PRPF6 is a 941 amino acid, 102 kDa protein which acts as a molecular bridge between the U5 snRNP and the U4/U6 di-snRNP and is essential for the assembly of the tri-snRNP, confirmed by recent atomic structures of the human tri-snRNP (Figure 1; Makarov et al., 2000; Liu, 2006; Agafonov et al., 2016; Ru°žičková and Staněk, 2017). The first autosomal dominant RP variant reported in PRPF6 was a heterozygous c.2185C > T (p.Arg729Trp) variant in exon 16 (Table 2; Tanackovic et al., 2011a). Arginine 729 is a highly conserved residue which lies within one of the HAT repeat domains at the C-terminal end of PRPF6. Mutant PRPF6 harbouring the p.Arg729Trp variant accumulates in Cajal bodies in patient lymphoblasts. Cajal bodies are nuclear structures involved in snRNP maturation, tri-snRNP regeneration and the site of defective snRNP accumulation (Tanackovic et al., 2011a; Staněk, 2017). Accumulation of mutant PRPF6 in Cajal bodies indicates impairment of tri-snRNP assembly in patients with the PRPF6 p.Arg729Trp variant. Indeed, the HAT domain containing Arg729 is in a region of PRPF6 known to interact with the U4/U6 di-snRNP and therefore likely affects PRPF6 interactions with this di-snRNP (Tanackovic et al., 2011a). Additionally, human cell lines with the p.Arg729Trp variant displayed inefficient splicing of a number of introns whose decreased splicing is associated with PRPF-linked RP in cell lines from various RP patients with PRPF3, PRPF31, and PRPF8 variants (Tanackovic et al., 2011a,b). Therefore, the PRPF6 p.Arg729Trp variant affects both spliceosomal composition and function.

A further two novel PRPF6 heterozygous missense variants were identified in a cohort of Chinese patients with RP by next-generation sequencing. These variants, c.514C > T, p.Arg172Trp and c.551A > G, p.Asp184Gly, co-segregated with the disease, were absent in population variation databases and were predicted as damaging, but no further characterisation of the variants was performed (Huang et al., 2015). Furthermore, another missense variant, c.1430A > G, p.Asn477Ser, was identified within PRPF6 co-segregating with the neurodegenerative Kufs disease. The patients with this PRPF6 p.Asn477Ser variant were reported with visual impairment (Velinov et al., 2012). However, further work is required to confirm whether this PRPF6 p.Asn477Ser variant has a role in either Kufs disease or the visual impairment (Ru°žičková and Staněk, 2017).

The 220 kDa PRPF8 is the largest known protein of the spliceosome. PRPF8 is a highly conserved protein which forms the catalytic centre of the spliceosome and interacts with the U5 snRNA and the 5′ and 3′ splice sites (Galej et al., 2014; Yan et al., 2015; Bertram et al., 2017; Ru°žičková and Staněk, 2017). PRPF8 interacts with numerous tri-snRNP components including the U5 proteins, EFTUD2 (Snu114p in Saccharomyces cerevisiae) and SNRNP200 (Brr2p in S. cerevisiae), which have essential roles in the splicing cycle (Figures 1, 2). EFTUD2 and PRPF8 regulate the activity of SNRNP200, which unwinds the U4/U6 snRNA duplex to activate the spliceosome (Kuhn et al., 1999; Small et al., 2006; Frazer et al., 2008; Maeder et al., 2009; Mozaffari-Jovin et al., 2012, 2013, 2014; Nancollis et al., 2013; Nguyen et al., 2013).

At least 19 variants in PRPF8 have been associated with autosomal dominant RP (Table 2; Malinová et al., 2017; Ru°žičková and Staněk, 2017). All the known variants cluster in the C-terminal Jab1/MPN domain of the PRPF8 protein which interacts with SNRNP200 (McKie et al., 2001; Kondo et al., 2003; Martinez-Gimeno et al., 2003; Ziviello et al., 2005; Towns et al., 2010; Malinová et al., 2017). The majority of the variants fall in exon 42, although a few PRPF8 variants (including p.Ser2118Phe) lie within exon 38. In yeast, RP variants in PRP8 (yeast homologue of PRPF8) lead to growth defects, although the growth defects were not completely consistent and may be related to the genetic background of the yeast strain (Boon et al., 2007; Maeder et al., 2009; Mozaffari-Jovin et al., 2013; Ru°žičková and Staněk, 2017). Inhibition of U5 snRNP assembly and disruption of the transition between the first and second steps of splicing were also observed in yeast models, with yeast containing Prp8p RP mutations interacting less efficiently with Snu114p and Brr2p and disrupting the regulation of Brr2p helicase activity, leading to defects in pre-mRNA splicing (Boon et al., 2007; Pena et al., 2007; Maeder et al., 2009; Mozaffari-Jovin et al., 2013; Ledoux and Guthrie, 2016; Mayerle and Guthrie, 2016). Reduced assembly of the spliceosome, inefficient splicing and differential alternative splicing was also observed in human cells derived from RP patients with PRPF8 variants (Tanackovic et al., 2011b). PRPF8 RP variants introduced into HeLa cells accumulated in the cytoplasm as well as displaying the normal PRPF8 nuclear localisation, while the majority of PRPF8 mutant proteins degraded more rapidly than the wildtype PRPF8 (Malinová et al., 2017). Furthermore, the majority of PRPF8 RP variants specifically affected spliceosome assembly via inhibition of tri-snRNP formation, reducing the number of fully assembled functional spliceosomes, and consequently splicing defects were observed (Malinová et al., 2017). However, two PRPF8 variants examined (p.Tyr2334Asn and p.Phe2314Leu) did not affect snRNP biogenesis but did affect splicing in vivo (Malinová et al., 2017). These p.Tyr2334Asn and p.Phe2314Leu variants inhibited SNRNP200 helicase activity and weakened association with SNRNP200, respectively, and the resulting SNRNP200 mis-regulation likely accounts for the splicing defects observed (Malinová et al., 2017).

SNRNP200 (Brr2p in yeast) is a 200 kDa protein which interacts with and is regulated by PRPF8 and EFTUD2 at the heart of the U5 snRNP (Figures 1, 2; Van Nues and Beggs, 2001; Liu, 2006; Frazer et al., 2008; Häcker et al., 2008; Mozaffari-Jovin et al., 2012; Nguyen et al., 2013). A recent 7Å cryo-electron microscopy (cryo-EM) structure of the human tri-snRNP revealed that SNRNP200 interacts with the PRPF8 Jab1 domain, the region of PRPF8 containing the majority of PRPF8 RP-linked variants (Figure 1; Agafonov et al., 2016). SNRNP200 is one of eight ATP-dependent DExD/H box RNA helicases in the spliceosome (Cordin and Beggs, 2013). SNRNP200 has an N-terminal domain of unknown function and two helicase modules, an active N-terminal helicase module which unwinds the U4/U6 snRNA duplex during spliceosome activation, and a C-terminal helicase module which acts as an intramolecular regulator (Pena et al., 2009; Zhang et al., 2009). Recent cryo-EM structures of the human spliceosomal pre-B and B complexes have indicated that as well as driving the pre-B to B complex transition during spliceosome activation, SNRNP200 undergoes a major conformational shift, resulting in a drastic change in the overall structure of the spliceosome (Zhan et al., 2018; Zhang et al., 2018).

A number of heterozygous variants in SNRNP200 have been linked to autosomal dominant RP (Table 2). The first two variants identified, p.Ser1087Leu and p.Arg1090Leu, were found in the Sec63-like domain of the N-terminal active helicase module. In yeast, both variants resulted in reduced U4/U6 snRNA unwinding (Zhao et al., 2009; Cvačková et al., 2014). In human cells, neither SNRNP200 variant affected snRNP assembly (unlike the PRPF6 and the majority of PRPF8 RP-linked variants) but did promote the use of cryptic splice sites (Cvačková et al., 2014). The authors proposed that SNRNP200 has an important role in 5′ splice site recognition and splicing fidelity, which is compromised by the two RP variants. Subsequently, additional SNRNP200 variants were identified in the Ski2-like helicase domain of N-terminal active helicase module, which likely disrupt the RNA helicase activity of SNRNP200 (Benaglio et al., 2011; Liu et al., 2012; Bowne et al., 2013; Zhang et al., 2013; Huang et al., 2015). Indeed, in a structural model of the SNRNP200 with a RP-associated p.Gln885Glu mutation, the affected residue is predicted to be a key nucleic acid interaction site and the change in electrostatic potential caused by the mutation would affect nucleic acid contact, disrupting U4/U6 snRNA unwinding and stalling the spliceosome (Liu et al., 2012).

These SNRNP200 RP variants suggest that mis-regulation or defects in SNRNP200 are detrimental to spliceosome activity and may disrupt RNA splicing (Ru°žičková and Staněk, 2017).

Mandibulofacial dysostoses (MFDs) are craniofacial disorders in which malar and mandibular hypoplasia are the core phenotypic features; hearing loss, dysplastic ears and eyelids, and cleft palate are also frequently observed in patients (Wieczorek, 2013). In MFDGA, patients display typical MFD features, microcephaly, external ear malformations and intellectual disability. Hearing loss, cleft palate, choanal atresia, oesophageal atresia, congenital heart defects and radial ray defects are also (less commonly) observed (Table 3; Wieczorek, 2013).

In 2012, using exome sequencing approaches, Lines et al. (2012) identified heterozygous pathogenic variants in the EFTUD2 gene as the cause of MFDGA in 12 unrelated individuals. This study, and subsequent reports, have revealed a variety of different EFTUD2 variants, including missense variants (some of which are pathogenic by affecting splicing of EFTUD2 pre-mRNA), nonsense variants, splice site variants and frameshifts, all of which are predicted to inactivate one allele and therefore reduce EFTUD2 expression, supporting haploinsufficiency as the mechanism of disease (Gordon et al., 2012; Lines et al., 2012; Luquetti et al., 2013; Voigt et al., 2013; Lehalle et al., 2014; Sarkar et al., 2015; Smigiel et al., 2015; Huang et al., 2016; Vincent et al., 2016; Matsuo et al., 2017; Yu et al., 2018; Lacour et al., 2019; Thomas et al., 2020).

EFTUD2 encodes a GTPase which is essential during multiple steps of the spliceosomal cycle, and is highly conserved across eukaryotes from yeast to humans (Fabrizio et al., 1997). Snu114p (yeast orthologue of EFTUD2) plays critical roles in spliceosomal remodelling and dynamics during pre-mRNA splicing (Frazer et al., 2008). Snu114p interacts genetically and physically with Brr2p and Prp8p (Brenner and Guthrie, 2005; Häcker et al., 2008; Nguyen T. H. D. et al., 2016). Similarly, in humans, yeast two-hybrid and in vitro binding assays have demonstrated physical interactions between EFTUD2, SNRNP200 and PRPF8 proteins (Figures 1, 2), and these interactions have been confirmed by recent cryo-EM structures of the human U4/U6.U5 tri-snRNP (Figure 1; Liu, 2006; Agafonov et al., 2016; Charenton et al., 2019). Prior to the first catalytic step of splicing, Snu114p is involved in the dissociation of the U4 and U6 snRNAs by regulating the activity of Brr2 (Bartels, 2002; Bartels et al., 2003; Liu, 2006; Small et al., 2006; Agafonov et al., 2016). After the splicing reactions are complete, Snu114p is also believed to regulate the dissociation of spliceosomal subunits (Small et al., 2006). Interestingly, while Snu114p/EFTUD2 interacts with both Brr2p/SNRNP200 and Prp8p/PRPF8, EFTUD2 is associated with a craniofacial phenotype while SNRNP200 and PRPF8 both cause RP (Figure 1). It is remarkable that variants in functionally and physically interacting proteins of the same spliceosome complex lead to two very different disease phenotypes.

Several groups have generated zebrafish models in which the eftud2 gene is disrupted that display severe, disease-relevant phenotypes, indicating a vital, conserved role for eftud2 in vertebrate development (Deml et al., 2015; Lei et al., 2017). In particular, Lei et al. (2017) developed a zebrafish model in which the eftud2 gene contains a nonsense mutation leading to reduced eftud2 expression. The authors found increased apoptosis and mitosis of neural progenitors but little effect on differentiated neurons in these mutants. RNA-Seq and functional analyses revealed that there was a transcriptome-wide splicing deficiency, with increased intron retention and exon skipping events, leading to inadequate nonsense-mediated decay (NMD) and activation of p53-dependent apoptosis in this eftud2 mutant zebrafish. The authors proposed that these non-degraded, aberrant, transcripts imposed particular stresses on neural progenitors because eftud2 expression is highly concentrated in the brain in zebrafish embryos after 36 hours post-fertilisation (36hpf) (although it was broadly expressed at earlier stages of development), promoting neural-specific apoptosis. Together, these findings demonstrated an important involvement of eftud2 in neural progenitor development, which may contribute to the neurological abnormalities observed in MFDGA patients (Lei et al., 2017). Beauchamp et al. (2019) recently used in situ hybridisation to characterise Eftud2 expression during mouse development and revealed that Eftud2 is expressed throughout development (as expected for a core spliceosome factor), but with a particularly enrichment in the developing head and craniofacial regions. The authors also generated a mouse model with a loss-of-function exon 2 deletion in Eftud2; however, heterozygous embryos did not model MFDGA, while homozygous mutant embryos were not observed post-implantation, confirming a requirement for Eftud2 expression for survival and viability of pre-implantation embryos (Beauchamp et al., 2019).

Recently, we generated an EFTUD2 knockdown HEK293 cell line modelling EFTUD2 haploinsufficiency in MFDGA (Wood et al., 2019). Reduction of EFTUD2 expression in these cells resulted in decreased proliferation, cell cycle defects and an increased sensitivity to endoplasmic reticulum (ER) stress. Furthermore, RNA-Seq analysis revealed widespread mis-expression and mis-splicing of genes, including transcripts relevant to embryonic and craniofacial development, with the mis-spliced genes sharing common cis-acting sequence properties thought to allow (by an as-yet-unknown mechanism) increased sensitivity to dysregulated splicing when EFTUD2 expression is lowered. We, and others, have proposed a mechanism in which an increased burden of mis-spliced pre-mRNAs in the ER resulting from reduced EFTUD2 expression ultimately activates p53-dependent apoptosis (Wood et al., 2019; Beauchamp et al., 2020). Why cells of the developing craniofacial region are particularly affected remains to be determined, although the apparent dependence of neural progenitors on EFTUD2 in animal models is a likely explanation. Indeed, neural progenitors have very high turnover of pre-mRNAs and high levels of alternative splicing (similar to the human retina), indicative of a high reliance on spliceosome function compared to other tissues (Gurok, 2004; Rosignoli et al., 2010; Burow et al., 2015; Su et al., 2018; Weyn-Vanhentenryck et al., 2018). Additionally, neural crest cells (NCCs), the key cells involved in vertebrate craniofacial development and central to the aetiology of a number of other similar craniofacial disorders, are particularly sensitive to activated p53 and are twofold more likely to undergo apoptosis when exposed to stabilised p53 (Calo et al., 2018; Merkuri and Fish, 2019). NCCs also express higher levels of p53 than other cell types during development (Rinon et al., 2011; Merkuri and Fish, 2019). Taken together, these findings would corroborate a mechanism for MFDGA involving NCC-specific apoptosis during development.

In RP it has been suggested that constant production of mis-folded snRNP proteins over time activates the unfolded protein response and creates long-lasting stress. Together with photo-oxidative damage common to retinal cells, this stress eventually triggers apoptosis leading to retinal degeneration later in life, especially as photoreceptor cells do not regenerate so protein defects and cell stresses accumulate over time (Figure 3; Ru°žičková and Staněk, 2017). This hypothesis links the potential mechanisms of RP and MFDGA and the susceptibilities of retinal cells and NCCs to apoptosis may help to explain why these are the primary tissues affected by spliceosome protein variants.

Burn-McKeown syndrome is an MFD in which affected individuals display a characteristic combination of choanal atresia, craniofacial anomalies, including cleft lip and/or palate, lower eyelid coloboma, short palpebral fissures, a prominent nasal bridge, large protruding ears and sensorineural deafness (Table 3). Cardiac defects and other extra-craniofacial phenotypes may also be observed, but intellectual development is usually normal (except in one reported case thus far) (Table 3; Burns et al., 1992; Toriello and Higgins, 1999; Wieczorek et al., 2003, 2014; Lehalle et al., 2015; Goos et al., 2017; Strang-Karlsson et al., 2017; Narayanan et al., 2020). BMKS is a rare human disorder – fewer than 20 individuals with the condition have been reported, including a large consanguineous Alaskan family who were initially diagnosed with oculo-oto-facial dysplasia (Hing et al., 2006).

Wieczorek et al. (2014) identified biallelic variants in the U5 snRNP gene TXNL4A as causative in BMKS. Most patients have a 34 bp deletion (known as type 1Δ) in the promoter region of one allele of TXNL4A in combination with a loss-of-function variant (microdeletion, splice site, nonsense or frameshift variant) on the other allele (Wieczorek et al., 2014; Goos et al., 2017). Some patients are homozygous for a slightly different 34 bp deletion, the type 2Δ, in the promoter region of TXNL4A (Wieczorek et al., 2014; Narayanan et al., 2020). The type 1Δ and type 2Δ promoter deletions led to a reduction in reporter gene expression in a dual luciferase assay, with the type 2Δ causing a more severe reduction in reporter gene expression (Wieczorek et al., 2014). This more severe reduction in expression for type 2Δ might explain why a homozygous type 2Δ is sufficient to cause BMKS, while a type 1Δ must be combined with a null allele. Nonetheless, it is considered that BMKS is the product of reduced dosage of TXNL4A in affected individuals, with more severe genotypes such as homozygous loss-of-function variants being incompatible with life.

Interestingly, a study by Goos et al. (2017) identified two cousins with a homozygous TXNL4A type 2Δ with choanal atresia and other minor facial anomalies but not the full features of BMKS, in contrast to the previously described type 2Δ patients with BMKS. This finding may indicate variable or incomplete penetrance. Compensatory genetic variants in TXNL4A or other genes may abrogate the reduction in TXNL4A expression to some extent and lead to a milder phenotype.

TXNL4A is one of eight core protein members of the U5 snRNP (Liu, 2006). The S. cerevisiae orthologue of TXNL4A, DIB1, encodes a small highly conserved protein which is absolutely required for pre-mRNA splicing in vivo, as demonstrated by genetic depletion experiments (Reuter et al., 1999). In S. cerevisiae, null mutations of DIB1 are lethal, as are deletions of the Schizosaccharomyces pombe orthologue DIM1. In haploid S. cerevisiae in which DIB1 was placed under the control of the GAL1 promoter, defective assembly of the U4/U6.U5 tri-snRNP was observed when DIB1 expression was blocked, which was predicted to affect downstream pre-mRNA splicing (Reuter et al., 1999; Wieczorek et al., 2014). Because DIB1 is highly evolutionarily conserved from yeast to humans, it is likely that reduced TXNL4A expression arising from the BMKS-associated variants in affected patients also leads to defective tri-snRNP assembly. These defects in spliceosome assembly could, in turn, lead to the altered splicing of a subset of pre-mRNAs, the downstream consequence of which is the clinical manifestation of BMKS.

Recent studies have suggested that Dib1p has an important role in preventing premature spliceosome activation, and the departure of Dib1p and other proteins from the spliceosome defines the transition from the B to B^act complex during the splicing cycle (Schreib et al., 2018). It has been proposed that Dib1p acts as a “placeholder” in the B complex, preventing formation of certain RNA-RNA interactions, the recruitment of other proteins and/or required movements to form the B^act complex. Schreib et al. (2018) generated a range of Dib1p mutants and found that Dib1p is robust and can tolerate many mutations, even at positions believed to be critical for folding stability, possibly through the compact structure of Dib1p. Dib1p also readily exchanged in splicing extracts, indicating accessibility of the Dib1p binding site in the spliceosome, despite Dib1p being a core protein of the U5 snRNP. The authors did identify two temperature-sensitive mutants which stalled in vitro splicing reactions before the first catalytic step of splicing and blocked spliceosome assembly at the B complex. It was proposed that the temperature-sensitivity resulted from altered interactions between Dib1p and other spliceosomal proteins, such as Prp6p and Prp8p, and not changes in Dib1p conformation (Figures 1, 2; Schreib et al., 2018). This study has provided insight into how Dib1p functions in the activation of the spliceosome. Furthermore, a recent cryo-EM structure of the human U4/U6.U5 tri-snRNP also revealed that TXNL4A is connected to EFTUD2 via PRPF8 and the U5 snRNA loop I, and it is hypothesized that EFTUD2 may catalyse the removal of TXNL4A from the tri-snRNP (Figures 1, 2; Fabrizio et al., 2009; Agafonov et al., 2016; Wan et al., 2016; Wood et al., 2019). Cryo-EM structures of the human spliceosome pre-B and B complexes also revealed that the N-terminus of PRPF6 forms two short α-helices on the surface of TXNL4A, and it is hypothesized that PRPF6 might block TXNL4A from exiting the spliceosome prematurely during the splicing cycle (Zhan et al., 2018). These findings indicate the RP- and craniofacial disorder-linked U5 snRNP proteins form an intricate network of physical and functional interactions at the heart of the spliceosome, making the phenotypic discrepancies all the more intriguing (Figures 1, 2).

Recently, we discovered that induced pluripotent stem cells (iPSCs) generated from a BMKS patient have defective differentiation to NCCs compared to maternal and unrelated control iPSCs, in particular revealing defects in the epithelial-to-mesenchymal transition (EMT) (Wood et al., 2020). RNA-Seq analysis revealed widespread differential gene expression and differential splicing in patient NCCs, with an enrichment for genes involved in processes involved in craniofacial development and the mis-splicing of a gene early in the WNT signalling pathway required for NCC specification. The mis-spliced genes shared common sequence features, although how these sequence properties render a pre-mRNA more vulnerable to mis-splicing when TXNL4A expression is reduced is unclear. Interestingly, the BMKS patient NCCs did not display increased apoptosis compared to maternal and control NCCs, indicating a different mechanism to that proposed for MFDGA (Wood et al., 2019, 2020). Furthermore, the mis-expressed and mis-spliced pre-mRNAs in HEK293 EFTUD2 knockdown cells and in BMKS patient-derived NCCs did not overlap to a great extent, although again this finding may be due, at least in part, to cell type specificity, and there were different sequence features associated with the mis-spliced pre-mRNAs in each disease model (Wood et al., 2019, 2020). It would be expected that the majority of the affected transcripts in MFDGA and BMKS would be the same, resulting in the overlapping clinical features in the two disorders. Nonetheless, the differences in affected transcripts could explain the non-identical phenotypes of MFDGA and BMKS, although for a true comparison between the disorders iPSCs should be generated from MFDGA patients and differentiated to NCCs as a disorder-relevant cell type-specific model.

Therefore, similar to the discussion for RP, the variants in the core U5 snRNP proteins linked to craniofacial disorders result in changes in pre-mRNA splicing. These defects in pre-mRNA splicing are presumed to result in the specific disease phenotype at least in part, by affecting transcripts involved in the relevant developmental processes. Additionally, global mis-splicing and cell type-specific apoptosis may also play a role in the disease mechanism, at least for MFDGA. However, distinct groups of genes are presumably mis-spliced in RP and craniofacial disorders which leads to the different phenotypic features of each disorder. Understanding why specific sequence features render certain pre-mRNAs more vulnerable to mis-splicing when different U5 snRNP proteins are mutated or have reduced expression may be the key to understanding how the distinct phenotypic differences arise. It is plausible that, as every pre-mRNA is unique, certain spliceosomal proteins have a more important role in the splicing of pre-mRNAs with particular features, making those pre-mRNAs more reliant on proper functioning or amount of that spliceosomal protein for normal splicing. However, there is still much work needed to unravel this hypothesis.

In addition to the role of PRPF8 in RP, recurrent somatic mutations and hemizygous deletions have been identified in PRPF8 in myeloid neoplasms including myelodysplastic syndrome (Table 4) (MDS). Kurtovic-Kozaric et al. (2015) screened a large cohort of patients with MDS and related conditions to identify a number of somatic missense and somatic nonsense mutations in PRPF8, as well as numerous cases containing the deletion of one copy of the PRPF8 locus exhibiting PRPF8 haploinsufficiency (Makishima et al., 2012; Kurtovic-Kozaric et al., 2015). The PRPF8 missense mutations are distributed throughout the length of the gene, including the Jab1/MPN domain, and were most frequently identified in primary and secondary acute myeloid leukaemia (AML), suggesting an association with more aggressive cancer phenotypes compared to low-risk MDS. PRPF8 mutations resulted in increased cellular proliferation, and PRPF8 mutations and deletions correlated with the presence of ringed sideroblasts (RS) and pseudo Pelger-Huet anomaly (PHA) (Kurtovic-Kozaric et al., 2015). The authors suggested that the identified PRPF8 mutations alter the internal dynamics of the spliceosome, and revealed that splicing patterns and splice site recognition were altered in both yeast and human cells carrying the MDS-associated PRPF8 mutations (Kurtovic-Kozaric et al., 2015). Gene expression patterns were also altered in PRPF8 mutated and deleted samples, with many of the differentially expressed genes and mis-spliced genes associated with mitochondrial function and haematopoietic differentiation (Kurtovic-Kozaric et al., 2015). More recent work has found that PRPF8 missense mutations in MDS patients are generally secondary mutations, and often co-occurred with other more common cancer splicing factor mutations (SRSF2, SF3B1, LUC7L2, U2AF1, and ZRSR2) (Adema et al., 2017).

Precisely how and why these PRPF8 mutations impact the overall function of PRPF8 protein resulting in neomorphic splicing activity and leading to the malignant phenotype of aggressive myeloid malignancies with increased RS is not yet understood (Ru°žičková and Staněk, 2017). It seems likely that the mis-expression and mis-splicing of genes involved in iron accumulation in the mitochondria and abnormal haematopoiesis has a central role in the cancer phenotype (Kurtovic-Kozaric et al., 2015). Furthermore, how different missense changes can cause such a dramatically different phenotype to other missense changes in the same protein (cancer versus retinitis pigmentosa) is not known. While both PRPF8 mutations in MDS and PRPF8 variants in RP alter RNA splicing and change splicing patterns, why different groups of genes (presumably with different physical properties) are specifically affected by the different missense mutations in the same protein and how these result in the very different disease presentations is not known. One hypothesis is the cancer-associated variants affect the entire length of the PRPF8 protein and so may affect different functions and/or interactions of PRPF8 than the RP-linked PRPF8 variants which are only found in the Jab1/MPN domain (Kurtovic-Kozaric et al., 2015; Ru°žičková and Staněk, 2017). However, mutations affecting the Jab1/MPN domain have been identified in some cancers as well, so this explanation cannot fully account for the different disease presentations (Kurtovic-Kozaric et al., 2015). Even within the Jab1/MPN domain, different amino acids may be involved in different functions so the exact identity of the altered residue likely has an important role in disease outcome.

In addition to its role in RP, overexpression or amplification of PRPF6 is a common oncogenic driver of proliferation in human primary and metastatic colon cancer (Table 4; Adler et al., 2014; Lokody, 2014). Knockdown of PRPF6 expression in human cancer cell lines with increased levels of PRPF6 inhibited cell growth in vitro, and inducible knockdown of PRPF6 in xenograft tumours led to tumour shrinkage only in tumour models with high PRPF6 expression (Adler et al., 2014). Reduced PRPF6 led to intron retention of a relatively small subset of genes, including an oncogenic long isoform of the ZAK kinase (ZAK-LF) (Adler et al., 2014). ZAK-LF levels correlated with PRPF6 expression in colon cancer cells, and PRPF6 was required for the alternative splicing of ZAK to produce ZAK-LF. Expression of ZAK-LF transformed immortalised murine fibroblasts and induced xenograft tumour formation in immunodeficient mice, while depletion of ZAK-LF reduced the growth of PRPF6-overexpressing colon cancer cells in vitro and in xenografts (Adler et al., 2014; Lokody, 2014).

From this study, it was suggested that overexpression of PRPF6 has an important role in driving colon cancer, via the altered splicing of gene isoforms related to growth and proliferation (Adler et al., 2014). Why different groups of genes may be differentially spliced from PRPF6 overexpression compared to RP missense variants in PRPF6, remains unclear. Characterising the sequence properties of the mis-spliced RNAs in each case may help to begin unravelling this difference.

The function of SNRNP40 in the U5 snRNP complex during pre-mRNA splicing is not well-understood. In 2016, Nguyen et al., identified clonal human breast cancer subpopulations with different levels of morphological and molecular diversity (which are associated with metastatic colonisation and chemotherapeutic survival), and identified genes with high inter-cell transcript expression variability (Nguyen A. et al., 2016). The authors found high variability in genes encoding splicing machinery proteins, including SNRNP40 (Table 4). The authors engineered cells with variable SNRNP40 expression and revealed that SNRNP40 depletion promoted systemic metastasis, with increased levels of unspliced pre-mRNAs in cells with low SNRNP40 expression. Clinically, low SNRNP40 expression was found associated with metastatic outcomes. It was proposed that deregulation of splicing factors, including SNRNP40, may amplify alterations of gene regulatory and expression networks and might lead to molecular and phenotypic diversity associated with metastasis (Nguyen A. et al., 2016). However, the underlying mechanisms of precisely how variable expression of splicing factors, including SNRNP40, promote metastatic progression and the overall contribution to cancer progression is not understood.

While reduced expression of EFTUD2 is associated with MFDGA, increased expression of EFTUD2 has been linked to human colitis-associated cancer (CAC) (Table 4). This contrast in phenotypes indicates that human cells are very sensitive to alterations in EFTUD2 expression levels. EFTUD2 plays a role in preventing hepatitis C virus (HCV) by upregulating expression via splicing of interferon-stimulated genes (ISGs) such as RIG-I and MDA5, suggesting EFTUD2 is a novel innate immune system regulator (Zhu et al., 2015). More recently, in mouse models of CAC, Eftud2 was overexpressed in colonic tissues and infiltrating macrophages (Lv et al., 2019). Myeloid-specific knockout of Eftud2 suppressed chronic intestinal inflammation and tumour development by decreasing inflammatory cytokine and tumorigenic factor production via compromised activation of NF-κB signalling. This impaired signalling activation resulted from changes in Eftud2-mediated alternative splicing of components of the NF-κB pathway in macrophages. The authors concluded that overexpression of Eftud2 is involved in the pathogenesis of CAC by modulating the inflammatory response of macrophages, highlighting the link between inflammation, cancer and alternative splicing in the innate immune system. Furthermore, this work emphasises how excessive EFTUD2 expression can also lead to distinct pathological consequences compared to reduced EFTUD2 expression in MFDGA. Why the processing and/or expression of different groups of genes is affected when EFTUD2 is overexpressed, resulting in such a different phenotype, compared to EFTUD2 knockdown cells modelling MFDGA remains to be determined (Wood et al., 2019). It is possible that different levels of EFTUD2 result in the formation of different splicing complexes and/or that more or less EFTUD2 protein allows either faster or slower regulation of SNRNP200, which in turn could affect the splicing of different pre-mRNAs.

DDX23 is an 820 amino acid DEAD-box RNA helicase protein of the U5 snRNP which is required for the formation of the spliceosomal B complex after its phosphorylation by SRPK2 (Mathew et al., 2008). However, the exact role of DDX23 in the spliceosome is not well-understood. Yin et al. (2015) identified DDX23 overexpression in glioma tissues, with high expression of DDX23 correlating with poor glioma patient survival (Table 4). The authors found that knockdown of DDX23 in vitro and in vivo suppressed glioma cell proliferation and invasion. Interestingly, the DDX23 protein promoted the post-transcriptional biogenesis of microRNA mir-21 via interaction with the Drosha complex (Yin et al., 2015). miR-21 upregulation was previously known to be strongly associated with proliferation, invasion and radiation resistance of glioma cells (Kwak et al., 2011; Ha and Kim, 2014). Mutagenesis demonstrated that the helicase activity of DDX23 was essential for processing of mir-21. Furthermore, inhibiting DDX23 activity chemically with the RNA helicase inhibitor ivermectin decreased miR-21 levels and blocked invasion and cell proliferation in glioma cell lines and decreased glioma growth in mouse xenografts (Yin et al., 2015). Thus, unlike other U5 proteins associated with cancer, the key role of DDX23 upregulation in glioma progression does not appear to stem from specific alterations in pre-mRNA splicing, but rather through additional functions of the protein, in this case microRNA processing. It is possible that additional, non-spliceosomal and as-yet-unknown functions of the other U5 snRNP proteins may play a role in the pathogenesis of cancer, retinitis pigmentosa or craniofacial disorders.

Interestingly, the phosphorylation of DDX23 by SPRK2 following pausing of RNA polymerase II during transcription plays an important role in suppressing R-loops, nucleic acid structures generated during transcription that can lead to genomic instability (Skourti-Stathaki and Proudfoot, 2014; Sollier and Cimprich, 2015). The absence of either SPRK2 or DDX23 leads to an accumulation of R-loops resulting in massive genomic instability, with the role of DDX23 in suppressing R-loops not requiring a functional U5 snRNP (Achsel et al., 1998; Makarov et al., 2000; Grainger and Beggs, 2005; Agafonov et al., 2016; Sridhara et al., 2017). DDX23 mutations and homozygous deletions have been identified in several different cancers, including adenoid cystic carcinoma (ACC), implicating DDX23 loss as a potential source of genomic instability which may have an important role in cancer development (Sridhara et al., 2017). Again, this link between DDX23 and cancer is not related to its function in the spliceosome and argues that extra-spliceosomal functions of certain U5 proteins could potentially play a role in their pathogenesis.