1. glutamic acid decarboxylase 1 (GAD1) and GAD2 were downregulated in both postmortem brains and organoids derived from patients with schizophrenia, suggesting a neurodevelopmental origin.

2. Somatostatin (SST) and parvalbumin (PVALB) were reduced in brain tissue but trended upward in organoids, indicating their downregulation may arise later in disease progression.

3. PVALB downregulation in blood suggests its potential as a biomarker.

Schizophrenia is a complex psychiatric disorder, affecting roughly 0.7% of the people worldwide [1]. The underlying cause of the disease remains unclear. Still, it is widely acknowledged to stem from an intricate interplay of genetic and environmental factors, with genetics playing a substantial role, accounting for around an 80% heritability [2, 3]. Genetic studies, such as genome-wide association studies (GWAS), have identified hundreds of common single-nucleotide polymorphisms (SNPs) associated with schizophrenia [4]. The majority of these schizophrenia associated SNPs are non-coding variants situated within regulatory DNA elements [5, 6]. This pattern strongly implies that the interplay between genetic variations and schizophrenia phenotypes is predominantly mediated through gene expression [7]. Therefore, the study of gene expression and the identification of differentially expressed genes are critical to improving our understanding of the biological basis of schizophrenia.

Identifying distinct abnormalities in schizophrenia is challenging due to its involvement in multiple biochemical pathways. The disease is linked to neurotransmitters like dopamine, serotonin, norepinephrine, gamma-aminobutyric acid (GABA), and glutamate [8].

GABA is the primary inhibitory neurotransmitter and plays an essential role in neural network oscillations, balancing excitation and inhibition, maturation of neural circuitry, and neuronal differentiation [9].

GABA has a regulatory effect on dopamine activity, and a decline in GABAergic activity could lead to hyperactivity of dopaminergic neurons, which is thought to be involved in schizophrenia [10, 11, 12].

GABA is synthesized by the enzymes glutamic acid decarboxylase (GAD) 67 and GAD65, encoded by the GAD1 and GAD2 genes, respectively. GAD1 is primarily expressed in the brain and is predominantly located in the cytoplasm, where it supports continuous GABA production for cell metabolism and tonic, non-vesicular GABA release. In contrast, GAD2 is expressed both in the brain and the pancreas and is preferentially localized in axon terminals, associated with synaptic vesicles, where it plays a critical role in activity-dependent, vesicular GABA release during intense synaptic activity [13, 14].

In genetic studies, there is evidence of a notable association between variants in the GAD1 gene and schizophrenia, which supports the hypothesis that it plays a causal role in the development of schizophrenia [15, 16]. However, the latest GWAS did not detect a significant association between GAD1 or GAD2 and schizophrenia, as demonstrated by the RICOPILI tool [4, 15].

Multiple studies have reported downregulation of GAD1 in schizophrenia, evidenced through the analysis of postmortem brain samples for messenger RNA (mRNA) and protein levels [16, 17]. Notably, a review of 25 postmortem studies shows a consistent reduction of GAD1 gene and the encoded GAD67 protein in the dorsolateral prefrontal cortex (DLPFC) [17, 18, 19]. For example, a postmortem study measured both GAD1 mRNA and GAD67 protein levels in the same DLPFC samples from individuals with schizophrenia, finding coordinated significant reductions in both measures. In contrast, GAD65 protein levels did not show a significant reduction [17]. Similar concordant reductions in GAD1 mRNA and GAD67 protein were reported in a second study of schizophrenia DLPFC samples [18].

GABAergic neurons can be subdivided into distinct types based on the specific molecular markers they express. Among these are parvalbumin (PVALB) expressing interneurons, characterized by the calcium-binding protein parvalbumin, and somatostatin (SST) expressing interneurons, marked by the neuropeptide somatostatin. These interneuron types differ significantly in their anatomical features, connectivity, and electrophysiological properties. SST interneurons predominantly target dendritic regions and enhance excitatory input processing, while PVALB interneurons specialize in providing rapid, synchronized inhibition at the soma and proximal dendrites, thus facilitating faster neural dynamics within cortical circuits [20, 21].

GAD1 expression was reduced in PVALB-expressing prefrontal cortex (PFC) interneurons, where around 50% lacked detectable levels of GAD1 in individuals with schizophrenia [22]. Interestingly, both SST and PVALB-expressing GABA neurons showed downregulated expression of SST and PVALB, respectively, in the DLPFC of individuals with schizophrenia [23]. In addition to the DLPFC, additional brain regions were shown to have downregulated expression of GAD1 [12, 19]. Fewer studies explored the levels of GAD2 and the encoded protein GAD65, and the findings were less consistent, as summarized in Table 1 (Ref. [17, 18, 24]) and Table 2 (Ref. [11, 12, 16, 17, 18, 22, 25, 26, 27, 28, 29]).

It should be noted that postmortem studies cannot determine whether differential expression is a causative factor or a consequence of the development of schizophrenia. In this context, recent technological advancements have significantly benefited in vitro organoid preparations, advancing developmental neuroscience [30]. Brain organoids, which are self-assembled three-dimensional aggregates derived from pluripotent stem cells, mimic embryonic human brain structures. This makes them valuable models for studying brain development and disorders [31]. Research on schizophrenia using patient-derived cerebral organoids has uncovered unique gene expression profiles, for example, in genes related to synaptic function and the regulation of excitation-inhibition balance, including significant downregulation of GAD1 and GAD2 [32].

In conclusion, while previous findings regarding GAD1 expression were consistent, there are inconsistencies regarding GAD2 gene expression in tissue samples from subjects with schizophrenia. Moreover, the interplay between these genes remains unclear.

We conducted a systematic participant data meta-analysis following the PRISMA 2020 guidelines (Supplementary Material-PRISMA_checklist) [33]. Publicly available gene expression datasets from brain samples were included to systematically calculate the differential expression of GAD1, GAD2, as well as SST and PVALB, and their correlation patterns in different brain regions of patients with schizophrenia. Additionally, we examined their expression in organoids derived from patients, offering insights into early neurodevelopmental stages, and in blood samples from patients to assess their potential as biomarkers.

In our study, we conducted a systematic participant data meta-analysis of seven brain samples gene expression datasets (overall 295 samples, 151 schizophrenia vs. 144 controls). Our analysis revealed a marked reduction in GAD1 (Hedges’ g = –0.55, 95% CI: –0.89 to –0.22, p = 0.0012), GAD2 (Hedges’ g = –0.56, 95% CI: –0.93 to –0.19, p = 0.0029), PVALB (Hedges’ g = –0.38, 95% CI: –0.71 to –0.06, p = 0.021) and SST (Hedges’ g = –0.71, 95% CI: –1.07 to –0.35, p = 0.00012) in brain samples of individuals with schizophrenia. In addition, GAD1 and GAD2 were significantly downregulated in brain organoids derived from individuals with schizophrenia (p = 0.00064, 0.0014, respectively), as was previously published [32], while SST and PVALB showed upregulated expression that did not reach statistical significance (p = 0.069, 0.062, respectively). While GAD1 was previously shown to be downregulated in brain samples of patients with schizophrenia, the results regarding GAD2 were inconsistent (summarized in Table 2). Our meta-analysis detects significant downregulation of GAD2 (g = –0.56, p = 0.0029) in both cortical and subcortical areas. In addition, as previously observed [32], GAD1 and GAD2 showed downregulation in cerebral organoids derived from patients with schizophrenia. This finding supports the hypothesis that gene expression alterations occur during neurodevelopment, rather than solely as a consequence of disease progression.

Beyond the downregulation in brain samples, we also detected downregulation of PVALB in blood samples of individuals with schizophrenia (p = 0.027), in a meta-analysis of 293 samples (160 schizophrenia and 133 controls). This suggests the PVALB gene’s potential role as a biomarker of schizophrenia.

The current study has certain limitations that should be acknowledged. Like other postmortem studies, it captures the neurobiological characteristics at the end of life, thus offering insights solely into the terminal stages. However, by examining also organoids that replicate embryonic human brain structures, we provide insights into gene expression changes occurring during early developmental stages. We note that the downregulation of GAD1 and GAD2 has already been detected in the same organoids dataset that we looked at [32], while the expression of SST and PVALB was not examined. In addition, the dataset we examined has a relatively small sample size (n = 16, comprising 8 cases and 8 controls). This small number of samples may have contributed to the lack of statistical significance when we examined SST and PVALB differential expression in organoid samples. Another limitation of this work is the relatively small number of included gene expression datasets, which reduces the robustness of our conclusions and constrains our ability to perform more refined analyses. Substantial heterogeneity (I² $>$50%) in the differential expression of GAD2 and SST was observed, which indicates variation in how these genes were differentially expressed between schizophrenia and controls across datasets (Table 4). For GAD2, three of the seven datasets demonstrated significant downregulation in schizophrenia (p 0.05), while the remaining four did not show significant differential expression (Fig. 2B). A similar pattern was observed for SST (Fig. 2C). Notably, two of the four datasets without significant differential expression, GDS1917- cerebellum and GSE37981- pyram3 STG, nonetheless showed downregulation that did not reach statistical significance. This heterogeneity may reflect both technical and biological factors, including differences in platforms, postmortem interval (PMI), pH, medication exposure, and brain regions analyzed. While we could not isolate the impact of each of these factors, they are all plausible contributors to the observed variability.

Moreover, the peripheral blood datasets included in our analysis were heterogeneous, both in terms of sample type (whole blood versus peripheral blood mononuclear cells (PBMCs)) and in gene expression platforms used (Affymetrix versus Illumina arrays). Such variability may have introduced technical and biological noise, potentially contributing to inconsistent findings across genes.

In addition, prolonged exposure to antipsychotics may affect gene expression. To address this bias, we conducted a linear regression analysis that accounted for the potential effect of antipsychotics and additional confounders. While information on the use of antipsychotics was available only for the parietal cortex dataset (GSE35978), all genes consistently exhibited downregulation in schizophrenia, even after accounting for the effects of antipsychotics (Supplementary Table 1). The analysis of gene expression of postmortem brain samples is exposed to various technical and biological noise. However, the significant correlation coefficient values we detected between the expression patterns of all four genes (Supplementary Table 2) support the results of our meta-analysis and reduce the likelihood of false positives resulting from arbitrary noise. Another limitation is the fact that we measured gene expression levels, which may not accurately reflect the levels of the encoded proteins. Consequently, further investigation is essential to elucidate the implications of the signal we detected in relation to protein levels and activity. Nevertheless, several postmortem studies have demonstrated that decreased GAD1 mRNA levels correspond to a proportionate reduction in GAD67 protein, supporting the biological relevance of transcript-level findings [17, 18, 27].

In conclusion, the significant reduction of GAD1 and GAD2 expression in both schizophrenia brain samples and organoids supports the hypothesis that decreased expression begins prior to disease onset, potentially contributing to its development. In contrast, the reduced expression of SST and PVALB, observed in brain samples but not in organoids, likely reflects changes occurring at later stages. The downregulation of PVALB in blood samples further suggests its potential as a peripheral biomarker for schizophrenia.

Further research is necessary to clarify the causal role of reduced GABA activity in schizophrenia and to investigate the utility of PVALB expression as a biomarker. Validation of GAD1 and GAD2 downregulation in independent iPSC-derived organoid models will be essential to confirm reproducibility. Integrating additional molecular layers, such as proteomics, could enhance our understanding of the underlying biological mechanisms. Likewise, assessing PVALB expression in larger and more heterogeneous blood cohorts will help establish its robustness as a biomarker for diagnosis, disease monitoring, or treatment response. Together, such efforts would strengthen the translational significance of our findings and help bridge the gap between transcriptomic alterations and clinical applicability.

Gene expression datasets are available through the Gene Expression Omnibus (GEO) database (GDS4523, GSE37981, GDS1917, GSE35978, GSE80655, GSE53987, GSE138082, GSE38481, GSE18312, GSE38484, GSE133534).

LH: Supervision, Conceptualization, Methodology, Writing - Reviewing and Editing. YS: Conceptualization, Methodology, Investigation, Writing- Original draft preparation. AY: Methodology, Software, Data Curation, Visualization. All authors contributed to editorial changes in the manuscript. All authors read and approved the final manuscript. All authors have participated sufficiently in the work and agreed to be accountable for all aspects of the work.

Informed consent was waived as the analysis relied exclusively on publicly accessible data.

We thank Professor Eytan Domany for his help with the infrastructure that enabled the performance of this project.

This research received no external funding.

The authors declare no conflict of interest.

Supplementary material associated with this article can be found, in the online version, at https://doi.org/10.31083/AP43531.