Gut microbiota variations in patients diagnosed with major depressive disorder—A systematic review

Abstract Objective The etiology of major depressive disorder (MDD) is multi‐factorial and has been associated with a perturbed gut microbiota. Thus, it is therefore of great importance to determine any variations in gut microbiota in patients with MDD. Methods A systematic literature search was conducted including original research articles based on gut microbiota studies performed in patients with MDD. Demographic and clinical characteristics, applied methodology and observed gut microbiota composition were compared between included studies. Results Seventeen studies were included with a total of 738 patients with MDD and 782 healthy controls using different DNA purification methods, sequencing platforms and data analysis models. Four studies found a reduced α‐diversity in patients with MDD, while gut microbiota compositions clustered separately according to β‐diversity between patients and controls in twelve studies. Additionally, there was an increase in relative abundance of Eggerthella, Atopobium, and Bifidobacterium and a decreased relative abundance of Faecalibacterium in patients with MDD compared with healthy controls. Conclusion Gut microbiota differs significantly when comparing patients with MDD and healthy controls, though inconsistently across studies. The heterogeneity in gut microbiota compositions between the studies may be explained by variations in study design.

. Gut microbiota has suggested to play a role in the bidirectional communication between the gastrointestinal system and the brain, also known as the gut-brain axis Cryan & Dinan, 2012). The study by Sudo et al. indi-cated an association between gastrointestinal bacteria and altered behavior (Sudo et al., 2004). In this study, colonization of germ-free mice with Escherichia coli resulted in a significantly altered behavior when exposed to stress compared to specific pathogen-free mice. The altered behavior in the mice was associated with a proinflammatory profile and inoculation with a strain of Bifidobacterium attenuated both behavioral and immunological changes. This experiment displayed how a negative behavioral effect can be produced by a specific bacterial pathogen, but also rescued by introduction of a beneficial species.
Other studies have shown that gut microbes interact with the central nervous system by signaling through bacterial components recognized by the immune system (Schroeder & Backhed, 2016), or by various derived metabolites (Fischbach & Segre, 2016). A study in rats showed that intravenous administration of lipopolysaccharide (LPS), a cell wall component of gram-negative bacteria, induced a depression-like phenotype as well as increased levels of white blood cells and pro-inflammatory cytokines (Wrotek et al., 2016).
Not all metabolites produced by the gut microbiota have detrimental effects on emotional and cognitive functions. Shortchain fatty acids (SCFAs), produced by beneficial species such as Faecalibacterium from indigestible fibers, can induce vagus nerve stimulation in the colon (Cawthon & Serre, 2018;Chun et al., 2016;Schroeder & Backhed, 2016), microglia maturation and activation (Sharon et al., 2016;Yang & Chiu, 2017), as well as produce brainderived neurotrophic factor (BDNF) (Sandberg et al., 2018), a neural growth hormone. In clinical trials, healthy volunteers consuming probiotic supplements, of which several species produce SCFAs, experienced improved cognition (Marotta et al., 2019;Tillisch et al., 2013) or mood (Benton et al., 2007;Marotta et al., 2019). Combined, these experiments suggest an association between gastrointestinal bacterial species and behavioral alterations, which may become potential therapeutic targets for MDD treatment.
Recently, several clinical studies have explored the association between a specific composition of gut microbiota and depressive features (Barandouzi et al., 2020;Cheung et al., 2019;Sanada et al., 2020). In the assessment of gut microbiota, however, the study design itself may affect the overall gut microbiota composition. Diet has been found to have a major impact on bacterial species in the gut microbiota (Conlon & Bird, 2014). In the experimental processing, choice of DNA purification methods (Costea et al., 2017) and primers targeting the 16S rRNA gene for sequencing (Albertsen et al., 2015;Hamady & Knight, 2009;Lozupone et al., 2013) can considerably affect the observed composition of the gut microbiota. In determining if the gut microbiota is altered in patients with MDD, it is imperative to assess the methods used to characterize the gut microbiota.
The aim of this review was to evaluate and compare studies of gut microbiota composition in patients with MDD compared with healthy controls.

| ME THODS
A protocol for this systematic review was uploaded and accepted into the (PROSPERO) server under the ID number CRD42018104925.

| Information sources
The databases PubMed, Embase (Ovid), and PsycINFO (Ovid) were searched for articles published up until November 13th, 2020.

The literature search was conducted according to the Preferred
Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) guidelines (Moher et al., 2009). The search strategies are described Supporting information 1. All fields (including title and abstract) were explored to ensure completeness of the literature search.

| Inclusion and exclusion criteria
Articles were included if they met the following inclusion criteria: • Clinical studies performed on patients diagnosed with MDD according to criteria in the Diagnostic and Statistical Manual of Mental Disorders (DSM) or the International Classification of Diseases (ICD).
• Assessment of the gut microbiota composition through genomic analysis, including both targeted and nontargeted approaches.
• Inclusion of a control group of nondepressed individuals.
Studies were excluded if they met the following exclusion criteria: • The focus of the study was inclusion of patients with known comorbidities, such as assessing the gut microbiota in patients with both MDD and inflammatory bowel disorders.
• Assessment of the effect of pro-, pre-, syn-or antibiotics on a group of patients with MDD with no baseline measurement of the original gut microbiota.

| Study selection
Studies identified in the systematic literature search were imported into the EndNote software (Clarivate Analytics) for removal of duplicates. Files generated from these databases were imported into the Systematic Review Facility (SyRF) app (http://app.syrf.org.uk). SyRF was then used for screening of papers and data were extracted manually. Authors JKK and CBN independently reviewed and selected studies based on title and abstract presented by the SyRF app, and later manually reviewing the full-text. Both reviewers consistently agreed upon which studies to include and there was thus no need to include a third reviewer. JKK extracted outcome measures as described below.

| Outcome measures
Demographic and clinical data were extracted from the patient and control groups. Sample processing and analyses were focused on the methods applied to analyze the bacterial community: fecal storage conditions, DNA extraction process, choice of primers, and platforms for DNA sequencing, bioinformatics analysis programs, and databases used for taxonomic classification. The gut microbiota composition results of each study were extracted and focused on α-and β-diversity measures and overall significant differences in composition between the two groups. Significantly altered bacteria was included for taxa phylum, family and genus level, only.

| Clinical information reveals highly heterogeneous study populations
Sixteen of the included studies were designed as cross-sectional, case-control studies comparing the gut microbiota of patients with MDD and healthy controls. One study was a longitudinal study, where each participant provided fecal samples at three different time points (Lin et al., 2017). The seventeen studies included a total of 738 patients diagnosed with MDD and 782 healthy controls (Table 1).
In the majority of the studies, patients were assessed according to the DSM criteria (Aizawa et al., 2016;American Psychiatric Association, 1994, 2000Chen et al., 2018Chen et al., , 2020Chung et al., 2019;Jiang et al., 2015;Kelly et al., 2016;Lai et al., 2019;Lin et al., 2017;Liu et al., 2020;Mason et al., 2020;Rong et al., 2019;Stevens et al., 2020;Zheng et al., 2016); the remaining studies used the ICD criteria (Hiller et al., 1994;Huang et al., 2018;Naseribafrouei et al., 2014;Vinberg et al., 2019) (Table 2). Eight of the studies excluded patients with comorbid psychiatric disorders (Chen et al., 2018(Chen et al., , 2020Chung et al., 2019;Jiang et al., 2015;Lai et al., 2019;Mason et al., 2020;Rong et al., 2019;Zheng et al., 2020), while six of the studies excluded patients with inflammatory bowel disorders (Chen et al., 2018;Huang et al., 2018;Jiang et al., 2015;Kelly et al., 2016;Vinberg et al., 2019;Zheng et al., 2020). Otherwise, there was no consensus between studies on exclusion criteria for specific psychiatric or somatic disorders. Four studies also examined a subset of patients with either bipolar disorder Vinberg et al., 2019;Zheng et al., 2020) or comorbid anxiety (Mason et al., 2020). Data from these subgroups are not included in this review. Some studies included antidepressant treatment-naïve patients exclusively (Chen et al., 2018;Huang et al., 2018), or a subset of patients not previously prescribed antidepressants (Naseribafrouei et al., 2014;Zheng et al., 2016Zheng et al., , 2020. In the remaining studies, the use and description of pharmacological treatment varied, with only a few studies limiting their patients to specific pharmacological treatment (Chung F I G U R E 1 PRISMA flow diagram for article selection. Three databases were used to identify relevant papers, and SyRF was used to screen abstracts

| Methodology and bioinformatics analyses of gut microbiota composition varied considerably between studies
In all studies, fecal samples were collected to determine the gut microbiota composition of both patient and control groups ( Table 3).
The majority of the studies used DNA sequencing to assess the gut microbiota: thirteen performed 16S ribosomal ribonucleic acid (rRNA) gene sequencing (Janda & Abbott, 2007;Jovel et al., 2016) and two used shotgun sequencing (Jovel et al., 2016). The remaining two studies used targeted reverse transcriptase-quantitative polymerase chain reaction (qRT-PCR) with species-specific primers (Aizawa et al., 2016) or a combination of 16S rRNA gene sequencing and qRT-PCR (Mason et al., 2020).
The studies performing 16S rRNA gene analysis or metagenomic sequencing primarily applied one of three distinct sequencing plat- Overall, the studies were highly heterogeneous regarding choice of nucleic acid extraction method, 16S rRNA gene target region and bioinformatics analysis program, as well as database for taxonomic classification.

| Both diversity and specific taxa were reported significantly different across studies
Despite methodological heterogeneity, the studies observed several variations in gut microbiota composition between patients and controls. Observations of differences in α-or β-diversity indices between patients and controls are presented in Table 4. Generally, results on bacterial diversity differed extensively between studies.
The majority of the studies did not find significant differences in α-diversity, using several diversity indices (Chen et al., 2018(Chen et al., , 2020;

TA B L E 3 (Continued)
Chung  (Jiang et al., 2015). Similarly, contradicting results were also found in the analyses of β-diversity. Distinguishing between patients and healthy controls was possible in twelve of the seventeen studies using a variety of different analytical methods (Chen et al., 2018(Chen et al., , 2020Chung et al., 2019;Huang et al., 2018;Lai et al., 2019;Lin et al., 2017;Liu et al., 2020;Naseribafrouei et al., 2014;Stevens et al., 2020;Zheng et al., 2016Zheng et al., , 2020. Overall, sixteen of the seventeen analyses found compositional differences using either α-or βdiversity indices in the gut microbiota between patients and healthy controls despite methodological heterogeneity.
Although not all studies agreed upon a different composition of the gut microbiota between patients and controls, sixteen out of seventeen studies showed variations in relative abundance of individual taxa between cases and controls. A list of all bacterial taxa significantly different in relative abundance in patients compared with controls is presented in Supporting information 2.
A selected fraction of these is also presented in Table 4. In total, 5 phyla, 36 families, and 78 genera of bacteria were found to be significantly altered in relative abundance between patients and healthy controls. At phylum level, Actinobacteria and Bacteroidetes were observed to be significantly altered in relative abundance, but often with opposing directions of change, thus making comprehensive conclusions difficult. On the other hand, Bifidobacteriaceae and Coriobacteriaceae, both belonging to the Actinobacteria phylum, were consistently increased in relative abundance in MDD in five (Chen et al., 2018;Chung et al., 2019;Lai et al., 2019;Rong et al., 2019;Zheng et al., 2020) and four (Chen et al., 2018(Chen et al., , 2020Rong et al., 2019;Zheng et al., 2016) studies, respectively. Three bacterial genera, belonging to either the Bifidobacteriaceae or the Coriobacteriaceae family, were also increased in relative abundance in patients; Eggerthella in six studies (Chen et al., 2018(Chen et al., , 2020Chung et al., 2019;Lai et al., 2019;Naseribafrouei et al., 2014;Rong et al., 2019) and Atopobium (Chen et al., 2018;Lai et al., 2019;Rong et al., 2019) and Bifidobacteria (Chen et al., 2018;Chung et al., 2019;Lai et al., 2019;Rong et al., 2019) in four study populations. Four of these studies were conducted in the People's Republic of China or the Republic of China and had similar study designs. There was a decreased relative abundance of one taxa belonging to the Firmicutes phylum, Faecalibacterium, in seven studies (Chen et al., 2018(Chen et al., , 2020Jiang et al., 2015;Lai et al., 2019;Rong et al., 2019;Stevens et al., 2020;Zheng et al., 2016Zheng et al., , 2020, but their study designs were not similar. For the remaining taxa showing differences between patients with MDD and controls, no similarities were found between studies; taxa were often observed to be increased in one study, but decreased in another, with most taxa only showing significant differences between groups in one or two studies. This suggests that TA B L E 4 Bacterial taxa obtained from next generation sequencing of the bacterial DNA comparing patients with MDD and healthy controls Note: Statistically significant differences observed between the patient and control group in the bioinformatical analyses are designated as either "Yes" or "No". Arrows specify the direction of change. Fields with a question mark correspond to studies, where the bioinformatics analysis was conducted according to the methods section, but no conclusions were reported in the results section. Bacterial taxa are listed as phylum -family -genus. Arrows here specify the increase (green) or decrease (red) in relative abundance of bacterial taxa in patients compared with controls, if this was observed in four or more study populations. If a study conducted analyses on several populations, such as dividing participants into sexes, each observation made was counted as an independent observation. "Total observations" is the cumulative number of times a specific direction of change was observed across the studies. Taxa significantly different between patients and controls were presented, when four or more studies agreed upon a specific direction of change.

No No
No these singular observations may be unique to their respective study populations, and not applicable to gut microbiota variations associated with MDD on a global scale. Opposite the other studies, Aizawa et al. (Aizawa et al., 2016) used targeted qRT-PCR and discovered less Bifidobacterium counts in their patient group. Mason et al. did not find any association between specific bacteria and MDD diagnosis using qPCR, but only between the Clostridium cluster IV and the severity of depression (Mason et al., 2020).
Overall, the gut microbiota composition of patients and controls clustered separately in two-thirds of the studies. Additionally, the relative abundance of Eggerthella, Atopobium, and Bifidobacterium was increased, while it was decreased for Faecalibacterium in a subset of studies.

| D ISCUSS I ON
Of the seventeen included studies, sixteen observed significant differences in gut microbiota composition between patients with MDD and healthy controls. An increase in relative abundance of Eggerthella, Atopobium, and Bifidobacterium and a decrease in relative abundance of Faecalibacterium was a frequent finding. Despite methodological heterogeneity, it was possible to distinguish between patients with MDD and healthy controls in almost all of the included studies based on either α-or β-diversity. These results were based on highly heterogeneous study designs, with various study populations, clinical assessments, and experimental setups.
Bacterial taxa increased in relative abundance may affect signaling pathways in MDD such as bile acids signaling in the brain.
Some strains of Eggerthella can facilitate bile acid oxidation (Harris et al., 2018), while Bifidobacteria can hydrolyse bile salts (Kumar et al., 2006). This was supported by the study by Chung et al., which showed perturbed bile acid metabolism in patients and a positive correlation between the relative abundance of Eggerthella and Beck Depression Index scores (Chung et al., 2019). Increased bile acid stimulation of the farnesoid X receptor inhibits the production of BDNF . This neural growth hormone has been observed to be decreased in MDD (Dwivedi, 2009), which was also observed by Jiang et al. (Jiang et al., 2015), linking gut microbiota alterations to bile acid metabolism and symptoms in patients with MDD.
In contrast to increases in relative abundance, a loss of relative abundance of Faecalibacterium was reported in seven studies.
The study by Jiang et al. reported that levels of Faecalibacterium correlated negatively with severity of depressive symptoms (Jiang et al., 2015). Lack of Faecalibacterium in patients with MDD may exacerbate chronic low-grade inflammation associated with the disorder (Beurel & Lowell, 2017;Furtado & Katzman, 2015;Otte et al., 2016). This was supported by the studies by Huang Faecalibacterium (Huang et al., 2018;Stevens et al., 2020). This taxa is known to have anti-inflammatory properties (Quevrain et al., 2016;Sokol et al., 2008). This genus produces SCFAs (Koh et al., 2016), which downregulate the production of proinflammatory cytokines (Lopez-Siles et al., 2017). The study by Kelly et al. found increased pro-inflammatory cytokines as well as a loss of the genus Prevotella (Kelly et al., 2016), another known producer of SCFA (Chen et al., 2017). This indicates that loss of SCFA producers may lead to loss of regulatory interactions with the immune system. This is supported by clinical studies where supplementation with probiotics, which produce SCFAs, reduced depressive symptoms in patients with MDD , presumably caused by limiting low-grade inflammation (Jakobsen et al., 2017;Park et al., 2018;Yu et al., 2017).
Overall, changes in these genera may be involved in the pathology of MDD. These changes were not observed across all studies in this review, and interpretation of the results should thus be made with caution. Despite heterogeneity concerning changes in taxa, twelve out of eighteen studies were able to distinguish patients from healthy controls based on β-diversity. This suggests that the gut microbiota as a whole, rather than singular bacterial taxa, differentiates patients from healthy controls. Reasons for the heterogeneity in results may lie in the variations in study design and populations. Firstly, it is well established that dietary, geographical, and cultural impacts influence the stability, functionality, and structure of the bacterial communities (Conlon & Bird, 2014;Singh et al., 2017;Yatsunenko et al., 2012). This might explain the consistent increase in the relative abundance of the three genera Eggerthella, Atopobium, Secondly, heterogeneity in clinical characteristics of patients may have resulted in gut microbiota composition differences between studies. The studies in this review focused on depression, but the diagnostic criteria used to determine the diagnosis differed between studies. For example, assessment of bipolar disorder was only performed in some studies (Chen et al., 2018(Chen et al., , 2020Chung et al., 2019;Jiang et al., 2015;Lai et al., 2019;Mason et al., 2020;Rong et al., 2019;Zheng et al., 2020). Interestingly, the study by Zheng et al. (2020) also examined patients with bipolar disorder, and found that patients with bipolar disorder and depression, respectively, were distinguishable based on gut microbial composition (Zheng et al., 2020). Likewise, comorbid inflammatory bowel disorder only led to exclusion in some studies (Chen et al., 2018;Huang et al., 2018;Jiang et al., 2015;Kelly et al., 2016;Vinberg et al., 2019;Zheng et al., 2020). Heterogeneity in patient characteristics may therefore have led to differences in gut microbiota composition between studies. The active pharmaceutical treatment can also have affected the gut microbiota composition in patients. Nonantimicrobial drugs have the potential to either influence the bacterial composition (Maier et al., 2018;Vich Vila et al., 2020), or be metabolized into bioactive compounds (Enright et al., 2016;Li et al., 2016).

| Standardization of study populations and applied methods may limit heterogeneity
Additionally, different types of antidepressants may give rise to different side effects such as altered appetite with subsequent diet and weight changes (Fava, 2000;Lee et al., 2016), thereby affecting the gut microbiota. Furthermore, antidepressant drugs have docu- Thirdly, the DNA purification method, choice of primer pairs used for amplification, and database used for taxonomic assignment have been shown to affect experimental outcomes (Costea et al., 2017;Voigt et al., 2015). However, it was not obvious if the choice of method affected the observed variability in bacterial compositions. Two studies applied the same laboratory setup and observed an increase in relative abundance of Eggerthella, Atopobium and Bifidobacterium Rong et al., 2019).
Three additional studies also had the exact same setup for characterization of gut microbiota composition (Chen et al., 2018(Chen et al., , 2020Zheng et al., 2016), but they only agreed upon an increase in relative abundance of Lachnospiraceae. Despite their similar molecular approaches, the discrepancies suggest that the choice of patient population may affect the gut microbiota composition to a greater extent than the DNA purification method and sequencing platform.
Fourthly, the choice of hypervariable regions of the 16S rRNA gene is particularly important, as different primer pairs have been shown to induce selective bias in the detection of bacteria (Albertsen et al., 2015;Hamady & Knight, 2009;Lozupone et al., 2013). Notably, studies using the 314F/805R primer pair did not consistently observe the same differences; they did, however, agree on a reduction in relative abundance of Faecalibacterium (Huang et al., 2018;Stevens et al., 2020). Additionally, the 16S rRNA gene sequencing method has limitations, as the 97% clustering of OTUs generally only allows taxonomic assignment at genus level (Poretsky et al., 2014).
This lack of sensitivity and specificity may account for the contrasting directions of change observed across studies as alterations at species level may obscure observations at genus level. Two studies performed shotgun metagenomics and found that species belonging to the Bifidobacterium genus, such as B. longum and B. dentium were increased in relative abundance in patients with MDD. This is in accordance with the other studies who observed a relative increase in Bifidobacterium at genus level, further strengthening the association between this genus and MDD.

| CON CLUS ION
Sixteen out of seventeen studies reported a difference in the gut microbiota composition between patients and controls. Several studies found either a higher relative abundance of Eggerthella, Atopobium and/or Bifidobacterium, or a lower relative abundance of Faecalibacterium or Dialister in patients with MDD. However, there was limited agreement between the studies, possibly due to heterogeneity in the experimental design.

| SUMMATIONS
Gut microbiota was observed significantly different in most of the included studies.
Eggerthella was increased while Faecalibacterium was decreased in patients with MDD.
Variability in methodology makes generalization across studies difficult.

| LI M ITATI O N S
This systematic review has some limitations. It was not possible to perform a meta-analysis due to laboratory methodological heterogeneity between the studies. We did not perform a quality assessment of the studies, but relied on the studies meeting eligibility criteria.
Publication bias was not investigated on the same grounds, resulting in a lack of statistical approximations.

ACK N OWLED G M ENTS
We would like to thank associate professor Manimozhiyan Arumugam, MD for providing expert insight into the bioinformatics processing of 16S rRNA sequencing data.