Psychobiotics and their possible effects on depression and anxiety through the gut-brain connection

The systematic search for systematic reviews, as well as the construction of the present review, was partially conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [56] and fully assessed for methodological quality using the AMSTAR 2 tool (a measurement tool to assess systematic reviews). This study was registered in the PROSPERO database under the number CRD420251164884.

Only systematic reviews with meta-analyses of randomized controlled trials (RCTs) were eligible for inclusion. To be included, the trials had to investigate the effects of psychobiotic interventions, such as probiotics, prebiotics, or synbiotics, administered with the aim of improving symptoms of depression and/or anxiety and/or stress. Secondarily, changes in biological markers, such as neuroinflammatory indicators, were also examined. Only studies conducted in humans and published in English were considered.

The following studies were excluded: (1) observational studies, standalone systematic reviews, case reports, and preclinical studies involving animals or in vitro models; (2) studies assessing gut microbiota modulation without considering depression or anxiety symptoms as primary or secondary outcomes; (3) studies using antibiotics or fecal microbiota transplantation as the main intervention; (4) studies including samples with other psychiatric disorders, such as schizophrenia, psychosis, attention-deficit/hyperactivity disorder, or post-traumatic stress disorder, as the primary focus of investigation; and (5) studies focusing on special conditions, such as pregnancy or the postpartum period, were also excluded.

Systematic reviews with meta-analyses that investigated individuals diagnosed with depressive and/or anxiety disorders were included, based on standardized diagnostic criteria, such as those outlined in the Diagnostic and Statistical Manual of Mental Disorders or the International Classification of Diseases. However, meta-analyses that included mixed populations, provided they also enrolled participants diagnosed with depression and/or anxiety, were retained but classified as secondary evidence and did not impact the primary analysis. This category included studies involving healthy individuals, students, or clinical populations with systemic or mental comorbidities not specifically related to mood or anxiety disorders or clinically affected by depression and anxiety. Additionally, only studies involving adults were considered. No restrictions were applied regarding sex, ethnicity, or disease duration.

Between May and June 2025, two reviewers (JOF and ASF) independently conducted a comprehensive literature search using the PubMed/MEDLINE, Web of Science, Scopus, Scielo, EBSCO, and Cochrane Database electronic databases. The search was restricted to studies involving human subjects and published in English up to September 2025. The reference lists of relevant studies (other sources) were also manually examined to identify additional eligible articles. The complete search strategy combined terms using Boolean operators across the different databases, as presented in Table S1 (Supplementary Materials). Additionally, the file related to the extraction of all databases was deposited in The Open Science Framework—OSF (https://osf.io/fq5c9/overview↗; access on 7 January 2026).

In general, we combined the following keywords, which varied in their inclusion format depending on the database: (psychobiotics OR probiotics OR prebiotics OR synbiotics) AND (depressive OR depressive disorder OR anxiety) AND meta-analysis.

The primary outcomes of this review included changes in symptoms of depression and/or anxiety, assessed through validated psychometric instruments. All scales available in the different meta-analyses were included, such as, the Beck Depression Inventory (BDI), the Depression, Anxiety and Stress Scales (DASS), the Hospital Anxiety and Depression Scale (HADS), the Montgomery–Åsberg Depression Rating Scale (MADRS), the Hamilton Depression Rating Scale (HAM-D), the Patient Health Questionnaire-9 (PHQ-9), the Beck Anxiety Inventory (BAI), the State–Trait Anxiety Inventory (STAI), the Perceived Stress Scale (PSS), the Hamilton Anxiety Rating Scale (HAM-A), and the Generalized Anxiety Disorder-7 (GAD-7). Changes in these scores from baseline to post-intervention were analyzed as indicators of improvement in mood and anxiety symptoms.

As secondary outcomes, biological markers and physiological mechanisms related to gut–brain axis modulation were considered, aiming to explore the biological basis of psychobiotic effects, although they were not included in the statistical analysis. These markers included the following: (a) biomarkers related to neurotransmitters, such as serum or plasma levels of serotonin, dopamine, gamma-aminobutyric acid (GABA), and noradrenaline, as well as tryptophan concentration and the kynurenine/tryptophan ratio; (b) metabolites produced by the gut microbiota, particularly short-chain fatty acids (SCFAs), including butyrate, propionate, and acetate; (c) inflammatory and immunological markers, such as interleukin-6 (IL-6), tumor necrosis factor alpha (TNF-α), C-reactive protein (CRP), and interleukin-10 (IL-10); (d) endocrine responses to stress, especially cortisol levels (in saliva, serum, or urine) and measures of hypothalamic–pituitary–adrenal (HPA) axis regulation; and, when available, data on gut microbiota composition.

In addition to the traditional methodological variables, an expanded set of bibliometric and psychometric characteristics was systematically extracted to map temporal, geographical, and conceptual patterns across the included studies.

Information regarding the geographical origin of each study (country and continent of conduct) was recorded to identify the spatial distribution of scientific production and potential regional asymmetries. Subsequently, all studies were organized according to their chronological order of publication, from which a timeline was constructed to illustrate the historical evolution of the evidence and periods of increased or reduced research activity. To characterize the conceptual structure of the literature, a citation-based connectivity map was generated, capturing the structural relationships among the included studies, revealing variations in citation density and the strength of inter-study links. This mapping approach elucidates how the evidence base is organized, identifying central nodes (hubs) and more peripheral contributions to the progression of research on psychobiotics.

Additionally, standardized extraction of the psychometric scales used in each study was performed, covering instruments for the assessment of depression, anxiety, stress, quality of life, clinical symptoms, and other psychological domains. All scales were categorized according to their purpose, psychometric properties, and frequency of use, enabling comparative synthesis across instruments and identification of methodological convergences among the included studies.

The methodological quality of the included systematic reviews was assessed using the AMSTAR 2 instrument. Two assessors independently carried out this process (JOF and ASF). AMSTAR 2 comprises 16 items addressing key methodological aspects such as the presence of a pre-registered protocol, adequacy of inclusion criteria, search strategy, assessment of risk of bias in primary studies, and appropriateness of statistical analyses. This tool does not generate a final numerical score but rather classifies the overall confidence in the review's findings as high, moderate, low, or critically low, allowing for a more qualitative interpretation of the methodological robustness of each evaluated study [57].

The screening of titles, abstracts, and full texts was independently performed by two reviewers, strictly following the predefined inclusion and exclusion criteria. To assess the reliability of the selection process, Cohen's Kappa coefficient was calculated, which measures the degree of agreement beyond that expected by chance. The coefficient was obtained from the proportion of observed agreement (P_o) and the proportion of expected agreement by chance (P_e) according to the following formula:κ = Po−Pe1−Pe

This procedure is recommended in systematic reviews and meta-analyses, as it ensures greater transparency and reproducibility in study selection while minimizing potential biases derived from subjective judgments.

Statistical analyses were conducted by extracting effect sizes as SMDs accompanied by their 95% confidence intervals (95%CI). Between-study variability was quantified using the I² statistic. The overall pooled effect was estimated using a random-effects model. Additionally, sensitivity analyses were carried out to assess the robustness of the findings by sequentially excluding studies judged to have a high variability. In this procedure, each meta-analytic estimate was sequentially removed, and the random-effects model was recalculated to determine whether any single study exerted disproportionate influence on the summary effect. Stability of the pooled SMD across iterations was taken as evidence that the findings were not driven by isolated data points or outlier effect sizes.

To assess the robustness of the pooled SMD, a leave-one-out sensitivity analysis was performed, in which the random-effects model was recalculated sequentially after removing each meta-analysis individually.

Cochran's Q test was used to assess the presence of heterogeneity, while the magnitude of inconsistency was quantified using the I² statistic, interpreted according to established thresholds (25% low, 50% moderate, and 75% high heterogeneity). In addition, the between-study variance (τ²) was estimated using the DerSimonian–Laird method to account for the dispersion of true effect sizes across studies. Because substantial heterogeneity was expected due to differences in population characteristics, psychometric scales, intervention duration, probiotic strains, and study design, all pooled estimates were computed using random-effects models.

The systematic search initially identified 529 articles related to the topic. After removing duplicates and reading by title and abstract, the screening observed 55 systematic reviews with meta-analyses across the selected databases and additional sources. Of these, 22 were excluded for not meeting the predefined inclusion criteria (e.g., scope outside the research question or incompatible population), leaving 33 articles for full-text assessment. In the end, only 30 systematic reviews with meta-analyses remained. A list of the included articles was presented separately, with the respective databases from which they were extracted in Table S2 (Supplementary Materials). Figure 1 details the inclusion and exclusion process for the selected articles. The Supplementary Materials (Table S3) detail the reasons for exclusion.

During the screening process, a high level of consistency was observed between the two independent reviewers. Cohen's Kappa coefficient was 0.80, a value corresponding to substantial to almost perfect agreement according to the criteria of Landis and Koch. This result confirms the robustness of the selection process and minimizes the risk of inappropriate inclusion or exclusion of studies in the present meta-analysis.

The included meta-analyses showed considerable methodological heterogeneity, both regarding search strategies and eligibility criteria, population, and strains used in the intervention, as well as in relation to the outcomes analyzed. The number of primary studies included within each meta-analysis ranged from 5 to 62, with pooled sample sizes varying from 365 to 5245 participants. Most of the reviews were published after 2020, in journals indexed primarily in the PubMed and Web of Science databases (22 meta-analyses; 66.6%), simultaneously, and the investigated populations were characterized by a diagnosis of mood and/or anxiety disorder, or healthy participants. However, a few studies added analyses for patients with other diagnoses. In our final analysis, we discarded the extraction of groups with varied diagnoses, focusing primarily on aspects of depression and anxiety.

North America accounted for three meta-analyses (9.1%), with studies from the United States [31,32] and Canada [39].

South America contributed one meta-analysis (3.0%), represented by Brazil [42], whereas in Europe contributed seven meta-analyses (21.2%), which were conducted in France [16], the United Kingdom [34,40,41,48], Germany [43], and Poland [35].

Asia showed the highest concentration of research, with 20 meta-analyses (60.6%), including China [27,28,30,36,37,44,46,50,52,54,58], Singapore [33], Malaysia [17], and Korea [29]. Finally, the Middle East contributed six meta-analyses (18.2%), from Iran [38,47,51,53,59] and Saudi Arabia [49]. Figure 2 presents the timeline of scientific production. Figure 3 presents the global distribution of the studies included in the review.

This map illustrates the geographical distribution of all clinical trials included in the systematic review. The studies were conducted across North America, South America, Europe, Asia, Africa, and Oceania, highlighting the multinational and heterogeneous nature of the available evidence.

Figure 4 presents the citation-based connectivity map of the meta-analyses included in this review. The visualization illustrates the extent to which individual papers share similar reference lists, which reflect patterns of methodological overlap rather than scientific influence in a causal sense. Accordingly, high citation density in this context indicates that two papers draw on a comparable pool of primary studies, often because they employed similar search strategies or targeted similar publication periods.

From this perspective, the high connectivity observed for papers such as Liu et al. [32] and Huang et al. [28] should not be over-interpreted as evidence of conceptual influence or foundational status. Instead, their large number of shared references likely reflects the fact that these reviews included broad search windows and extensive primary-study lists, increasing the probability of overlap with other meta-analyses. So, high connectivity among more recent papers may indicate redundancy; when a new meta-analysis cites nearly the same set of primary studies already synthesized in earlier work, this suggests that its analytical contribution may be limited.

In contrast, more peripheral papers (e.g., Zhang et al. [27], Zhao et al. [50], and Cheng et al. [52]) that display low inter-study connectivity may be incorporating newer datasets, diverging populations, or more selective inclusion criteria—features that can enhance their added value despite having fewer shared citations.

Thus, our citation-based connectivity map provides a structural overview of how similarly (or differently) the included meta-analyses assembled their evidence bases. This distinction is important to avoid conflating methodological overlap with scientific impact.

The map reveals a clear hierarchical structure: a small number of highly cited, conceptually influential studies anchor the network, intermediate studies provide bridging links, and emerging publications expand the field in more specialized directions.

The included reviews employed a wide range of psychometric instruments, reflecting the conceptual and methodological heterogeneity of the field. The scales identified encompassed assessments of depressive symptoms, anxiety, perceived stress, mood, general psychological distress, and well-being. A predominance of traditional measures of depression and emotional disorders was observed, alongside instruments tailored to specific clinical populations or sensitive periods, such as pregnancy and the postpartum phase. This diversity highlights both the complexity of the constructs under investigation and the lack of standardization across studies, which challenges direct comparisons and underscores the need for analytical approaches that account for conceptual and sensitivity differences among instruments (Figure 5).

Finally, according to the initial extraction plan, we present the number of participants, age and population status, intervention (mono-strain or multi-strain), and duration, as well as the outcomes reported. General characteristics of the 33 eligible meta-analyses are summarized in Table 1.

Probiotics accounted for the largest proportion, with 30 studies (100%) examining their effects on psychological outcomes [16,17,27,28,29,30,31,32,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54]. Prebiotics represented seven studies (23.3%), reflecting a smaller body of evidence focusing on isolated modulation of fermentable fibers [27,32,39,41,45,51,53]. Synbiotics were evaluated in four studies (13.3%), indicating more limited interventions [27,39,45,51]. Although several studies appeared in more than one category due to overlapping formulations or multifactorial designs, these proportions provide a clear depiction of the relative emphasis placed on each intervention type within the current evidence base.

The methodological appraisal using the AMSTAR 2 framework revealed substantial variability in the confidence attributed to the included systematic reviews and meta-analyses, highlighting marked heterogeneity in their overall robustness. Among the 31 reviews assessed, only 12 (38.7%) achieved a rating of high confidence [16,17,27,28,30,31,33,39,44,48,53,54] characterized by the absence of critical flaws and by strong adherence to core methodological standards, particularly comprehensive search strategies, rigorous risk-of-bias assessments, transparent synthesis protocols, and registration in the PROSPERO database or similar. We highlight that, despite the generally high adherence to methodological quality criteria and the absence of critical flaws, some reviews exhibited shortcomings in adequately reporting specific AMSTAR 2 checklist items. Consequently, their overall compliance did not reach 100% [17,28,31,33].

A smaller subset, four reviews (12.9%), was classified as moderate confidence, typically due to the presence of isolated non-critical weaknesses, yet maintaining an overall coherent and well-structured methodological approach. Although none of these reviews exhibited critical flaws according to the AMSTAR 2 criteria, several demonstrated non-critical weaknesses that limited the completeness of their methodological reporting. For example, Zhang et al. [37] did not adequately describe the procedures for duplicate study selection and duplicate data extraction, which are essential for minimizing operational errors and bias. Similarly, Zhao et al. [45] also failed to clearly report duplicate study selection, compromising methodological transparency; El Dib et al. [42] presented gaps in the detailed description of included studies. In addition, Zhao et al. [50] did not explicitly address the assessment of heterogeneity or the potential sources of variability.

Strikingly, eleven studies were rated as low confidence (35.4%), reflecting recurrent deficits across multiple AMSTAR2-critical domains [32,34,35,36,40,41,43,46,47,49,52]. The most frequent shortcomings included lack of protocol registration (Item 2), inadequate or poorly justified search strategies (Item 4), failure to account for risk of bias in primary studies when interpreting findings (Item 13), and omission of publication bias assessment (Item 15). These recurrent methodological gaps suggest that substantial portions of available syntheses provide limited reliability for decision-making, despite often reporting statistically significant results.

Finally, four reviews (12.9%) scored critically low confidence, primarily because they contain multiple critical flaws [29,38,51,58]. These studies, in particular, exhibit failures in fundamental reporting processes, such as the following: (a) disclosure of funding sources and conflicts of interest; (b) insufficient or low-quality reporting of risk of bias; (c) inadequate assessment of heterogeneity; and (d) lack of protocol registration in the PROSPERO database. This pattern not only reflects variability in methodological maturity among research groups and publication periods but also emphasizes the importance of cautious interpretation when integrating findings from reviews with suboptimal quality. Table S4 details the items classified for the AMSTAR 2 scale. Figure 10 presents the judgment criteria for the AMSTAR2 checklist.

The overall confidence in the results of each systematic review was rated according to AMSTAR 2 guidelines, based on the presence and number of critical and non-critical weaknesses. Critical domains include the following: (2) protocol registration, (4) adequacy of the search strategy, (7) listing of excluded studies with justification, (9) risk of bias assessment of included studies, (11) appropriateness of meta-analytic methods, (13) consideration of risk of bias when interpreting results, and (15) assessment of publication bias. High confidence: No critical flaws and at most one non-critical weakness; moderate confidence: No critical flaws but more than one non-critical weakness; low confidence: One critical flaw with or without non-critical weaknesses; and critically low confidence: Two or more critical flaws with or without other weaknesses. The green bar represents items fully met according to the AMSTAR-2 criteria. The yellow bar indicates items that were only partially met or presented with limited methodological detail. The red bar denotes items that were not met and correspond to potential critical and/or non-critical weaknesses in the methodological quality of the included reviews.

The assessment of heterogeneity among the primary studies within each meta-analysis revealed substantial variability, with I² values ranging from minimal (0.8%) to extremely high (99.7%). According to the interpretation framework recommended by the Cochrane Handbook, only five meta-analyses (16.1%) demonstrated low heterogeneity (I² ≤ 25%), indicating a high degree of consistency across their primary studies [17,32,41,42], whose findings are less influenced by methodological or clinical diversity.

Moderate heterogeneity (I² = 25% to 50%) was observed in five studies (16.1%), including [27,28,36]. This level of variability is generally acceptable and often reflects moderate differences in sample characteristics, interventions, or outcome measures. In contrast, the largest proportion of meta-analyses fell within the category of substantial heterogeneity (I² = 50% to 75%), comprising nine studies (29.0%), such as [30,31,43,46,49,58]. This degree of inconsistency suggests that moderate-to-marked differences in methods, populations, or analytical approaches contributed significantly to the observed variance.

Finally, a considerable number of meta-analyses (twelve studies; 38.7%) exhibited very high heterogeneity (I² > 75%), indicating pronounced disagreement among the included primary studies. This group includes [16,33,34,35,38,48,53], with values exceeding 75% suggesting that methodological, clinical, or statistical factors strongly influence the consistency of aggregated effects.

It is important to contextualize these findings within the broader pharmacological landscape of depression treatment, particularly regarding the widespread use of antidepressants such as selective serotonin reuptake inhibitors (SSRIs). SSRIs remain first-line therapy for major depressive disorder; however, their clinical effectiveness is frequently limited by delayed onset of action, incomplete remission rates, substantial interindividual variability in response, and the occurrence of adverse effects, notably gastrointestinal symptoms [73,74].

Evidence indicates that antidepressants themselves may interact bidirectionally with the gut microbiota [74]. Several SSRIs exhibit intrinsic antimicrobial activity capable of altering microbial composition and diversity, which may influence both therapeutic efficacy and tolerability [2,62,64,72,75]. These microbiota alterations may partially explain antidepressant-induced gastrointestinal side effects and, potentially, variability in clinical response [76]. Conversely, baseline dysbiosis has been associated with altered antidepressant metabolism and reduced treatment responsiveness, suggesting that the gut microbiome may act as a moderator of pharmacological efficacy [7,76,77].

Probiotics and prebiotics may complement antidepressant therapy through convergent modulation of the microbiota–gut–brain axis [1,10,23,64,72,78,79]. Proposed pathways include increased production of short-chain fatty acids, regulation of inflammatory and immune signaling, modulation of tryptophan metabolism and serotonergic neurotransmission, reinforcement of intestinal barrier integrity, and attenuation of HPA axis hyperactivity [62,80]. Several of these mechanisms overlap with biological targets indirectly influenced by antidepressant drugs, supporting a biologically plausible rationale for combined or adjunctive approaches [62].

Confirming this, a recent high-quality systematic review and network meta-analysis by Zhao et al. [50] directly compared microbiota-targeted therapies with antidepressants in adults with MDD. Among 42 RCTs, probiotics demonstrated significant reductions in depressive symptom severity compared with placebo and were non-inferior to multiple antidepressants, including venlafaxine, vortioxetine, duloxetine, and citalopram. Notably, probiotics ranked second only to escitalopram in the treatment hierarchy and showed comparable tolerability to antidepressants when administered for ≥8 weeks. Moreover, probiotic interventions used as adjuncts to antidepressant therapy were superior to several pharmacological agents alone.

So, preclinical and clinical studies provide support for this interaction, demonstrating that specific probiotic strains can exert antidepressant-like effects and possibly modulate stress-related neuroendocrine and inflammatory responses, sometimes comparable to conventional antidepressants [1,28,38,41,46,48,52,60,68,81].

It is important to highlight that a critical characteristic of the trials included in this review is the predominance of Lactobacillus and Bifidobacterium strains, administered either as single strains or in multispecies formulations. This observation is relevant because the biological effects attributed to probiotics are not uniform but rather depend on strain-specific and functional properties. Psychobiotic effects arise from specific microbial capacities related to metabolite production, immune modulation, and host–microbe signaling, rather than from probiotic exposure per se [6,49,62,82].

Experimental evidence provides robust support for neural communication as one of the pathways linking specific Lactobacillus strains to the regulation of stress and affective states. Chronic administration of Lactobacillus rhamnosus (JB-1) has been shown to induce region-specific alterations in GABA receptor expression and to attenuate stress-related behaviors; notably, these effects are abolished following vagotomy, identifying the vagus nerve as a critical conduit for gut–brain signaling [81]. Cryan et al. [62] position this vagal pathway as one of the most biologically coherent routes of gut–brain communication. However, despite its strong biological plausibility, direct confirmation of vagal mediation in the improvement of depressive and anxiolytic symptoms in humans remains limited, as few randomized controlled trials have simultaneously assessed vagal biomarkers.

In addition, the low-grade systemic inflammation commonly observed in patients with depressive disorders constitutes a biologically plausible target for microbiota-mediated interventions. Both Lactobacillus and Bifidobacterium strains have demonstrated anti-inflammatory properties, including reductions in pro-inflammatory cytokines and markers of oxidative stress. Clinical trials have reported concurrent improvements in depressive symptom scores alongside reductions in hs-CRP, IL-6 expression, and oxidative stress indices following probiotic supplementation [1,38]. Sulaiman et al. [49] identify immune modulation as one of the most consistently supported mechanisms across human studies, suggesting that attenuation of peripheral inflammation may indirectly influence neuroinflammatory processes and HPA activity. Among the proposed pathways, immune–brain signaling currently represents one of the strongest translational links between probiotic exposure and reductions in depressive symptoms.

Another frequently cited mechanism involves the regulation of tryptophan availability. Certain Bifidobacterium strains have been associated with alterations in tryptophan metabolism and increased peripheral serotonin availability in clinical trials [83]. This pathway provides a coherent intersection between immune activation, microbial function, and neurotransmission. However, it is now well-recognized that depression and its comorbidities do not depend exclusively on monoaminergic pathways, and, therefore, this mechanism should be interpreted with caution [61,84].

Emerging neuroimaging data further suggest that probiotic interventions may influence brain structure and functional connectivity within frontolimbic circuits implicated in mood regulation. Changes in hippocampal activation patterns, fractional anisotropy, and gray matter volume have been reported in small clinical samples following supplementation with multispecies probiotics [85,86]. Although these findings are intriguing, further studies are required to confirm and extend these observations.

During the process of collecting and synthesizing the various meta-analyses, we identified potential sources of confusion, including the directionality of reported outcomes in some studies. When evaluating effects based on psychometric instruments such as the BDI, HAM-D, or DASS, one expects results to be expressed as reductions in symptom scores, given that lower scores correspond to clinical improvement. Consequently, when reporting effect estimates, the direction of the effect should mathematically be negative (post-intervention values must be lower than baseline for the intervention to be interpreted as beneficial).

However, the meta-analyses by Nikolova et al. [34], Nikolova et al. [40], and Hofmeister et al. [39] reported clusters of positive SMDs, resulting in an overall pooled effect that was also positive (SMD = 0.86 [95%CI: −0.57 to 2.17], p > 0.05; SMD = 0.58 [95%CI: 0.19 to 0.97], p < 0.05; and SMD = 0.78 [95%CI: 0.19 to 1.37], p < 0.05, respectively). When these data were synthesized alongside the remaining literature, we observed an irregular pattern: the direction of the effect in these three studies favored the control or placebo group, despite the authors' conclusions being favorable to supplementation. Accordingly, it was necessary to adjust the direction of these effect sizes to ensure consistent alignment and appropriate aggregation within our analysis.

Additionally, the meta-analysis by Nikolova et al. [34] appeared to contain several methodological inconsistencies. For instance, Kazemi et al. [60] were included alongside Akkasheh et al. [1] and Romijn et al. [79]. The estimated effect size for Kazemi et al. [60] was strikingly large (SMD = 1.99 [95% CI: 1.43 to 2.55]), which would only be plausible if the authors had reported mean differences rather than standardized mean differences, which they did not. When cross-referencing these values with more recent meta-analyses published between 2020 and 2025, such as Misera et al. [43], the same trial yields an SMD of −0.39 [95% CI: −0.85 to 0.07], i.e., a markedly different effect estimate. This discrepancy strongly suggests a potential error in the earlier review, which, in turn, compromises the validity of its conclusions.

Finally, the inclusion of multiple interventions, such as probiotics, prebiotics, and synbiotics, within the same dataset, without appropriate subgroup analyses or with missing subgroup information, compromises the interpretability of findings in subsequent research. Zandifar et al. [51], for example, examined the effects of prebiotics and probiotics on depression, anxiety, and cognition; however, the authors only reported a combined effect estimate, pooling prebiotics, probiotics, and synbiotics. Although they stated that subgroup analyses had been performed, these results were not adequately reported (there is no Supplementary Materials section). Other reviews similarly lack essential information when dealing with multiple intervention types, particularly in the context of synbiotic supplementation, further complicating the synthesis and comparison of effects across studies. In our case, this issue eliminated the possibility of adequately processing the data to estimate the effects of synbiotics.

The methodological assessment conducted in this umbrella review revealed a consistent pattern of substantial weaknesses among most of the included systematic reviews and meta-analyses, which is in line with the limitations previously identified by Morán et al. [23]. Using the AMSTAR 2 instrument, we found that the majority of reviews were classified as low or critically low in overall confidence. This distribution closely mirrors the findings of Morán et al. [23], who also reported a predominance of low-confidence ratings, highlighting Hofmeister et al. [39] as the only review meeting the high-quality standard.

One of the most frequent shortcomings was the absence of a pre-registered protocol (Item 2), which compromises transparency and increases the risk of post hoc decision-making. Recurrent issues also included inadequate reporting of funding sources for the included studies (Item 10) and the absence or insufficiency of publication bias assessments (Item 15). Additionally, as noted by Morán et al. [23], only a portion of the reviews appropriately incorporated risk-of-bias evaluations into the interpretation of their findings (Item 13). Even when such assessments were performed, their implications were seldom discussed with the necessary depth.

Another recurring problem was the high level of statistical heterogeneity, with I² values frequently exceeding 70%. Although some reviews conducted subgroup or sensitivity analyses to mitigate this issue, substantial heterogeneity persisted. Key contributors included notable differences in population characteristics (healthy individuals vs. clinically diagnosed patients), baseline symptom severity, intervention dosage and duration, and variability in the psychometric instruments used to assess outcomes.

So, as indicated by the AMSTAR 2 assessment, several of the included studies exhibited critical or high-risk methodological weaknesses, which constrain the confidence that can be placed in pooled estimates. These limitations are largely inherent to the existing evidence base and cannot be corrected retrospectively within the present synthesis. Consequently, while observed effects may indicate the potential benefits of psychobiotic interventions, the findings should be interpreted with caution and should not be considered definitive evidence of efficacy.

A major source of heterogeneity among the included studies likely arises from the substantial variability in probiotic and prebiotic formulations rather than from differences in outcome assessment alone. Across primary trials, interventions differed markedly with respect to microbial strains, combinations of species, dosages, intervention duration, and delivery formats. Such compositional heterogeneity may exert a stronger influence on observed effects than the choice of psychometric instruments used to assess depressive or anxiety symptoms.

Although some recent studies have attempted to address this issue by employing standardized or well-characterized strains, the majority of available trials continue to use heterogeneous formulations, limiting the ability to attribute effects to specific microbial profiles. The present review did not perform strain-specific or dose–response analyses, which represents an important limitation and reflects the current constraints of the evidence base. As illustrated in Table 1, the wide diversity of probiotic compositions precludes firm conclusions regarding the relative efficacy of individual strains or combinations. Future research should prioritize standardized formulations and strain-specific designs to reduce heterogeneity and enable more precise mechanistic and clinical inferences.

Additionally, while the adoption of a single universal outcome measure could, in theory, enhance comparability, such an approach is unlikely to be feasible or desirable given differences in study objectives, populations, symptom severity, and clinical versus subclinical contexts. From a practical and methodological standpoint, a more realistic solution lies in the standardization of analytical frameworks rather than in the exclusive use of a single instrument. In this review, comparability was addressed by prioritizing effect estimates derived from clinically relevant subgroups when available and by selecting SMDs based on instruments with stronger psychometric support and greater prevalence among trials. Global SMDs were retained when populations were homogeneous or when further stratification would compromise statistical robustness.

Future research should move beyond highly generalized analyses that pool heterogeneous populations, symptom profiles, and probiotic formulations, as such approaches may obscure strain-specific effects and attenuate clinically meaningful signals. Accumulating evidence suggests that probiotic effects on mental health outcomes are not uniform [47,64,69,82,83], but instead depend on clearly defined strain characteristics and well-characterized participant profiles. Accordingly, future trials should prioritize more phenotypically and biologically specific study designs, including the evaluation of single strains, that are well-justified.

In parallel, greater emphasis should be placed on recruiting clinically defined populations, such as individuals with established diagnoses of depressive or anxiety disorders, or on clearly stratifying participants by symptom severity and baseline microbiota characteristics. It is worth highlighting that only a limited number of existing studies have adopted such stringent designs, focusing on specific strains or clinically diagnosed populations (observed in more recent studies) [47,64,69,82,83], and yet these studies tend to provide clearer and more interpretable effect estimates.

Finally, despite growing interest in microbiota-targeted interventions, the current evidence base remains heavily skewed toward prebiotic formulations, limiting the ability to differentiate the effects on depressive and anxiety symptoms. Future progress in this field will require methodologically rigorous primary studies specifically designed to evaluate prebiotics and synbiotics, with clearly defined compositions.