantly spp. (C. freundii and C. werkmanii), Phascolarctobacterium spp.,
associated with the FC of the thalamus and striatum ('network- Yersinia spp. (Y. frederiksenii and Y. aleksiciae), Enterobacter spp. (E.
20'), auditory areas ('network-7'), language areas ('network-3', cloacae and E. kobei) and Flavobacterium spp. co-varied with
'network-4' and 'network-5'), dorsal-attention network ('network- multiple B-vitamins in serum, indicated by the significant
12'), and hippocampal formation and amygdala ('network-18' and deficiencies in folic acid (B9), riboflavin (B2) and pantothenic acid
'network-19'), suggesting that the identified dysregulation of the (B5), and excess in pyridoxine (B6) among unmedicated BD
neuroactive metabolites in serum may affect specific brain patients. In addition to dietary access, human gut microbial
functions, implicated to language, emotion and reward processing communities have been reported to synthesise vitamins, which
in BD (Fig. 6c, Supplementary Fig. 17 and Supplementary are subsequently absorbed by the host in the large intestine
Table 11). [71, 75]. The B-vitamins serve as pivotal micronutrients to maintain
Likewise, 78.33% (470/600) of gut microbes were significantly brain function and mental health [71]. We observed that BD-
correlated (Supplementary Table 12, PERMANOVA test, p < 0.05) related alterations in identified B-vitamins and relevant gut
with at least one individual connection. Specifically, micro- microbiota were associated with consistently weaker FC in
organisms that were associated with neuroactive metabolites of 'network-18' (hippocampus), 'network-19' (amygdala) and 'net-
serum were also related to specific connectivity networks. For work-20' (thalamus and striatum), and stronger FC in 'network-10'
instance, Akkermansia spp. (mostly A. muciniphila), C. freundii, (inferior temporal gyrus) and 'network-16' (sensorimotor cortex)
Yersinia spp. (Y. frederiksenii and Y. aleksiciae), Phascolarctobac- though the latter lacked clear-cut evidence (Supplementary
terium spp., Flavobacterium spp. and Enterobacter spp. (E. cloacae Fig. 17). Similarly, GABA, SCFAs and kynurenic acid were also
and E. kobei) were significantly associated with the FC of the detected as key BD-related neuroactive metabolites that can be
language areas ('network-3' and 'network-4'), thalamus and produced by A. muciniphila, Citrobacter spp. (C. freundii and C.
striatum ('network-20'), sensorimotor areas ('network-10' and werkmanii), Phascolarctobacterium spp., Yersinia spp. (Y. frederiksenii
'network-16'), and hippocampal formation and amygdala ('net- and Y. aleksiciae) and Flavobacterium spp. Abnormal levels of GABA
work-18' and 'network-19') (Fig. 6d, Supplementary Fig. 17 and in arginine and proline metabolism have been reported to be
Supplementary Table 12). associated with BD [76]. Present literature indicates that neuroac-
These results implied that the gut microbiota affected BD possibly tive steroids acting at inhibitory GABA receptors might be
by affecting the metabolism of certain neuroactive metabolites, candidate modulators of BD [77]. Apart from the robustly
which might, in turn, regulate the cognitive, emotional and diminished FC of 'network-18' and 'network-19' in BD, alluding
interoceptive function of the bipolar brain. to the importance of the hippocampus and amygdala as neural

Molecular Psychiatry
 Z. Li et al.
11
nodes of BD, heightened FC in the language and auditory areas may hold the particular promise for therapeutic intervention for
(i.e. superior temporal gyrus; 'network-3' and 'network-5') was more targeted clinical management in the near future.
detected in the case of depleted GABA and SCFAs (i.e. In conclusion, our study has identified BD-associated microbes
2-hydroxybutyric acid). Increased FC harbouring the sensorimotor (A. muciniphila, Citrobacter spp. [C. freundii and C. werkmanii],
areas ('network-12' and 'network-16') was more pronounced in the Phascolarctobacterium spp., Yersinia spp. [Y. frederiksenii and Y.
kynurenic acid-centred analysis (Supplementary Fig. 17). aleksiciae], Enterobacter spp. [E. cloacae and E. kobei] and
Notably, this study has employed the latest gut genome Flavobacterium spp.), neuroactive metabolites (B-vitamins, kynure-
catalogue [36] which is used as the reference for the gut nic acid, GABA and SCFAs) and functional connectivity networks
microbiome, and our average mapped reads ratio has reached a (language processing, emotion regulation and interoception). We
high level of 91.9 ± 0.03%. Based on this catalogue, we have comprehensively demonstrated interplays between the gut
identified a few microorganisms exhibiting a strong correlation microbiota at the species level and serum metabolites in
to BD, including A. muciniphila, Citrobacter spp. (C. freundii and C. unmedicated BD patients and proposed new insights into the
werkmanii), Yersinia spp. (Y. frederiksenii and Y. aleksiciae) and MGB axis and the relationship between the gut microbiota, host
Enterobacter spp. (E. cloacae and E. kobei). Compared with the metabolism, and the dynamic bipolar brain.
previous 16 S RNA report [78–80], the current study had notably
higher sensitivity and accuracy.
It is important to note that our study has been limited by the DATA AVAILABILITY
small sample size, imbalanced case-control ratio, the mere Metagenomic and neuroimaging data have been deposited into the CNGB Sequence
inclusion of depressive bipolar patients, and the lack of additional Archive (CNSA; https://db.cngb.org/cnsa/) [85] of China National GeneBank DataBase
validation data. Arguably, the power calculation showed that the (CNGBdb) [86] with accession number CNP0002003. The datasets generated by this
study are available from the corresponding authors upon request.
current sample size and case-control ratio were acceptable to
achieve relatively satisfying statistical power (>0.8). Nevertheless,
the results we reported in this study can be generalised to bipolar
depression only, where we cannot preclude the possibility of CODE AVAILABILITY
All codes used for data analysis including serum metabolome, gut microbiome and
patients with a current manic episode exhibiting distinct rs-fMRI are available on https://github.com/lizhiming11/BD_project.
microbiome and metabolic profiles. Ideally, future studies with a
larger sample size should include BD patients in both manic and
depressive phases with a more balanced number of controls. It REFERENCES
would also be of interest to compare the multi-omics differences 1. Lynch SV, Pedersen O. The human intestinal microbiome in health and disease. N
with regard to the subtypes of BD if more type-I patients are to be Engl J Med. 2016;375:2369–79.
recruited. Another issue is the non-targeted analysis of serum 2. Michaudel C, Sokol H. The gut microbiota at the service of immunometabolism.
metabolomics, where more than 90% of the metabolites could not Cell Metab. 2020;32:514–23.
be identified. This has hampered the research for merely focusing 3. Wu J, Wang K, Wang X, Pang Y, Jiang C. The role of the gut microbiome and its
on metabolites that are known to be relevant to the neuronal metabolites in metabolic diseases. Protein Cell. 2020:12, 360–73.
processes, namely the neuroactive metabolites. It is likely that 4. Simpson CA, Diaz-Arteche C, Eliby D, Schwartz OS, Simmons JG, Cowan CS. The
gut microbiota in anxiety and depression–A systematic review. Clin Psychol Rev.
some serum metabolites that we have overlooked may be crucial
2020:83;101943.
in the physiology of BD, which warrants further investigations. 5. Morais LH, Schreiber HL, Mazmanian SK. The gut microbiota–brain axis in
Based on the current associative evidence, it is not possible to behaviour and brain disorders. Nat Rev Microbiol. 2020:19;241–55.
determine the causation of alterations in the gut microbiome or 6. Lai J, Jiang J, Zhang P, Xi C, Wu L, Gao X, et al. Gut microbial clues to bipolar
serum metabolome to the development of BD symptoms, which is disorder: state‐of‐the‐art review of current findings and future directions. Clin
a common limitation shared by studies of observational nature Transl Med. 2020;10:e146.
[81]. Although the exact mechanisms of how the gut microbiota 7. Guest PC, Guest FL, Martins-de Souza D. Making sense of blood-based pro-
affects the brain or vice versa are far from clear, our findings of teomics and metabolomics in psychiatric research. Int J Neuropsychopharmacol.
specific microbes as the sources of neuroactive metabolites of 2016;19.
8. Nikolova VL, Hall MR, Hall LJ, Cleare AJ, Stone JM, Young AH. perturbations in gut
interest have pointed to key signalling pathways and compounds
microbiota composition in psychiatric disorders: a review and meta-analysis.
involved in the interactions between the gut and the brain. Our JAMA Psychiatry. 2021;78:1343–54.
future research will address this issue by utilising animal models 9. McGuinness A, Davis J, Dawson S, Loughman A, Collier F, O’Hely M, et al. A
and intervention studies to further narrow the gaps in our systematic review of gut microbiota composition in observational studies of
understanding of the MGB axis in higher-order human cognition major depressive disorder, bipolar disorder and schizophrenia. Mol Psychiatry.
and behaviour. For example, faecal transplantation of gut 2022:1–16.
microbiota from drug-naïve or drug-free BD patients to germ- 10. Sublette ME, Cheung S, Lieberman E, Hu S, Mann JJ, Uhlemann AC, et al.
free mice may allow us to investigate whether the BD-associated Bipolar disorder and the gut microbiome: a systematic review. Bipolar Disord.
gut microbial ecosystem could induce BD-like characteristics in 2021:23;544–64.
11. Huang T-T, Lai J-B, Du Y-L, Xu Y, Ruan L-M, Hu S-H. Current understanding of gut
mice models, an experimental practice that has been reported in
microbiota in mood disorders: an update of human studies. Front Genet. 2019;10:98.
schizophrenia research [82, 83]. Similarly, a well-controlled long- 12. Rong H, Xie X-h, Zhao J, Lai W-t, Wang M-b, Xu D, et al. Similarly in depression,
itudinal intervention design makes it possible to follow and nuances of gut microbiota: evidences from a shotgun metagenomics sequencing
scrutinise the progression and changes of symptoms along with study on major depressive disorder versus bipolar disorder with current major
the timeseries-based assessment of gut microbiome and serum depressive episode patients. J Psychiatr Res. 2019;113:90–9.
metabolome in BD patients. 13. Bartoli F, Misiak B, Callovini T, Cavaleri D, Cioni RM, Crocamo C, et al. The
The last decade has seen a rapidly growing field of microbiome kynurenine pathway in bipolar disorder: a meta-analysis on the peripheral blood
research incorporating neuroimaging techniques and document- levels of tryptophan and related metabolites. Mol Psychiatry. 2021;26:3419–29.
ing how gut microbiota may influence brain structure and 14. Lai W-t, Deng W-f, Xu S-x, Zhao J, Xu D, Liu Y-h, et al. Shotgun metagenomics
reveals both taxonomic and tryptophan pathway differences of gut microbiota in
function [84]. Our new analytical protocol of studying the gut
major depressive disorder patients. Psychological Med. 2021;51:90–101.
microbial ecosystem and serum metabolome in tandem with 15. Fakhoury M. Revisiting the serotonin hypothesis: implications for major depres-
whole-brain functional networks provides insights for research sive disorders. Mol Neurobiol. 2016;53:2778–86.
exploiting the gut microbiota and neuroactive metabolites, e.g. B- 16. Yoshimi N, Futamura T, Kakumoto K, Salehi AM, Sellgren CM, Holmén-Larsson J,
vitamins, kynurenic acid, GABA and SCFAs, as informative et al. Blood metabolomics analysis identifies abnormalities in the citric acid cycle,
biomarkers of diagnostic and prognostic potentials. This approach urea cycle, and amino acid metabolism in bipolar disorder. BBA Clin. 2016;5:151–8.

Molecular Psychiatry
 Z. Li et al.
12
17. Burghardt KJ, Evans SJ, Wiese KM, Ellingrod VL. An untargeted metabolomics 48. Lovelace MD, Varney B, Sundaram G, Lennon MJ, Lim CK, Jacobs K, et al. Recent
analysis of antipsychotic use in bipolar disorder. Clin Transl Sci. 2015;8:432–40. evidence for an expanded role of the kynurenine pathway of tryptophan
18. Sussulini A, Prando A, Maretto DA, Poppi RJ, Tasic L, Banzato CEM, et al. Metabolic metabolism in neurological diseases. Neuropharmacology. 2017;112:373–88.
profiling of human blood serum from treated patients with bipolar disorder 49. Marx W, McGuinness AJ, Rocks T, Ruusunen A, Cleminson J, Walker AJ, et al. The
employing 1H NMR spectroscopy and chemometrics. Anal Chem. 2009;81:9755–63. kynurenine pathway in major depressive disorder, bipolar disorder, and schizo-
19. Yang J, Yan B, Zhao B, Fan Y, He X, Yang L, et al. Assessing the causal effects of phrenia: a meta-analysis of 101 studies. Molecular Psychiatry. 2020:26;4158–78.
human serum metabolites on 5 major psychiatric disorders. Schizophrenia Bull. 50. Cervenka I, Agudelo LZ, Ruas JL. Kynurenines: Tryptophan’s metabolites in
2020;46:804–13. exercise, inflammation, and mental health. Science. 2017;357:eaaf9794.
20. Hamilton M. A rating scale for depression. J Neurol Neurosurg Psychiatry. 51. Mora JR, Iwata M, Von, Andrian UH. Vitamin effects on the immune system:
1960;23:56. vitamins A and D take centre stage. Nat Rev Immunol. 2008;8:685–98.
21. Müller M. Differentiating moderate and severe depression using the Montgomery– 52. Gevi F, Zolla L, Gabriele S, Persico AM. Urinary metabolomics of young Italian
Åsberg depression rating scale (MADRS). J Affect Disord. 2003;77:255–60. autistic children supports abnormal tryptophan and purine metabolism. Mol
22. Hamilton M. The assessment of anxiety states by rating. Br J Med Psychol. Autism. 2016;7:1–11.
1959:32;50–5. 53. Wikoff WR, Anfora AT, Liu J, Schultz PG, Lesley SA, Peters EC, et al. Metabolomics
23. Young RC, Biggs JT, Ziegler VE, Meyer DA. A rating scale for mania: reliability, analysis reveals large effects of gut microflora on mammalian blood metabolites.
validity and sensitivity. Br J Psychiatry. 1978;133:429–35. Proc Natl Acad Sci USA. 2009;106:3698–703.
24. Feng Q, Li Y, Yang Y, Feng J. Urine metabolomics analysis in patients with nor- 54. Zhu F, Ju Y, Wang W, Wang Q, Guo R, Ma Q, et al. Metagenome-wide association
moalbuminuric diabetic kidney disease. Front Physiol. 2020;11:578799. of gut microbiome features for schizophrenia. Nat Commun. 2020;11:1–10.
25. Chong J, Soufan O, Li C, Caraus I, Li S, Bourque G, et al. MetaboAnalyst 4.0: towards 55. Liu R, Hong J, Xu X, Feng Q, Zhang D, Gu Y, et al. Gut microbiome and serum
more transparent and integrative metabolomics analysis. Nucleic Acids Res. 2018;46: metabolome alterations in obesity and after weight-loss intervention. Nat Med.
W486–94. 2017;23:859–68.
26. Whitfield-Gabrieli S, Nieto-Castanon A. Conn: a functional connectivity toolbox for 56. Qin J, Li Y, Cai Z, Li S, Zhu J, Zhang F, et al. A metagenome-wide association study
correlated and anticorrelated brain networks. Brain Connectivity. 2012;2:125–41. of gut microbiota in type 2 diabetes. Nature. 2012;490:55–60.
27. Penny WD, Friston KJ, Ashburner JT, Kiebel SJ, Nichols TE. Statistical parametric 57. Yan Q, Gu Y, Li X, Yang W, Jia L, Chen C, et al. Alterations of the gut microbiome in
mapping: the analysis of functional brain images. Burlington: Elsevier; 2011. hypertension. Front Cell Infect Microbiol. 2017;7:381.
28. Whitfield-Gabrieli S, Nieto-Castanon A, Ghosh S. Artifact detection tools (ART) 58. Evans SJ, Bassis CM, Hein R, Assari S, Flowers SA, Kelly MB, et al. The gut
2011;7:11. microbiome composition associates with bipolar disorder and illness severity. J
29. Behzadi Y, Restom K, Liau J, Liu TT. A component based noise correction method Psychiatr Res. 2017;87:23–9.
(CompCor) for BOLD and perfusion based fMRI. Neuroimage. 2007;37:90–101. 59. Osadchiy V, Martin CR, Mayer EA. The gut–brain axis and the microbiome:
30. Desikan RS, Ségonne F, Fischl B, Quinn BT, Dickerson BC, Blacker D, et al. An mechanisms and clinical implications. Clin Gastroenterol Hepatol. 2019;17:322–32.
automated labeling system for subdividing the human cerebral cortex on MRI 60. Strandwitz P, Kim KH, Terekhova D, Liu JK, Sharma A, Levering J, et al. GABA-
scans into gyral based regions of interest. Neuroimage. 2006;31:968–80. modulating bacteria of the human gut microbiota. Nat Microbiol. 2019;4:396–403.
31. Nieto-Castanon A. Handbook of functional connectivity magnetic resonance 61. Tran SM-S, Mohajeri MH. The role of gut bacterial metabolites in brain devel-
imaging methods in CONN. Boston: Hilbert Press; 2020. opment, aging and disease. Nutrients. 2021;13:732.
32. Jafri MJ, Pearlson GD, Stevens M, Calhoun VDJN. A method for functional network 62. Perry A, Roberts G, Mitchell PB, Breakspear M. Connectomics of bipolar disorder: a
connectivity among spatially independent resting-state components in schizo- critical review, and evidence for dynamic instabilities within interoceptive net-
phrenia. Neuroimage. 2008;39:1666–81. works. Mol Psychiatry. 2019;24:1296–318.
33. Yang J, Zheng P, Li Y, Wu J, Tan X, Zhou J, et al. Landscapes of bacterial and 63. Altinay MI, Hulvershorn LA, Karne H, Beall EB, Anand A. Differential resting-state
metabolic signatures and their interaction in major depressive disorders. Sci Adv. functional connectivity of striatal subregions in bipolar depression and hypo-
2020;6:eaba8555. mania. Brain Connectivity. 2016;6:255–65.
34. Li Z, Xia J, Jiang L, Tan Y, An Y, Zhu X, et al. Characterization of the human skin 64. Stoddard J, Gotts S, Brotman M, Lever S, Hsu D, Zarate C Jr, et al.
resistome and identification of two microbiota cutotypes. Microbiome. 2021;9:1–18. Aberrant intrinsic functional connectivity within and between corticostriatal
35. Li R, Yu C, Li Y, Lam T-W, Yiu S-M, Kristiansen K, et al. SOAP2: an improved and temporal–parietal networks in adults and youth with bipolar disorder.
ultrafast tool for short read alignment. Bioinformatics. 2009;25:1966–67. Psychological Med. 2016;46:1509.
36. Almeida A, Nayfach S, Boland M, Strozzi F, Beracochea M, Shi ZJ, et al. A unified 65. Skåtun KC, Kaufmann T, Brandt CL, Doan NT, Alnæs D, Tønnesen S, et al. Thalamo-
catalog of 204,938 reference genomes from the human gut microbiome. Nat cortical functional connectivity in schizophrenia and bipolar disorder. Brain
Biotechnol. 2021;39:105–14. Imaging Behav. 2018;12:640–52.
37. Kang DD, Li F, Kirton E, Thomas A, Egan R, An H, et al. MetaBAT 2: an adaptive 66. Dalile B, Van Oudenhove L, Vervliet B, Verbeke K. The role of short-chain fatty
binning algorithm for robust and efficient genome reconstruction from meta- acids in microbiota–gut–brain communication. Nat Rev Gastroenterol Hepatol.
genome assemblies. PeerJ. 2019;7:e7359. 2019;16:461–78.
38. Le Chatelier E, Nielsen T, Qin J, Prifti E, Hildebrand F, Falony G, et al. Richness of 67. Liu L, Liu Z, Li Y, Sun C. Integration of metabolomics and proteomics to highlight
human gut microbiome correlates with metabolic markers. Nature. 2013;500:541–6. altered neural development related pathways in the adult offspring after
39. Oksanen J, Blanchet FG, Kindt R, Legendre P, Minchin PR, O’hara R, et al. Package maternal folic acid supplement. Clin Nutr. 2021;40:476–87.
‘vegan’. Community Ecol Package. 2013;2:1–295. 68. Gao L, Liu X, Yu L, Wu J, Xu M, Liu Y. Folic acid exerts antidepressant effects by
40. Dray S, Dufour A-B. The ade4 package: implementing the duality diagram for upregulating brain-derived neurotrophic factor and glutamate receptor 1
ecologists. J Stat Softw. 2007;22:1–20. expression in brain. Neuroreport. 2017;28:1078–84.
41. McArdle BH, Anderson MJJE. Fitting multivariate models to community data: a 69. Enderami A, Zarghami M, Darvishi-Khezri H. The effects and potential mechan-
comment on distance‐based redundancy analysis. Ecology. 2001;82:290–297. isms of folic acid on cognitive function: a comprehensive review. Neurological
42. Feng Q, Liang S, Jia H, Stadlmayr A, Tang L, Lan Z, et al. Gut microbiome Sci. 2018;39:1667–75.
development along the colorectal adenoma–carcinoma sequence. Nat Commun. 70. Valles-Colomer M, Falony G, Darzi Y, Tigchelaar EF, Wang J, Tito RY, et al. The
2015;6:1–13. neuroactive potential of the human gut microbiota in quality of life and
43. Karlsson FH, Tremaroli V, Nookaew I, Bergström G, Behre CJ, Fagerberg B, et al. depression. Nat Microbiol. 2019;4:623–32.
Gut metagenome in European women with normal, impaired and diabetic glu- 71. Kennedy DO. B vitamins and the brain: mechanisms, dose and efficacy—a review.
cose control. Nature. 2013;498:99–103. Nutrients. 2016;8:68.
44. Benjamini Y, YJJotRsssB Hochberg. Controlling the false discovery rate: a practical 72. Agirman G, Hsiao EY. SnapShot: the microbiota-gut-brain axis. Cell. 2021;
and powerful approach to multiple testing. J R Stat Soc Ser B (Methodol). 184:2524.
1995;57:289–300. 73. Goldstein BI, Young LT. Toward clinically applicable biomarkers in bipolar disorder:
45. Wang X, Yang S, Li S, Zhao L, Hao Y, Qin J, et al. Aberrant gut microbiota alters focus on BDNF, inflammatory markers, and endothelial function. Curr Psychiatry
host metabolome and impacts renal failure in humans and rodents. Gut. Rep. 2013;15:425.
2020;69:2131–42. 74. de Sousa RT, Machado-Vieira R, Zarate CA Jr, Manji HK. Targeting mitochondrially
46. Breiman L. Random forests. Mach Learn. 2001;45:5–32. mediated plasticity to develop improved therapeutics for bipolar disorder. Expert
47. Pu J, Liu Y, Zhang H, Tian L, Gui S, Yu Y, et al. An integrated meta-analysis of Opin Therapeutic Targets. 2014;18:1131–47.
peripheral blood metabolites and biological functions in major depressive dis- 75. Das P, Babaei P, Nielsen J. Metagenomic analysis of microbe-mediated vitamin
order. Mol Psychiatry. 2020:26;4265–76. metabolism in the human gut microbiome. BMC Genomics. 2019;20:1–11.

Molecular Psychiatry
 Z. Li et al.
13
76. Brady RO Jr, McCarthy JM, Prescot AP, Jensen JE, Cooper AJ, Cohen BM, et al. analysis (Z.L.), gut microbiota analysis (Z.L.) and neuroimaging analysis (J.D.); C.L., S.L.,
Brain gamma‐aminobutyric acid (GABA) abnormalities in bipolar disorder. Bipolar J.L., D.W., H.Z., M.H., H.Z., Y.S., L.W., Y.L., Z.J., S.L., W.Z. and H.Y. facilitated scientific
Disord. 2013;15:434–9. discussion and gave suggestions; S.H., C.N., X.S. and K.K. supervised these activities; Z.L.,
77. Marx CE, Stevens RD, Shampine LJ, Uzunova V, Trost WT, Butterfield MI, et al. J.D. and J.L. wrote the paper, which was revised by S.B., K.K. and S.H.; and all authors
Neuroactive steroids are altered in schizophrenia and bipolar disorder: relevance to read and approved the manuscript.
pathophysiology and therapeutics. Neuropsychopharmacology. 2006;31:1249–63.
78. Zheng P, Yang J, Li Y, Wu J, Liang W, Yin B, et al. Gut microbial signatures can
discriminate unipolar from bipolar depression. Adv Sci. 2020;7:1902862. FUNDING
79. Rhee SJ, Kim H, Lee Y, Lee HJ, Park CHK, Yang J, et al. Comparison of serum This study was supported by grants from the National Natural Science Foundation of
microbiome composition in bipolar and major depressive disorders. J Psychiatr China (Grant numbers: 81971271 and 81971253), the Zhejiang Provincial Key
Res. 2020;123:31–8. Research and Development Programme (Grant number: 2021C03107), the Leading
80. McIntyre RS, Subramaniapillai M, Shekotikhina M, Carmona NE, Lee Y, Mansur RB, Talent of Scientific and Technological Innovation - 'Ten Thousand Talents
et al. Characterizing the gut microbiota in adults with bipolar disorder: a pilot Programme' of Zhejiang Province (Grant number: 2021R52016), the Zhejiang
study. Nutr Neurosci. 2021:24;173–80. Provincial Natural Science Foundation (Grant number: LQ20H090013), the Pro-
81. Wang DD, Nguyen LH, Li Y, Yan Y, Ma W, Rinott E, et al. The gut microbiome gramme from the Health and Family Planning Commission of Zhejiang Province
modulates the protective association between a Mediterranean diet and cardi- (Grant number: 2020KY548) and Zhong yuan Technological Innovation leading
ometabolic disease risk. Nat Med. 2021;27:333–43. Talents (Grant number: 204200510019).
82. Zheng P, Zeng B, Liu M, Chen J, Pan J, Han Y, et al. The gut microbiome from
patients with schizophrenia modulates the glutamate-glutamine-GABA cycle and
schizophrenia-relevant behaviors in mice. Sci Adv. 2019;5:eaau8317.
83. Zhu F, Guo R, Wang W, Ju Y, Wang Q, Ma Q, et al. Transplantation of microbiota CONFLICT OF INTEREST
from drug-free patients with schizophrenia causes schizophrenia-like abnormal The authors declare no competing interests.
behaviors and dysregulated kynurenine metabolism in mice. Mol Psychiatry.
2020;25:2905–18.
84. Cryan JF, O’Riordan KJ, Cowan CS, Sandhu KV, Bastiaanssen TF, Boehme M, et al. ADDITIONAL INFORMATION
The microbiota-gut-brain axis. Physiol Rev. 2019:99;1877–2013. Supplementary information The online version contains supplementary material
85. Guo X, Chen F, Gao F, Li L, Liu K, You L, et al. CNSA: a data repository for archiving available at https://doi.org/10.1038/s41380-022-01569-9.
omics data. Database. 2020;2020:baaa055.
86. Chen FZ, You LJ, Yang F, Wang LN, Guo XQ, Gao F, et al. CNGBdb: china national Correspondence and requests for materials should be addressed to Susanne Brix,
genebank database. Yi Chuan. 2020;42:799–809. Karsten Kristiansen, Xueqin Song, Chao Nie or Shaohua Hu.

Reprints and permission information is available at http://www.nature.com/

AUTHOR CONTRIBUTIONS reprints
S.H., X.S. and J.L. designed the study; Participants were recruited by J.L., P.Z., J.J., H.H.,
C.X., L.W., X.G., Y.F., D.Z., Y.C., Y.Z., X.Y., X.L. and L.P.; J.L. and P.Z. coordinated the data Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims
processing and availability; Data analysis was performed as follows: serum metabolome in published maps and institutional affiliations.

Molecular Psychiatry

View publication stats