**3. Results**

## *3.1. Microbiota Compositions*

General characteristics of patients are shown in Table 1. There was no significant change in alpha diversity as measured by Chao1 and Shannon indexes (*p* = 0.955 and *p* = 0.808, respectively; Figure 2A). The PCoA with Bray–Curtis dissimilarity is presented in Figure 2B. Samples were separated into distinct regions, mainly along the PC1 (Axis.1) that explained 42.5% of the intersample variance. The gu<sup>t</sup> microbiome was individually specific, and the Bray–Curtis distances between the same samples were significantly smaller than those between all W0 samples (*p* = 0.00006; Figure 2C). The direction of change along the PC1 was not consistent (Supplementary Figure S2). The mean change in the PC1 was not significantly different from 0 (0.0012, (95% confidence interval: −0.0946, 0.0970), *t* = −0.03, df = 19, *p* = 0.979), suggesting that the gu<sup>t</sup> microbial community composition does not change after six weeks of treatment. In line with this observation, no OTUs were differentially abundant (from the genus to phylum level) between W0 and W6 (Supplementary Figures S3–S5). There was no change in the ratio of Firmicutes to Bacteroidetes (F/B) in the whole group, as well as in males and females (Supplementary Figure S6). In addition, there were no significant differences in the abundance of the KEGG orthologs, modules, and pathways between W0 and W6 samples in the whole group, as well as in men and women (Supplementary Figure S7).


**Table 1.** Clinical characteristics of patients included in the study (*n* = 20).

BMI—Body Mass Index; a Ordinal variables (per day): 1, non-smokers; 2, up to 10 cigarettes; 3, up to 20 cigarettes; 4, up to 40 cigarettes; 1st Q, first quartile; 3rd Q, third quartile, BMI—body mass index.

**Figure 2.** (**A**) Alpha diversity measures at baseline (W0) and after six weeks of hospitalization (W6). The boxplots represent the diversity measures (center line, median; lower and upper hinges correspond to the first (Q1) and third (Q3) quartiles; whiskers, 1.5 \* IQR (Q3–Q1). Grey lines connect samples from the same patients. (**B**) Genus level resolution analysis of gu<sup>t</sup> microbiota in patients diagnosed with paranoid schizophrenia treated with olanzapine during six weeks of hospitalization. The principal coordinate analysis was based on Bray–Curtis dissimilarities calculated using relative abundance data. Samples are colored according to time points (W0 and W6). Grey lines connect samples from the same patients. Ellipses correspond to 95% confidence intervals for two timepoints (W0 and W6) with a multivariate normal distribution. (**C**) The boxplot shows Bray–Curtis dissimilarities calculated in the same patients (within (W0 vs. W6), 0.29 (0.19–0.42)) and in different patients (between subjects (W0), 0.44 (0.36–0.56), *p* = 0.00006, Wilcoxon rank-sum test) (center line: median, lower, and upper hinges correspond to the first (Q1) and third (Q3) quartiles; whiskers: the upper whisker is located at the smaller of the maximum Bray–Curtis measures and Q3 + 1.5 \* IQR (Q3–Q1); the lower whisker is located at the larger of the minimum Bray–Curtis measures and Q1—1.5 \* IQR). W0 and W6 represent time points.

Despite the lack of a consistent shift along the PC1, we examined whether the PC1 changes are associated with demographic, clinical, and environmental factors. The mean PC1 changes did not di ffer between men (0.011 (−0.118–0.140)) and women (0.004 (−0.036–0.016)) (Wilcoxon rank-sum test FDR adjusted *p* (*q*) = 0.649). Demographic, clinical, and environmental factors were not correlated with the change in the PC1 (Supplementary Table S2, Supplementary Figure S8). There was no association between the dose of OLZ and the shift in the gu<sup>t</sup> microbial composition (Supplementary Figure S9). However, the change in the PC1 was significantly greater in patients consuming alcohol (1–3 unit of alcohol; 0.16 (0.03–0.32)) than in non-alcohol drinkers (−0.01 (−0.19–0.01)) (Wilcoxon rank-sum test *q* = 0.036). To further explore the distinct regions revealed by the ordination of samples by PCoA (Figure 2B), we conducted an unsupervised hierarchical clustering using an average linkage algorithm of the Bray–Curtis dissimilarity. The clustering analysis showed the presence of two clusters (Type 1: 9 samples; Type 2: 31 samples) that matched with distinct regions revealed in the PCoA. A heatmap displaying the relative abundances for the genera annotated with two resulting clusters is shown in Figure 3A. Di fferential abundance testing revealed that the *Bacteroides*, *Blautia*, *Clostridium*, *Anaerostipes*, *Bilophila*, *Anaerotruncus,* and *Eggerthella* were enriched in the Type 2 cluster, whereas *Prevotella* was enriched in the Type1 cluster (Figure 3B). Clusters Type 1 and 2 seemed to correspond to enterotypes 2 (*Prevotella*) and 1 (*Bacteroides*) described by Arumugam et al. in 2011 [43], respectively. To explore these enterotypes in more detail, our cluster Type 2 was analyzed more thoroughly, as it seemed not to be completely homogenous. Cluster Type 2 was divided into Type 2A and Type 2B, and then the relative abundances of the main contributors from each enterotype in the resulting three clusters (Type 1, Type 2A and 2B) were ascertained (Supplementary Figure S10) Cluster Type 1 had an abundance pattern similar to enterotype 2 (Figure 2d in Arumugam et al. [43]), and that for cluster type 2A to enterotype 1 (Figure 2d in Arumugam et al. [43]). However, cluster type 2B seemed not to be similar to any enterotype. Two genera (*Prevotella* and *Bacteroides*) exhibited similar abundance which was greater than of *Ruminococcus*. The pattern Type 2B seemed to be a type of a mixture from clusters Type 1 and Type 2A. The addition of *Blautia* made no di fference to this assessment (Supplementary Figure S11).

**Figure 3.** (**A**) Genus level resolution analysis of gu<sup>t</sup> microbiota in patients diagnosed with paranoid schizophrenia treated with olanzapine during six weeks of hospitalization. Unsupervised average linkage hierarchical clustering of gu<sup>t</sup> microbiota at the genus level was conducted. Two resulting clusters (Type 1, blue and Type 2, red) are shown as the top annotation. Both samples (W0 and W6) of 15 patients were found in either Type 1 or Type 2 cluster (two patients in Type 1 and 13 patients in Type 2). Samples of the five patients (3SJP, 4SMA, 10SUG, 11SAS, and 12SMS) belonged to di fferent clusters. (**B**) Di fferential abundance testing at the genus level between Type 1 and Type 2 clusters. a two sided Wilcoxon signed-rank test, FDR adjusted p, the genera with the relative abundance >1% in at least one cluster are shown, *Eggerthella* not shown due to low abundance.

Taken together, our results sugges<sup>t</sup> that the gu<sup>t</sup> microbiota is highly individually specific, and the microbial community compositional changes during six weeks of OLZ treatment are not consistent across the patients.

#### *3.2. Clinical Improvement and BMI Changes*

We found that OLZ treatment was associated with significantly improved treatment e fficacy as measured by PANNS, 36-item short form survey (SF36), and CGI-S scales (Supplementary Table S3). We further investigated whether these improvements are correlated with the change in microbiota compositions (as measured by a change in the PC1 component) and with demographic and clinical characteristics. No significant correlations were observed between clinical improvements and changes in microbiota composition (Supplementary Figure S12) or demographic and clinical characteristics, except the duration of untreated psychosis (DUP) (Supplementary Table S4).

In contrast to changes in the symptom severity of schizophrenia (Supplementary Table S3), there was no significant change in the patients' BMI during OLZ treatment (*q* = 0.763). However, the BMI change (W6 vs. W0 di fference) was significantly higher in women than in men (Supplementary Figure S13) but did not correlate significantly with age, OLZ average dose per day, OLZ maximum dose, disease duration, or duration of untreated psychosis.

Because we found clear di fferences in gu<sup>t</sup> microbiome compositions in all 40 samples (Figure 3), we next sought to determine whether similar di fferences in microbial community compositions and metabolic potentials exist in baseline samples and whether those di fferences could a ffect the patients' clinical improvement and change in BMI within six weeks. We performed the unsupervised average linkage hierarchical clustering of the Bray–Curtis dissimilarity among the baseline samples (W0, Supplementary Figure S14), as well as that of the relative abundances of the predicted KEGG orthologs, modules, and pathways (Supplementary Figures S15–S17). Regarding the microbiome compositions, we were able to demonstrate di fferent groups of patients (clusters) using hierarchical clustering of KEGG features in the W0 samples: KEGG orthologs (Supplementary Figure S15), modules (Supplementary Figure S16), and pathways (Supplementary Figure S17). Di fferential abundance testing revealed that only the *Prevotella* genus di ffered between the two clusters (Type 1, 0.01% (0.006–0.004) vs. Type 2, 27.4% (17.7–43.1); two-sided Wilcoxon signed-rank test, FDR adjusted *p* = 0.033; Supplementary Figure S14). To identify di fferentially abundant genes, modules, and pathways between clusters, we conducted a linear discriminant analysis with e ffect size (LEfSe) method (Figure 4).

Subsequently, we compared the baseline symptom scales and BMI between Type 1 and Type 2 clusters. We found significant di fferences in the baseline PANNS, PANNS G, and CGI-S between the groups created from the clustering of the pathway abundance (Table 2). The patients classified into a Type 2 cluster had significantly more severe symptoms at baseline. The improvement in symptom severity after OLZ treatment assessed by PANNS, SF36, and CG1I was not associated with microbial community compositions (Supplementary Figure S14, Table S5) or KEGG features at baseline (Table 2; Supplementary Figures S15–S17 and Tables S6 and S7). Likewise, no associations were found between baseline gu<sup>t</sup> microbiota (Supplementary Figure S14, Supplementary Table S5) or its metabolic potentials (Table 2 and Supplementary Figures S15–S17 and Supplementary Tables S6 and S7) and the BMI change in the whole group or separately in women or men.

**Figure 4.** Unsupervised average linkage hierarchical clustering was carried out for each KEGG feature category. (**A**) KEGG orthologs; (**B**) KEGG modules; (**C**) KEGG pathways. K06147, ATP-binding cassette, subfamily B, bacterial; K02025, K02026, multiple sugar transport system permease proteins; K09687, antibiotic transport system ATP-binding protein; K02014, iron complex outer-membrane receptor protein; K03088, RNA polymerase sigma-70 factor, ECF subfamily; M00239, peptides/nickel transport system; M00237, branched-chain amino acid transport system; M00299, Spermidine/putrescine transport system; M00051, Uridine monophosphate biosynthesis, glutamine (+ PRPP) → UMP; M00222, phosphate transport system; M00002, glycolysis, core module involving three-carbon compounds; M00004, pentose phosphate pathway (pentose phosphate cycle); M00115, NAD biosynthesis, aspartate → NAD; M00126, Tetrahydrofolate biosynthesis, GTP → THF; M00006, pentose phosphate pathway, oxidative phase, glucose 6P → ribulose 5P; M00123, Biotin biosynthesis, pimeloyl-CoA → biotin; ko00511, other glycan degradation; ko00531, glycosaminoglycan degradation; ko00540, lipopolysaccharide biosynthesis; ko00780, biotin metabolism; ko00790, folate biosynthesis; ko00130, ubiquinone and other terpenoid–quinone biosynthesis; ko00460, cyanoamino acid metabolism; ko00020, citrate cycle (TCA cycle); ko00600, sphingolipid metabolism; ko00940, phenylpropanoid biosynthesis; ko00908, zeatin biosynthesis; ko00250, alanine, aspartate and glutamate metabolism; ko00430, taurine and hypotaurine metabolism; ko00040, pentose and glucuronate interconversions; ko00030, pentose phosphate pathway; ko02060, phosphotransferase system (PTS); ko00290, valine, leucine and isoleucine biosynthesis; ko00072, synthesis and degradation of ketone bodies; ko00860, porphyrin and chlorophyll metabolism; ko04122, sulfur relay system; ko02010, ABC transporters; ko00061, fatty acid biosynthesis; ko02030, bacterial chemotaxis.


**Table 2.** Associations of KEGG pathways with BMI changes and clinical improvements (PANNS, SF36, and CGI).

a Two-sided Wilcoxon rank-sum test, median with lower and upper quartiles in parentheses; BMI, PANNS, and SF36—changes from baseline (W0); CGI-I—an improvement from baseline; KEGG, Kyoto Encyclopedia of Genes and Genomes; BMI, body mass index; PANNS, positive and negative syndrome scale; SF36, 36- item short form survey; CGI, clinical global impression-improvement scale.

To further explore the gu<sup>t</sup> microbiota and OLZ treatment interactions, we classified the included patients as responders and non-responders as follows: Early responders, early non-responders, late responders, and late non-responders using the PANNS total score and responders and non-responders using the CGI-I scale. Subsequently, microbial community compositions and KEGG features were compared between responders and non-responders. Phylogenetic compositions of the samples at the phylum level in the responders and non-responders are shown in Supplementary Figure S18. The phyla were not differentially abundant in responders and non-responders, regardless of the definition of clinical improvement. There were no differences in gu<sup>t</sup> microbiome compositions at other taxonomic levels (Supplementary Figure S19), as well as in the KEGG orthologs, modules, and pathways (Supplementary Figure S20). Sex-specific results are shown in Supplementary Figures S21 and S22 (bacterial community composition) and S23 and S24 (KEGG features).
