*2.2. Study Protocol*

All participants were subjected to the same daily activities, including physical exercise (daily morning exercise and a walk with a therapist), occupational therapy, and psycho-educational activities. Two senior psychiatrists performed the psychiatric and basic physical examinations, and a gastroenterologist conducted a comprehensive physical examination.

Patients received a standard hospital diet (i.e., 2995 ± 93 kcal, 106 ± 14 g total protein, 420 ± 24 g carbohydrates, and 102 ± 10 g fat per day), balanced by a hospital dietician, in accordance with the Polish standards for hospitalized patients [33]. Detailed nutritional data on the diet during hospitalization, including fiber consumption, are presented in Supplementary Table S1.

This study included 20 patients, with 11 males and 9 females. After admission to the hospital ward, they were all subjected to a 7-day washout from psychiatric medications, received the standard hospital diet, and had a similar hospital routine. The first stool samples were collected after the washout period (W0), and subsequently, OLZ treatment was administered (initially 5 mg/day; doses were individually adjusted up to 20 mg/day). After 6 weeks of treatment, the second stool samples (W6) were collected (Supplementary Figure S1).

Clinical responses were defined as follows: Early responders, 30% reduction in positive and negative syndrome scale (PANNS) total score at 4 weeks; late responders, 40% reduction in PANNS total score at end-point [34]; Clinical global impression-improvement scale (CGI-I) responders, score of 3 points (much improvement); and non-responders, clinical global impression-severity (CGI-S) scores of 4 (minimal improvement) or 5 (no improvement).

#### *2.3. Processing of Raw Data and Statistical Analysis*

Sequencing of the V4 region of 16S rRNA gene was performed by the uBiome, Inc. (San Francisco, CA, USA). The 16S amplicons from each sample were individually barcoded and sequenced in the multiplex in the NextSeq 500 platform in a 150 bp (base pair) paired-end modality. The initial quality check of the 16S sequences was conducted using the AfterQC (version 0.9.7) software with default settings [35]. Subsequently, forward and reverse reads were, respectively, capped at 125 and 124 bp and then joined together with an in-between padding sequence (8 of "Ns" with a base score quality of 40). Each sequence was assigned the number of expected errors, and the sequences were filtered to have a maximum expected error of 1.0. The above steps were conducted using the VSEARCH (2.8.0) tool [36]. The sequences were processed using mothur (v.1.41.3) [37]. Briefly, sequences were aligned to the SILVA bacterial reference alignment (release 132), and were then screened to drop those not aligning to positions 13,148 and 25,277 of the SILVA alignment and were pre-clustered to allow two differences between sequences. The chimeras were identified and removed using VSEARCH implemented in mothur. Subsequently, sequences were classified using a Wang method with the Greengenes 16S rRNA Database version 13.8. Finally, sequences were clustered into OTUs using opticlust algorithm and Matthews correlation coefficient metric.

Metagenomic predictions from 16S rRNA marker genes (corrected for predicted 16S rRNA copy number) were carried out using PICRUSt (version 1.1.3) [38], and a list of the KEGG (Kyoto Encyclopedia of Genes and Genomes) functional orthologs (KO) was created. Reference genome coverage of the samples was calculated using the weighted nearest sequenced taxon index (NSTI) [38]. The PICRUSt predicted a median NSTI score of 0.11 (interquartile range, IQR of 0.05). The predicted metagenomes were analyzed with HUMAnN [39] and LEfSe [40]. The KO list was submitted as input data to HUMAnN, which generated KEGG modules and KEGG pathway abundances.

Downstream data analysis was performed using the R software (version 3.5.1, https://cran.rproject.org/), R based tools (such as Phyloseq package (version 1.24.2)) [41] and ComplexHeatmap [42], and custom-made scripts. Before calculating alpha diversity, the samples were rarefied to 3680 sequences per sample. Prior to beta diversity analysis, the taxa with the prevalence of less than 5% were removed (the prevalence of taxa was defined as the proportion of samples in which the taxa appeared at least once). Beta diversity was analyzed using principal coordinate analysis (PCoA) on Bray–Curtis distance matrices generated from the relative OTU abundances. To analyze the changes in bacterial community composition, a change in the principal coordinate 1 (PC1) was examined. The statistical analysis methods included the Wilcoxon rank-sum test, paired Wilcoxon signed-rank test, *t*-test for one sample, and Spearman rho correlation coefficient. *p*-values were adjusted using the Benjamini–Hochberg's false discovery rate (FDR) controlling procedure. Numerical data are presented as median, lower quartile, and upper quartile.
