Alternative splicing is an essential molecular mechanism underlying risk of schizophrenia

Schizophrenia is a severe disabling mental illness with a global lifetime prevalence of ~1% [1]. Accumulating studies have indicated a strong genetic component in schizophrenia pathogenesis [2], and genetic analyses including genome-wide association studies (GWASs) have reported multiple single-nucleotide polymorphisms (SNPs) associated with this illness [3,4,5]. There is a growing consensus that genetic risk of schizophrenia tends to affect mRNA expression in human brains [6,7,8,9,10]. During the past few years, mRNA expression analyses using multiple approaches (such as real-time quantitative PCR, microarray or RNA-sequencing (RNA-seq)) have identified many dysregulated genes associated with schizophrenia. While the functional outcomes of altered mRNA expression related to schizophrenia pathology are being investigated, the impact of the much more complicated regulatory network of RNA processing during transcription has drawn growing attention, and the role of alternative splicing in schizophrenia has been explored [11,12,13]. In the human genome, majority of the multi-exon genes are alternatively spliced during transcription [14], resulting in cassette exons, microexons, intron retention, alternative 5′ and 3′ splice sites, alternative promoters, and alternative untranslated regions (UTRs) (Fig. 1) [15], and thereby producing diverse transcriptomes, proteomes, and phenomes [16]. In this perspective, we mainly discuss the contributions of alternative splicing to neuropathology of schizophrenia, highlighting potential mechanisms by which miss-splicing events related to schizophrenia genetic risk facilitate its pathogenesis.

Fig. 1: Alternative splicing patterns.
figure 1

The solid and dash V-shaped lines represent two distinct splicing options, respectively. a Cassette exons: the most common pattern of alternative splicing is the inclusion or exclusion of a cassette exon in the mRNA. b Microexons: a special type of cassette exon with 3–27 nucleotides shows enrichment in neuron-specific transcripts. c Intron retention: the omission of intron exclusion leaves the retained intronic sequence (shown as purple block) in mature mRNA transcript. d, e Alternative 5′/3′ splice sites: through selecting different combinations of the 5′ (donor)/3′ (acceptor) splice sites, exons can be extended or shortened in length. f, g Alternative 5′/3′ exons: different transcriptional initiation or termination sites generate alternative 5′-terminal exons or 3′-terminal exons with alternative polyadenylation sites (shown as gray blocks).

Full size image

Alternative splicing in brain is highly complex [17], suggesting that essential roles of appropriate RNA splicing in myriad neuronal development and functions [14], and aberrant splicing of particular genes may underlie the pathogenesis of brain disorders [18]. For example, Gomafu, a long noncoding RNA whose expression is decreased in postmortem brains of schizophrenia patients, acts as a scaffold for splicing factors such as serine/arginine-rich SF 1, and knockdown of Gomafu increases the expression of schizophrenia-associated isoforms of ErbB4 and DISC1 [19]. Similarly, compared with longer exons, alternative microexons (i.e., a class of exons comprising 3–27 nucleotides, are strongly conserved and usually frame-preserving [20]) are preferentially brain-enriched and regulate neuronal differentiation [21, 22], and Ganda et al. found significant enrichment of switched isoforms with microexons in schizophrenia patients [23]. In addition, cumulative studies have also reported dysregulated mRNA levels of alternatively spliced isoforms of schizophrenia risk genes in brains of patients, which will be described in the following sections.

Alternative splicing promotes understanding of schizophrenia pathogenesis

Neurotransmitter dysfunction (including dopamine, glutamate, and γ-aminobutyric acid (GABA) systems) and neurodevelopmental disturbances are central hypotheses of schizophrenia pathology [24]. However, many questions still remain regarding how genetic risks contribute to these hypothesized models. Intriguingly, several essential genes involving in these schizophrenia hypotheses are alternatively spliced (Table 1), producing different isoforms with distinct functions, which may in part explain the disease pathogenesis. We herein briefly discuss the alternative splicing patterns of several genes in human brains, including DRD2 (encoding dopamine 2 receptor), GRM3 (encoding metabotropic glutamate receptor 3 (mGluR3)), GAD1 (encoding glutamic acid decarboxylase), DISC1 (encoding the disrupted in schizophrenia 1 scaffold protein), NRG1 (encoding neuregulin 1), and ErbB4 (encoding erb-b2 receptor tyrosine kinase 4).

Table 1 Alternative spliced isoforms in the schizophrenia risk genes.
Full size table

Dopamine hypothesis of schizophrenia

The dopamine hypothesis of schizophrenia is originated from the observations that neuroleptic drugs (e.g., chlorpromazine) could block brain dopamine receptor dates back to the ~1960s [25,26,27]. Until now, many major antipsychotic drugs are still designed to antagonize dopamine 2 receptor and are proven effective in alleviating positive symptoms among schizophrenia patients. The dopamine 2 receptor is a central molecule of dopamine signaling involved in schizophrenia and a major antipsychotic drug target [28, 29], and SNPs spanning DRD2 showed genome-wide associations with the illness [3]. The DRD2 gene is transcribed primarily into two isoforms, D2-short receptor (D2SR, skipping exon 6) and D2-long receptor (D2LR, inclusion of exon 6) [30]. D2SR encodes a presynaptic receptor of dopamine, whereas the protein encoded by D2LR mainly mediates postsynaptic dopamine signaling [30, 31], and multiple studies have examined changes of these DRD2 isoforms in schizophrenia. A recent study found elevated mRNA expression of D2SR and simultaneously reduced expression of D2LR in the dorsolateral prefrontal cortex (DLPFC) of schizophrenia patients compared with controls [32], although inconsistent results have also been reported [33]. In addition, studies have consistently found significant associations between these isoforms and schizophrenia genetic risk. For example, the mRNA expression of D2SR has been found to be significantly associated with a schizophrenia risk SNP rs1076560 in the intron 6 of DRD2 [33, 34]. Furthermore, Cohen et al. showed that rs1076560 affected binding affinity for a splicing regulator ZRANB2 in an in vitro oligonucleotide assay, and this SNP was correlated with changes of D2SR/D2LR ratio in a minigene assay when ZRANB2 was co-expressed [34]. Intriguingly, rs1076560 was significantly associated with activity and functional connectivity of striatum/DLPFC during working memory tasks [33, 35], as well as that of amygdala/DLPFC during emotion processing [36], providing hints for the physiological impact of D2SR. Therefore, schizophrenia genetic risk (e.g., rs1076560) likely modulates the balance between D2SR and D2LR, and thereby affects D2 receptor-mediated signaling and physiological consequences (Fig. 2). However, the transcription of DRD2 is likely also affected by other schizophrenia risk factors in addition to this SNP, as its disease risk allele is unlikely the causal factor for the increased DRD2 expression in schizophrenia [12].

Fig. 2: The association between schizophrenia genetic risk and alternative splicing of DRD2.
figure 2

a The schizophrenia risk allele at rs1076560 (T-allele) facilitates the inclusion of exon 6 and is also associated with higher mRNA level of D2LR by abolishing the binding site of ZRANB2, while G allele has opposite effect. D2SR mRNA lacks exon 6 comparing with D2LR mRNA, which results in a deletion of 29 amino acids (represented by gray dash lines) at IL-3 (represented by red lines) of D2SR. b Different IL-3 structures of D2SR and D2LR affect their binding properties with G-protein subunits and the downstream signaling pathways activated. Upper figure: D2SR predominantly locates at presynaptic dopaminergic neurons and acts as auto-receptor to provide a negative feedback for modulation of neuron firing and dopamine neurotransmission. Activation of D2SR reduces dopamine synthesis via regulation of TH activity, facilitates dopamine reuptake partially through increasing the surface expression of DAT, and represses neuron excitability. Bottom figure: D2LR mainly locates at postsynaptic dopaminoceptive neurons. Activation of D2LR inhibits cAMP production, thereby mediating the phosphorylation state of DARPP-32, which is a major target for dopamine in striatum and also a potent inhibitor of a multifunctional PP1. The D2LR-triggered DARPP-32/PP1 cascade shows impacts on a wide range of downstream effectors including neurotransmitter receptors and ion channels in dopaminoceptive neurons, e.g., striatal medium spiny neurons. Moreover, D2LR signaling regulates intracellular calcium levels involving Gβγ activation of PLCβ-IP3-calcineurin cascade. It is noteworthy that activation of D1-class and D2 heterodimer receptors modulates calcium signaling mediated by Gαq and PLCβ/IP3 pathway. AC adenylyl cyclase, β, γ, α/io, αq, subunits of G-protein complex, cAMP 3′,5′-cyclic adenosine monophosphate, DA dopamine, DAG diacylglycerol, DARPP-32 dopamine- and cAMP-regulated phosphoprotein 32 kDa, DAT dopamine transporter, GTP guanosine triphosphate, IL-3 third intracellular loop, IP3 inositol 1,4,5-trisphosphate, p phosphorylated, PKA protein kinase A, PKC protein kinase C, PLC-β phospholipase C-β, PP1 protein phosphatase 1, TH tyrosine hydroxylase.

Full size image

Glutamate and GABA hypothesis of schizophrenia

Apart from the dopamine hypothesis for schizophrenia pathogenesis, it has also been known for ~30 years that antagonists of N-methyl-D-aspartate receptor (NMDAR), such as phencyclidine and ketamine, could induce schizophrenia-like negative symptoms and subtle cognitive impairments [37, 38], promoting the formation of glutamate hypofunction hypothesis [39, 40]. Indeed, the lost dendritic spines in postmortem brains of schizophrenia patients were primarily characterized as excitatory glutamatergic synapses [41]. The mGluR3 is a type of G-protein-coupled receptor (GPCR) modulating glutamate neurotransmission and synaptic plasticity through enhancing glutamate uptake [42], and previous studies have reported significant associations between SNPs in GRM3 (e.g., rs2228595) and risk of schizophrenia, cognition, prefrontal, and hippocampal physiology, as well as glutamate neurotransmission [3, 43]. Several lines of evidence have also demonstrated the effects of mGluR2/3 agonists, such as LY354740 and LY379268, on ameliorating NMDAR antagonist-induced behavioral defects in animals [44,45,46,47]. Corti et al. reported decreased dimeric form of mGluR3 in schizophrenia patients [48]. Notably, abundant presence of a spliced isoform lacking exon 4 in mGluR3 (mGluR3Δ4) was identified in human brain. This isoform was predicted to translate into a truncated protein that lacked the transmembrane domain while contained a novel intracellular C-terminal and a partially truncated extracellular ligand binding domain [49]. Intriguingly, mRNA expression of mGluR3Δ4 in human brain, rather than mGluR3, was associated with the schizophrenia risk SNP rs2228595 [50]. The mGluR3Δ4 also possesses altered functions compared with mGluR3. Despite lacking the transmembrane region, mGluR3Δ4 still locates at the plasma membrane in cultured hippocampal neurons [49] probably due to its heterodimerization with mGluR3. As dimerization between some full-length GPCRs and their truncated isoforms has been commonly seen, the resultant complexes generally show unique subcellular localizations, ligand binding properties, and pharmacological potentials compared with the canonical homodimers formed by full-length GPCRs [51,52,53]. Likewise, a recent study showed physical interactions between mGluR3Δ4 and mGluR3, and transfection of mGluR3Δ4 could cause deficits in ligand binding availability and decreased membrane mGluR3 abundance [54]. Together, these results suggest a possible mechanism underlying schizophrenia risk through negative modulation of mGluR3 function by mGluR3Δ4 [50].

In addition, dysfunction of GABAergic neurotransmission has been implicated in schizophrenia through studies of human postmortem brain tissues and in vivo levels of GABA [55,56,57], and alternative splicing of GAD1 gene likely contributes to GABA dysfunction in the brains of schizophrenics [58]. There are two primary isoforms transcribed by GAD1, a full-length isoform encoding the active protein glutamic acid decarboxylase 67 (GAD67, 67 kDa) and a shorter isoform GAD25 (25 kDa) encoding a protein lacking the enzymatic domain [59]. As the most abundant GAD1 transcript in human brain, GAD67 regulates the synthesis of GABA [60], and the mRNA and protein levels of this isoform were both decreased in the postmortem brains of schizophrenia patients compared with normal controls [58, 61,62,63,64]. Further analyses by Hyde et al. and Tao et al. respectively demonstrated an increased ratio of GAD25 to GAD67 in schizophrenics compared with healthy controls [63, 64]. Intriguingly, the altered expression ratio of GAD1 isoforms could be predicted by a schizophrenia risk SNP (rs3749034) within this gene. Given that transcription of GAD25 might have jeopardized appropriate expression of GAD67, the molecule necessary for mature GABA signaling, this mis-splicing-mediated developmental deficit might underlie pathogenesis of schizophrenia [64]. Remarkably, Tao et al. also discovered ten novel GAD1 transcripts in human brains, among which four of them showed a life span trajectory expression pattern that is anticorrelated with the expression of GAD67 [63]. These results provided novel hints into schizophrenia pathogenesis, which not only highlighted potential effects of the reduced mature isoform (GAD67), but also emphasized influences of the truncated transcripts (e.g., GAD25).

Neurodevelopmental hypothesis of schizophrenia

Neurodevelopmental hypothesis of schizophrenia suggests that perturbation of early central nervous system development may be the key etiology for later onset of schizophrenia symptoms. This theory has been supported by early epidemiological and circumstantial data as well as more recent brain-specific molecular and genetic findings [65,66,67]. It is noteworthy that some schizophrenia risk genes related to dysregulation of neurotransmitter systems (e.g., GAD1) also play essential roles in regulation of neurodevelopment [68]. In addition, there are multiple schizophrenia risk genes (like DISC1 and NRG1) exerting pivotal functions in embryonic and postnatal neurodevelopmental processes. DISC1 gene is genetically associated with schizophrenia [69, 70], and encodes a scaffold protein regulating neuron proliferation, migration, neurite outgrowth, synaptogenesis, and integration of newborn neurons in multidimensional pathways [71,72,73,74,75]. Intriguingly, Nakata et al. reported that human DISC1 mRNA underwent extensive alternative splicing, and expression of several DISC1 isoforms were changed during brain development and in schizophrenia patients [76]. Among these DISC1 transcripts, the one that missing exon 3 (Δ3), or exons 7 and 8 (Δ7Δ8), as well as the one with an insertion variant in exon 3 (extra short variant-1, Esv1) exhibited higher mRNA levels in the brain of schizophrenics than healthy controls [76]. Moreover, expressions of DISC1 Δ3 and Δ7Δ8 isoforms were associated with schizophrenia risk SNPs (rs821616, rs6675281, and rs821597). The full-length DISC1 protein has been found to exert functions in early neuronal development and synapse maturation via interacting with other proteins [71], whereas Newburn et al. found that the truncated DISC1 proteins respectively encoded by these spliced isoforms (Δ3, Δ7Δ8, and Esv1) showed altered binding abilities with DISC1-interacting proteins, e.g., reduced binding for NDEL1 and PDE4B, and intact binding for FEZ1 and GSK3B [77]. Moreover, all the truncated DISC1 proteins could be co-immunoprecipitated with full-length DISC1, suggesting that they likely formed a complex and modulate the biological function of DISC1 through affecting its protein interaction network [77].

NRG1-ErbB4 signaling plays essential roles in axon/dendrite development and synapse formation/plasticity [78, 79], and both NRG1 and its receptor ErbB4 were perturbed in schizophrenia [80,81,82,83]. NRG1 gene is primarily transcribed into six types of isoforms (I–VI) with distinct amino-terminal regions via alternative 5′ flanking regulatory elements usage, and these isoforms exhibit diverse expression patterns, properties, and features in neurodevelopment [82, 84]. For example, the most mature NRG1 isoforms are soluble and act as chemoattractants, but the membrane-bound NRG1-III acts in a contact-dependent manner [82]. NRG1 has been confirmed as a schizophrenia susceptibility gene via linkage analyses and early candidate studies [85,86,87,88]. Subsequently, several risk SNPs (SNP8NRG221132, rs6994992, rs7014762, SNP8NRG221533, and SNP8NRG241930) have been reported to affect expression of NRG1 isoforms (NRG1-I, NRG1-IV, and NRG1-III), suggesting their potential roles in schizophrenia pathogenesis [89,90,91,92,93]. Consistently, schizophrenia-like behavioral (e.g., deficits in social interaction) and endophenotypic deficits (e.g., decreased prepulse inhibition) shown in mice with modified expression of NRG1-I/IV/III isoforms corroborated their involvement in schizophrenia, whereas mice overexpressing different isoforms exhibited subtly different deficits, suggesting possible isoform-specific biological mechanisms [94,95,96,97]. Meanwhile, four primary isoforms of ErbB4 have been characterized. They are generated by alternative splicing at the juxtamembrane (JM) and cytoplasmic (CYT) locus, and are therefore named JM-a, JM-b, CYT-1, and CYT-2 [98]. A splicing shift from major JM-b/CYT-2 isoforms to minor JM-a/CYT-1 isoforms in schizophrenia patients was associated with a risk haplotype containing three SNPs (rs7598440, rs707284, and rs839523) in ErbB4 [99, 100]. This shift, which leads to suppressed ErbB4 kinase activity and a subsequent reduction of excitatory synapses on parvalbumin-containing GABAergic interneurons, likely contributes to defects in synapse pruning and cognitive deficits in schizophrenia [101, 102].

Alternative splicing analyses are urgently needed in the current omics era post GWAS

While earlier candidate risk gene studies have significantly strengthened the understanding of schizophrenia pathogenesis, the panorama of alternative splicing in schizophrenia remains less clear. With the advancements in high-throughput genetic analyses of schizophrenia, the number of potential disease risk loci has also rapidly grown. Unlike the previously defined genes with well-characterized functionality and clinical applications, many of these newly-identified disease risk loci have been less-studied, let alone their transcription and splicing patterns that are potentially essential in schizophrenia pathogenesis. Therefore, further endeavors defining schizophrenia risk transcripts within these loci are urgently needed to interpret the current massive genetic data at the post GWAS era.

To reveal additional schizophrenia risk genes and their disease-associated isoforms, multiple methods have been developed to uncover disease relevant transcriptomic characteristics on gene-, isoform-, exon-, junction- to single base-levels using the high-throughput RNA-seq approach [103]. Gandal et al. recently conducted gene-level and annotation-guided isoform-level analyses using the RNA-seq data from DLPFC tissues of schizophrenia patients and normal controls [23]. Both methods highlighted generally similar pathways and cell-type enrichment in schizophrenics compared with controls, despite that isoform-level analyses identified multiple differentially expressed transcripts that were not significant in gene-level analyses. Intriguingly, these transcripts exhibited significant overlap with excitatory neuron clusters and enrichment for neuron projection development, mRNA metabolism, and synaptic pathways. In addition, networks built by the differentially expressed isoforms exhibited increased resolution in the disease-specific biological insights [23], and isoform-level changes likely provided extra clues for the neuronal and synaptic characteristics of schizophrenia. Therefore, they concluded that changes at the isoform level, compared with the gene level, showed larger effect sizes and genetic enrichment and a greater disease specificity. Another advantage of isoform-level analyses compared with gene-level analyses is the consideration of potentially distinct impact of certain isoforms in different illnesses. Although GWAS has identified multiple shared genetic risk genes across psychiatric disorders, different alternatively spliced isoforms of a single risk gene might exert unique functions in each illness, probably resulting from their specific expression pattern in distinct developmental stages and brain regions/cell types [104]. This speculation is concordant with the isoform-level analyses in Gandal et al. study [23], which emphasized the importance of splicing and isoform-level gene regulatory mechanisms in defining cell type and disease specificity. A study by Yang et al. further implied that isoforms of a single gene could function like different genes through interacting with extremely different protein networks [105]. Adding more complexity to the functional impact of alternative splicing in schizophrenia, developmental stage-specific expression patterns of isoforms have been implicated by several studies as well. For example, Jaffe et al. compared the human cortex transcriptome differences across developmental stages, and found that genes with developmental stage-related isoforms shifts were more likely to locate at schizophrenia GWAS risk loci than those without isoforms shifts [106]. In addition, Walker et al. conducted splicing QTL analyses using mid-gestational human brain and found that schizophrenia-associated SNPs were significantly enriched for prenatal splicing QTL loci [107]. Therefore, future endeavors are called to reveal potential effects of schizophrenia genetic risk on splicing events in developing brains, as current studies are mainly based on adult postmortem brain tissues.

Local splicing analysis of RNA-seq data

To gain more comprehensive insights into the expression patterns of known and unknown isoforms in schizophrenia, the “local splicing” analysis, which circumvents the limitations of imputation and assembling from short reads guided by existing transcriptomic annotations (e.g., inaccurate quantifications, incomplete annotations [108], loss of sequencing coverage, statistical analysis bias, etc.), is also adopted for RNA-seq analysis. For example, Gandal et al. investigated “local splicing” events using de novo aligned RNA-seq reads of human DLPFC [23], and they observed multiple types of “local splicing” changes in schizophrenia, such as exon skipping, alternative 5′ exon inclusion, and alternative 3′ splice-site usage. They found that genes with altered “local splicing” in schizophrenia showed significant enrichment for cell communication, actin cytoskeleton, synapse and neuronal development, as well as guanosine triphosphatase receptor activity [23]. In addition, Jaffe et al. detected numerous previously unannotated splice junctions tagging potential transcripts with alternative exonic boundaries or exon skipping using brain DLPFC RNA-seq samples, suggesting the incomplete annotations of human brain transcriptomes in existing databases [109]. They also identified numerous schizophrenia GWAS risk SNPs associated with these novel junctions (or unannotated transcribed sequences) [106]. Takata et al. analyzed RNA-seq data of DLPFC tissues [110], and identified many alternative splicing events including exon skipping, alternative usage of splice sites, and intron retentions. Their further analyses of these splicing events revealed that the splicing QTL SNPs were significantly enriched at schizophrenia GWAS risk loci [110]. Similarly, using RNA-seq data of postmortem samples across 13 brain regions, Ma et al. investigated the exon–exon splice junctions (for exon skipping events), and found that some schizophrenia GWAS risk SNPs were significantly associated with the expression of exon skipping junctions in several genes including CYP2D6 and SNX19 [111]. Notably, the exon skipping junction in SNX19 has also been detected in independent brain RNA-seq samples [112]. We herein also briefly summarize several representative “local splicing” events associated with schizophrenia GWAS risk SNPs that were retrieved from two DLPFC RNA-seq studies (Table 2) [106, 110].

Table 2 Local splicing events associated with schizophrenia risk SNPs.
Full size table

Despite identifying “local splicing” events associated with risk of schizophrenia through RNA-seq analyses, experimental validations using molecular approaches are necessary to verify such “local splicing” events in organisms. For example, Li et al. conducted RNA-seq analysis followed by experimental validations and identified a human-unique isoform within the AS3MT gene, which lacks exon 2 and 3 (named AS3MTd2d3) compared with the full-length AS3MT transcript (AS3MTfull) [113]. Briefly, to explore the potential mechanisms underlying the schizophrenia GWAS locus at 10q24.32, which contains numerous genome-wide significant risk SNPs spanning multiple genes, Li et al. performed junction analysis of RNA-seq data in human DLPFC tissues. They showed that the junction skipping exon 2 and 3 of AS3MT (referred as AS3MTd2d3) was significantly associated with the schizophrenia risk SNP rs7085104 [113]. This eQTL association was further confirmed in independent studies [114, 115]. The mRNA level of AS3MTd2d3 isoform was elevated in the DLPFC of schizophrenia patients compared with normal controls [113], and overexpression of AS3MTd2d3 in neurons resulted in a significant reduction of mushroom dendritic spine density [114], mimicking the endophenotypes observed in the brains of schizophrenia patients [116,117,118,119]. Cai et al. also found that the variable number of tandem repeat in high linkage disequilibrium with rs7085104 regulated the mRNA expression of AS3MTd2d3 using an in vitro minigene splicing assay [114]. Altogether, these studies have identified an alternatively spliced isoform AS3MTd2d3, which likely accounts for (at least part of) the molecular mechanisms underlying genetic risk of schizophrenia in the 10q24.32 GWAS locus.

Besides exon skipping, other types of alternative splicing events, such as alternative 5′ exon inclusion, have generated transcripts associated with schizophrenia risk in human brains. For example, Tao et al. discovered a novel spliced transcript in the schizophrenia risk gene ZNF804A in human brains [120]. This novel isoform arises from a new 5′ UTR in the intron 2 of this gene and lacks the first 2 exons compared with the wild-type ZNF804AFull transcript (named ZNF804AE3E4). Predictive analysis suggested that ZNF804AE3E4 might encode a protein lacking the zinc finger domain, the vital functional element in ZNF804AFull [121]. Intriguingly, the schizophrenia risk SNP rs1344706 was significantly associated with lower mRNA expression of ZNF804AE3E4 rather than ZNF804AFull in human fetal brains, corroborating the reduced ZNF804AE3E4 mRNA in brains of schizophrenia patients relative to healthy controls [120]. It has been reported that ZNF804A protein is present in dendrites and synapses, and knockdown of the aggregated isoforms reduces dendritic spine density and inhibits neurite formation [122, 123]. Surprisingly, overexpression of ZNF804AE3E4 in cultured neurons provokes more mushroom dendritic spines than overexpression of ZNF804AFull, suggesting a pronounced and specific functional effect of this isoform on mature spines compared to wild-type transcript [124]. Therefore, decreased ZNF804AE3E4 expression during early brain development likely explains the molecular mechanism underlying genetic risk of schizophrenia in this GWAS locus.

Long-read sequencing analysis

Despite the recent enlightening discoveries of alternative splicing in schizophrenia using RNA-seq data, short-read sequencing has also brought challenges to accurate assembling and quantification of isoforms since certain sequences are likely shared by multiple transcripts of a single gene [103]. Meanwhile, although RNA-seq supplemented with molecular characterizations have identified novel schizophrenia risk isoforms, these analyses usually require extensive efforts. Therefore, feasible strategies for efficiently characterizing more target genes and isoforms, e.g., long-read sequencing, are needed.

Long-read sequencing allows increased sequencing resolution and transcript assembling accuracy that are required for deciphering the alternative splicing profiles. A recent nanopore long-read sequencing analysis characterized the alternative splicing of CACNA1C in human brains [125], a psychiatric risk gene encoding the CaV1.2 voltage-gated calcium channel alpha1 subunit [3]. The authors applied two complementary approaches, exon-level analysis and splice-site-level analysis, which incorporated novel exons, novel junctions between annotated exons, and new combination of known junctions, and they identified 241 novel transcripts within the CACNA1C gene. Notably, many of the novel transcripts were abundant in human brain and were predicted to have functional impact on CaV1.2. These transcripts might reveal novel mechanisms of schizophrenia pathogenesis that were previously hidden by the incomplete annotation of CACNA1C transcripts [125].

The schizophrenia risk gene NRXN1 can be transcribed into NRXN1-α and NRXN1-β depending on the usage of two alternative promoters [126,127,128], and Jenkins et al. previously found that mRNA levels of NRXN1-β were significantly higher in the DLPFC of schizophrenia patients compared with controls, whereas NRXN1-α levels were consistent between diagnostic groups [129]. Intriguingly, a recent PacBio long-read Iso-seq analysis has described alternatively spliced NRXN1 isoform patterns in the induced pluripotent stem cell (iPSC)-derived neurons from psychiatric patients with non-recurrent NRXN13′/5′ heterozygous deletions [130]. Apart from the altered mRNA levels of some known NRXN1-α isoforms, there were dozens of novel isoforms identified in the iPSC-derived neurons carrying the mutant allele in 3′-NRXN1+/−. The authors then performed comparative analysis of the novel and conventional isoforms, and found generally similar exon inclusion patterns except for exons encompassed by the 3′-deletion. As a result, the NRXN1 deletion perturbs NRXN1-α isoform repertoire, which might alter the protein expression profile of this gene and thereby participate in psychiatric illnesses [130].

Functional analyses of alternatively spliced isoforms in brain and in schizophrenia

So far, transcriptomic analyses have identified numerous schizophrenia-relevant alternative splicing events, majority of which occur in genes playing key roles in neurodevelopment, synaptic plasticity, and cognition. Therefore, it is of great importance to uncover the functional complexity of these diverse alternative splicing events and relevant isoforms in schizophrenia. Although many of these isoforms are yet to be functionally characterized, several groups have reported inspiring findings highlighting specific roles of disease-associated isoforms (distinct from the full-length transcripts) in brain development aberrations linked with schizophrenia. We herein briefly summarize the functional analyses of two alternatively spliced isoforms with critical roles in brain development and potentially schizophrenia pathogenesis.

Alternatively spliced isoforms encoding proteins with unique functional characteristics

KCNH2 encodes the human ether-a-go-go-related voltage-gated potassium channel controlling neuronal firing patterns with characteristic electrophysiological properties: slow activation, fast inactivation, and slow and voltage-dependent deactivation [131, 132]. Huffaker et al. reported that SNPs in the KCNH2 gene were significantly associated with risk of schizophrenia, cognition, as well as brain structure and physiology [133], but the risk SNPs did not affect expression of KCNH2-1A (full-length transcript) or KCNH2-1B (a minor isoform). Notably, they identified a primate-specific and brain-enriched isoform (KCNH2-3.1) lacking the first two exons but containing a previously undescribed 5′ extension from exon 3 compared with KCNH2-1A. Moreover, KCNH2-3.1 mRNA expression was associated with schizophrenia risk alleles and was significantly increased in patients, suggesting that its elevated expression likely confers risk of the illness [133]. Intriguingly, individuals carrying schizophrenia risk alleles associated with increased KCNH2-3.1 expression showed better response to antipsychotics [134, 135], indicating the potential of this isoform in clinical applications.

Accordingly, a transgenic mouse model that mimics the increased expression of KCNH2-3.1 in schizophrenia patients was used to interrogate the functional impact of KCNH2-3.1 on cortical and hippocampal circuit [136, 137]. Electrophysiological recordings revealed faster ERG channel deactivation kinetics and increased firing rate in neurons of prefrontal cortex slices prepared from KCNH2-3.1 overexpressed mice compared with wild-type mice [136]. Interestingly, in KCNH2-3.1 transgenic mice, long-term potentiation induced in hippocampal CA1 synapses by theta burst stimulation is impaired, which is in accordance with the hippocampal-dependent memory deficits measured in object location task [136]. These data suggest that KCNH2-3.1 isoform might regulate information processing in prefrontal cortex and hippocampal microcircuit. Moreover, KCNH2-3.1 transgenic mice display impaired synaptic connectivity and transmission in ventral hippocampus-medial PFC long-range projection [137].

Alternatively spliced isoforms exerting noncoding function

While many spliced isoforms encode proteins with significant physiological impact, some are also found to exert functions in a coding-independent manner, such as Ube3a1. Ube3a1 refers to an alternatively spliced transcript of the UBE3A gene in the chromosome 15q11.2 locus, and duplications of this genomic region confer risk of schizophrenia in diverse populations [138,139,140,141,142]. Ube3a originally encodes a ubiquitin E3 ligase that plays an important role in dendrite and spine development, synaptic plasticity, and excitatory/inhibitory imbalance [143,144,145]. There are three isoforms identified for Ube3a. Both Ube3a2/3 transcripts are translated into the full-length protein, while Ube3a1 contains a unique 3′UTR producing a truncated isoform due to the alternative polyadenylation [146]. Interestingly, Valluy et al. found that Ube3a1 and Ube3a2/3 transcripts have opposite effects on dendrite complexity. Knockdown of Ube3a1 with shRNA specifically targeting its 3′UTR in cultured neurons increased dendritic complexity, whereas knockdown of Ube3a2/3 RNA reduced dendritic complexity. In addition to increasing dendrites complexity, Ube3a1 RNA knockdown reduced sizes of dendritic spines and amplitude of miniature excitatory postsynaptic currents, suggesting that Ube3a1 could promote spine maturation but prevent the overgrowth of dendrites [146]. Surprisingly, only the mRNA form of Ube3a1 has been detected in neurons, and the unique 3′UTR of Ube3a1 RNA might contain signals for localization at the dendritic compartments. Intriguingly, further investigations indicated that Ube3a1 RNA acted as a competing endogenous RNA and sequestered miR-134 from its natural target dendritic genes, resulting in abnormal translation and function of these genes [146]. Therefore, the coding-independent functionality of spliced isoforms are also of potentially great significance in schizophrenia.

Conclusions and perspectives

So far, as the importance of alternative splicing in pathogenesis of multiple diseases, such as autism, amyotrophic lateral sclerosis, and Parkinson’s disease, has been acknowledged, and therapeutic strategies based on correcting splicing defects are being investigated. For example, the clinical usage of antisense oligonucleotides or CRISPR/Cas9 that could recognize specific RNA splicing regulatory elements has been extensively studied [147]. However, such therapies are likely more effective for Mendelian diseases than for complex disorders such as schizophrenia. As the polygenic nature of schizophrenia requires intervention of hundreds or thousands of genes involved in its pathogenesis, treatment with oligonucleotides or CRISPR/Cas9-based methods is not feasible currently. Hence, novel phenomics strategy might offer new insights into the complex interaction network between risk genes and clinical features of schizophrenia patients, and thereby revealing potential hub pathways underlying the endophenotypes in schizophrenia pathology. Although intervention of altered alternative splicing in schizophrenia is impractical, targeting the spliced mRNAs or proteins of schizophrenia-associated isoforms might be possible with a handful of new techniques. For example, Liu et al. recently developed a genetic method called “isoTarget” for in vivo isoform functionality characterization, which could knock out or tag an isoform in a cell-type-specific manner through inserting a cassette sequence into an exon [148]. It is of great interest to explore the possibility of applying this technique to agonize or antagonize GPCRs (e.g., DRD2, mGluR3, etc.) to ameliorate schizophrenia symptoms while reduce side effects and declined effects during chronic treatment [39, 149]. Considering the splicing diversity of these genes, better understanding of the structure and expression patterns of their isoforms in vivo could benefit the discovery of compounds specifically targeting the pharmacologically effective sites without activation elsewhere [150]. Overall, greater attentions into altered alternative splicing of schizophrenia risk genes are necessary for efficient translation of genetic discoveries into the understanding of controlling of the illness.