Abstract

   Identifying genes whose expression is associated with schizophrenia
   (SCZ) risk by transcriptome-wide association studies (TWAS) facilitates
   downstream experimental studies. Here, we integrated multiple published
   datasets of TWAS, gene coexpression, and differential gene expression
   analysis to prioritize SCZ candidate genes for functional study.
   Convergent evidence prioritized Propionyl-CoA Carboxylase Subunit Beta
   (PCCB), a nuclear-encoded mitochondrial gene, as an SCZ risk gene.
   However, the PCCB’s contribution to SCZ risk has not been investigated
   before. Using dual luciferase reporter assay, we identified that
   SCZ-associated SNPs rs6791142 and rs35874192, two eQTL SNPs for PCCB,
   showed differential allelic effects on transcriptional activities. PCCB
   knockdown in human forebrain organoids (hFOs) followed by RNA
   sequencing analysis revealed dysregulation of genes enriched with
   multiple neuronal functions including gamma-aminobutyric acid
   (GABA)-ergic synapse. The metabolomic and mitochondrial function
   analyses confirmed the decreased GABA levels resulted from inhibited
   tricarboxylic acid cycle in PCCB knockdown hFOs. Multielectrode array
   recording analysis showed that PCCB knockdown in hFOs resulted into
   SCZ-related phenotypes including hyper-neuroactivities and decreased
   synchronization of neural network. In summary, this study utilized
   hFOs-based multi-omics analyses and revealed that PCCB downregulation
   may contribute to SCZ risk through regulating GABAergic pathways,
   highlighting the mitochondrial function in SCZ.

   Subject terms: Stem cells in the nervous system, Molecular
   neuroscience, Schizophrenia
     __________________________________________________________________

   Identifying schizophrenia risk genes is essential for illuminating the
   disease etiology. Here, authors prioritized Propionyl-CoA Carboxylase
   Subunit Beta (PCCB) as a schizophrenia-associated gene, and linked PCCB
   to GABAergic pathways using human forebrain organoids-based
   transcriptomic and metabolomic analysis.

Introduction

   Schizophrenia (SCZ) is a complex polygenic psychiatric disorder with
   risk contributed by environmental and genetic factors^[45]1. Genetic
   studies such as genome-wide association studies (GWAS) have identified
   hundreds of common single nucleotide polymorphisms (SNPs) associated
   with SCZ^[46]2,[47]3. Most of the SCZ-associated SNPs are noncoding
   variants located in regulatory DNA elements^[48]4–[49]6, suggesting
   that gene expression mediates the connection between genetic variants
   and SCZ phenotypes^[50]7. Identifying genes whose expression is
   associated with SCZ phenotypes facilitates discovering SCZ risk genes
   for downstream functional studies.

   By integrating SCZ GWAS and brain expression quantitative trait loci
   (eQTL) data, several approaches, which are collectively described as
   transcriptome-wide association studies (TWAS) have been used to
   identify SCZ risk genes. These TWAS approaches, including
   FUSION^[51]8–[52]10, PrediXcan^[53]11, summary-data-based Mendelian
   randomization (SMR)^[54]2,[55]10,[56]12,[57]13, and joint-tissue
   imputation approach with Mendelian randomization (MR-JTI)^[58]14, aimed
   to identify the association between predicted gene expression and SCZ
   risk. Though MR-JTI could improve gene expression prediction
   performance in TWAS and provide a causal inference framework^[59]15,
   experimental validation is still needed.

   Here we integrated results from MR-JTI^[60]14 and other published SCZ
   TWAS datasets^[61]2,[62]8–[63]13 to prioritize promising SCZ risk genes
   for functional study. Through this integrative analysis, we identified
   Propionyl-CoA Carboxylase Subunit Beta (PCCB), a protein-coding gene
   that plays important roles in mitochondrial metabolism^[64]16,[65]17,
   as an SCZ risk gene with the most supporting evidence in our analysis.
   However, the PCCB’s contribution to SCZ risk has not been investigated
   before. Using human forebrain organoids (hFOs), three-dimensional cell
   cultures that recapitulate key aspects of the human brain^[66]18, we
   found that PCCB knockdown in hFOs resulted into SCZ pathology-related
   cellular phenotypes. We also identified that SCZ-associated common SNPs
   rs6791142 and rs35874192 may regulate PCCB expression, supporting that
   PCCB expression may mediate the genetic effects on SCZ risk.

Results

PCCB is prioritized as a promising SCZ risk gene

   To obtain reliable SCZ risk genes for downstream functional study, we
   integrated multiple published TWAS datasets (Supplementary Data [67]1)
   to prioritize genes with sufficient supporting evidence. We also
   checked whether the prioritized genes are located in SCZ
   risk-associated gene coexpression module or dysregulated in postmortem
   SCZ brains. These analyses prioritized PCCB, GATAD2A, and GNL3 as the
   top three SCZ risk genes (Table [68]1). Notably, PCCB was also
   identified as an SCZ risk gene in the gene-based MAGMA analysis^[69]19.
   Moreover, PCCB was located in the gene coexpression module (M2)
   downregulated in SCZ based on the PsychENCODE data^[70]10. PCCB was
   also found to be nominally downregulated in postmortem SCZ brains
   (P = 0.01, FDR = 0.14) by checking the SZDB database^[71]20, which
   integrated transcriptome data from the CommonMind consortium^[72]21.
   These lines of evidence suggested that PCCB expression mediated the
   genetic effects on SCZ risk. Therefore, we focused on studying how PCCB
   contributes to SCZ risk in this study.

Table 1.

   Prioritized top three SCZ risk genes
   Gene MR-JTI TWAS SMR Differential expression Coexpression module
   FUSION PrediXcan FUSION FUSION
   Wu et al.^[73]14 Hall et al.^[74]9 Huckins et al.^[75]11 Gusev et
   al.^[76]8 Gandal et al.^[77]10 Yang et al.^[78]13 Gandal et al.^[79]10
   Li et al.^[80]12 Goncalves et al.^[81]19 Gandal et al.^[82]10
   PCCB P[Bonferroni] = 3.17E-11 P[TWAS] = 5.39E-12 P = 2.05E-08
   P[TWAS] = 3.07E-10 P[Bonferroni] = 2.42E-05 – P[SMR] = 3.74E-10
   P[SMR] = 4.17E-15 ↓ P = 0.01 ↓ FDR = 6.71E-03
   GATAD2A P[Bonferroni] = 1.87E-08 P[TWAS] = 8.67E-11 P = 2.18E-10
   P[TWAS] = 8.83E-07 P[Bonferroni] = 6.98E-09 P[SMR-multi] < 1.00E-05
   P[SMR] = 2.21E-10 – – ↑ FDR = 5.03E-03
   GNL3 P[Bonferroni] = 1.58E-18 – P = 1.39E-11 P[TWAS] = 6.00E-07
   P[Bonferroni] = 8.24E-03 P[SMR-multi] < 1.00E-05 P[SMR] = 4.71E-09
   P[SMR] = 4.53E-13 – –
   [83]Open in a new tab

   ↓, Downregulation in SCZ; ↑, Upregulation in SCZ. P values are adjusted
   by the Benjamini-Hochberg or Bonferronid methods.

PCCB eQTL SNPs rs6791142 and rs35874192 affect transcriptional activities

   Since PCCB expression is genetically associated with SCZ, we
   investigated the functional impacts of SCZ-associated SNPs on PCCB
   expression. Based on the TWAS results used in this study, we retrieved
   the top SNPs (rs7432375, rs7427564, rs527888, rs66691851) and their
   linkage disequilibrium (LD) SNPs that were associated with PCCB
   expression. To narrow down to the putatively causal variants, we
   focused on those eQTL SNPs (eSNPs) that are likely to affect PCCB
   expression in the brain. Since opening chromatin facilitates gene
   expression activation, we used brain ATAC-seq data from the PsychENCODE
   consortium^[84]22 to identify PCCB eSNPs located in active
   transcription regions. By integrating PsychENCODE ATAC-seq data and SNP
   annotation information from the Roadmap Epigenetics Consortium^[85]23,
   we prioritized six eSNPs (rs6791142, rs35874192, rs900818, rs7616204,
   rs570621, and rs7349597) (Table [86]2) that were located in genomic
   regions strongly suggested as enhancers or promoters in the human brain
   tissues or neural cell cultures.

Table 2.

   Prioritized six eQTL SNPs for PCCB
   Source References Best GWAS SNP LD eSNP LD (r^2) Ref/Alt Chromatin