RNA mis-splicing in disease

Common cause: pre-mRNA mutations and mis-splicing

The most common type of mutations that alter splicing patterns are cis-acting and are located in either core consensus sequences (5′ss, 3′ss and branch point (BP)) or the regulatory elements that modulate spliceosome recruitment, including exonic splicing enhancer (ESE), exonic splicing silencer (ESS), intronic splicing enhancer (ISE) and intronic splicing silencer (ISS) elements⁸ (BOX 1). Mutations in these regulatory elements have been documented in multiple diseases that have characteristic effects on many tissues (TABLE 1, cis). An early splicing mutation described soon after the discovery of splicing was a point mutation that generates an alternative 3′ss in HBB, which encodes β-globin, resulting in β⁺-thalassaemia, a condition that is characterized by reduced β-globin protein levels and anaemia^14–16. More recent examples include: splice site mutations in dystrophin (DMD), which result in loss of dystrophin function and Duchenne muscular dystrophy^17,18 (discussed in further detail below); polymorphic UG and U tracts near the 3′ss of CFTR (cystic fibrosis transmembrane conductance regulator) exon 9, which modify the severity of cystic fibrosis^19,20; and ESE, ESS and 5′ss mutations in MAPT (microtubule-associated protein tau) exon 10, which cause frontotemporal dementia with parkinsonism linked to chromosome 17 (FTDP-17)²¹.

In contrast to the above gene mutations that cause a single type of disease, mutations in several types of sequence elements in LMNA, the gene encoding lamins A, C, Δ10 and C2 result in multiple pathological phenotypes²². Lamins are type V intermediate filament proteins of the nucleus that have crucial roles in differentiated cell nuclear architecture (peripheral lamins) and gene expression (nucleoplasmic lamins). Laminopathies comprise a heterogeneous group of over 14 diseases, including cardiomyopathies, hereditary peripheral neuropathies, lipodystrophies, muscular dystrophies and premature ageing (progeroid) syndromes²³.

Interestingly, 5′ss mutations in LMNA cause two progressive but distinct disorders (FIG. 1): limb girdle muscular dystrophy 1B (LGMD1B) primarily affects the proximal muscles of the shoulders and hips, whereas familial partial lipodystrophy type 2 (FPLD2) is characterized by a selective loss and abnormal distribution of body fat^24,25. Both 5′ss mutations lead to intron retention (albeit for different introns), frameshifting and the generation of a premature termination codon (PTC) that should activate NMD and increase LMNA RNA turnover. However, the different disease presentations suggest that distinct truncated LMNA proteins may be produced by intron 8 versus intron 9 retention (FIG. 1a,b). Moreover, an unrelated premature ageing disease, Hutchinson–Gilford progeria syndrome (HGPS), is caused by the utilization of an alternative 5′ss in LMNA exon 11, resulting in a 150 nucleotide deletion that generates progerin, a carboxy-terminal truncated protein that is constitutively farnesylated and remains associated with the nuclear membrane during mitosis^26,27 (FIG. 1c). Other mutations at the exon 4 3′ ss cause the addition of an extra three amino acids at the exon 3–4 border and lead to dilated cardiomyopathy (DCM), a heart condition that is caused by thinning and dilation of the ventricles and progressive atrioventricular block, a cardiac conduction defect²⁸ (FIG. 1d). Thus, splicing mutations result in abnormal LMNA proteins, which themselves lead to a range of pathological effects.

Core spliceosome mutations in retinitis pigmentosa

In addition to mutations that alter precursor RNA sequence elements which regulate splicing, mutations in core constituents of the spliceosome also underlie a discrete set of diseases, including retinal degenerative disorders and cancer (TABLE 1, spliceosome). Retinitis pigmentosa (RP), a progressive disease that is characterized by initial night blindness and loss of peripheral vision followed by central retinal degeneration, is the most prevalent form of hereditary blindness²⁹. Multiple gene mutations cause autosomal dominant RP (adRP), but six of these genes encode components, or associated factors, of the spliceosomal U4/U6.U5 tri-small nuclear ribonucleoprotein particle ( tri-snRNP; namely pre-mRNA processing factor 3 (PRPF3), PRPF4, PRPF6, PRPF8, PRPF31 and SNRNP200 (also known as BRR2)) (FIG. 2). Among surveyed tissues, the human retina expresses the highest levels of some housekeeping genes and major, as well as minor, spliceosomal snRNAs. Therefore, PRPF mutations might cause global splicing dysregulation that manifests in the retina because of enhanced splicing activity³⁰.

Studies on patient lymphoblasts with PRPF3, PRPF8 and PRPF31 mutations show that these mutant cells also display impaired constitutive and alternative splicing together with changes in snRNA stoichiometry and tri-snRNP composition. Variable expressivity and/or incomplete penetrance have also been reported for RP-associated mutations in U4/U6.U5 tri-snRNP components^31–33. For PRPF31, incomplete penetrance in some families with adRP results from increased expression of the wild-type PRPF31 allele in asymptomatic carriers owing to reduced expression of the repressive modifier gene CNOT3 (CCR4-NOT transcription complex, subunit 3)³⁴. These results support a haploinsufficiency model for spliceosome-associated adRP and highlight the importance of optimal U4/U6.U5 tri-snRNP function in the retina.

However, gain-of-function pathogenesis models have also been proposed. For example, some PRPF3 mutations lead to large aggregate formation and apoptosis in photoreceptor, but not epithelial or fibroblast cell lines, suggesting that these aggregates may directly interfere with normal cellular activities³⁵. Loss-of-function mutations may also occur in adRP. This possibility is exemplified by PRPF8, a highly conserved and large protein that is a component of the U5 snRNP and is positioned in the catalytic core of the spliceosome. PRPF8 C-terminal mutations cause an early onset and severe form of adRP, and structural analysis of this region shows that it inserts into the RNA-binding tunnel of the SNRNP200 helicase. This insertion event intermittently blocks, and thus controls, the ATP-dependent U4/U6 unwinding activity of this helicase. Thus, loss of the PRPF8 RNA-blocking activity owing to some adRP-linked mutations would be expected to cause mis-regulation of U4/U6 unwinding due to premature helicase activity³⁶.

Different mutations in a single gene may also result in haploinsufficiency and dominant-negative effects in adRP. Heterozygous single-nucleotide variants and deletion variants in the PRPF4 coding and promoter regions, respectively, have been reported recently³⁷. PRPF4 encodes a 60 kDa protein that is important for U4/U6 di-snRNP stability. Whereas the promoter deletion causes decreased PRPF4 expression in patient fibroblasts, the coding region variant (p.Pro315Leu) results in the upregulation of PRPF4 together with several U4/U6.U5 tri-snRNP components (PRPF3, PRPF6, PRPF8 and PRPF31) and other splicing factors (SRSF1 and SRSF2). In addition, overexpression of human PRPF4 in which the Pro at position 315 is mutated to Leu (PRPF4^Pro315Leu) in zebrafish results in larval deformity and retinal phenotypes. Although the central question of why the U4/U6.U5 tri-snRNP is particularly important for normal retinal function remains unanswered, several intriguing clues are starting to emerge. The initial events in RP include loss of rod photoreceptors, the cells responsible for vision under low light conditions, and of the retinal pigment epithelium (RPE), the monolayer of cells that carry out functions such as the phagocytosis of the outer segments of photoreceptors (~10% of rod cell volume)³⁸. Thus, the RPE has a high rhythmic metabolic burden, and recent results suggest that the RPE may be the primary cell type affected by PRPF mutations. Mouse RPE morphology is sensitive to Prpf mutations, and RPE phagocytic function is inhibited in Prpf3^{Thr494Met/Thr494Met} and Prpf8^{His2309Pro/His2309Pro} knock-in, as well as Prpf31^+/− hemizygous RPE cell cultures³⁹. Further studies designed to investigate how these Prpf mutations alter splicing patterns in these mutants should provide mechanistic insights into U4/U6.U5 tri-snRNP dynamics in normal versus diseased retinal cells.

Spliceosome dysregulation in cancer

Mis-regulation of alternative splicing is an important factor in several types of cancer⁴⁰. In addition, somatic mutations that affect the expression of core spliceosome components have an important role in cancer progression. For example, PRPF6, a U5 snRNP protein that mediates interactions between U5 and the U4/U6 di-snRNP to form the U4/U6.U5 tri-snRNP (FIG. 2), is overexpressed in colorectal carcinoma owing to chromosomal instability, copy number gain and possibly other factors⁴¹, and this promotes cancer cell proliferation. The increased PRPF6 expression in cancer cell lines correlates with an alternative splicing event that generates an oncogenic form of the stress-activated kinase ZAK. A direct role for PRPF8 somatic mutations and hemizygous deletion has also been proposed for myelodysplastic syndromes (MDS), which are the most prevalent forms of adult myeloid malignancies and are characterized by abnormal growth or development of blood cells⁴². In contrast to PRPF6, reduced expression of PRPF8 correlates with increased cell proliferation, and PRPF8 heterozygous mutations or hemizygous deletions result in widespread alternative splicing defects owing to enhanced activation of suboptimal splice sites⁴³.

Exome and whole-genome sequencing studies have also uncovered frequent somatic mutations in a key group of spliceosome-associated components, including SF3B1, U2AF1 and U2AF2 (see FIG. 2) in several types of myeloid neoplasms^44,45. High-throughput RNA sequencing (RNA-seq) results indicate that U2AF1 mutations alter haematopoiesis and cause changes in 3′ss recognition, resulting in the mis-splicing of hundreds of gene transcripts^46–48. Mutations in some splicing factor genes also occur frequently in myelodysplastic syndromes and chronic myelomonocytic leukaemia. One example is SRSF2, which encodes a Ser/Arg-rich (SR) splicing factor (see BOX 1). Similar to U2AF1 mutations, these SRSF2 mutations alter the RNA-binding characteristics of SRSF2 and result in extensive changes in splicing patterns and impairment of haematopoietic cell differentiation^49–51. Importantly, antitumour drugs that target the spliceosome have been described, including for cancers that are driven by overexpression of the MYC oncogene and by increased levels of nascent RNAs^52,53.

Development and stress: key roles for the minor spliceosome

New roles for minor spliceosomal snRNAs and the U4atac/U6atac.U5 tri-snRNP during fetal development have been reported (FIG. 2). Homozygous mutations in RNU4ATAC, which codes for the U4atac snRNA, leads to microcephalic osteodysplastic primordial dwarfism type 1 (MOPD I; also known as Taybi–Linder syndrome), an autosomal recessive developmental disorder that is characterized by intrauterine growth retardation and multiple tissue abnormalities that lead to early death^54–56 (TABLE 1). Most MOPD I mutations disrupt the U4atac snRNA 5′ stem-loop, inhibit binding of U4atac/U6atac di-snRNP proteins and decrease the levels of U4atac/U6atac.U5 tri-snRNP so that minor intron splicing is impaired⁵⁷. Alternatively, a transition mutation (124G>A) near the Sm protein-binding site results in reduced levels of U4atac snRNA.

The idea that minor spliceosome levels have key regulatory roles in gene expression is also supported by the observation that minor introns act as molecular switches that modulate stress-induced expression of their host genes⁵⁸. Under normal conditions, U6atac is unstable (t_1/2 <2 hours) relative to other snRNAs. However, the stress-induced kinase p38 mitogen-activated protein kinase (MAPK) increases U6atac snRNA stability, resulting in enhanced splicing of genes that contain minor introns, including the tumour suppressor PTEN⁵⁸.

Splicing factor dysregulation: perils of imbalance

Recruitment of the spliceosome to splice sites is modulated by the dynamic association of splicing factors with RNA regulatory elements on nascent transcripts. Alternative splicing is particularly important in the brain, and mutations in factors connected with splicing regulation cause a range of neurological diseases, from spinal muscular atrophy (SMA) to amyotrophic lateral sclerosis (ALS; also known as motor neuron disease) (both discussed in more detail below). RNA-binding proteins (RBPs) are generally multi-functional and many have key roles in splicing regulation. Moreover, ultraviolet light-induced RNA–protein crosslinking protocols (such as high-throughput sequencing of RNA isolated by crosslinking immunoprecipitation ( HITS-CLIP; also known as CLIP-seq)⁵⁹, photoactivatable-ribonucleoside-enhanced-CLIP (PAR-CLIP)⁶⁰ and individual-nucleotide resolution-CLIP (iCLIP)⁶¹) have been used to identify hundreds of RBPs in HeLa and mouse embryonic stem cells^62,63, as well as other cell types⁶⁴. Thus, it is not surprising that mutations in the genes encoding RBPs have been linked to multiple disorders.

An abundant class of nuclear RBPs are factors that associate with both the RNA Pol II carboxy-terminal domain (CTD) and nascent transcripts (FIG. 3). Early electron microscopy studies on the co-transcriptional assembly of large pre-mRNA–protein complexes indicated that mRNA biogenesis involves an orchestrated series of RBP–RNA assembly and remodelling transitions⁶⁵. These, and other studies suggested that both stable and transient protein–protein and protein–RNA interactions play key parts in splicing regulation. Some of the most abundant RBPs in the nucleus contain structural motifs — such as RNA recognition motifs (RRMs), heterogeneous nuclear RNP (hnRNP) K homology domains, zinc fingers and zinc knuckles — that bind RNA, as well as auxiliary, or low complexity (LC), regions composed of limited amino acid diversity (for example, [Gly/Ser]Tyr[Gly/Ser] tripeptide repeats in the RBP FUS). In some RBPs, LC regions also contain prion-like domains that can function as self-templating protein conformers and which are hotspots for disease-associated mutations^66,67. The importance of LC regions for protein–protein interactions has been highlighted by studies using the small molecule biotinylated isoxazole (b-isox)⁶⁸, treatment with which results in the precipitation of numerous RBPs from cell lysates. The LC regions of these proteins are necessary and sufficient to drive this insolubility, and these regions alone undergo a concentration-dependent phase transition to a hydrogel that consists of polymerized amyloid-like fibres. This observation suggests that dynamic interactions between the LC regions of RBPs are a fundamental principle underlying the co-transcriptional assembly of RBP–RNA protein complexes during splicing (FIG. 3a). These association–dissociation events may also be modulated by signal-induced post-translational modifications, such as phosphorylation of Ser-rich LC regions⁶⁹.

If protein–protein interaction dynamics are crucial for the assembly and disassembly of RBP–RNA complexes, then mutations in these protein interaction regions might result in coordinate perturbations in protein homeostasis (proteostasis) and RNA splicing (FIG. 3b,c). Indeed, this scenario is exemplified by studies designed to understand the molecular aetiology of ALS, a neurodegenerative disease that affects motor neurons and that leads to paralysis, respiratory failure and death generally within a few years of symptom onset⁷⁰. Although ~90% of ALS cases are sporadic (sALS), the remaining 10% are familial (fALS), and a number of genes have been implicated in fALS by either linkage analysis or candidate gene studies. Relevant to this Review, several of these genes encode RBPs, including FUS (also known as ALS6, TLS and HNRNPP2), TDP‑43 (TAR DNA-binding protein; also known as TARDBP and ALS10), ATXN2 (ataxin 2; also known as ALS13), HNRNPA1, HNRNPA2B1 and MATR3 (matrin 3). In addition, microsatellite GGGGCC expansions (G4C2^exp) in C9orf72 (REF. 71) cause chromosome 9p21-linked ALS and frontotemporal dementia (C9ALS/FTD), and one disease model proposes that these expansions are transcribed into rG4C2^exp RNAs that disrupt splicing by sequestering splicing factors^72,73. Indeed, a recent transcriptome analysis of autopsied brains (specifically, the cerebellum) from patients with C9ALS/FTD reported thousands of alternative splicing changes in fALS, and to lesser extent in sALS, and motif analysis indicated that many of the alternative cassette exon splicing changes might be regulated by hnRNP H, which has previously been implicated in C9ALS/FTD pathogenesis^74,75. Mutations in TDP‑43 and FUS account for 4% and 5%, respectively, of fALS cases, and most of these mutations occur in the prion-like domains of the encoded proteins (FIG. 3b), including the TDP-43 Gly-rich C-terminal region and the FUS Gln-, Gly-, Ser-, Tyr-rich and Gly-rich amino-terminal regions, although additional FUS mutations occur in the C-terminal Arg- and Gly-rich and nuclear localization signal regions^67,72. Whereas earlier studies have demonstrated that TDP-43 plays a direct part in the splicing of specific exons, TDP-43 is a member of the hnRNP family of proteins that function in multiple cellular pathways, so it is not clear how many of these ALS-associated splicing changes are directly regulated by TDP-43 versus secondary effects of TDP-43 nuclear depletion and/or cytoplasmic accumulation⁷⁶. For example, transgenic mice expressing human TDP-43^Gln331Lys and TDP-43^Met337Val mutants at levels similar to endogenous TDP-43 develop progressive motor neuron degeneration with target-specific splicing alterations in the absence of TDP-43 nuclear depletion or aggregation⁷⁷. FUS also recognizes a GU-rich motif (GUGGU) and binds to thousands of gene transcripts to regulate splicing in the CNS^78,79. Interestingly, FUS targets conserved introns within genes encoding RBPs that are important for splicing regulation, such as intron 7 in SNRNP70, which encodes the U1 snRNP-associated 70K protein (U1–70K).

In addition to splicing factors, snRNP components may also be prone to aggregation in some diseases. A recent surprise is the potential connection between U1 snRNP activity, splicing regulation and Alzheimer disease⁸⁰. Mass spectrometry analysis of the sarkosylinsoluble proteome from the brains of patients with Alzheimer disease, which includes Aβ peptide and tau protein, revealed that several U1 snRNP proteins, including SNRNP70/U1–70K and SNRPA/U1-A, form tangle-like cytoplasmic inclusions that associate with tau neurofibrillary tangles. This aggregate formation correlates with the accumulation of unspliced precursor RNAs. Both SNRNP70 knockdown and blocking U1 snRNA with an antisense oligonucleotide (ASO) leads to increased levels of the amyloid precursor protein, suggesting that loss of U1 snRNP splicing activity may be an important feature of Alzheimer disease.

Large introns and microexons in neurological disorders

Although introns and exons are highly variable in length, current studies have shown that long (>100 kb) introns and small (≤51 nucleotides⁸¹) exons present particular challenges for the splicing machinery. For example, recursive splicing (RS) — the processing of long introns by sequential events that regenerate a 5′ss — was described originally in the Drosophila melanogaster Ultrabithorax (Ubx) gene. Recently, RS has also been documented in nine human genes by identifying RNA-seq sawtooth read patterns within introns^82,83. In contrast to D. melanogaster, human RS requires the definition of an RS exon downstream of the RS site, and it has been suggested that mutations near these sites might contribute to disease. In this regard, it is noteworthy that the RS read pattern is similar to the FUS-binding distribution determined by iCLIP. However, the relationship between large intron splicing and disease has been brought to the forefront by studies on TDP-43 in ALS. High-throughput strategies (HITS-CLIP and iCLIP) have been used to map binding sites for TDP-43 on RNAs in mouse and human brains^84,85. The preferred binding motif for this protein is UG/GU-rich clusters, and the mouse brain contains thousands (>6,000) of genes with TDP-43-binding sites, often in distal regions of introns. Transcriptome analysis — using RNA-seq and splicing-sensitive microarrays — of control versus ASO-induced TDP-43 knockdowns in the striatum of the brain revealed that TDP-43 regulates the alternative splicing of target RNAs and is also important for maintaining wild-type levels of some transcripts with large introns⁸⁴. A mechanistic connection between these observations has recently emerged from the demonstration that a normal function of TDP-43 is to repress the splicing at non-conserved cryptic splice sites, which are often located in distal introns, as inclusion of the resulting cryptic exons often makes the RNA susceptible to NMD⁸⁶. Because TDP-43 is a member of the hnRNP A/B protein family, it will be interesting to determine whether depletion of other hnRNPs results in splicing of additional non-conserved cryptic exons.

At the other end of the scale, mis-splicing of microexons is involved in autism spectrum disorders (ASDs). ASDs are a clinically heterogeneous group of neurodevelopmental disorders distinguished by impaired social interactions and communication combined with repetitive behaviours, possibly due to cortical circuit hyperexcitability⁸⁷. Microexons are characterized by a high level of evolutionary conservation and have a prominent regulatory function during neurogenesis of the mammalian brain^81,88. These exons encode peptides that modulate interactions between neurogenesis factors during brain development, and alternative splicing of microexons is regulated by the SR-related protein SRRM4 (Ser/Arg repetitive matrix protein 4; also known as nSR100), RBFOX (RNA-binding protein fox-1 homologue 3) and PTBP1 (polypyrimidine tract-binding protein 1)⁸⁸. Importantly, RBFOX1 regulates the alternative splicing of genes that are important for neuronal function, and point, translocation and copy number mutations in RBFOX1 occur in several neurological disorders, including ASDs^89–91. Moreover, HITS-CLIP analysis has determined that RBFOX targets splicing events for multiple autism-susceptibility genes in mice, including Shank3 (SH3 and multiple ankyrin repeat domains 3) and Tsc2 (tuberous sclerosis 2)⁹².

Antisense oligonucleotides

Duchenne muscular dystrophy (DMD) is a progressive muscle disease that affects ~1 in 3,500 newborn males. It is caused by mutations, often deletions, in the largest annotated human gene (2.4 Mb, 79 exons), DMD, which encodes dystrophin⁹⁴. This protein is a key factor in muscle maintenance because it provides an essential link between the dystroglycan complex within the muscle cytoplasmic membrane, or sarcolemma, and the intracellular actin network. Thus, loss of dystrophin results in continuous cycles of myofibre necrosis, satellite cell activation and muscle regeneration, ultimately leading to premature muscle wasting and death. DMD mutations are often multiexon deletions that cause frameshifts, and a common deletion results in a frameshift at exon 51. However, the reading frame can be restored by skipping of exon 51, mediated by ASOs that target an exon 51 ESE. This leads to the production of internally deleted DMD proteins that retain partial function (FIG. 4a). To induce exon 51 skipping, two ASO drug candidates, drisapersen (a 2′O-methylphosphorothioate ASO (2′OMePS)) and eteplirsen (a phosphorodiamidate morpholino oligomer, (PMO)), have progressed through clinical trials, although inflammatory responses to drisapersen have been noted⁹⁵. A similar strategy has been used to reduce abnormal progerin expression and increase lifespan in a mouse model of HGPS by dual targeting of LMNA exon 10 and cryptic exon 11 5′ splice sites with vivo-morpholino ASOs⁹⁶.

Recent studies have also supported the efficacy of ASOs for treating SMA (also known as proximal/5q SMA), an autosomal recessive neuromuscular disorder that is characterized by progressive degeneration of spinal cord anterior horn α-motor neurons⁹⁷. SMA is the leading genetic cause of infant mortality (1 in ~10,000 live births) and is clinically subdivided by age-of-onset and severity. It is caused by loss-of-function mutations and/or deletions in the survival of motor neuron 1 (SMN1) gene, which encodes the SMN protein required for the assembly of Sm proteins onto snRNAs to form functional snRNPs⁹⁸. A paralogous gene, SMN2, also encodes SMN. However, it varies in sequence from SMN1 by a C>T transition in exon 7, which abrogates ESE splicing mediated by the splicing factor SRSF1 (REF. 99), thus promoting skipping of this exon and resulting in an unstable SMN isoform, denoted SMNΔ7 protein that is expressed at low levels (FIG. 4b). To activate SMN2 exon 7 splicing, a 2′-O-methoxyethyl (MOE) ASO (ASO-10-27), which blocks an ISS in SMN2 intron 7 (REF. 100), has been used to increase SMN levels in type 1 (severe) SMA infants and children following intrathecal injection and is currently being tested in Phase III clinical trials⁹⁷. Interestingly, systemic administration of ASO-10-27 in neonates in a mouse model of SMA is effective at rescuing the mutant phenotype, suggesting that SMA is also a peripheral tissue splicing disease¹⁰¹.

For DMD and SMA, mutations result in reduced levels of the encoded proteins. By contrast, another class of diseases, the microsatellite expansion disorders, are associated with unusual RNA structures that alter splicing indirectly (FIGS 3c, 4c). Microsatellites are tandem repeats of 2–10 base pairs in length that comprise ~3% of the human genome and are prone to instability due to DNA replication, recombination and repair errors¹⁰². Although microsatellites undergo both expansions and contractions, expansions cause >30 hereditary diseases. When these repeats occur in non-coding regions, such as untranslated regions (UTRs) and introns, and expand beyond a particular threshold length, they gain a dominant-negative function at the RNA level by sequestering splicing factors, which results in the mis-splicing of hundreds to thousands of RNAs^103,104. A prominent case is myotonic dystrophy (dystrophia myotonica; DM), which is the most common adult-onset muscular dystrophy and is associated with either a CTG expansion (CTG^exp) in the DM protein kinase (DMPK) 3′ UTR (for DM1 disease) or an intronic CCTG^exp in CCHC-type zinc finger, nucleic acid binding protein (CNBP; for DM2 disease)⁷³. Following transcription, these non-coding repeats fold into stable hairpin structures that sequester the muscleblind-like (MBNL) proteins while also triggering CELF1 hyperphosphorylation. MBNL and CELF proteins are alternative splicing factors that act antagonistically during development¹⁰⁵, and MBNL loss-of-function due to sequestration by expansion RNAs activates fetal splicing patterns in adult tissues, leading to the characteristic pathophysiology of DM. Although CAG25 morpholinos have been used to displace MBNL proteins from CUG^exp RNAs, disperse RNA foci and reverse mis-regulated RNA splicing in a mouse transgenic model of DM¹⁰⁶, systemic uptake requires coupling to cell-penetrating peptides, such as peptide-linked morpholinos (PPMOs)¹⁰⁷. Another approach, currently being tested in Phase II clinical trials, is the use of ASO gapmers, which are composed of arms with MOE modifications for stability together with a central gap of ten unmodified nucleotides that are susceptible to RNase H-mediated cleavage. These gapmers are designed to target sequences outside the CUG expansion region in the DMPK transcript¹⁰⁸. This is an effective strategy, as mutant allele transcripts are preferentially targeted because RNase H is localized in the nucleus (cytoplasmic activity is confined to mitochondria), and mutant DMPK mRNAs accumulate in nuclear RNA foci whereas normal mRNAs are efficiently transported into the cytoplasm.

Small molecules

A variety of small molecule strategies have been reported that target disease-relevant mis-splicing. For DMD, the ASOs that promote DMD exon 51 skipping result in low levels of dystrophin production in humans (~2–16% of normal levels) so small molecule screens have been used to identify drug candidates that increase ASO-induced skipping¹⁰⁹. One example is dantrolene, which modulates ryanodine receptor activity and is currently used to treat malignant hyperthermia and muscle spasticity. In one study, dantrolene administered together with suboptimal ASO dose was found to increase exon 51 skipping ~10-fold compared with the vehicle (dimethyl sulfoxide) dose in DMD myotubes (patient fibroblasts that have been re-reprogrammed following MYOD1 expression)¹⁰⁹.

For SMA, a number of compounds that increase SMN protein levels have been identified by high-throughput screens^97,110. Analogous to ASO splice-switching strategies, a recent study uncovered potential drug splicing modifiers that enhance SMN2 exon 7 inclusion¹¹¹. Using an SMN2 minigene reporter cell-based assay, the study found that treatment of SMA type 1 fibroblasts and induced pluripotent stem cell-derived motor neurons with several compounds (SMNC1, SMNC2 and SMNC3) resulted in increased full-length SMN protein levels. Moreover, RNA-seq analysis demonstrated that these compounds are fairly selective and do not cause widespread transcriptome changes¹¹¹. Importantly, these compounds increased SMN protein levels and improved motor function in a mouse model of severe SMA¹¹¹. A similar high-throughput screen of NSC34 motor neurons expressing a SMN2 splicing reporter was used to identify other compounds (NVS-SM1, NVS-SM2, NVS-SM3 and NVS-SM4) that can increase SMN protein levels. One of these, NVS-SM1, achieved a dose-dependent elevation of SMN protein levels and increased the lifespan of SMNΔ7 mice — which are deleted for their single endogenous Smn gene but express SMNΔ7 from human SMN2 trans-genes — from ~15 to >35 days¹¹². The structurally similar NVS-SM2 acts by sequence-dependent stabilization of U1 snRNP bound to the SMN2 exon 7 5′ss.

A wide array of small molecule compounds also block the toxic effects of non-coding microsatellite expansion RNAs. For DM, high-throughput screens as well as more targeted screens have identified a number of compounds that block MBNL1 sequestration and rescue mis-splicing (FIG. 4c). These include a substituted naphthyridine that interacts with UU loops in CUG^exp RNA (DM1)¹¹³, a kanamycin A derivative (multivalent K-alkyne) that binds CCUG^exp RNA (DM2)¹¹⁴, the antifungal pentamidine¹¹⁵, the natural antimicrobial lomofungin¹¹⁶ and ligand 1 (REF. 117).