A systematic review of gut microbiota composition in observational studies of major depressive disorder, bipolar disorder and schizophrenia

This systematic review adheres to the relevant criteria of the Preferred Reporting Items for Systematic reviews and Meta-analyses (PRISMA) statement [64] and was registered on PROSPERO (#CRD42020189823).

The research question and inclusion and exclusion criteria were determined a priori and developed using a PICOS structure (Patient, Intervention/Exposure, Comparators, Outcome, Study Design). Only peer-reviewed, full-text studies published in English were included. Inclusion criteria were: (1) observational study designs including cross-sectional, case-control, and prospective and retrospective cohort studies or intervention studies with baseline data; (2) adults aged 18 years or older; (3) participants with a clinical diagnosis of MDD, BD or SZ as the outcome variable compared to non-psychiatric controls; and (4) reporting of gut microbiota composition data as an exposure, for example measures of specific bacterial taxa, diversity and ordination techniques. We grouped studies by disorder when comparing cases to controls, and then we compared gut microbiota results across the pre-specified mental disorders.

We conducted a systematic search using PubMed, EMBASE, CINAHL, CENTRAL and PsycINFO for articles published from database inception through to 3rd December 2021 using the search strategy (microbiome OR microbiota) AND (depression OR depressive OR schizophrenia OR psychosis OR bipolar OR mania OR manic OR “severe mental illness”).

Primary screening was conducted independently by two reviewers (AJM, JD) using the web application Rayyan [65]. Full-text secondary screening was also performed in duplicate (AJM, JD) to assess eligibility and exclude studies that did not meet inclusion criteria. Reference lists of relevant publications were examined for studies not identified in the database search. Two reviewers independently extracted the data from eligible studies (AJM, JD). Where there were conflicts, consensus was achieved through discussion.

Extracted aggregated data included: publication information; study design; participant demographics and characteristics; covariates and potential confounding variables; psychiatric disorder information and severity; blood biomarkers; gut microbiota data collection, sequencing and analysis methods; and gut microbiota outcome data (e.g. diversity, specific taxa, ordination).

We were interested in associations between gut microbiota and psychiatric outcomes rather than predictive performance. Hence, where studies used a discovery dataset to determine gut microbiota composition in cases then tested its diagnostic performance for predicting the mental disorder on a validation dataset, we extracted the results from the discovery set only.

We included studies that used 16S rRNA gene sequencing (hereafter 16S) and whole-genome shotgun metagenomic sequencing (hereafter metagenomics), which aim to profile the entire faecal microbiota rather than focusing on specific taxa of interest (e.g. polymerase chain reaction, culture-based methods). 16S profiles faecal bacteria based on variations in their 16S rRNA gene, which is an important housekeeping gene present in all bacteria. Metagenomics sequences all the genetic material in the faecal sample and generates greater sequencing depth and resolution. Where studies reported differing results for both 16S and metagenomics, both sets of results were included.

Alpha-diversity (α-diversity) is a measure of gut microbiota diversity within a single sample, and its metrics are single numbers that describe the gut microbiota environment based on the number (i.e. richness) and/or distribution (i.e. evenness) of bacterial species present within an individual sample [66]. Studies in this review investigated whether people with a mental disorder had a higher or lower number or distribution of bacteria (α-diversity) compared to controls. Beta-diversity (β-diversity) derives from pairwise measures of similarity or dissimilarity in gut microbiota communities between groups. Ordination plots are used to display β-diversity statistics for visual inspection of data and allow researchers to observe whether samples from different groups cluster together or separately, suggesting compositional divergence of the gut microbiota between groups [67]. Studies included here investigated whether people with or without a mental disorder had a different gut microbiota composition from each other, as indicated by a between-group difference in β-diversity, and/or observed differences in visual clustering.

Bacteria are categorised into taxa based on their traits, such as their phylogeny, common metabolic potential, preferred growth environments, morphology and their genetic sequence. The broadest level of classification is called a ‘phylum’. Bacteria are progressively classified together into ranked groups called class, order, family, genus and species, in order of increasing similarity. Studies in this review identified taxa at various ranks that were different in their abundances in the gut microbiota of those with a mental disorder compared to controls. Some studies also identified bacteria that were best able to discriminate between those with and without mental disorders. For this review, we included both differentially abundant and discriminatory taxa in the taxonomy synthesis.

An aim of this review was to identify consistencies in gut microbiota across MDD, BD and SZ. Therefore, we identified and reported taxa that were reported as differentially abundant or discriminatory in at least 20% of the studies that reported data at that taxonomic rank. This allowed us to report taxa present in at least seven studies at phylum level, two studies at class level, three studies at the order level, seven studies at family level and nine studies at genus level. Due to the low resolution at the species level afforded by 16S, only studies using metagenomics were included in the species-level synthesis. Due to the heterogeneity and incomparability of processing and sequencing methodology, a meta-analysis was not conducted.

We used the National Institutes of Health (NIH) National Health, Lung and Blood Institute Study Quality Assessment Tool for Observational Cohort and Cross-sectional Studies [68] to assess the internal validity and potential bias of included studies. Study quality was rated as ‘Good’, ‘Fair’ or ‘Poor’ by two reviewers (AJM, CS), with discrepancies addressed through discussion. We considered body mass index, diet, anti-depressant use, or similar psychotropic medication use as the key confounding variables used in the quality assessment.

Our database search yielded a total of 3516 potentially relevant studies, with 2591 studies remaining after duplicates were removed. After primary screening, 106 studies were selected for full-text review (Fig. 1). Of the 44 studies that met criteria for inclusion in this systematic review (including a total of 2510 psychiatric cases and 2407 controls), 24 investigated gut microbiota composition in MDD (comprising data on 1038 MDD cases and 1048 non-MDD controls [69–92]), seven investigated BD (comprising data on 527 BD cases and 477 non-BD controls [81, 83, 93–97]) and 15 investigated SZ (comprising data on 945 SZ cases and 882 non-SZ controls [98–112]). Full summaries of study characteristics, results, covariates and methodologies are provided in Supplementary Tables S1–S8. Some studies conducted analyses across subgroups; this resulted in a total of 56 comparison groups across studies that, for simplicity, will be referred to as individual studies when reporting the results.

For α-diversity metrics and differential abundances of taxa, most studies defined a p value of <0.05 for significance in the methods and thereafter stated that results were ‘significantly different’ without a specific p value. For β-diversity metrics, differences were most commonly determined using permutational multivariate analysis of variance, with a p value < 0.05 deemed as significant.

A total of 173 α-diversity analyses were conducted across 43 studies (Figs. 2A and S1). Reported metrics included measures of richness, phylogenetic diversity, evenness and composite metrics of evenness and richness. The most commonly reported α-diversity metric was the Shannon Index (27% of total analyses; N = 46/173).

The majority of analyses reported no difference in α-diversity between the psychiatric cases and controls (77% of total analyses; N = 134/173). Of the studies that did report a difference (N = 23%; 39/173 analyses), most reported lower α-diversity in cases (17% of total analyses; N = 29/173), and few reported higher α-diversity (6% of total analyses; N = 10/173).

In a breakdown of the separate mental disorders, there was no evidence of a difference in α-diversity for the majority of the MDD studies (79%; N = 68/86 analyses). However, a small number reported lower (13%; N = 11/86 analyses) or higher (8%; N = 7/86 analyses) α-diversity in MDD cases compared to controls. A similar pattern was observed for analyses in SZ, with no difference in most SZ studies (81%; N = 51/63 analyses); however, there was evidence of lower (16%; N = 10/63 analyses) or higher (3%; N = 2/63 analyses) α-diversity in a small number of studies. Whilst the results for BD were similar (63% showed no difference; N = 15/24 analyses), a greater proportion of studies reported lower alpha diversity compared to the other disorders (33%; N = 8/24 analyses) and only one study reported higher α-diversity (4%) (Fig. 2A).

A total of 62 β-diversity analyses were reported across 37 studies (Figs. 2B and S2), most commonly using the unweighted UniFrac distance (34%; N = 21/62 analyses). However, nearly a third of these analyses (29%; N = 18/62 analyses) presented only the visual ordination plots and did not report any statistical testing.

Of the β-diversity analyses that were tested statistically (71% of total analyses; N = 44/62), most reported dissimilarity in the gut microbiota composition of cases compared to controls (80% of statistically tested analyses; N = 35/44 analyses). Differences in gut microbiota composition were reported in 87% of MDD β-diversity analyses (N = 13/15), 70% of BD analyses (N = 7/10) and 79% of SZ analyses (N = 15/19) (Fig. 2B). Of the studies that visually observed their data using ordination plots (52% of total studies; N = 32/62), most reported clear clustering of participants with a mental disorder separately to controls (59% of total ordination plots; N = 19/32).

Bacteria at various taxonomic ranks were reported. At the highest level, the phylum most commonly reported to be different between cases and controls was Bacteroidetes (N = 18/32 studies), although it was equally reported to be both higher (N = 10 studies) and lower (N = 12 studies) in abundance in cases (Fig. S3). Of the studies that reported their differential abundance, the phyla most frequently reported to have a higher abundance in psychiatric cases were Actinobacteria (N = 10/13 reported studies), Fusobacteria (N = 5/6 reported studies) and Proteobacteria (N = 10/12 reported studies), whereas Firmicutes was more frequently reported as lower (N = 12/19 reported studies).

Taxa at the class (N = 9 total studies) and order (N = 11 total studies) levels were infrequently reported across studies and few consistent patterns were observed. At the class level, studies reported higher abundances of Coriobacteriia (N = 3/9 reported studies) and Deltaproteobacteria (N = 3/9 reported studies), and lower abundances of Bacteroidia (N = 3/9 reported studies). At the order level, Actinomycetales was frequently reported as higher (N = 5/11 reported studies) and Bacteroidales lower (N = 4/11 reported studies) in cases compared to controls (Fig. S3).

At the family level, we identified ten families that were reported as being different between cases and controls (Fig. S3). The most commonly reported families were Lachnospiraceae (N = 21/36 reported studies) and Ruminococcaceae (N = 17/36 reported studies). Of the studies that reported their differential abundance, results for Lachnospiraceae were directionally heterogeneous, whereas Ruminococcaceae was predominantly lower in psychiatric cases (N = 13/17 reported studies). The bacterial family Enterobacteriaceae was consistently reported to be in higher abundance in mental disorders (N = 7/7 reported studies). No bacterial families were consistently reported as lower in cases.

Forty-eight studies identified differences in the abundances of taxa at the genus level, and we identified 21 bacterial genera that were consistently different in those with and without a mental disorder (Fig. S4). The most commonly reported genera observed to be differentially abundant in cases were Bacteroides (N = 20/48 studies), Faecalibacterium (N = 20/48 studies), Prevotella (N = 17/48 studies) and Blautia (N = 15/48 studies). Of the studies that reported on the following taxa, every study reported a higher abundance of Eggerthella (N = 12/12 reported studies), Flavonifractor (N = 9/9 reported studies) and Veillonella (N = 8/8 reported studies) in psychiatric cases compared to controls. No genus was always lower in cases; however, the majority of the studies that reported on Faecalibacterium (N = 17/20 reported studies), Coprococcus (N = 12/13 reported studies), Haemophilus (N = 8/9 reported studies) and Ruminococcus (N = 11/14 reported studies) observed lower abundances in psychiatric cases.

Differences in the abundances of the above genera were commonly reported across all three mental disorders, however some differences were more pronounced in individual disorders: higher Alistipes and Parabacteroides and lower Prevotella were observed in MDD; higher Bifidobacterium and Oscillibacter were observed in BD; and higher Prevotella, and lower Bacteroides, Haemophilus, and Streptococcus, were observed in SZ (Fig. S4).

Differences in the abundances of some genera were shared between disorders: higher Escherichia/Shigella and Veillonella were common to both MDD and SZ; higher Megasphaera and lower Roseburia were common to both SZ and BD; and higher Enterococcus, Flavonifractor and Streptococcus and lower Faecalibacterium and Ruminococcus were commonly observed in both BD and MDD. Finally, Eggerthella and Lactobacillus were frequently higher, and Coprococcus frequently lower, in all three mental disorders compared to controls (Fig. S4).

Twelve metagenomic studies reported differences at the species-level and we identified 18 species as consistently different between cases and controls (Fig. S5). Almost all reported species were higher in abundance in mental disorders (N = 16/18 reported species). However, Haemophilus parainfluenzae was more commonly reported as lower in mental disorders (N = 3/4 reported studies) compared with controls, and Bacteroides helcogenes was only reported as lower in mental disorders (N = 3/3 reported studies).

Functional potential was measured in twelve studies of MDD using either the PICRUSt software [113] for 16S data (N = 8) or KEGG database mapping [114] for metagenomic data (N = 4). Very few pathways were reported in more than one study. One study only investigated pathways associated with tryptophan metabolism and biosynthesis, identifying differences between cases and controls [75]. Another study observed correlations between specific bacterial taxa and functional pathways within the KEGG orthologue database; however, they did not report what the specific pathways were, only providing the KEGG identification number [88]. Of the studies that examined multiple pathways and reported differences (N = 9), MDD was associated with an enrichment of: glycan biosynthesis and metabolism (33%; N = 3/9 reported studies) [86, 89]; lipopolysaccharide biosynthesis and biosynthesis proteins (33%; N = 3/9 reported studies) [72, 85]; and transport and catabolism pathways (33%; N = 3/9 reported studies) [86, 89], compared to controls. Conversely, cell motility and secretion (33%; N = 3/9 reported studies) [71, 89] and membrane transport (33%; N = 3/9 reported studies) [86, 89] were pathways enriched in controls compared to cases. One study did not observe any differences between MDD cases and controls across multiple investigated pathways [81].

Three studies of BD reported functional potential using either PICRUSt (N = 1) or KEGG (N = 2). The first study only investigated pathways associated with tryptophan metabolism and biosynthesis, finding differences between cases and controls [97]. The second study identified 31 pathways that were different between cases and controls; pathways associated with BD included tetrahydrofolate biosynthesis, pentose phosphate pathway and ornithine biosynthesis, whereas pathways associated with controls included peptides and nickel transport systems, branched chain amino acid transport system and putative sugar transport system [81]. The third study did not identify any pathways associated with either cases or controls across multiple investigated pathways [81]. Therefore, insufficient data were available to detect common functional pathways differentially reported across BD studies.

Six studies of SZ reported functional potential using PICRUSt (N = 5) or KEGG (N = 1). One of these studies reported differences across functional pathways relating to trimethylamine-N-oxide reductase, Kdo₂-lipid A biosynthesis, and glycerol degradation to 1,2-propanediol [107]. Of the studies that reported on specific functional pathways (N = 4), those commonly enriched in SZ included: alpha-linolenic acid metabolism (50%; N = 2/4 reported studies) [101, 102]; ascorbate and aldarate metabolism (50%; N = 2/4 reported studies) [98, 102]; geraniol degradation (50%; N = 2/4 reported studies) [101, 102]; nucleotide metabolism (50%; N = 2/4 reported studies) [98, 102]; and pertussis (50%; N = 2/4 reported studies) [101, 102] pathways. On the other hand, pathways commonly enriched in controls included: phenylpropanoid biosynthesis (50%; N = 2/4 reported studies) [98, 102]; RNA transport (50%; N = 2/4 reported studies) [98, 102]; and starch and sucrose metabolism (50%; N = 2/4 reported studies) [98, 102] pathways. In addition, one study of SZ investigated gut brain modules (GBMs) [106]; GBMs are a framework developed to investigate microbial pathways believed to have neuroactive metabolic potential [115]. This study identified pathways associated with short-chain fatty acid (SCFA) synthesis (acetate, propionate, butyrate and isovaleric acid), tryptophan metabolism, and synthesis of neurotransmitters such as glutamate, GABA, and nitric oxide to be enriched in SZ cases compared to controls. One study did not report any differences in functional pathways between either first-episode or treated SZ cases compared to controls [99].

Of the studies that investigated associations between psychiatric symptom severity and differentially abundant taxa in MDD (39%; N = 11/28 MDD studies), very few taxa were commonly associated with depressive symptoms across multiple studies. Taxa belonging to Blautia (27%; N = 3/11 reported studies), Parabacteroides (18%; N = 2/11 reported studies) and Ruminococcus (18%; N = 2/11 reported studies) were reported as positively associated with depressive symptoms, whereas Faecalibacterium (36%; N = 4/11 reported studies), Roseburia (18%; N = 2/11 reported studies), and Veillonella (18%; N = 2/11 reported studies) were inversely associated with depressive symptoms (Table S4). Only two studies of MDD investigated associations between psychiatric symptom severity and diversity measures; one study observed no associations between depressive symptoms and measures of α- or β-diversity [79], whereas the other study reported contradictory associations between different measures of α-diversity and anxiety symptoms [81].

Few studies investigated associations between gut microbiota composition and psychiatric symptom severity in BD (38%; N = 3/8 BD studies). There were no commonalities in the results observed across BD studies, however one study reported the relative abundance of Faecalibacterium (33%; N = 1/3 reported studies) as inversely associated with depressive symptoms in BD patients, which was also observed in our synthesis of MDD. Ten studies examined associations between gut microbiota composition and symptom severity in SZ (45%; N = 10/22). Bacteria belonging to Firmicutes (20%; N = 2/10 reported studies), Haemophilus (20%; N = 2/10 reported studies), and Lachnoclostridium (20%; N = 2/10 reported studies) were positively correlated with SZ symptom severity, whereas Bifidobacterium (20%; N = 2/10 reported studies), Coprococcus (20%; N = 2/10 reported studies) and Ruminococcaceae (20%; N = 2/10 reported studies) were inversely associated with symptom severity (Table S4).

Most studies were rated as ‘Fair’ (39%; N = 17/44 studies) or ‘Poor’ (43%; N = 19/44 studies) quality. The major sources of potential bias were lack of adequately specified and defined study populations, poor reporting of recruitment details and inclusion and exclusion criteria consistency, and lack of consideration of the pre-specified key potential confounding variables (Table S9). Only eight studies (18%) were categorised as ‘Good’ [73, 77, 86, 89, 90, 95, 106, 111], and only two studies (5%) clearly considered all of the key potential confounding variables [86, 106].

Our syntheses provided no evidence for discrimination between mental disorder cases and healthy controls based on the richness and evenness (i.e. α-diversity) of their gut microbiota. Previously, higher α-diversity has been considered a marker of ‘better’ gut health. The assumption is that greater species number and diversity may increase gut ecosystem resilience and stability due to increased functional redundancy for metabolic functions, and a more robust resistance to pathogenic invasion [116, 117]. However, growing evidence from gut microbiota research in humans suggests that α-diversity metrics are of limited utility as a measure of gut health or to discriminate between disease cases and controls. This has been demonstrated in similar reviews of the gut microbiota’s associations with obesity [118], type 2 diabetes mellitus [119], IBS [120] and ulcerative colitis [121]. These reviews align with our syntheses, and are also concordant with recent findings in other neurological and psychiatric conditions such as Parkinson’s disease [122], multiple sclerosis [123], autism spectrum disorder [124] and anxiety, which also report equivocal α-diversity findings [56].

Our findings are further corroborated by a recent systematic literature review and meta-analysis of gut microbiota composition across psychiatric disorders [59]. Concordant with our observations, the authors reported no differences in α-diversity composite indices (i.e. Shannon Index, Simpson Index) or phylogenetic diversity across mental disorders. With regards to richness, meta-analyses identified a pooled reduction in richness; however, when looking at individual diagnoses this reduction was only observed in BD, which we also observed. To date, very few studies have investigated gut microbiota composition in BD compared to the other psychiatric disorders included in this review, and no studies have investigated state-associated compositions across different phases of this disorder. Additional research into the gut microbiota in BD is required in order to confirm and further elucidate the relevance of reduced richness in this mental disorder.

The assumption that lower α-diversity is a biomarker of a poorer gut ecosystem is problematic. For example, two individuals may have equivocal α-diversity of their gut microbiota, but one individual’s ecosystem may be abundant in taxa with potential pro-inflammatory metabolism and pathogenicity, whereas the other may be abundant in microbes associated with the production of beneficial metabolites. Despite similar α-diversity, the implications of these two gut microbiota compositions for host health may be completely different. The basic metrics of α-diversity are derived from ecology frameworks; therefore, their applicability to complex and dynamic human physiological systems may be inadequate. In this sense, α-diversity metrics may have greater utility in studies that employ a common intervention across participants, such as antibiotic or probiotic treatment strategies, where change in α-diversity may be a useful marker.

Seven of the included studies reported higher α-diversity associated with mental disorders. Of these studies, all but one was very recently published (2020/21) and used either the SILVA or RDP Classifier for taxonomic assignment for 16S sequencing. These two databases are iteratively updated to reflect advances in taxonomic discoveries. This is in comparison to the Greengenes database, which has not been updated since 2013. Despite this, Greengenes has been widely used for gut microbiota research for almost a decade, including for nine of the studies included in this review. It is possible that earlier studies, using Greengenes or less complete databases, may have been missing or misidentified many bacterial taxa, impacting on the accuracy of α-diversity measures as a reflection of gut microbiota richness and evenness. Moreover, one third of the studies that reported higher α-diversity used metagenomic sequencing. A recent study comparing taxonomic characterisation using 16S and metagenomics revealed that 16S was only able to detect a part of the gut microbiota community that was identified using metagenomics, and that metagenomics was superior for detecting low abundance, yet biologically relevant, taxa [125]. Estimates of α-diversity are also inherently biased by sampling depth, which is the number of sequences yielded per sample [126]. As sampling depth is rarely adjusted for in analyses, comparisons of α-diversity between studies are difficult to interpret [126]. In order to elucidate the variations in α-diversity that are possibly being obscured by the use of previously outdated databases and limitations of 16S, such as poorer sequencing depth and resolution, these results suggest that the use of α-diversity metrics should ensure they use the most recently updates databases or, ideally, use metagenomic sequencing, and that caution should be applied in using α-diversity as a proxy of host health.

On the other hand, β-diversity analyses consistently suggested that individuals within each mental disorder category harboured a more similar gut microbiota composition to each other than to the clustered compositions observed in ‘healthy’ controls. Our findings are somewhat discordant with those of several previous systematic reviews [53, 54, 56, 59, 61], which report less consistent differences in β-diversity; however, this may be due to the greater number of studies included in our review. Whilst β-diversity analyses can indicate whether differences exist between groups, they do not indicate what the differences are, or at which level of taxonomy they are occurring. Therefore, the causes of these divergences in gut microbiota composition, the clinical or functional meaning of these differences, or how these differences might reflect disruptions in physiological processes, all remain unclear. Whilst metagenomics can suggest functional potential of bacteria, further research combining multi-omics approaches, including meta-transcriptomics (to identify which genes are being expressed), proteomics (to identify which proteins are being formed), and metabolomics (to identify which metabolites are being produced), are likely to provide greater insight into functional differences and assist in understanding how these compositional differences are associated with pathological processes potentially influencing mental health.

Our review identified some bacterial genera that were commonly differentially abundant in cases compared to controls across all three mental disorders (Fig. 3); these were higher Eggerthella and Lactobacillus, and lower Coprococcus. In addition, some bacterial genera were overlapping between two disorders. Both MDD and SZ were also associated with higher Escherichia/Shigella and Veillonella, and SZ and BD shared similarities of higher Megasphaera and lower Roseburia. The two mood disorders, MDD and BD, had more commonalities, including higher Enterococcus, Flavonifractor, and Streptococcus, and lower Faecalibacterium and Ruminococcus. The potential functional relevance of these differentially abundant taxa is discussed below.

Whilst there were common and overlapping taxa differentially abundant between cases and controls across all three mental disorders, there were some taxa for which differential abundances were specific to each mental disorder (Fig. 3). MDD was often characterised by higher Alistipes and Parabacteroides and lower Prevotella; BD was often characterised by higher Bifidobacterium and Oscillibacter; and SZ was often characterised by higher Prevotella and lower Bacteroides, Haemophilus, and Streptococcus. Again, the potential relevance of these differentially abundant taxa is discussed below.

Our current review has several methodological characteristics that need to be considered when interpreting these results. Firstly, included studies were comprised of cross-sectional data, which cannot infer causality nor account for temporal variations in the gut microbiome. Secondly, geographical distribution of studies shows a clear overrepresentation of studies conducted in China. Considering different geographic regions are associated with different microbial compositions [164], this imbalance in sampling region may have influenced the results of our synthesis. Thirdly, this systematic review focused only on characterising the bacterial members of the gut microbiome, due to their documented association with host mood and behaviour. As the GI tract harbours a myriad of other microorganisms including archaea, viruses, bacteriophages, and fungi, their potential influence on host mental health, which is newly being investigated [104, 165], should not be overlooked. Fourth, the focus on compositional data within the field of gut microbiome research (primarily using 16S) is a significant limitation due to acknowledged factors such as limited resolution and lower sensitivity [166, 167]. Despite these limitations, these data provide an important foundation for our understanding of the gut microbiome in psychiatry. As the field continues to grow and develop, a greater number of studies using other omics techniques (such as metagenomics, metabolomics, and meta-transcriptomics) will be produced, and thus our ability to explore the gut microbiome beyond its composition will be possible. This may be of particular benefit to the field of psychiatry, which currently lacks biomarkers for diagnosis and prognosis, as well as a clear understanding of disorder aetiology. Fifth, our syntheses may be biased by unmeasured confounding. The collection of covariate data was heterogenous and inconsistent across studies, and very few studies adjusted for potential confounding in their analyses. Finally, this review identified many methodological differences between studies. A comprehensive discussion of these differences is outside the scope of this review, however we have provided a summary of the key differences in the supplementary material.