Depression & the Microbiome-Gut-Brain Axis: Changing Your Gut
Microbiome to Improve Your Mental Health, Will it be Possible One Day?
Maaike van der Rhee
Student Number: 12644781
University of Amsterdam
Literature Thesis Brain & Cognitive Sciences
Supervisor: A.K. Groen.
Second reader: W.J. Wiersinga
Abstract
The microbiome-gut-brain axis has increasingly been studied as an important factor in depression. In this paper, putative forms of communication within this axis are discussed briefly. Next, eight cross-sectional studies on human microbiome in depressed versus non-depressed individuals are analyzed. The results from these studies are diverse and often contradictory. This paper emphasizes how methodological inconsistencies and sample differences may have affected results, which greatly hinders the comparability between the studies. Collectively, this limits the possibility to form a uniform conclusion about the type and degree of changes in the microbiome of depressed individuals. In order to minimize heterogeneity the following recommendations for methodology and sample selection were made: (1) using non-Bayesian classifiers and moving towards open databases for increased comparability and reproducibility (2) using the 97% threshold for operational taxonomic unit (OTU) picking (3) open-reference method of OTU picking for its ability to identify known and unknown sequences (4) exclusion of IBS patients (5) exclusion if antibiotics were taken in the prior month (6) consistent documentation of antidepressant medication status, and preferably a homogenous antidepressant profile (7) including diet as a covariate in analysis (8) being mindful of the geographical and ethnic profile of the sample. Additionally, more research should be conducted on the effect of the selected variable region and DNA extraction kits. Notwithstanding the limitations of the studies, it is evident that some significant changes are present in the microbiome of depressed individuals in comparison to healthy controls. Future well-standardized research should elaborate on these changes, as well as the underlying mechanisms.
Depression & the Microbiome-Gut-Brain Axis: changing your gut to improve your mental health, will it be possible one day?
Depression is a mental disorder characterized by abnormal depressed mood (dysphoria) and diminished pleasure (anhedonia; WHO, 2016). Diagnosis of Major Depressive Disorder (MDD) is based on depressive symptoms that persist most of the day for at least two weeks (American Psychiatric Association, 2013), although most episodes last considerably longer (6 months on average; WHO, 2016). Episodes involve clear-cut changes in affect, cognition, and functionality, such as sleep disturbance, lack of appetite, poor concentration, and possibly reoccurring thoughts about death and suicide (American Psychiatric Association, 2013). MDD is highly recurrent, with 40% to 60% of individuals who suffer from a first depressive episode experiencing a subsequent episode (Burcusa & Iacono, 2007; Eaton et al., 2008). According to the WHO, 322 million people are affected globally by depressive disorders, with comparable prevalence around the continents (WHO, 2017).
Models of Depression
Historically, the aetiology of MDD has predominantly been researched in the context of an imbalance in neurotransmitters (Van Praag, 2001). It has been proposed that depression is caused by a functional deficit in monoamines: the concentration of neurotransmitters such as serotonin, noradrenaline, and dopamine are reduced (Hirschfeld, 2000). Indeed, in major depression, hypoactivity in serotonin, noradrenaline, and dopamine has been reported in the midbrain (Ruhe, Mason, & Schene, 2007). One of the limitations of this monoamine hypothesis is that it does not explain why regular antidepressant drugs increase monoamine levels within 1 or 2 days, while it can take up to 2 or 3 weeks of exposure to alleviate depressive symptoms (Hindmarch, 2001). Thus, this theory cannot fully explain the latency nor the mechanism of antidepressants and might rely too heavily on one process to provide a comprehensive understanding of the neurobiology of depression.
Other biological models of depression have been presented, such as the inflammation model of depression. This theory postulates that altered cytokine activity can influence brain circuits directly and indirectly and subsequently cause depressive symptoms (Otte et al., 2016). Peripheral cytokines can act directly on the central nervous system (CNS), if they are transported across the blood-brain barrier. In addition, inflammatory signals can be delivered to the CNS through cellular mechanisms or signaling via the vagus nerve (Otte et al., 2016). Meta-analysis supports the notion of altered cytokine response, as depressed subjects had
significantly higher concentrations of pro-inflammatory cytokines (TNF-α and IL-6) compared to control subjects (Dowlati et al., 2010). However, other studies have failed to confirm this association (Steptoe, Kunz-Ebrecht & Owen, 2003;), or found that this association was attenuated when other factors were included, such as gender (Ford & Erlinger, 2004; Danner, Kasl, Abramson & Vaccarino, 2003) or BMI (Haack et al., 1999). Based on these inconsistencies, a strong stance on the role of the immune system in depression might be premature, and inflammation may contribute to some, but not all, cases of depression (Raison, Capuron & Miller, 2006).
Another body of research on the aetiology of depression has focused on an overactive hypothalamus-pituitary-adrenal axis (HPA axis). The HPA axis plays an important role in the regulation of body peripheral functions such as metabolism and immunity through the production of glucocorticoids (Pariante & Lightman, 2008). In addition, the HPA axis also affects the brain directly, where it can alter cognition and neural plasticity (Belanoff, Gross, Yager & Schatzberg, 2001). With respect to depression, two large meta-analyses indicate that cortisol levels in patients with MDD were increased (Stetler & Miller, 2011; Knorr, Vinberg, Kessing & Wetterslev, 2010), though these results should be interpreted with caution as considerable heterogeneity was present between sample groups. Interestingly, Hinkelmann and colleagues (2009) found that elevated cortisol levels were correlated with impaired cognitive function in depressed participants. Furthermore, a prospective study has shown that increased levels of cortisol are a risk factor for MDD in women at 13 months’ follow up (Harris et al., 2000). However, theories on the HPA axis have been criticized for having little clinical utility: despite strong (pre)clinical data of a dysregulation of the HPA axis, no drug or diagnostic tests have been approved that target the HPA axis (Menke, 2019).
Impact
Depression can profoundly affect all aspects of life, including productivity at work, performance at school, relationships with family and friends, and the ability to participate in a community (WHO, 2016). Greater depressive symptoms have also been linked to poorer quality of life (as measured by the Health-related quality of life index, HRQOL; Trivedi, 2010). Additionally, consideration of the disease burden of depression shows worrisome results: among all medical conditions worldwide, MDD has been a leading contributor to chronic disease burden (as measured by YLD: “years lived with a disability”) for nearly three decades (James et al., 2018).
Treatment
In the treatment of MDD, three main treatment recommendations exist: psychotherapy, antidepressant pharmacotherapy, or the combination of both (Gelenberg, 2010). Meta-analyses on a large number of randomized controlled trials show that psychotherapy is effective in treating MDD and that there is no consistent or clinically meaningful difference between different types of psychotherapy (Cuijpers et al., 2013; Cuijpers, Van Straten, Andersson & Van Oppen, 2008). Pharmacotherapy usually involves antidepressant drugs that target the monoamine imbalance (as explained above). Selective serotonin reuptake inhibitors (SSRIs) are the most commonly prescribed class of antidepressants (Forns et al., 2019). The efficacy of antidepressant drugs has been heavily debated. In general, large meta-analyses have concluded that antidepressants have a significant, yet moderate, effect when compared to control, although it is uncertain to what extent the placebo effect is responsible for a large proportion of antidepressants’ efficacy (for an overview on the debate, see Bschor & Kilarski, 2016).
Many of the aforementioned studies focused on improvements on depressive scales only. However, when full remission is considered the ultimate goal of treatment (rather than just response), only two-thirds of patients experience successful treatment based on standard pharmacological and psychological treatment (Rush et al., 2006). This data included patients who had switched antidepressant medication multiple times (a practice that is quite common). Moreover, analysis of Australian data revealed that treatment is generally bad at lowering the disease burden (as measured by YLD; Andrews, Issakidis, Sanderson, Corry & Lapsley, 2004). These data regarding the prevalence of unsuccessful treatment and inefficiency of decreasing the disease burden underscore the need to identify more efficacious treatment approaches by including a wider range of possible ‘causal’ factors than simply neurotransmitter depletion (Winter, Hart, Charlesworth & Sharpley, 2018).
The Microbiome-Gut-Brain Axis
It has long been recognized that gastrointestinal and mental health disorders co-occur at remarkably high rates, and a substantial body of research has investigated associations between the community of microbes that inhabit the gastrointestinal tract, or gut microbiome, and the health of the human host (Simpson, Mu, Haslam, Schwartz & Simmons, 2020). The human gut microbiome consists of 40 trillion cells of hundreds of different species, carrying
250 to 800 times more genes than humans (Lozupone, Stombaugh, Gordon, Jansson, Knight, 2012). Microbiota of the gastrointestinal tract is emerging as a putative player in depression. The microbiome presents a complex active network that has the potential to affect host metabolic phenotypes, as well as be influenced by the host itself (Lozupone et al., 2012). Researchers have coined this connection the Microbiome-Gut-Brain Axis (MGB). The MGB
refers to the bidirectional system in which the microbiota can exert influence on host physiology in a bottom-up manner, and the brain can influence the community of microbiota in a top-down fashion (Isaiah, Du Toit Loots, van der Kuip, Van Furth & Mason, 2020). There is an ongoing debate about the different modes of communication within the MGB (which will be discussed in more detail later), but in general, it is believed to involve the nervous, endocrinal, and immune systems (Lerner, Neidhöfer & Matthias, 2017). It is observed that diseases that are known to have dysregulated microbiome often co-occur with depression (Simpson et al., 2020). This indicates that depression and the microbiome may be related. For this reason, considering the MGB axis and its role in depression could enhance understanding of the disorder.
Relevance of the MGB Axis in Depression
As explained, depression affects millions of people globally and has a severe impact on the quality of many lives. The economic costs associated with depression are staggering in the European Economic Area alone: the estimated economic costs of depression amounted to €136.3 billion, of which the largest share stems from reduced productivity and health care costs (data for 2007; WHO, 2016). Additionally, the treatment of depression has suboptimal outcomes as a significant proportion of patients does not go into remission (Rush et al., 2006). Evidently, there is a socially relevant window for improvement of treatment. Therefore, this paper will focus on the MGB axis as a putative therapeutical target to improve the treatment of depression. It aims to investigate to which extent the MGB axis plays a role in depression and whether it is manipulatable for treatment of depressive symptoms. It will do so by first examining the putative ways in which the MGB axis functions. Second, cohort studies in humans will be discussed, including serious methodological challenges. Finally, recommendations for methodology and sample selection are made in order to improve comparability and reproducibility.
Communication within the MGB Axis
Researchers have started to investigate how alterations in normative gut-brain axis communication could be involved in disease pathophysiology. Essential terms with respect to this topic are intestinal permeability and dysbiosis. The former refers to the control over the extent to which material in the gut lumen can translocate across the intestinal barrier. Naturally, the intestinal wall allows for some permeability in order to absorb nutrients, while maintaining a barrier for potentially harmful substances (Arrieta, Bistritz & Meddings, 2006). Dysbiosis describes changes in the gut microbiome composition (Robles-Alonso, Guarner & Ehrlich, 2014), often linked to a decrease in diversity (Makris, Karianaki, Tsamis & Paschou, 2020). Communication within the MGB axis involves both bottom-up and top-down processes, which will be discussed separately in the next sections.
Bottom-up Communication
The microbiome produces a variety of metabolites. These metabolites allow for communication between microbes but are often recognized by the human system as well, as they are based on a shared neurochemical language that exists between host and microbe (Cryan et al., 2019). These microbiome-derived “neuromodulators” include tryptophan precursors, serotonin, GABA, catecholamines (noradrenaline, adrenaline, and dopamine), and “immunomodulators” such as Short Chain Fatty Acids and histamines. Due to the shared neurochemical language, these microbiome-derived modulators can communicate with the host. Exactly how this communication occurs or what impact it has on the host’s physiology remains to be uncovered for the majority of these modulators (Cryan et al., 2019). Nevertheless, there are various theories on the modes of communication, most prominently the idea that there are two possible mechanisms via which communication between microbiome-derived modulators and the host could occur: Firstly, via direct stimulation of the nervous system. It has been indicated that the autonomic nerve cells in the gut can carry sensory information directly to the brain upon local microbial metabolite stimulation (Cryan et al., 2019). This most likely occurs via the vagus nerve, which is the most direct route between the gut and the brain, has extensive innervation, and consists predominantly of afferent nerves (Cryan et al., 2019). It has been argued that histamines in particular use this mechanism, as the afferent vagal nerves have shown to be sensitive to histamines (Kreis, Jiang, Kirkup & Grundy, 2002). Secondly, communication could occur via translocation and subsequent entering of the systemic circulation. Intestinal permeability is an important determinant of the degree of
translocation. With higher integrity in the intestinal barrier and lower permeability, microbiome-derived metabolites are less likely to be able to translocate (Viana et al., 2010).
Perhaps the most researched microbiome-derived metabolites are the Short Chain Fatty Acids (SCFA). SCFAs are produced by a microbial fermentation process of dietary fibers, generally absorbed by the epithelium, and used as an energy source. SCFAs have one to six carbon atoms and the major SCFAs include acetate, propionate, and butyrate (Ratajczak et al., 2019). The three major SCFA may exert considerably different effects upon host physiology due to two factors (Louis, & Flint, 2017): firstly, they differ in their respective relative production rates and concentrations. Secondly, they interact distinctively with proteins and receptors. Consonantly, the different types of SCFA seems to have diverse and complex effects on host physiology. For example, butyrate can bind to epithelial cells to stimulate cytokine production, but also promotes anti-inflammatory properties in macrophages during an immune response (Chang, Hao, Offermanns, & Medzhitov, 2014). Additionally, it promotes integrity of the gut barrier and thereby decreases permeability, stimulates the expression of anti‐ inflammatory IL‐10 in macrophages and intestinal dendritic cells, and enhances the anti‐ inflammatory activity of Tregs (Singh et al., 2014). On the other hand, propionate has been shown to stimulate intestinal gluconeogenesis through direct stimulation of enteric–CNS pathways (De Vadder et al., 2014) and acetate has been linked to appetite regulation (Frost et
al., 2014). In general, acetate and propionate stimulate while butyrate inhibits immune cell
function (Bolognini, Tobin, Milligan & Moss, 2016). This presents a complex scheme of action for SCFAs, including both pro- and anti-inflammatory effects (Li et al., 2018). Additionally, it warrants caution when mixtures of SCFAs are used in research, as the overall results of such research might not be precise or applicable to all subtypes of the SCFA.
It is of importance to note that strains of bacteria produce different metabolites (Berdy, 2005). A particular composition of the microbiome generates a distinct profile in metabolites and can therefore exert a specific effect on the host.
Top-down Communication
In the previous section, it was discussed how microbiome-derived metabolites affect other systems in the body. This section will cover how top-down communication occurs within the MGB axis.
As mentioned in the introduction, the HPA axis and elevated cortisol levels are thought to be related to the aetiology of depression. Concerning the MGB axis, when a stressful event activates the HPA axis, the intestinal motility and epithelial permeability are expected to increase. Indeed, corticotropin-releasing hormone (CRH) has been linked to increased intestinal motility in humans (Fukudo, Nomura, & Hongo, 1998). Similarly, stress has been found to increase intestinal permeability in humans (Vanuytsel et al., 2014). When the intestinal permeability is heightened, more (potentially harmful) molecules could seep through the barrier and enter circulation. For example, endotoxins may translocate and initiate an immune response termed “endotoxemia” (de Punder & Pruimboom, 2015). Additionally, when a stressful event activates the HPA axis, this is thought to change the environment of the gastrointestinal tract, possibly making it less attractive for habitation by certain bacteria and more attractive for other species (Frankiensztajn, Elliott & Koren, 2020). Hereby, gut dysbiosis might occur, and as metabolites are produced by the microbiota, dysbiosis can consequently alter metabolite profiles (Kim, 2018). Collectively, this presents a complex feedback loop: acute stress may increase gastrointestinal motility and permeability, which changes the microbiome composition. Altered composition, in its turn, can lead to distinct production of metabolites, which exert diverse effects on host physiology and changes the bottom-up communication. Modified bottom-up communication could ultimately affect the activity of the HPA axis.
Immune system-HPA axis interactions are common in a variety of psychological disorders (Cryan et al., 2019), and therefore it is not surprising that these interactions might affect the MGB axis as well. The precise mechanism via which the HPA axis and the immune system interact is beyond the extent of this paper. In brief, there is a negative feedback loop between the HPA axis and immune system. Within this loop, cytokines can stimulate the HPA axis (ThyagaRajan & Priyanka, 2012). In turn, glucocorticoids produced by the HPA axis can give negative feedback to immune cells. This suppresses the further synthesis and release of cytokines, thereby protecting the host from the detrimental consequences (i.e. tissue damage) of an overactive immune response (Silverman, Pearce, Biron & Miller, 2005). With respect to the MGB axis, it has been demonstrated that the afferent vagus nerve can sense peripheral pro-inflammatory cytokines and activate the HPA axis to dampen the immune response (Borovikova et al., 2000). Therefore, the vagus nerve might play an essential role in sensing the enteric conditions and regulating immunity-HPA interactions (Peirce & Alviña, 2019). The HPA-axis has been mentioned as an immunosuppressant, however, it has been found that the
HPA axis may also activate a mucosal immune response and cause local inflammation in the gut (Dinan & Cryan, 2012). This effect deviates from the general immunosuppressant role of the HPA-axis. Furthermore, this local gut inflammation can spread to the central nervous system and cause neuro-inflammation (for an overview of ways this occurs, see Peirce & Alviña, 2019). Finally, neuro-inflammation can disrupt the synthesis of the monoamine neurotransmitters (Miller & Raison, 2016). This is especially interesting in light of the monoamine theory behind depression, which depicts a functional deficit in monoamine neurotransmitters.
Sequencing the Human Gut Microbiome
Previously, physiological pathways that were likely to be related to depression and the microbiome were discussed. This has provided a broad perspective on putative pathways and functional mechanisms. Moving from a theoretical to experimental context, next the study of the composition of the human gut microbiome will be discussed. This may establish to what extent expected differences in the gut microbiome of depressed individuals are actually present. Sequencing Techniques
Within Next Generation Sequencing techniques, the majority of microbiome studies have relied on 16S rRNA sequencing (Clooney et al., 2016). This method targets the ubiquitous 16S rRNA gene in prokaryotes and utilizes polymerase chain reaction (PCR) to amplify it. The 16S rRNA gene consists of both evolutionary highly conserved and hypervariable regions. The highly conserved stretches are suitable for primer binding, while nine hypervariable regions harbor phylogenetical differences useful for prokaryotic identification purposes (Neefs, Van de Peer, De Rijk, Chapelle & De Wachter, 1993). Primers can target one or more variable regions (Shahi, Freedman & Mangalam, 2017). Once the targeted amplicon sequences have
been multiplied, the downstream steps of taxonomic classification involve a similarity search within the sequenced DNA fragments (referred to as reads). Reads are clustered based on some predefined similarity threshold and binned in an operational taxonomic unit (OTU). This predefined similarity threshold is artificially (and arguably arbitrarily) set, usually at 97% similarity (Koeppel & Wu, 2013).
The clustering of OTUs can occur de novo, closed- and open-reference. In de novo OTU picking, reads are compared against one another and clustered without any external reference database. In contrast, closed-reference OTU picking uses a reference database to
cluster known reads. Any reads which do not match with a reference sequence are excluded from downstream analyses. Naturally, this can impose serious limitations based on the selected reference database. In practice, open-reference OTU picking is frequently used, in which reads are clustered against a reference database and any reads which do not provide a match are subsequently clustered de novo (Yadav, Dutta & Mande, 2019). Advantages of 16S sequencing are its relatively cheap cost and insensitivity to contamination (because 16S sequencing relies on a subunit unique to prokaryotes, host contamination is not an issue; Cryan et al., 2019). It should be noted, however, that 16S sequencing only provides the “who is there” and does not address functionality (Claesson, Clooney & O'toole, 2017). Additionally, it often lacks the power to differentiate at the species taxonomic level, as the evolutionary rate is relatively low between species (Boers, Jansen & Hays, 2019). This may lead to polyphyletic OTUs, which are similarity-based OTUs that contain sequences with mixed phylogenetic signal (i.e. sequences of different strains; Koeppel & Wu, 2013). In conclusion, researchers frequently choose 16S sequencing for its relatively cheap cost, insensitivity to contamination, and de novo clustering abilities.
Cohort studies on Human Gut Microbiome
This paper has identified eight human case-control studies that have examined the fecal microbiome (quantified from stool samples) from depressed individuals in comparison to healthy controls (Aizawa et al., 2016; Chung et al., 2019; Huang et al., 2018; Jiang et al., 2015; Kelly et al., 2014; Lin et al., 2017; Naseribafrouei et al., 2014; Zheng et al., 2016). These studies were found through a literature search which combined the terms “depression,” “depressive disorder,” “microbiome,” “microbiome composition,” “fecal,” “gut,” “gut brain axis,” and “microbiome gut brain axis”. These studies will be discussed in detail next and an overview can be found in Table 1.
Firstly, one study has found that the order Bacteroidales was significantly overrepresented, while the family Lachnospiraceae was underrepresented in a cohort of 37 depressed patients (versus 18 controls; Naseribafrouei et al., 2014). Additionally, at lower taxonomic levels, the overrepresentation of the genera Alistipes and Oscillibacter were associated with depression. However, binning the data at the genus level provided a poor classification model, with misclassification of about half of the individuals. This suggests that although there were some significant differences at distinct taxonomic levels, there were no
global patterns evident at the genus level when considering the entire microbiome of depressed individuals. To the researchers’ surprise, closely related OTUs within this study had opposite correlations. Due to this, the taxonomy level that is considered could seriously affect the results, as it dictates to what extent OTUs are considered together (because they fall under the same family or phylum). If many OTUs with opposite correlations are examined as one group, it might nullify or strengthen the overall correlation of the group. This, in turn, may lead to misleading findings. Lastly, no difference in diversity between depressed patients and controls was evident in this cohort.
In contrast, in another study increased microbiome diversity was present in patients with active depression, but not in those that had responded to treatment (Jiang et al., 2015). This study compared 46 individuals with depression to 30 matched healthy controls. Within the depressed patients a distinction was made between those with active depression, and those with a responding depression, meaning they had shown a 50% reduction of symptoms (as measured by the Montgomery–Asberg Depression Rating Scale) after 4 weeks of treatment. At the phylum level, two phyla Bacteroidetes and Proteobacteria were increased in (both active and responding) depression, while the Firmicutes and Actinobacteria were underrepresented (Jiang et al., 2015). Interestingly, Fusobacteria levels were increased in active depression, but decreased in responding depression. Additionally, at the familial level Acidaminococcaceae,
Enterobacteriaceae Porphyromonadaceae, and Rikenellaceae were increased in both types of
depressed subjects, while Lachnospiraceae, Ruminococcaceae, and Veillonellaceae were decreased. Interestingly, Bacteroidaceae had an opposite effect in responding and active depression, respectively a higher and lower abundance compared to healthy controls. Lastly, at the genus level, more than 70 genera differed significantly between the microbiome from depressive and healthy individuals, including 15 predominant strains. Perhaps the study’s most striking finding is the presence of a negative correlation between the relative abundance of
Faecalibacterium and the severity of depression in patients (measured by the Hamilton’s
Depression Scale, HAMDS; Jiang et al., 2015). However, the researchers did note that they did not use HAMDS to measure the depressive symptoms of the healthy controls, and that there might be a risk of overestimation of this correlation in the small cohort.
Phyl um O rde r Fa m ily G enus N as er ib af ro ue i e t al . B ac te ro id et es UP NA Lac hn os pi ra ce ae DO W N A lis tipe s UP O sc ill ib ac te r UP Ji an g et al . B ac te ro id et es UP F us ob ac te ri a U P ( A -MD D ) P rot eoba ct er ia UP ( A -MD D ) A ct inob ac te ri a DOW N Fi rm ic ut es DO W N F us ob ac te ri a D O W N ( R -MD D ) P rot eoba ct er ia DOW N ( R -MD D ) NA R ik ene ll ac ea e U P P or ph yr om onada ce ae U P A ci dam in oc oc cac eae UP E nt er oba ct er iac eae UP B ac te ro ida cea e UP ( R -MD D ) F us ob ac te ri ac eae UP ( A -MD D ) Lac hn os pi ra ce ae DO W N V ei ll one ll ac ea e DO W N B ac te ro ida ce ae DO W N (A -MD D ) P re vo te llac eae DOW N (A -MD D ) E ry si pe lo tr ic hac eae DOW N ( A -MD D ) R um ino coc cac eae DOW N (R -MD D ) A lis tipe s UP P ar ab ac te ro id es UP R os eb ur ia UP P has col ar ct ob ac te ri um UP B ac te ro id es UP ( R -MD D ) Lac hn os pi ra ce ae in ce rt ae s ed is UP ( A -MD D ) B lau tia UP (A -MD D ) O sc ill ib ac te r UP (A -MD D ) C los tr id ium (X IX +I X ) UP (A -MD D ) M eg am ona s UP ( A -MD D ) P ar as ut te re ll a UP ( A -MD D ) P re vo te lla DO W N F ae ca libac te ri um DO W N R um ino coc cus DOW N D ial is te r DO W N B ac te ro id es DOW N ( A -MD D ) O sc ill ib ac te r DOW N (R -MD D ) E sc he ri ch ia /Sh ige lla DO W N (R -MD D ) A iz aw a et al . NA NA B ifi do bac te ri um DO W N Zh en g et a l NA C or io bac te ri al es U P C los tr id ial es U P C or io bac te ri ac eae U P Lac hn os pi ra ce ae UP R um ino coc cac eae UP U nc la ss if ie d U T O ’s U P A nae ros ti pe s U P B lau tia UP A lis tipe s DO W N R os eb ur ia DO W N F ae ca libac te ri um DO WN N ot e. B lue hi gl ight s in di ca te tha t a f indi ng is c ont ra di ct ed by at le as t one ot he r st ud y. G re en hi ghl igh ts ind ic at e tha t t he f indi ngs is c onf ir m ed by at le as t one s tudy , but not c ont ra di ct ed by a not he r. T ab le 1 R es ul ts fr om the e ight s tudi es quant if yi ng gut m ic robi om e fr om fe ce s of de pr es se d vs . non -de pr es se d hum ans , s how n at di ff er ent tax onom ic le ve ls
Phyl um O rde r Fa m ily G enus K el ly et a l NA NA The rm oanae rob ac te ri ac ea e UP P re vo te llac eae DOW N E gge rt he lla UP H ol de m ani a UP Ge lr ia UP Tur ic ibac te r U P P ar ap re vot el la U P A nae rof ilum U P P re vo te lla DO W N D ial is te r DO W N Li n et a l. Fi rm ic ut es UP B ac te ro id et es DO W N NA P re vo te lla UP C los tr id ium (X IX + IX ) UP St re pt oc oc cu s UP Kl eb si el la U P H uan g et a l. Lac hn os pi ra ce ae DO W N R um ino coc cac eae DOW N C los tr id iac eae DOW N O xal obac te r UP P se ud om ona s UP P ar vi m ona s UP B ul le idi a UP P ept os tr ept oc oc cus UP Ge me lla UP C opr oc oc cus DOW N B lau tia DO W N D or ea DOW N C hu ng et a l. A ct inob ac te ri a UP Fi rm ic ut es UP B ac te ro id et es DO W N P rot eoba ct er ia DOW N B ifi do bac te ri ac ea e UP Lac hn os pi ra ce ae UP P ept os re pt oc oc ca ce ae UP P or ph yr om onada ce ae UP St re pt oc oc ca ce ae UP A lc al ige nac eae DOW N P re vo te llac eae DOW N A dl er cr eu tz ia UP B ifi do bac te ri um UP B lau tia UP C los tr id ium X I UP E gge rt he lla UP H ol de m ani a UP P ar ab ac te ro id es UP R um ino coc cus UP St re pt oc oc cu s UP M eg anom as DO W N Pr ev ot el la DO W N Sut te re lla DOW N N ot e. B lue hi gl ight s in di ca te tha t a f indi ng is c ont ra di ct ed by at le as t one ot he r st ud y. G re en H ighl igh ts indi ca te tha t t he f indi ngs is c onf ir m ed by a t l ea st one st udy, but not c ont ra di ct ed by a not he r. T ab le 1 C on ti n u ed
Similarly, to Jiang et al. (2015), Huang and colleagues also found a change in diversity in depressed individuals (2018), only Huang and colleagues found a decrease in diversity. Additionally, a number of changes in microbiome composition was found between 27 depressed individuals and 27 controls in this study. At the familial level three families were underrepresented in depressed patients (Lachnospiraceae, Ruminococcaceae &
Clostridiaceae). At the genus level six gene genera were increased (Oxalobacter, Pseudomonas, Parvimonas, Bulleidia, Peptostreptococcus & Gemella) while three decreased
(Coprococcus, Blautia & Dorea; Huang et al., 2018). As the majority of these changes are part of the Firmicutes, the researchers concluded that Firmicutes is the most important phylum that is correlated to depression.
In a slightly larger cohort, principal coordinate analysis (PCoA) of unweighted UniFrac distances showed an obvious difference in gut microbiotic composition between major depressive disorder (MDD) patients and healthy controls (n = 58/MDD, n = 63/HC; Zheng et
al., 2016). When the researchers sought out which OTUs were responsible for the distinction
between MDD and controls, a total of 54 discriminate OTUs was found, of which 29 OTUs were overrepresented in depression, while 25 OTUs were underrepresented. These discriminative OTUs were mainly assigned to the phyla Firmicutes, Actinobacteria, and
Bacteroidetes. The relative abundances of Actinobacteria were increased in MDD subjects
compared with healthy controls, while those of Bacteroidetes were decreased (Zheng et al., 2016). In contrast to the conclusion by Huang et al. (2018), which highlights the importance of
Firmicutes, no significant differences were found in the overall abundances of Firmicutes by
Zheng et al. (2016). The researchers argued this is due to the fact that some members of the
Firmicutes OTUs were increased, while others decreased in MDD. This complements the
findings by Naseribafrouei and colleagues (2014), who also indicated that closely related OTUs have opposing correlations. Zheng and colleagues did not elaborate on any significant differences at lower taxonomy levels.
In another study, the microbiome of 34 depressed patients was compared to 33 healthy controls matched for gender, age, and ethnicity (Kelly et al., 2016). Although there were significant differences in beta diversity between healthy and depressed groups, a PCoA analysis was unable to separate the groups (Kelly et al., 2016). Furthermore, at the phylum level, there were no statistically significant differences in the relative abundances between the groups. However, at the family level, the relative proportions of Prevotellaceae in depression were decreased, whereas Thermoanaerobacteriaceae was increased. Statistics based on random permutations of the redundancy analysis (RDA) showed that the depressed group is
significantly separated at the genus level from the control group (Kelly et al., 2016). The relative proportions of genera Eggerthella, Holdemania, Gelria, Turicibacter, Paraprevotella, and Anaerofilum were increased in the depressed group, whereas Prevotella and Dialister were decreased. This finding conflicts with Naseribafrouei and colleagues, as they found that binning at the genus level gave poor classification (2014). It is unclear what this difference is due to. Nevertheless, it can be hypothesized that this discrepancy might be caused by a difference in the control group, both in size and nature. Naseribafrouei et al. had 18 controls with diffuse neurological symptoms (2014), while Kelly et al. used 33 controls who were matched for gender, age, and ethnicity (2016).
To continue with another study, Lin and colleagues detected more phylum Firmicutes in MDD patients than in control subjects, and fewer Bacteroidetes (2017). The current study also illustrated significant differences in four genera Prevotella, Klebsiella, Streptococcus, and
Clostridium XI between the patient and the control group. Interestingly, the researchers link
these differences in genera levels to the function of these bacteria and theorize that increased bacterial translocation via these genera may play a role in the pathophysiology of depression.
Similarly to Lin et al. (2017), Chung and colleagues found that Firmicutes increased while Bacteroidetes decreased in 36 MDD patients in comparison to 37 controls (2019). At the phylum level, they also observed an increase in Actinobacteria and a decrease in
Proteobacteria. Additionally, they found 7 differences at the familial level, among which an
overrepresentation of Lachnospiraceae in depression, and an underrepresentation of
Prevotellaceae. At the genus level, numerous genera were increased in depression (such as
Bifidobacterium, Eggerthella, Holdemania, Parabacteroides, Ruminococcus &
Streptococcus), while others were decreased (such as Prevotella & Meganomas). Interestingly,
these researchers inquired about the diet of their participants, with results indicating that depressed patients had a significantly lower fat intake compared to controls (Chung et al., 2019). The aforementioned changes in microbiome composition are after an adjustment for fat intake, which makes the research of Chung et al. unique, as they have solely revealed microbiota changes for depression that are independent of fat intake. Another intriguing aspect of this study is the finding that a few of the significantly altered genera in depression are correlated to a severity score (as measure by Beck Depression Inventory, BDI; Chung et al., 2019). To be exact, the family Peptostreptococcaceae showed moderate correlations with BDI score, with similar trends in patients and controls (Chung et al., 2019). At genus level, Blautia
Lastly, in one study, researchers wanted to specifically investigate relatively minor bacteria Bifidobacterium and Lactobacillus, as they argue these bacteria might suffer from poor quantification accuracy in some of the studies that have been mentioned above (Aizawa et al., 2016). In this cohort of 43 patients and 57 controls, it was found that depressed patients had reduced Bifidobacterium rates and tended to have lower Lactobacillus counts.
Compare and Align Results
The results from eight studies that have sequenced the human gut microbiome have been described in detail above. Table 1 recapitulates the findings with an overview on phylum, order, family, and genus levels. What stands out most is the diverse and contradictory nature of the different findings. It is not easy to reconcile the variations seen across the studies, as there is limited consensus between them in relation to the type and extent of observed changes.
At almost all taxonomic levels partial confirmation and contradiction between the studies are evident. For example, three studies find increased phylum Bacteroidetes in depressed subjects (Chung et al., 2019; Jiang et al., 2015; Naseribafrouei et al., 2014), whereas one reports a decrease in this phylum (Lin et al., 2017). Jiang et al. state they found less
Firmicutes in depression (2015), while two other studies report more (Chung et al., 2019; Lin et al., 2017). Similarly, three studies report lower abundances of the family Lachnospiraceae
(Huang et al., 2018; Jiang et al., 2015; Naseribafrouei et al., 2014), while two others found higher abundances (Chung et al., 2019; Zheng et al., 2016). The genus Alistipes was more prominent in depression in two studies (Jiang et al., 2015; Naseribafrouei et al., 2014), but less prominent according to another (Zheng et al., 2016). Likewise, fewer Prevotella (Chung et al., 2019; Jiang et al., 2015; Kelly et al., 2016), as well as more has been linked to depression (Lin
et al., 2017). Of course, some findings are not contradicted, such as a decrease in the genus Faecalibacterium (Jiang et al., 2015; Zheng et al., 2016) and less of family Prevotellaceae
(Chung et al., 2019; active depression in Jiang et al., 2015; Kelly et al., 2016). Lastly, there are many genera or families that were linked to abnormal abundance in depression in only one of the studies.
It is unclear what underlies the diversity in findings in the human gut composition. However, the inconsistencies may be reflective of methodological inconsistencies and/or confounding factors. The next section will argue how methodological choices for experimental procedures, sample selection and bioinformatics analyses can influence results and may therefore have contributed greatly to the observed heterogeneity in results.
Methodological & Sample Differences
The researchers from the eight aforementioned studies have selected related but distinct methodologies. This section will describe how the different choices may be a source of bias and could cause inconsistencies between the studies. For this, six methodological choices will be covered, for an overview see Table 2. Next, sample differences will be investigated. Methodological Discrepancies
Firstly, all studies used 16S rRNA gene sequencing, but with different region specification: region V1– V3 (Jiang et al., 2015), region V3–V4 (Chung et al., 2019; Huang et
al., 2018; Lin et al., 2017), region V3–V5 (Zheng et al., 2016), or no specified region (Aizawa
et al., 2016; Kelly et al., 2016; Naseribafrouei et al., 2014). It has been found that sequencing
data is affected by the region of the 16S rRNA gene that is targeted (Fouhy, Clooney, Stanton, Claesson & Cotter, 2016; Haas et al., 2011). In other words, some regions provide more accurate estimates than others (Kim, Morrison & Yu, 2011). Evidently, targeting various regions across different studies may therefore seriously limit comparability, as well as potentially explain discrepancies.
Table 2
The methodology of the eight reviewed studies on human microbiome sequencing
16S Region Extraction kit OTU Similarity Threshold
OTU Picking Database for classification
Pipeline analysis
Naseribafrouei
et al. Unspecified MagTM mini kit 99% Closed-reference RDP QIIME Jiang et al. V1–V3 QIAamp DNA
Stool Mini Kit 97% Unspecified RDP Mothur
Zheng et al V3–V5 PowerSoil DNA Kit
97% Unspecified RDP Mothur
Kelly et al Unspecified QIAamp DNA
Stool Mini Kit 97% Unspecified BLAST classifier SILVA v.111
QIIME
Lin et al. V3-V4 Tiagen DNA
Stool Mini Kit
97% Unspecified Bayesion
classifier Silva v.119
Mothur Aizawa et al. Unspecified Unspecified Unspecified Unspecified Unspecified Unspecified
Huang et al. V3-V4 PowerSoil
DNA Kit
97% Open-reference UCLUST
GreenGenes
QIIME
Chung et al. V3-V4 QIAamp DNA
Stool Mini Kit
Secondly, laboratory kits used to extract DNA from the sample have been indicated to influence results as well (Kennedy et al., 2014). Different extraction methods and kits are linked to a different microbiome composition and community structure (Wesolowska-Andersen et al., 2014). Five different kits were used in the studies covered in this paper, which could pose a serious problem for the comparability of the results. It should be noted that Fouhy and colleagues found that the extraction method had a lesser effect on over microbiome composition than primer choice, with samples extracted using different extraction procedures, but amplified with the same primers yielding similar results (2016). Nevertheless, it has been argued that a standardized DNA extraction method should be established in order to improve the comparability of human gut microbiome studies (Costea et al., 2017).
Thirdly, different methods and thresholds for OTU binning were applied. The OTU similarity threshold was 97% in most studies (Chung et al., 2019; Huang et al., 2018; Jiang et al., 2015; Kelly et al, 2014; Lin et al., 2017; Zheng et al., 2016), in one study the threshold was unspecified (Aizawa et al., 2016), and in another, it was 99% (Naseribafrouei et al., 2014). It has been shown that different threshold levels significantly affect which OTU’s are formed (Chen, Zhang, Cheng, Zhang & Zhao, 2013), and from there on seriously affect downstream results. Furthermore, the threshold can change the degree to which OTU’s are monophyletic (meaning “one of a kind”, that all members of the OTU share a single common ancestor; Koeppel & Wu, 2013). Additionally, the way OTUs are selected (de novo, open- or closed-reference) is also an important determinant (Westcott & Schloss, 2015), yet all but two studies did not specify which method they used (Aizawa et al., 2016; Chung et al., 2019; Jiang et al., 2015; Kelly et al, 2014; Lin et al., 2017; Zheng et al., 2016). This is a serious limitation for comparability as well as reproducibility.
Next, the type of classifier is considered. Three of the studies rely on the Ribosomal Database Project (RDP) database (Jiang et al., 2015; Naseribafrouei et al., 2014; Zheng et al., 2016). Two studies operated GreenGenes (Chung et al., 2019; Huang et al., 2018), while Lin et al. use SILVA with a Bayesian classifier (2017), and Kelly et al. use SILVA with BLAST (2014). One major drawback of two of these approaches (RDP and Bayesian classifier) is that they rely on Naïve Bayes for the assigning of taxonomy. Naïve Bayes works on the assumption that all genera are equally probable (Wang, Garrity, Tiedje & Cole, 2007). However, it is known that the gut microbiota of adults is dominated primarily by members of the Bacteroidetes and Firmicutes phyla, representing approximately 90% of the adult microbiota
(Tremaroli & Backhed, 2012). Considering this fact, it seems somewhat ignorant to use a classification model that does not take this prior probability into account. Additionally, the overall classification accuracy with Naïve Bayes is very different at the genus level based on the sequence length (ranging from only 50% for 50 base pairs to 90% for 400 base pairs; Wang et al., 2007). Besides the clear limitation for correct classification, this might also intensify divergence between studies that use different variable regions. Hypervariable regions are of different lengths, which means an additional difference in classification might arise between studies simply because one study chose a longer variable region.
Lastly, it is important to consider the pipeline used in the different studies. Roy and colleagues define a pipeline as bioinformatics algorithms executed in a predefined sequence to process NGS data (2018). Four studies used Quantitative Insights Into Microbial Ecology (QIIME; Chung et al., 2019; Huang et al., 2018; Kelly et al., 2014; Naseribafrouei et al., 2014), and three used Mothur (Jiang et al., 2015; Lin et al., 2017; Zheng et al., 2016). It has been indicated that pipelines such as QIIME and Mothur can produce significantly different results (Siegwald et al., 2017).
In sum, multiple steps in methodological set-up and data analysis have been examined. At each step, researchers from the eight studies have made distinct choices, making each step susceptible to different biases and errors. This highlights that results should be interpreted in the context of a combination of factors, including DNA extraction procedure, primer choice, and sequencing platform (Fouhy et al., 2016).
Sample Differences
In addition to the methodological differences, it is important to consider the samples in different studies. For an overview of the factors that will be discussed, see Table 3. First and foremost, the inclusion and exclusion criteria for the studies varied. All but two studies used a diagnosis based on the DSM IV (Aizawa et al., 2016; Jiang et al., 2015; Kelly et al., 2014; Lin et al., 2017; Zheng et al., 2016) or DSM V (Chung et al., 2019). However, some included an additional minimum score on the Hamilton Depression Rate Scale (HAMD) of equal to/higher than 17 (Kelly et al., 2014), 20 (Jiang et al., 2015), or 23 (Lin et al., 2017). As the HAMD measures severity of depression, this could mean that the studies investigated different subsamples within the depressed populations, each with a distinct phenotype. While these
phenotypes likely shared (some) overlap, it is nevertheless important to be mindful of this and acknowledge it might explain part of the inconsistencies between the studies.
Table 2
The sample characteristics of the eight reviewed studies on human microbiome sequencing
Diagnosis tool % of participants with IBS Exclusion based on Antibiotics Antidepressant Geographical/cultural context Controlling for Diet Naseribafrouei
et al. ICD-10 Unspecified No exclusion 6% of controls 73% of patients Unspecified type Norway NA Jiang et al. DSM IV HAMD > 20 0% Exclusion if 1 month prior 0% of controls A-MDD: 72% SSRIs or SNRIs 24 % Atypical antipsychotic R-MDD: 100% on SSRIs or SNRI, 29% on atypical antipsychotic China NA
Zheng et al DSM IV Unspecified Exclusion if currently taking 0% of controls 33% of MDD various anti-depressants China NA Kelly et al DSM IV HAMD > 17 0% Exclusion if 1 month prior 0 % of controls 100 % of patients on SSRIs Ireland Macronutrients via food-frequency questionnaire (FFQ) Lin et al. DSM IV HAMD > 23 Unspecified in patients* 0% in controls* Exclusion if 1 month prior 0% of controls 100% patients on standard dose of 10 mg/day SSRIs China “subject selection strategies including diet control” – unspecified Aizawa et al. DSM IV 33% in patients 12% in controls Exclusion if 1 “recent” -Unspecified 0% of controls 65% of patients Unspecified type Japan NA
Huang et al. ICD-10 0% Exclusion if
6 months prior 0% China Exclude if dietary changes 6 months prior
Chung et al. DSM-V 0% Exclusion if
1 month prior 0% of controls 86% of patients Unspecified type Taiwan Macronutrients via food-frequency questionnaire (FFQ)
Note. *Lin and colleagues (2017) did not specify whether or to what extent IBS was present in patients. For controls they
Concerning exclusion criteria, some researchers were relatively strict, others more partial. Important to note is that only four research groups excluded those with irritable bowel syndrome (IBS) or inflammatory bowel disease (IBD; Chung et al., 2019; Huang et al., 2018; Jiang et al., 2015; Kelly et al., 2014). In contrast, three other studies did not exclude those with IBS or IBD (Aizawa et al., 2016; Naseribafrouei et al., 2014; Zheng et al., 2016), and Lin et al. only excluded controls with gut problems (2017). Only Aizawa et al. reported the frequency of IBS in their sample (33% in patients versus 12% in controls; 2016), for the other studies it is unclear to what extent IBS is present in their sample. This lack of information is relevant, as it’s been shown that IBS and IBD are correlated with distinct microbiome compositions (Casen et al., 2015; Lo Presti et al., 2019). Furthermore, comorbidity has been observed between depression and IBS and IBD (Mikocka-Walus, Knowles, Keefer & Graff, 2016). The combination of these two observations suggests an increased probability that part of the samples had IBS or IBD, and that some of the observed difference between depressed patients and controls may be due to IBS or IBD, rather than depression. Furthermore, differences in IBS or IBD frequencies in different samples might be able to explain the inconsistencies between studies. However, it is difficult to theorize to what extent this was the case, as it has not been reported by three studies if and how many of their subjects suffer from IBS and IBD (Lin et al., 2017; Naseribafrouei et al., 2014; Zheng et al., 2016).
Similarly, a difference in exclusion criteria can be seen regarding antibiotic and (manufactured) probiotic use. Some studies excluded those that used anti- or probiotics in the month prior to the study (Chung et al., 2019; Jiang et al., 2015; Lin et al., 2017), or the six months prior (Huang et al., 2018). One study excluded those that were currently taking anti- or probiotics (Zheng et al., 2016). Aizawa et al. excluded individuals with “recent” antibiotic use (but did not define “recent”), but not those with probiotic use (2016), while Naseribafrouei et al. did not take antibiotics nor probiotic consumption into account at all (2014). It has been shown that antibiotic treatment can have profound and rapid effects on the human microbiome, including a decrease in diversity (Raymond et al., 2016). After a week, the microbiome tends to recover, but this return can often be incomplete (Dethlefsen & Relman, 2011). Therefore, comparing studies with potential recent antibiotic use with those who have not can be problematic. Additionally, the exclusion of those who used antibiotics in the last month seems like a strong methodological choice to prevent confounding effects. A similar line of reasoning can be used for artificial prebiotic, as it recently has been suggested that prebiotics can improve dysbiosis and its inflammatory-associated state (Paiva, Duarte-Silva & Peixoto, 2020).
Another important difference within the samples is the use of antidepressant medication. Only three studies presented a depressed sample with a homogenous antidepressant profile: in one study no antidepressant medication was used at all (Huang et al., 2018), and in two studies all patients used SSRIs (Kelly et al., 2014; Lin et al., 2017). In Lin et al. dose was even standardized to 10 mg/day (2017). In the remaining studies, a heterogenous antidepressant profile was present, in which the extent of antidepressant intake ranges from one-third (Zheng et al., 2016) to 86% of depressed subjects taking antidepressants (Chung et
al., 2019). The most common antidepressant medication form was SSRIs, but especially Jiang
et al. report a significant proportion of atypical antipsychotic drugs (2015). Three studies did not report on the type of antidepressant used (Aizawa et al., 2016; Chung et al., 2019; Naseribafrouei et al., 2014). Although both Aizawa et al. (2016) and Zheng et al. (2016) found that antidepressant use did not significantly influence results in their sample, a recent mouse study has indicated that various types of antidepressants can alter gut microbiome composition and richness (Lukić et al., 2019). This finding suggests that differences in antidepressant may have influenced results in some studies, and that differences between samples hinder comparability. Further studies are required to more comprehensively evaluate antidepressant effects on specific microbiota targets. Until research provides a clearer idea on if and how antidepressant might affect gut composition, comparison of studies with different antidepressant profiles should occur with caution.
Next, the geographical and cultural context of the studies must be considered. Four of the studies took place in China (Huang et al., 2018; Jiang et al., 2015; Lin et al., 2017; Zheng et al., 2016), one in Japan (Aizawa et al., 2016), one in Taiwan (Chung et al., 2019), one in Norway (Naseribafrouei et al., 2014) and lastly one in Ireland (Kelly et al., 2014). A recent study on microbiome composition in different Chinese districts found that microbiome variation is overwhelmingly explained by one’s geographical location (He et al., 2018). Furthermore, these researchers describe a need for localized models, and indicate that one model for multiple geographical locations might function poorly. In addition to geographical location, subtle but significant differences in the microbiome have been linked to ethnicity
(Brooks, Priya, Blekhman, & Bordenstein, 2018). However, this sample consisted of people
throughout the entire USA. Especially interesting in this regard is the research conducted by Deschasaux and colleagues, as they investigated six ethnic groups living in one urban area (2018). With this experimental set-up, they were able to isolate the effect of ethnicity. Similarly
explanation of the interindividual (dis)similarities in gut microbiota composition, indicating that ethnicity strongly determines specific gut taxa (2018). Additionally, they found that individuals tend to have similar microbiome characteristics with others of a different ethnic background when they live in the same city (Deschasaux et al., 2018). In conclusion, geographical location and ethnicity are important to consider in microbiome research and caution is warranted when comparing findings on the microbiome from geographically and ethnically distinct groups.
Lastly, dietary intake has been linked to changes in the microbiome composition (De Filippo et al., 2010; Flint, Scott, Louis & Duncan, 2012). Most of the studies did not control for the diet of their participants (Aizawa et al., 2016; Jiang et al., 2015; Naseribafrouei et al., 2014; Zheng et al., 2016). Lin et al. wrote they used “subject selection strategies including diet control” but did not define what this diet control entails (2017). Similarly, Huang et al. excluded those with a dietary change in the last six months but did not specify further (2018). Within the studies that did not control for diet, it was argued multiple times that the subjects in a study were culturally homogenous and most likely had a similar (“traditional Norwegian” or “typically Chinese”) diet and that diet, therefore, most likely did not affect results (Naseribafrouei et al., 2014; Zheng et al., 2016). As none of the studies inquired on the average diet of their participants, it is hard to estimate to what extent this is a reasonable assumption. Even if one assumes equal diet-related “noise” within one sample, it is evident this can be problematic when comparing results between the samples. With a vast divergence in diet between the studies, it is hard to distinguish whether similarities or differences are (partly) due to depression or diet.
Kelly et al. (2014) and Chung et al. (2019) were the only ones to check an aspect of diet between the depressed and control group. To be precise, Kelly and colleagues assessed daily macronutrient consumption and established that apart from Trans fats, consumption was similar in depressed and control groups (2014). However, they only compared the average of the two groups and neglected to use the information on individual consumption in further analysis. In this sense the information was used predominantly as a data check. It could have been interesting to use the individual macronutrient consumption as a predictor for microbiome composition. In contrast to Kelly et al., Chung and colleagues found dissimilarities in diet composition between patients and healthy controls in carbohydrates, fat, short chain fatty acid and medium chain fatty acid intake (2019). The researcher then selected fat intake as a dietary
covariate, as they argued its relevance for changes in microbiota compositions. The fact that
these researchers controlled for fat intake adds to their findings and is a step forward compared to the other studies. However, refinement of analysis to incorporate more food items and include cluster-based methods to capture a broader aspect of diet is advised in the future.
Conclusion Based on Cohort Studies
In the previous section, multiple factors were discussed that could have influenced each study distinctively. Collectively, these factors make it difficult to distinguish the extent of “noise” within which the depression and microbiome signal were to be detected. Perhaps differences in the aforementioned factors (and subsequent differences in noise) drive the discrepancies between the studies. With the aforementioned limitation in mind, the next paragraphs will theorize about five findings. These findings (for example, a reduction of a particular genus in depressed individuals) were confirmed by at least one other study, and not contradicted by any other study.
Firstly, a decrease in the genus Faecalibacterium was observed in depressed patients (Jiang et al., 2015; Zheng et al., 2016). This is especially interesting, as Faecalibacterium has been shown to produce butyrate (Louis & Flint, 2009). A decrease in Faecalibacterium can therefore contribute to a reduction in butyrate concentration, which might lead to greater intestinal permeability. This in turn may allow neuroactive molecules to enter systemic circulation. This presents an idea that adequate levels of Faecalibacterium might hold a beneficial effect. Indeed, a recent study has shown that administering oral supplementation of a mixture of acetate, propionate and butyrate in drinking water can reduce anxiety- and depressive-like behavior in mice, as well as ameliorate hyperactivity in the HPA axis and intestinal permeability (van de Wouw et al., 2018). The use of a mixture of SCFAs is certainly a limitation, nevertheless this study provides an interesting idea that SCFAs are ultimately linked to behavior in mice. Intriguingly, in humans one study reported that the genus
Faecalibacterium is negatively correlated with depressive severity (Jiang et al., 2015), though
it should be noted that they only used scores from the depressed (and not control) participants for this analysis. In a large population human study, an association was found between bacteria that produce SCFA (Faecalibacterium and Coprococcus) and higher quality of life indicators (Valles-Colomer et al., 2019). However, not all research has supported this, for example, Kelly et al. did not find increased fecal SCFA levels in depressed patients (2016). However, this
finding may be explained by the lack of significant differences in SCFA-producing bacteria between depressed and healthy controls in their cohort.
Secondly, a reduction of the family Prevotellaceae was present in three studies (Chung
et al., 2019; active depression in Jiang et al., 2015; Kelly et al., 2016). Considering Prevotellaceae have been linked to increased mucosal inflammation (Iljazovic et al., 2020),
this is quite a surprising finding. To the author’s knowledge there is currently no theory that can explain the observed reduction in Prevotellaceae and its role in depression. This topic may thus require future research. Thirdly, two studies found a decrease in Dialister (Jiang et al., 2015; Kelly et al., 2016). This finding is complemented by the Flemish gut project that found that Dialister was positively associated with Quality-of-Life scores (Valles-Colomer et al., 2019).Fourthly, some studies observed an increase in Clostridium (Chung et al., 2019; active depression Jiang et al., 2015; Lin et al., 2017). Interestingly, Jiang and colleagues found that
Clostridium was negatively correlated with the serum Brain-derived neurotrophic factor
(BDNF) level (2015). BDNF plays an important role in the maintenance and survival of neurons and in synaptic plasticity and several lines of evidence suggest that BDNF is involved in depression (Dwivedi, 2009). BDNF levels are found to be significantly lower in depressed patients than in controls and negatively correlated to severity scores (Karege et al., 2002). Furthermore, antidepressants successfully increase the attenuated BDNF levels in depressed patients (Gonul et al., 2005). Collectively, an increase in Clostridium might be related to the observed lower levels of BDNF in depressed individuals. Future research should investigate the link between the Clostridium and BDNF more thoroughly in order to assess its role in depression and the MGB axis.
Lastly, an increase in the genus Streptococcus was found in depressed patients (Chung
et al., 2019; Lin et al., 2017). Streptococcus has been linked to neurotoxins and may therefore
cause inflammation (Li et al., 2017). Additionally, Streptococcus has been associated with γ-aminobutyric acid (GABA) production (Yang et al., 2008). In combination with evidence that suggests that central nervous system GABA concentrations are reduced in patients with MDD (Pehrson & Sanchez, 2015), an increase in GABA-producing bacteria is surprising. However, it is unclear what effects local differences in GABA production would have on systemic concentrations. Additional research might be able to elucidate the relation between
