Poison exons in neurodevelopment and disease

Abstract

Poison exons are naturally occurring, highly conserved alternative exons that contain a premature termination codon. Inclusion of a poison exon in a transcript targets the transcript for nonsense mediated decay, decreasing the amount of protein produced. Poison exons are proposed to play an important role in tissue-specific expression, development and autoregulation of gene expression. Recently, several studies that performed systematic investigations of alternative splicing in the brain have highlighted the abundance of transcripts containing poison exons, some of which are spliced in a cell type-specific manner. Pathogenic variants in or near poison exons that result in aberrant splicing have been identified in several genes including FLNA, SCN1A and SNRPB. Improved understanding of the role of poison exons in development and disease may present opportunities to solve previously undiagnosed disease and to develop therapeutic approaches in the future.

Introduction & Terminology

Alternative splicing is a mechanism that increases the diversity of transcripts and protein isoforms that are expressed in a given organism or in a specific tissue. The majority of multi-exon genes undergo alternative splicing [1], which can result in proteins of various sizes – and functions – by alternatively splicing modular exons that result in in-frame inclusion or exclusion of amino acids. Generation of some alternatively spliced exons likely arose from tandem, intragenic exon duplication followed by mutation during evolution, leading to increased functional diversity [2].

Some genes have naturally occurring alternative exons that contain a premature termination codon (PTC); alternative splicing of such an exon is predicted to lead to nonsense mediated decay (NMD). We will refer to these exons as “poison exons” (PEs) throughout this review, though there are various synonymous terms including PTC exon and NMD exon. The name “poison exon” stems from the fact that when the exon is spliced into the transcript, it “poisons” the host transcript by targeting it for NMD, a surveillance system that detects and degrades RNA transcripts that harbor PTCs (Figure 1). During pre-mRNA processing, exon-exon junctions are ‘marked’ by exon junction complexes which are displaced by the ribosome during the first round of translation. In the event that an exon junction complex remains after this pioneering translation event, components of the NMD complex are recruited, resulting in transcript degradation [3]. In this manner, inclusion of PE ultimately most likely induces NMD, though the presence of truncated proteins has not been thoroughly evaluated. Alternative splicing of these PEs is controlled by RNA-binding proteins (RBPs) that recognize consensus RNA sequence motifs that facilitate binding.

Figure 1.

Alternative splicing of poison exons (red) can occur under four primary mechanisms. In all instances inclusion of the PE leads to premature truncation (red dash) of the transcript and nonsense mediated decay (NMD). This alternative splicing can occur through a myriad of RNA-binding proteins (RBPs) that can suppress/promote PE inclusion/exclusion, the exact mechanisms and RBPs involved are not well-understood, though at least in neuronal contexts RBPs such as PTBP1, NOVA and RBFOX are likely to play a role.

Naturally occurring PEs were first identified computationally by mapping of expressed sequence tags to the human genome, revealing that ~35% of alternative splicing events and 23% of all isoforms overall contain PEs[1]. Since then, a number of studies have identified additional PEs, mostly through analysis of transcriptomic data in human and mice [4–7]. These PEs are proposed to play an important role in tissue-specific expression, development and autoregulation of gene expression. An important feature of PEs is that they are naturally occurring; this is in contrast to disease-associated genetic variants that introduce (1) a PTC (nonsense variant), (2) a frameshift (e.g. an indel), or (3) aberrant splicing due to disruption of the splice donor/acceptor site that ultimately results in an aberrant transcript that is out-of-frame (i.e. a frameshift).

PEs have several important properties that are consistent with a role in development and regulation of protein levels. The majority of PEs lie within highly conserved or ultraconserved elements [8,9]. The conservation of PEs across species, rather than their elimination through selective pressure over time, suggests an important function. They tend to be smaller than constitutively spliced exons [5]. The presence of a PTC and the associated NMD mechanism indicates that there is no protein-coding potential when the exon is included, suggesting they do not add to functional diversity. Finally, certain classes of genes are enriched for containing a PEs including genes encoding splicing factors and other RBPs, as well as chromatin remodelers[5,9]. Many RBPs, including splicing factors, auto-regulate their expression by binding their own transcript to increase alternative splicing of a PE thereby introducing a PTC and inducing NMD to decrease protein levels [10].

PEs during neurodevelopment

Several studies have investigated the role of alternative splicing in brain development [5–7]. Using deep RNA-sequencing of mouse and human cortex, Yan and colleagues [5] identified 3,058 highly conserved, alternatively spliced cassette exons, 1,014 (33%) of which are predicted to induce NMD upon either PE inclusion or exclusion. The highest number of cassette exons were detected in neurons, with nearly three times the number in other cell-types in the brain. In a similar study comparing RNA-seq data from sorted mouse neural precursor cells (NPCs) and mouse neurons, Zhang and colleagues identified 622 exons that were differentially spliced during this developmental process, including a PE in Flna [6]. Similarly, fetal human single cell analysis of radial glia (progenitors) and neurons revealed 296 cassette exons, including the FLNA PE that is included in mature neurons in both humans and mice. Both of these studies demonstrated that the alternative splicing of PEs is controlled by cell-type specific RBPs, including Rbfox, Nova and Ptbp1 (Figure 2A). Finally, Steward and colleagues used long-read human brain-derived RNA-sequencing to identify previously uncharacterized exons in 191 genes implicated in epilepsy and identified a number of PEs in these genes, including three in SCN1A [7].

Figure 2. Alternative splicing of PEs in the brain and disease.

(A) PEs are preferentially skipped in cells that require the full-length protein to function. For instance, for many ion channels, including SCN1A, the PE is incorporated in non-neuronal cells and prevents expression of this ion channel. However, in neurons which require this voltage gated sodium channel to ensure proper neuronal function the PE is skipped and full-length transcript is produced. The converse is also true, full length FLNA is required in NPCs, but not in neurons where the FNLA PE is included leading to NMD. This alternative splicing includes a complex interaction of RBPs that can include Ptbp1/Nova binding upstream of the PE to prevent inclusion[19]. Similarly, Rbfox1/Nova binding downstream of the PE can promote PE inclusion, triggering NMD[20,21].

(B) The presence of genetic variants (purple diamond) within or flanking the PE can prevent RBP binding (e.g. PTBP1) by destroying the recognition sequence, leading to aberrant inclusion of the PE in cells where it should usually be excluded, this can ultimately lead to NMD and reduced protein levels.

(C) In other rarer instances, neurological disorders can lead to aberrant PE inclusion independent of a DNA sequence variant in or near the PE. This process is not well-studied but may be due to mislocalization or dysregulation of RBPs as a result of the disease process. Examples include the mislocalization of NOVA after seizures[15].

Genetic variants that cause aberrant inclusion of PEs.

Aberrant PE inclusion due to a genetic variant that destroys an RBP consensus motif can lead to NMD in a cell type where the protein should usually be expressed (Figure 2B). Such aberrant PE inclusion had now been reported for FLNA and SCN1A, two genes where loss-of-function pathogenic variants are associated with clinically defined neurodevelopmental disorders. Pathogenic variants in the X-linked gene, FLNA, are associated with periventricular nodular heterotopia (PVNH), seizures and intellectual disability (ID) in females and are usually lethal in males [11]. Flna contains a PE that is skipped in mouse neural progenitor cells (NPCs), but included in neurons where it introduces PTC that leads to NMD. By targeted resequencing of this PE in the human FLNA, as well as the flanking intron, Zhang et al identified a C>T variant upstream of the FLNA PE in a cohort of 221 individuals with PVNH of unknown genetic etiology [6]. This variant segregated with PVNH in the family, though with reduced penetrance, milder clinical presentation in females and included affected males. The increased inclusion of the FLNA PE was confirmed at the RNA level in blood samples of affected individuals but not their non-carrier relatives, as well as in a minigene assay. This ectopic inclusion is likely due to disruption of the RBP, PTBP1, consensus motif.

Dravet syndrome (DS) is a clinically recognizable developmental and epileptic encephalopathy (DEE) characterized by febrile seizures at onset, followed by multiple refractory seizures and ID. De novo SCN1A pathogenic variants occur in >80% of individuals with DS, and haploinsufficiency is the predominant disease mechanism. SCN1A variants are also associated with milder seizure phenotypes including genetic epilepsy with febrile seizures plus (GEFS+). In five individuals with DS but no pathogenic coding variants, targeted resequencing of a highly conserved intronic region containing a PE identified rare variants [12]. These variants arose de novo in two cases and segregated with GEFS+ phenotypes in an additional two familial cases; segregation analysis was incomplete in the fifth case. The majority of these variants (3/5) led to aberrant inclusion of the SCN1A PE in a minigene assay, possibly due to disruption of the consensus motifs of the RBPs, SRSF1 and PTBP1. This SCN1A PE and a PE in another DEE-related gene, SCN8A, were first detected in rat astrocytes and neuroblastoma cells [13]. Similarly Steward and colleagues performed targeted resequencing of novel exons in a subset of known DEE genes in 122 individuals with unsolved DEE. [7]. They identified two de novo variants in different SCN1A PEs, and it was hypothesized that the presence of these variants leads to aberrant inclusion of the SCN1A PE by disruption of an hnRNP A1 RBP binding site. In contrast to the first study, neither of these two SCN1A PEs are highly conserved.

Aberrant PE inclusion due to an underlying genetic variant is not only limited to neurological disorders. Haploinsufficiency of SNRPB causes cerebro-costo-mandibular syndrome (CCMS), a disorder characterized by multiple malformations of the ribs and craniofacial features. Five affected individuals have been identified with pathogenic variants in a PE that lead to increased PE inclusion and NMD [14]. This study suggests that usage of PE in a cell-specific manner is not limited to neurodevelopment and the brain and that this mechanism may account for at least some instances of unsolved mendelian diseases.

Of note, incomplete penetrance, milder clinical presentation and variable expressivity are all characteristic features of aberrant PE inclusion associated with disease. While speculative, this could be explained by variable level of PE inclusion, with more severe cases having more PE inclusion (and thus lower protein levels) and milder cases less PE inclusion. Also, while all the aforementioned examples demonstrate aberrant PE inclusion, it is also possible, perhaps even likely, that aberrant exclusion of these PEs may be associated with genetic disease.

Aberrant splicing of poison exons in disease

In addition to genetic variants that disrupt the binding of RBPs to RNA consensus motifs, aberrant PE inclusion has also been associated with neurological disease in the absence of a genetic variant in a specific transcript, but rather due to RBP dysfunction(Figure 2C). For instance, double knockout of the neuronal-specific RBPs, Nova1/2 in mice leads to spontaneous generalized seizures[15]. At the molecular level, these physiological changes were associated with changes in gene expression that were mediated by PE inclusion/exclusion of synaptic proteins, including Dlg3 (increased PE inclusion, decreased mRNA) and Scn9a (decreased PE inclusion and increased mRNA). Moreover, during pilocarpine-induced seizures in wildtype mice, NOVA is mislocalized to the cytoplasm and a small number of transcripts have increased PE inclusion and are targeted by NMD [15]. This NOVA mislocalization was limited to the cortex, and not observed in the cerebellum, consistent with the known cortical nature of seizures. These findings suggest that NOVA is important for maintaining homeostasis after seizures, at least in part by PE inclusion/exclusion regulation. Moreover, increased NMD in general is observed in both rodent epilepsy models and resected tissue from patients with epilepsy, though the role of PEs in this process is not known [16]. These studies provide a possible mechanism by which synaptic protein abundance is regulated after seizure activity and warrants follow up study in additional models. Aberrant PE inclusion or exclusion is also prevalent in certain cancers; some PEs have tumor-suppressor effects and others are correlated with poor survival [17].

Challenges and opportunities

The greatest limitation to determining the scope of aberrant PE usage in genetic disease is our narrow knowledge of where in the genome these PEs reside. Indeed, neither of the SCN1A nor FLNA PEs are annotated in the most commonly used gene annotations, RefSeq or GENCODE. As highlighted above, deep RNA-sequencing of fetal brain tissue and cells during mouse and human development have greatly expanded our appreciation for the occurrence of PEs [5–7], though this is likely an underestimation, as the PE-containing transcripts are degraded by NMD and thus are not reliably captured. Inhibition of NMD using compounds that inhibit translation (cycloheximide and emetine) or by knockdown of key NMD proteins (UPF1[18]) in a variety of primary and iPSC-derived cell lines followed by deep short- and long-read RNA-sequencing are necessary to identify additional putative PEs. Moreover, most studies to date have focused on brain-expressed genes, but PEs are likely to be important across multiple tissue types, as evidenced in the CCMS and cancer studies described above [14,17]. Finally, coordinated efforts are required to ensure these PEs are validated and incorporated into the annotation of the human genome.

In addition to identifying these PEs, it is crucial that new approaches are developed to validate and determine the function of putative PEs in a high-throughput manner. In a recent study, investigators used a highly multiplexed approach to systematically delete 556 human PEs (465 highly conserved, 91 poorly conserved) to investigate the functional relevance of PEs for cell growth [17]. They found that 43% of targeted PEs were depleted in their screen, supporting an essential role in cellular fitness. Extending their studies to a lung cancer xenograft model, they found that deletion of PEs in 61 genes, including SR and hnRNP gene families, resulted in pro-tumorigenic effects in vivo, again highlighting their important role in growth regulation.

An additional challenge lies in the reporting of these variants in clinical genetic testing reports both in terms of (1) the nomenclature used to annotate these variants and (2) ensuring adequate training of genetic counselors and clinicians for interpretation and return of results. This is evidenced in the FLNA and SCN1A studies, where these deeply intronic variants are annotated relative to the nearest constitutive exon and the relevance of the PE is not apparent. It will also be important to develop computational algorithms to predict which variants are most likely to disrupt RBPs and lead to aberrant PE splicing. In addition to the human genetics community in general, addressing these issues will require buy-in from groups such as ClinVar, Human Genome Variation Society (HGVS) and the American College of Medical Genetics (ACMG).

Despite these challenges, the existence and manipulation of PEs also presents a unique opportunity for novel therapeutic approaches. For instance, antisense oligonucleotides (ASOs) that prevent the inclusion of PEs could increase the amount of full-length protein and pose a treatment strategy for haploinsufficient conditions. Pre-clinical studies in an Scn1a mouse model of DS using an ASO that targets an Scn1a PE are currently underway. Conversely, ASOs that increase the inclusion of PEs may be a therapeutic avenue for tackling gain-of-function disease mechanism.

Summary

Although PEs were first identified nearly 20 years ago, improvements in technology now allow high-throughput identification and characterization of these highly conserved (and challenging to detect) elements. As a result, the role of PEs in human development and disease is becoming increasingly appreciated. Better understanding and detection of PEs provides unique opportunities to solve previously undiagnosed disease, to understand mechanisms of development and disease more broadly, and to develop unique therapeutic approaches in the future.

Abbreviations

CCMS

cerebro-costo-mandibular syndrome

DS

Dravet syndrome

DEE

developmental and epileptic encephalopathy

NMD

nonsense mediated decay

NPC

neural precursor cell

PE

poison exon

PTC

premature termination codon

PVNH

periventricular nodular heterotopia

RBP

RNA-binding protein