Gut microbiota alterations in schizophrenia might be related to stress exposure: Findings from the machine learning analysis

Specific mechanisms underlying gut microbiota alterations in schizophrenia remain unknown. We aimed to compare gut microbiota between patients with schizophrenia and controls, taking into consideration exposure stress across lifespan, dietary habits, metabolic parameters and clinical manifestation. A total of 142 participants, including 89 patients with schizophrenia and 52 controls, were recruited. Gut microbiota were analyzed using the 16 S rRNA sequencing. Additionally, biochemical parameters related to glucose homeostasis, lipid profile and inflammation were assessed. Increased abundance of Lactobacillus and Limosilactobacillus as well as decreased abundance of Faecalibacterium and Paraprevotella were found in patients with schizophrenia. The machine learning analysis demonstrated that between-group differences in gut microbiota were associated with psycho-social stress (a history of childhood trauma, greater cumulative exposure to stress across lifespan and higher level of perceived stress), poor nutrition (lower consumption of vegetables and fish products), lipid profile alterations (lower levels of high-density lipoproteins) and cognitive impairment (worse performance of attention). Our findings indicate that gut microbiota alterations in patients with schizophrenia, including increased abundance of lactic acid bacteria (Lactobacillus and Limosilactobacillus) and decreased abundance of bacteria producing short-chain fatty acids (Faecalibacterium and Paraprevotella) might be associated with exposure to stress, poor dietary habits, lipid profile alterations and cognitive impairment.


Introduction
Schizophrenia represents one of severe mental disorders that is associated with a significant impact on social functioning and reduced life expectancy (Correll et al., 2022). From the clinical perspective, it is a highly heterogeneous mental disorder characterized by multiple psychopathological symptoms, including, i.e., hallucinations, delusions, negative and mood symptoms as well as cognitive impairment (Jauhar et al., 2022). Accumulating evidence shows that schizophrenia is a multi-system neurodevelopmental disorder (Misiak et al., 2014;Pillinger et al., 2018). Indeed, there are various subclinical or clinically relevant alterations observed in the peripheral blood and cerebrospinal fluid of individuals with schizophrenia. These include subclinical inflammation (Miller et al., 2011), oxidative stress (Flatow et al., 2013), lipid profile disturbances (Misiak et al., 2017), impaired glucose homeostasis (Perry et al., 2016), altered hormonal regulation of appetite (Lis et al., 2020) and dysfunction of the hypothalamic-pituitary-adrenal (HPA) axis . Importantly, most of them can be detected in drug-naïve individuals with first-episode psychosis or even those at clinical risk of psychosis. However, specific mechanisms underlying these observations are still the subject of intensive research activity.
The recognition of the gut-brain axis has provided another perspective that might help to understand the pathophysiology of schizophrenia . Microogranisms initiate colonization of the gastrointestinal system approximately 10 weeks after conception, i.e., when the fetus starts to swallow the amniotic fluid (Golofast and Vales, 2020). The final composition of gut microbiota is unique and characterized by various functions related to the metabolism of nutrients, maintenance of the mucosal barrier integrity, immunomodulation and protection against pathogens (Rinninella et al., 2019). Intestinal microbes can communicate with the central nervous system through the enteric nervous system and vagal afferents (Carabotti et al., 2015). They might also interact with the central nervous system by releasing neurotransmitters, short-chain fatty acids (SCFAs) and specific microbial antigens (Mittal et al., 2017;Silva et al., 2020). Some of these metabolites can operate through the bloodstream and cross the blood-brain barrier. Moreover, gut microbes might stimulate production of pro-inflammatory cytokines, prostaglandins and ileal corticosterone that might further impact functioning of the central nervous system .
To date, a number of studies have investigated differences in gut microbiota composition between individuals with schizophrenia and healthy controls. Recent systematic reviews demonstrated that certain gut microbiota alterations in subjects with schizophrenia might overlap with those observed in patients with depression and bipolar disorder (Borkent et al., 2022;McGuinness et al., 2022). These include increased relative abundance of the genera Streptococcus, Eggerthella and Lactobacillus as well as decreased abundance of the Faecalibacterium genus that is responsible for the production of SCFAs. In turn, increased abundance of Prevotella together with decreased abundance of Bacteroides and Haemophilus might be more specific for individuals with schizophrenia. It should also be noted that certain studies have revealed that the fecal microbiota transplantation from drug-naïve individuals with schizophrenia to germ-free mice leads to altered neurotransmission, hyperactivity, impairments of learning and memory processes as well as up-regulation of the kynurenine-kynurenic acid pathway in recipient mice (Zheng et al., 2019;. Although a number of gut microbiota alterations have been demonstrated in patients with schizophrenia, specific mechanisms underlying these observations remain unclear. As noted by the authors of recent systematic reviews, studies comparing gut microbiota composition between individuals with severe mental disorders and healthy controls have rarely controlled for the effects of potential confounding factors, such as lifestyle characteristics or medication effects (Borkent et al., 2022;McGuinness et al., 2022). Moreover, little is known about the impact of known risk factors of schizophrenia on gut microbiota. For instance, there is compelling evidence that traumatic life events across the lifespan increase a risk of psychosis (Beards et al., 2013;Varese et al., 2012). It has been shown in animal model studies that early-life stress alters gut microbiota, and these alterations persist until adulthood (Jašarević et al., 2015(Jašarević et al., , 2017. For instance, rats exposed to neonatal stress have been demonstrated to show increased abundance of Bacteroides and Enterobacteria together with elevated plasma levels of corticosterone and increased inflammatory response lipopolysaccharide in adulthood (García-Ródenas et al., 2006). Similar findings were demonstrated by human studies. Zhang et al. (2022) found 27 taxonomic differences in gut microbiota composition between patients with depression and a history of adverse childhood experiences (ACEs), patients with depression without a history of ACEs and healthy controls. Another study revealed that the combination of four genera (Mitsuokella, Odoribacter, Catenibacterium and Olsenella) could differentiate individuals with posttraumatic stress disorder with the accuracy of 66.4% (Malan-Muller et al., 2022). However, the association of exposure to stress with gut microbiota alterations in schizophrenia has not been explored so far. Taking into consideration these research gaps, the present study had the following aims: 1) to compare gut microbiota composition between a homogenous sample of remitted outpatients with schizophrenia and healthy controls, taking into consideration dietary habits; 2) to investigate whether differences in gut microbiota composition between patients with schizophrenia and healthy controls might be attributed to the effects of various categories of psychosocial stress (ACEs, lifetime stressors and perceived stress within the preceding month) and 3) to explore whether between-group differences in gut microbiota are related to clinical manifestation of schizophrenia.

Participants
The present study included 89 individuals with schizophrenia and 53 healthy controls. They were recruited at two university hospitals in Wroclaw and Szczecin (Poland) as the convenience sample. All participants were non-consanguineous and represented Caucasian ethnicity. The inclusion criteria were: 1) age 18-65 years; 2) a diagnosis of schizophrenia according to the DSM-IV criteria validated using the Operational Criteria for Psychotic Illness (OPCRIT) checklist (McGuffin, 1991); 3) stable psychopharmacological treatment over the period of at least 6 preceding months and 4) remission of positive and disorganization symptoms based on the Positive and Negative Syndrome Scale (PANSS) items (P1 -delusions, P2 -conceptual disorganization, P3hallucinatory behavior, G5 -mannerisms/posturing and G9 -unusual thought content rated ≤ 3) (Andreasen et al., 2005). Current use of mood stabilizers and antidepressants as well as the dosage of antipsychotics expressed as chlorpromazine equivalents (CPZeq) were recorded for all individuals with schizophrenia.
Healthy controls were recruited through advertisements. They reported negative family history of psychotic and affective disorders in first-and second-degree relatives. All of them were screened for psychiatric disorders using the Mini-International Neuropsychiatric Interview (M.I.N.I.) (Sheehan, 1998).
Both groups of participants were matched for age, sex and the level of parental education. The later one was used as a proxy indicator of socioeconomic position (Aarø et al., 2009). The study was approved by the Bioethics Committees at Wroclaw Medical University (Wroclaw, Poland) and Pomeranian Medical University (Szczecin, Poland). All participants gave written informed consent.

Assessment of clinical manifestation
Psychopathology was recorded using the following scales: 1) the Positive and Negative Syndrome Scale (PANSS) (Kay et al., 1987); 2) the Social and Occupational Functioning Assessment Scale (Smith et al., 2011) and 3) the Calgary Depression Scale for Schizophrenia (Addington et al., 1993). Two items (N5 -difficulty in abstract thinking and N7stereotyped thinking) were not included to calculate the PANSS score of negative symptoms due to evidence that they measure other psychopathological constructs (N5 -cognitive symptoms and N7 -thought disorganization) (Galderisi et al., 2021). The associations with positive symptoms were not analyzed as the study was based on the sample of patients during remission of positive and disorganization symptoms.
Cognitive performance was examined using the Repeatable Battery for the Assessment of Neuropsychological Status (RBANS) (Randolph et al., 1998). The RBANS includes 12 tasks measuring performance of five cognitive domains, including immediate memory, visuospatial/constructional abilities, language, attention and delayed memory. Higher RBANS scores refer to better cognitive performance.

Dietary intake
The Food Frequency Questionnaire 6 (FFQ-6) was used to assess dietary habits (Niedzwiedzka et al., 2019). It is based on 62 items that record self-reported food frequency consumption over the preceding 12 months. Each item records the frequency of consuming specific products on a 6-point scale: 1 -'never or almost never'; 2 -'once a month or less'; 3 -'several times a month'; 4 -'several times a week'; 5 -'daily' and 6 -'several times a day'. In the present study, we analyzed consumption of products that allow to measure adherence to Mediterranean diet (aMED) (Hawrysz et al., 2020;Krusinska et al., 2018): 1) vegetables; 2) fruits; 3) whole grains; 4) fish; 5) legumes; 6) nuts and seeds; 7) the ratio of vegetable oils to animal fat and 8) red and processed meat.

Cigarette smoking
Cigarette smoking was assessed using the Fagerstrom Test for Nicotine Dependence (FTND) (Pomerleau et al., 1989). It includes six items that assess the quantity of cigarettes smoked per day, the compulsion to use nicotine and dependence. Yes-or-no items are scored from 0 to 1, while multiple-choice items are scored from 0 to 3. The total score ranges between 0 and 10; with higher scores indicating greater physical addiction to nicotine.

Assessment of stress exposure
The Perceived Stress Scale was administered to record perception of stress over the preceding month (Cohen et al., 1983). It includes 10 questions that are rated on a 5-point Likert scale (0 -never to 4 -very often). The total PSS score ranges from 0 to 40; higher scores indicate greater level of perceived stress.
The Traumatic Experiences Checklist (TEC) was used to record a history of traumatic life events (Nijenhuis et al., 2002). It consists of 29 statements referring to several categories of stressful experiences. In this study, we analyzed a history of emotional neglect, emotional abuse, physical abuse and sexual abuse. The later one was analyzed collectively based on the subscales of "sexual harassment" and "sexual abuse". A history of ACEs was analyzed as exposure under the age of 18 years.

Biological material
Biochemical parameters were determined in fasting serum samples and included: 1) glucose; 2) insulin; 3) total cholesterol; 4) low-density lipoproteins (LDL); 5) high-density lipoproteins (HDL); 6) non-HDL; 8) triglycerides and 9) high-sensitivity C-reactive protein (hsCRP). Stool samples were obtained using the home collection kits. For detailed description of methodology applied to measure biochemical parameters and obtain microbiome data see Supplementary Material.

Microbiome data processing
Paired-end 16 S rRNA gene sequencing was carried out using the MiSeq Reagent Kits v3 (600 cycles). The sequencing data were processed using the cutadapt (version 3.5) and the DADA2 from the Bioconductor package (version 1.22.0) (Callahan et al., 2016). In brief, data preprocessing, cutting of phased QIAseq primers for V3 -V4 region and demultiplexing were performed using a custom script based on the cutadapt for all raw FASTQ sequences. Afterwards, quality filtering, error correction, dereplication and chimeras filtering were applied using the DADA2. We applied the DADA2 filtering settings proposed by Reitmeier et al. (2021). This resulted in the reconstruction of 31983 amplicon sequence variants (ASVs). The ASVs were then classified using the naïve Bayesian classifier with the minimum bootstrap confidence set at 0.8 for assigning the taxonomic level . We used four various reference sequence databases: the SILVA, version 138.1 (McLaren and Callahan, 2021); the Genome Taxonomy Database (GTDB), version 207 (Ali et al., 2021); the RefSeq combined with Ribosomal Database Project (RDP), version 16 (Ali et al., 2021) and the Ribosomal Database Project (RDP), version 18 (Callahan, 2020). Subsequently, data were transformed into the phyloseq objects. The ASVs that could not be classified at the genus level were removed. Resulting taxa were collapsed by their taxonomic assignment into the genus level. Finally, low-frequency taxons (< 10) were removed. The final datasets included 169 taxa (GTDB, version 207), 100 taxa (RDP, version 18), 89 taxa (RefSeq combined with RDP, version 16) and 123 taxa (SILVA, version 138). The alpha diversity measures were calculated at the level of unfiltered ASVs, whereas the analysis of beta-diversity was carried out at the genus level (Cao et al., 2021;Nearing et al., 2022).

Statistical analysis and bioinformatics
Between-group differences in general characteristics of the sample were assessed using the χ 2 test or the Fisher's exact test (in case of categorical variables) and the Mann-Whitney U test or t-tests (in case of continuous variables). These analyses tested the null hypothesis that both groups of participants did not differ significantly with respect to general characteristics of the sample. The level of significance in bivariate tests of general characteristics was set at p < 0.05. This part of data analysis was performed using the SPSS, version 28.
To examine differences in alpha diversity measures between groups we calculated three indexes (Chao, Shannon and Simpson) and compared them with the Kolmogorov-Smirnov test using the RDPv18 data (Amezquita et al., 2020). Possible confounding effects of age, sex, BMI, smoking, the aMED indices and the level of education on differences in alpha diversity were examined with generalized linear regression models in the chest R package by comparing effect estimates from various models with different combinations of variables (Wang, 2007).
As reported recently by Nearing et al. (2022), tools to identify differentially abundant ASVs provide different numbers and sets of significant ASVs. Therefore, to examine the differences in beta-diversity between patients with schizophrenia and healthy controls, we performed the differential abundance analysis using 11 various tools: the LEfSe, the metagenomeSeq, the ANCOM, the ANCOMBC, the ALDEx2, the edgeR, the DESeq2, the limma-voom, the SIAMCAT, the Random Forest (RF)-based feature selection and the Support Vector Machine (SVM)-based feature selection (Cao et al., 2022;Fernandes et al., 2014;Kuhn, 2008;Law et al., 2014;Lin and Peddada, 2020;Love et al., 2014;Mandal et al., 2015;Paulson et al., 2013;Robinson et al., 2009;Segata et al., 2011;Wirbel et al., 2021). We used default parameters and input data for each method (Cao et al., 2022). Specific taxon with the false discovery rate (FDR)-corrected p-value < 0.05 indicated by any method was defined as significantly differentially abundant. For two machine learning feature selection approaches, we used a 10-fold and 10-times repeated cross-validation to select top 5 genera. Possible confounding effects of variables, such as age, sex, BMI, cigarette smoking, the aMED indices and the level of education on differences in abundances of specific taxa were examined by the constrained correspondence analysis (CCA) and the princeplot function in the swamp R package (Cao et al., 2022;Lauss et al., 2013). We defined specific variables as potentially confounding if it has been reported to influenced microbiome variability and it appeared to differ significantly between both groups of participants.
The Multiblock sparse Partial Least Squares Discriminant Analysis (the MsPLS-DA, DIABLO) integrative tool implemented in the mixOmics Bioconductor package was used to select discriminative clinical variables and provide correlation estimates between microbiome data (the CLR-transformed relative abundances of RDP, version 18) and selected variables (Singh et al., 2019). The DIABLO is a multi-omics method that simultaneously identifies highly correlated variables during the integration process that discriminate phenotypic groups (patients with schizophrenia and healthy controls). In brief, clinical variables were divided into the following groups: 1) biochemical parameters (the levels of glucose, insulin and hsCRP as well as lipid profile parameters); 2) cognition (the RBANS scores of cognitive performance); 3) diet (the aMED indices for the consumption of fish, legumes, nuts and seeds, vegetables, fruits and and whole grains as well as the plant oils/animal fats consumption ratio) and 4) the measures of psychosocial stress (the TEC general score of lifetime stressors, a history of any ACEs and the PSS score) and 4) the microbiome (differentially abundant genera).
Variables that best-discriminated both groups of participants were selected based on sparse Partial Least Squares Discriminant Analysis (sPLS-DA) using a 10-fold cross-validation repeated 50 times. Subsequently, variables were included into the final DIABLO model. Resultant correlation matrix was used as the input for the chordDiagram function in the circlize R package (Gu et al., 2014). The relationships between differentially abundant genera and clinical variables specific for patients with schizophrenia (the use of antidepressants and mood stabilizers, CPZq, the PANSS score of negative symptoms, the CDSS score and the SOFAS score) were analyzed using the Spearman correlation coefficients followed by the FDR procedure and visualized in the circlize R package.

Results
General characteristics of the sample are reported in Table 1. Both groups did not differ significantly in terms of age, sex, individual level of education and the level of parental education. However, patients with schizophrenia presented significantly lower cognitive performance across all RBANS domains. The aMED scores for all food products, except of the score of whole grains, were significantly lower in patients with schizophrenia. A history of any ACEs and specific categories of all ACEs was significantly more frequent in patients with schizophrenia. Similarly, the PSS score and the TEC general score of lifetime stressors were significantly higher in patients with schizophrenia. The levels of insulin, triglycerides and hsCRP were significantly higher, while the levels of HDL were significantly lower in patients with schizophrenia. The FTND score and BMI were also significantly higher in the group of patients compared to healthy controls.
There were no significant between-group differences with respect to alpha-diversity measures (Fig. 1) regardless of the use of specific reference sequence databases (data not shown). No significant effects of potential confounding factors, such as age, sex, BMI, the FTND score, the aMED indices and the level of education were found using the generalized linear regression models (data not shown).
To identify bacterial genera that were differentially abundant in patients with schizophrenia and healthy controls, we first focused on the existence of possible confounders using two independent approaches. Both of them indicated age as a variable that has impact on the microbiome variability (at genus level). However, no significant betweengroup differences with respect to age were found. Consequently, age was not included as a covariate in differential abundance analyses. Fig. 2 summarizes the results of univariate differential abundance analysis at the genus level. Results were aggregated based on the results of 11 differential abundance analysis tools according to 4 distinct taxonomic reference databases. In each of four taxonomic data, we observed very low concordance of differentially abundant bacterial genera by various tools. Highest concordance was observed in the RDPv18 data for Paraprevotella which was indicated by 7 out of 11 tools. Differential abundance of Paraprevotella was also most frequently selected in the RefSeq and the RDPv16 (7 out of 11 tools) and the SILVA, version 138 (6 out of 11 tools) datasets. In general, we observed very low intersection between distinct taxonomic reference databases and lists of differentially abundant genera. We found that significant differences in the abundance of Paraprevotella, Faecalibacterium and Limosilactobacillus had been indicated at least by one tool in all four datasets. Between-group differences in the abundance of Lactobacillus, Eisenbergiella and Alistipes were indicated by at least one tool in three datasets. However, in each dataset, significant differences in the abundance of Lactobacillus were most frequently selected by various tools (e. g., by 5 out of 11 tools in the RDPv18). In turn, other genera were found to be differentially abundant by 2 or 3 out of 11 tools. Consequently, we decided to select 4 genera for subsequent analyses, including Paraprevotella (significantly enriched in healthy controls), Faecalibacterium   (significantly enriched in healthy controls), Limosilactobacillus (significantly enriched in patients with schizophrenia) and Lactobacillus (significantly enriched in patients with schizophrenia). Results of subsequent analyses are shown in Fig. 3. In this analysis, we aimed to find correlations of differentially abundant taxa with sample characteristics operationalized within specific layers of variables (biochemical parameters, cognition, diet, and stress) in all participants. No strong correlations were found between the abundance of the Faecalibacterium genus and other variables. The abundance of Paraprevotella was strongly and positively correlated with fish consumption, and inversely correlated with the PSS score. The abundance of Lactobaccillus was strongly and negatively correlated with the consumption of vegetables, the RBANS attention score and HDL levels. The highest number of significant correlations was found for the abundance of the Limosilactobacillus genus: 1) positive correlations with a history of any ACEs and the TEC general score of lifetime stressors; 2) negative correlations with the levels of HDL, the RBANS attention score and the consumption of vegetables.
Finally, we analyzed correlations of four differentially abundant genera with clinical variables specific for patients with schizophrenia. We found weak, but significant, positive correlation between the use of mood stabilizers and the abundance of Paraprevotella. Moreover, there were weak, but significant, positive correlation of the CDSS and SOFAS scores with the abundance of Faecalibacterium (Fig. 4).

Discussion
In this study, we used multiple methods of differential abundance analysis and taxonomic databases to detect differentially abundant genera in patients with schizophrenia compared to healthy controls (Nearing et al., 2022;Wallen, 2021). Findings of the present study indicate that individuals with schizophrenia show greater abundance of Lactobacillus and Limosilactobacillus together with lower abundance of Paraprevotella and Faecalibacterium. The results with respect to differential abundance of Lactobacillus and Faecalibacterium are in agreement with findings from recent systematic reviews addressing gut microbiota composition in severe mental disorders (Borkent et al., 2022;McGuinness et al., 2022). Both genera with lower abundance in patients with schizophrenia (Paraprevotella and Faecalibacterium) represent bacteria producing one of SCFAs -butyrate. SCFAs are produced in the gut by anaerobic fermentation of dietary fiber. They might interact with vagal afferents by stimulating the secretion of gut hormones such as glucagon-like peptide 1 and peptide YY (PYY) as well as neurotransmitters such as γ-aminobutyric acid and serotonin (Silva et al., 2020). Fig. 3. The chord diagram illustrating the correlation between various groups of variables selected by the DIABLO approach and CLR-transformed relative abundances of selected genera. Correlation matrix has been calculated by the circosPlot function in the mixOmics package. Correlations with co-efficients ≥ 0.7 are shown. All of them were significant after adjustment for multiple testing (the False Discovery Rate procedure). Red links indicate positive correlations, whereas blue links indicate negative correlations. Abbreviations: ACEs, adverse childhood experiences; aMED, adherence to Mediterranean Diet; HDL, high-density lipoproteins; PSS, the Perceived Stress Scale; RBANS, the Repeatable Battery for the Assessment of Neuropsychological Status; TEC, the Traumatic Experiences Checklist; vis cons, visuospatial/constructional functions.
To our knowledge, none of previous studies investigated whether dietary habits contribute to gut microbiota alterations in subjects with schizophrenia. However, it is now widely known that patients with schizophrenia show poor dietary habits in terms of high intake of refined carbohydrates and saturated fats, together with low intake of fiber, certain vitamins and minerals (Aucoin et al., 2020). Our study demonstrated that dietary intake of certain food products might be related to gut microbiota alterations in schizophrenia. Specifically, greater abundance of Lactobacillus and Limosilactobacillus was associated with lower intake of vegetables, while lower abundance of Paraprevotella was related to lower intake of fish. Greater abundance of Lactobacillus and Limosilactobacillus was also associated with significantly lower levels of HDL. Importantly, the consumption of both categories of food products and the levels of HDL were significantly lower in subjects with schizophrenia, and differentiated both groups of participants according to the machine learning analysis. These observations might be perceived as contrary to findings from previous studies investigating the associations of various Lactobacillus species with diet and metabolic parameters. Indeed, it has been reported that lactic acid producing bacteria (including Lactobacillus and Bifidobacterium species) can exert a number of biological activities that might be beneficial in terms of modulating cholesterol metabolism (Vourakis et al., 2021). Similarly, it has been reported that polyphenols and fiber that are typical components of plant foods increase the abundance of lactic acid bacteria (Tomova et al., 2019). However, the Lactobacillus genus is also relatively heterogenous with certain species having pro-inflammatory properties (Rocha-Ramírez et al., 2017). Also, Zhu et al. (2020b) identified certain strains of Lactobacillus in patients with schizophrenia that do not occur in the healthy gut environment. Therefore, our findings indicate the necessity to explore the abundance of specific strains of Lactobacillus in this population.
It is important to note that differentially abundant genera were associated with clinical manifestation in patients with schizophrenia from our sample. Specifically, the abundance of Lactobacillus and Limosilactobacillus was negatively correlated with performance of the RBANS attention domain. Moreover, individuals with schizophrenia from our sample showed significantly lower scores on the RBANS attention domain. At this point, it should be noted that the RBANS attention score is the sum of scores from two tasksdigit coding and digit span. The first one also measures psychomotor speed, while the latter one is the measure of working memory. Thus, it should be concluded that greater abundance of Lactobacillus and Limosilactobacillus might be related to broader impairments of cognition in terms of attention, processing speed and working memory. However, the relevance and specific mechanisms of these correlations should be interpreted with caution. Indeed, these correlations are in agreement with Fig. 4. The chord diagram illustrating the correlation between clinical characteristics of schizophrenia and the CLR-transformed relative abundances of selected genera. Correlation matrix has been calculated using the Spearman coefficients. Only correlation coefficients ≥ 0.2 are shown. All of them were significant after adjustment for multiple testing (the False Discovery Rate procedure). Red links indicate positive correlations. Abbreviations: CDSS, the Calgary Depression Scale for Schizophrenia; PANSS N, the Positive and Negative Syndrome Scalescore of negative symptoms; SOFAS, the Social and Occupational Functioning Assessment Scale.
findings from previous studies that show increased abundance of Lactobacillus in patients with Alzheimer's disease (Li et al., 2019;Zhou et al., 2021). However, several lines of evidence from preclinical studies and clinical trials indicate that probiotics that are frequently based on specific strains of Lactobacillus might be beneficial in terms of improving cognition (Ruiz-Gonzalez et al., 2020). These observations also point to the heterogeneity of the Lactobacillus genus in terms of its biological activity. Similarly, it is difficult to interpret correlations of the Faecaliabacterium abundance with other clinical characteristics. Indeed, findings from our study suggest that decreased abundance of the Faecalibacterium genus is weakly but significantly related to lower severity of depressive symptoms and worse general functioning in patients with schizophrenia. This observation is not in agreement with the prior findings. Indeed, the previous studies have demonstrated decreased abundance of Faecalibacterium in patients with major depression (for a systematic review see Knudsen et al. (2021)). Also, the study by Jiang et al. (2015) revealed a negative correlation between the abundance of Faecalibacterium and the severity of depressive symptoms in patients with major depression. Potential explanations underlying this difference remain unknown. However, it should be noted that schizophrenia and major depression are different mental disorders. In this regard, both disorders might also differ in terms of specific mechanisms underlying the development of depressive symptoms.
Most importantly, the present findings, for the first time, indicate that gut microbiota alterations in patients with schizophrenia might be associated with exposure to stress. We found that increased abundance of Limosilactobacillus is related to a history of any ACEs and higher lifetime exposure to stressors. Also, individuals with schizophrenia from our sample reported higher rates of exposure to any categories of ACEs and lifetime stressors. Limosilactobacillus was splitted from the Lactobacillus genus in 2020 due to the property of most strains to produce exopolysaccharide from sucrose (Zheng et al., 2020). Nevertheless, the mechanisms linking the abundance of Limosilactobacillus with exposure to stress remain unknown. However, there is some evidence that multi-hit early-life adversities (maternal immune activation, maternal separation and maternal unpredictable chronic mild stress) have sex-specific effects on the abundance of Lactobacillus in mice (increased abundance in males and decreased abundance in females) (Rincel et al., 2019).
There are important limitations of the present study that should be considered. Our sample was not large, and thus false negative and false positive findings cannot be excluded. Medication effects should be considered, especially with respect to the abundance of Paraprevotella, for which the association with the use of mood stabilizers was found. Another point is that the short-read 16 S rRNA sequencing provides only limited insight into the composition of gut microbiota, and thus we were not able to analyze specific bacterial strains. Therefore, our study focused on data aggregated to the genus level as recently advised by Nearing et al. (2022). Our results linking psychosocial stress with gut microbiota should also be interpreted with caution. Indeed, it cannot be excluded that the effects of ACEs simply reflect the impact of other factors related to social disadvantage in childhood. According to the social defeat hypothesis of schizophrenia, ACEs might co-occur with other risk factors of schizophrenia, including urban upbringing, substance abuse, low education and low socioeconomic status (Selten et al., 2013). Although we controlled for parental education, it cannot be excluded that other components of socioeconomic status, including parental income, family structure, household quality and the use of healthcare facilities better explain socioeconomic status (Sankar et al., 2019). Family socioeconomic status has been associated with gut microbiota composition in infants and children (Lewis et al., 2021). Altogether these factors, including ACEs, might shape dietary habits that contribute to the composition of gut microbiota (Aquilina et al., 2021). Another limitation is that the study was based on self-reports of exposure to stress and dietary habits, and thus a recall bias needs to be taken into consideration. Finally, longitudinal design would be needed to establish causal associations.
In sum, the present findings largely confirm previous findings showing increased abundance of lactic acid producing bacteria and decreased abundance of bacteria producing SCFAs in patients with schizophrenia. However, for the first time, the present findings indicate that dietary habits and exposure to stress at various life periods might contribute to these alterations. Given that some of them might account for clinical manifestation and lipid profile disturbances, the development of interventions targeting the mechanisms underlying gut microbiota alterations in patients with schizophrenia is needed. However, it is still necessary to provide insights into specific mechanisms underlying the effects of psychosocial stress on gut microbiota.

Funding
This study received funding from the OPUS grant awarded by National Science Centre, Poland (grant number: 2018/31/B/NZ5/00527).

Declaration of Competing Interest
None to declare.