This cohort included 10 schizophrenia patients (6 males, 4 females) and 10 healthy controls (3 males, 7 females). All participants read a written description of the study objectives and provided written informed consent.

Qualified psychiatrists at both hospitals diagnosed schizophrenia using the Diagnostic and Statistical Manual of Mental Disorders, 5th Edition (American Psychiatric Association, 2013). The severity of psychotic symptoms was assessed using the 30-item Positive and Negative Symptom Scale (PANSS). Clinical data, including age, gender, race, PANSS score, and BMI, were collected from both patients and healthy controls by evaluating psychiatrists. Schizophrenia patients were receiving a single antipsychotic medication at the time of the study: eight were on risperidone and two on olanzapine. Excluded from the study were patients with recent antibiotic use, substance addiction, alcohol consumption, HIV infection, severe neurologic problems, or mental impairment. Healthy control individuals were screened using the Structured Clinical Interview for Non-Patient Edition (SCID-NP) and were excluded if they had a history of medical or mental illnesses, drug or alcohol misuse, or recent antibiotic use.

Each participant received a fecal collection kit with explicit instructions for self-collection of feces following defecation. The kit included disposable gloves, a sterile 50 ml Falcon tube with a wooden tongue depressor for stool collection, a 15 ml Falcon tube containing 5 ml of RNA Protect Bacteria reagent (Qiagen, USA) for sample preservation, and a stand to hold the tube upright during storage, transport, and laboratory analysis. Fecal samples were collected on the same day or within two days, and participants notified researchers once the sample was collected. The samples were either analyzed immediately upon arrival at the laboratory or frozen at −80 °C for later analysis. Blood samples (2 ml) were collected via venipuncture, stored in EDTA anticoagulant tubes at −80 °C, and analyzed later.

ELISA was used to measure plasma levels of dopamine, serotonin, and BDNF. Competitive ELISA kits (Elabscience, USA) measured dopamine and serotonin, while the Human BDNF ELISA Kit PicoKine™ Pre-Coated ELISA kit (Boster Bio, USA) assessed BDNF levels. All tests followed the manufacturer’s instructions. The Mann–Whitney U test was used to compare neurotransmitter levels between groups, and the Spearman rank correlation test assessed correlations between dopamine and serotonin levels.

RNA extraction was performed using the RNeasy Power Microbiome Kit (Qiagen, USA) with modifications in the lysis step. A total of 0.25 g of feces was transferred to a tube prefilled with 0.1 mm glass beads. The tube was filled with 650 μl of lysis solution PM1 (chemical lysis solution) and 6.5 μl of beta-mercaptoethanol, then vortexed using the Omni Bead Ruptor 4 Bead Mill Homogenizer (Omni International, USA) at 3 m per second (m/s) for 6 min and 5 m per second (m/s) for 8 min. The remaining steps of the protocol were carried out according to the manufacturer’s instructions. rRNA was removed using the Ribo-Zero rRNA Removal kit (Illumina, USA), and library quality was controlled with Qubit 2.0 (Thermo Scientific, USA). Libraries were sequenced on the Illumina HiSeq 2000 platform (Illumina, USA).

Bioinformatics analysis was performed using the ASaiM pipeline [13]. Initial sequence quality control was executed with FastQC v0.11.8 (2018, Babraham Bioinformatics, United Kingdom) to assess overall read quality. Data cleaning was carried out using Cutadapt v2.8 (2019, Marcel Martin, Germany), which removed adaptors and low-quality reads, ensuring a minimum read length of 30 base pairs and a Phred quality score threshold of 30. To differentiate between ribosomal and non-ribosomal RNA, SortmeRNA v2.1 (2016, Evguenia Kopylova, Laurent Noé, and Hélène Touzet, France) was employed, focusing on non-ribosomal RNA to represent the functional microbial community.

Microbiota community identification was conducted using MetaPhlAn2 v2.6.0 (2015, Nicola Segata, Harvard T.H. Chan School of Public Health, USA), which utilizes clade-specific marker genes from a comprehensive database of 17,000 reference genomes (including 13,500 bacterial and archaeal, 3500 viral, and 110 eukaryotic genomes). Functional analysis of microbial communities was performed using HUMaN2 v0.11.1 (2012, Curtis Huttenhower, Harvard T.H. Chan School of Public Health, USA).

The sequencing effort generated a total of 636,206,364 unprocessed sequences across 20 samples, with individual sequence counts ranging from 26,645,328 to 37,197,780. Quality metrics were assessed with Q20 and Q30 percentages ranging from 97.41 to 99.03% and 92.84 to 95.76%, respectively. These metrics corresponded to base-calling accuracies between 97.67 and 99.22%, with minimal base-calling errors ranging from 0.02 to 0.03%. All samples passed the quality control criteria for Q20 and Q30, and the GC content of all raw data was within the acceptable range of 40–60%.

Statistical analysis of clinical variables, including age, BMI, gender, neurotransmitter levels, and bacterial species abundance, was conducted using GraphPad Prism (version 9.0.0, 2020; GraphPad Software, San Diego, CA, USA). The modalities employed from GraphPad Prism included the Shapiro–Wilk test to assess whether the data followed a normal distribution, the Anderson–Darling test to provide a measure of goodness-of-fit for the distribution, and the Mann–Whitney U test to compare differences in non-normally distributed data between schizophrenia patients (SCZ) and healthy controls (HC).

Bacterial profiles of HC and SCZ were visually examined at the family, genus, and species levels using a TSS-normalized OTU count table. Alpha diversity metrics, including Observed, Chao1, Shannon, and Simpson indices, were calculated to quantify the diversity within each sample. These metrics were analyzed using the R function estimate richness and visualized using GraphPad Prism version 9 (2020, GraphPad Software, USA).

Beta diversity, which measures differences in microbial community composition between HC and SCZ, was assessed using the MicrobiomeAnalyst v1.0 (2017, McGill University, Canada). Bray–Curtis and Jaccard indices were calculated and visualized using non-metric multidimensional scaling (NMDS). The PERMANOVA test was applied to determine the statistical significance of observed differences in microbial community composition between the two groups.

To identify significant differences in microbial abundance and functional metabolisms between HC and SCZ, we used LEfSe v1.1.01 (2011, Harvard T.H. Chan School of Public Health, USA (Linear Discriminant Analysis Effect Size) with a Linear Discriminant Analysis (LDA) threshold >0.2 and p-value <0.05. LEfSe is designed to detect taxonomic features and functional biomarkers that differ significantly between groups, accounting for both biological significance and effect size. This analysis identified differentially abundant microbial taxa and functional pathways associated with schizophrenia.

To evaluate the diagnostic potential of different microbial species, Receiver Operating Characteristic (ROC) curves were generated using GraphPad Prism version 9 (2020, GraphPad Software, USA). ROC curves assessed the ability of specific microbial features to distinguish between HC and SCZ, providing insights into their potential as diagnostic biomarkers.

To explore the relationships between bacterial species abundance, neurotransmitter levels, and functional pathways, we conducted Spearman rank correlation analysis. This non-parametric method assessed the strength and direction of monotonic associations among these variables, providing insights into how changes in microbial communities relate to neurotransmitter levels and functional pathways.