Direct and indirect effects of spliceosome disruption compromise gene regulation by nonsense-mediated mRNA decay

ABSTRACT

Pre-mRNA splicing, carried out in the nucleus by a large ribonucleoprotein machine known as the spliceosome, is functionally and physically coupled to the mRNA surveillance pathway in the cytoplasm called nonsense-mediated mRNA decay (NMD). The NMD pathway monitors for premature translation termination, which can result from alternative splicing, by relying on the exon junction complex (EJC) deposited on exon–exon junctions by the spliceosome. Recently, multiple genetic screens in human cell lines have identified numerous spliceosome components as putative NMD factors. Using publicly available RNA-seq datasets from K562 and HepG2 cells depleted of 18 different spliceosome components, we found that natural NMD-targeted mRNA isoforms were upregulated when catalytic spliceosome members were reduced. While some of this increase could be due to widespread pleiotropic effects of spliceosome dysfunction (e.g. reduced expression of NMD factors due to missplicing of their mRNAs), we identified that AQR, SF3B1, SF3B4, and CDC40 may have a more direct role in NMD. We also tested the hypothesis that increased production of novel NMD substrates may overwhelm the pathway to find a direct correlation between the amount of novel NMD substrates detected and the degree of NMD inhibition observed. Finally, similar transcriptome alterations and NMD substrate upregulation were observed in cells treated with spliceosome inhibitors and in cells derived from retinitis pigmentosa patients with mutations in PRPF8 and PRPF31. Overall, our results show that regardless of the cause, spliceosome disruption upregulates a broad set of NMD targets, which could contribute to cellular dysfunction in spliceosomopathies.

Introduction

Pre-mRNA splicing has a profound impact on mRNA substrates that are generated and translated into proteins. While alternative splicing generates multiple mRNA isoforms from a single gene to diversify the proteome, it also has the potential to impact open reading frame integrity and, hence, compromise protein expression. It is therefore not surprising that pre-mRNA splicing in the nucleus is functionally coupled to the translation-linked nonsense-mediated mRNA decay (NMD) mechanism in the cytoplasm, which identifies and rapidly degrades mRNAs containing premature translation termination codons (PTC) [1,2]. The influence of this connection has been widely documented in previous studies. For example, in Saccharomyces cerevisiae mutations in NMD components cause increased accumulation of erroneously spliced mRNAs [3], whereas in human cell lines, numerous transcripts with disrupted open reading frames that are normally degraded and suppressed by NMD can be detected in the nucleus or in the pool of pre-translated mRNAs [4]. A particularly notable example of the functional connection between splicing and NMD is the process of alternative splicing-coupled NMD (AS-NMD), where regulated alternative splicing of poison exons either introduces a PTC upon their inclusion or creates one via frameshifting if the exon is excluded [1,2,5,6]. AS-NMD ties splicing and NMD together in a complex regulatory network used to fine-tune gene expression, often via evolutionarily conserved poison exons in developmentally important genes [5,6].

Splicing and NMD are physically connected via the deposition of the exon junction complex (EJC) 24 nucleotides (nt) upstream of exon–exon junctions during the catalytic steps of splicing [7]. During translation, the presence of an EJC downstream of a stop codon is the most prominent sensor of premature termination (reviewed in [8–10]). Such 3’-untranslated region (UTR)-bound EJCs engage with key NMD factors including UPF1, UPF2 and UPF3, to mark the termination event as aberrant. Following the sensing of aberrant termination, NMD is activated when UPF1 is phosphorylated by the SMG1 kinase to subsequently recruit SMG5, SMG6, and SMG7 proteins that either directly initiate mRNA degradation via SMG6-mediated endonucleolytic cleavage or by recruiting other mRNA decay enzymes through SMG5 and SMG7 [11–13].

Through their role in EJC deposition, many spliceosome components have been identified as playing a direct role in NMD, thereby extending the connection between splicing and NMD. CWC22 and CWC27, two proteins recruited to the activated spliceosome, directly mediate the recruitment and deposition of the EJC anchor, EIF4A3 [14–17], and thereby play an important role in NMD. AQR, also known as Intron Binding Protein 160 (IBP160), is another spliceosome protein implicated in EJC deposition through an unknown mechanism and has a documented role in NMD [18]. The structures of the spliceosome [19,20] that contain the EJC (or pre-EJC) show other spliceosome proteins, such as EFTUD2, which come in direct contact with EJC subunits and may have a role in EJC assembly or deposition.

Two lines of evidence suggest that the connection between splicing and NMD may be more extensive than currently understood. First, human patients with mutations in several spliceosome components and the EJC proteins have similar phenotypic effects. Mutations in pre-mRNA splicing components cause numerous disorders collectively referred to as spliceosomopathies [21], which can be classified into five broad categories: craniofacial disorders, neurodevelopmental deficits, limb defects, myelodysplastic syndrome (MDS), and retinitis pigmentosa (RP), the latter of which is the most common. Curiously, mutations in EJC subunits, notably the core protein EIF4A3, also cause cranio-facial disorders and neurodevelopmental defects [21–24]. Moreover, mutations in EIF4A3 and RBM8A, another EJC core protein, like those in spliceosome components SNRNPA and SF3B4, lead to limb defects [24,25]. Second, several recent genetic screens for potential NMD factors have identified numerous spliceosome components among the top hits [26–29]. The identification of spliceosome proteins as potential NMD factors in these screens, which were performed using different NMD reporter RNAs and employed different candidate gene inactivation strategies, underscores the possibility that spliceosome components beyond CWC22, CWC27, and AQR may have yet to be determined roles in NMD.

To uncover the extent and modes of connection between the spliceosome and NMD, we analysed publicly available RNA-seq datasets from human cell lines depleted of several spliceosome proteins, those treated with drugs that alter spliceosome function, and cells derived from patients with spliceosome mutations that cause retinitis pigmentosa. In these samples, we quantified changes in the abundance of NMD-targeted transcripts and splicing patterns to identify changes in the transcriptome when the spliceosome was disrupted. Our results show that the depletion of many catalytic spliceosome components leads to an increased abundance of endogenous EJC-dependent NMD targets. In several cases, we observed a similar increase in the other non-canonical isoforms. These results indicate that depletion of spliceosome components broadly changes the transcriptome, resulting in the upregulation of NMD-targeted transcripts through mis-splicing, reduction in NMD efficiency, or both. Interestingly, depletion of four spliceosomal components, AQR, CDC40, SF3B1, and SF3B4, similar to the knockdown of EJC core protein EIF4A3, causes higher upregulation of NMD-targeted isoforms as compared to other non-canonical isoforms, suggesting that their effects on NMD could be more direct. Taken together, we show that disruption of pre-mRNA splicing has direct as well as pleiotropic effects on gene expression, which also results in increased expression of NMD-targeted transcripts. Although the precise reason for this effect on NMD targets remains unresolved, altered levels of NMD-regulated genes may contribute to the molecular phenotypes observed in spliceosomopathies [30].

Results

Several components of catalytic spliceosome are overrepresented in genetic screens for novel NMD factors in human cell lines

In the search for novel NMD factors, several genome-wide genetic screens have been recently conducted in human cell lines using CRISPR-Cas9 mediated gene knockouts [26,28,29] or siRNA-mediated knockdowns [27]. Even though these screens employed different NMD reporters and gene knockdown/knockout methodologies, a comparison of the top 200 factors identified in each of these four screens showed a substantial overlap among the factors that influence NMD (Figure 1(A)). A total of 691 genes are present in the top 200 list of the four screens. Functional protein association analysis between these 691 proteins using STRING [31] showed that gene ontology terms related to mRNA metabolism, including pre-mRNA splicing and NMD, were enriched in this set (Table S2). Furthermore, spliceosome factors, as defined by the spliceosome database [32], constituted 170 of the 691 proteins (Table S3). Surprisingly, among a more stringent list of 65 proteins that are present in the top 200 hits of two or more screens, 43 are spliceosome factors (Figure 1(B)), which form an interconnected network with known and novel NMD factors (Figure 1C). The spliceosome undergoes a number of remodelling steps during pre-mRNA splicing, meaning that not all components are present during all stages of the splicing cycle [33]. We categorized the identified spliceosome factors based on the last spliceosome complex they are associated with as defined in the spliceosome database (Figure 1(C,D)). Based on these classifications, we find that components of the P complex, the spliceosome that results immediately after exon-ligation step and before spliceosome disassembly, are overrepresented in NMD screens compared to their abundance in the spliceosome database (p-value < 2.2 × 10⁻¹⁶, Χ² goodness-of-fit test) (Figure 1(B-D)).

Figure 1.

Spliceosome components identified in NMD factor screens are predominantly from catalytic spliceosome complexes.

A) A Venn diagram showing overlaps between the top 200 hits in the indicated screens for NMD factors. Sixty-five proteins were common in at least two of the four screens.

B) A bar plot showing what fractions of spliceosome components identified in the NMD screens group into the spliceosomal complexes shown on the x-axis. The spliceosomal complex assignment was based on when a particular protein leaves the spliceosome, as defined by the spliceosome database. Early spliceosome components are coloured red, catalytic spliceosome components are teal, and components with no annotated leaving time are grey.

C) STRING network of protein–protein interactions of all factors identified in the top 200 of any two of the four screens (65 proteins from A). Nodes are coloured according to when the protein leaves the spliceosome: red for early spliceosome components, teal for catalytic spliceosome components, grey for spliceosome components with no annotated leave point, and yellow for proteins that are not part of the spliceosome. Rings around nodes indicate gene ontology biological processes: RNA splicing (GO:0000398) in light blue; and nonsense-mediated decay (GO:0000184) in dark blue.

D) The spliceosome components under investigation, grouped into splicing subcomplexes, are arranged around the splicing cycle where they leave the spliceosome.

Given the tight coupling between pre-mRNA splicing and NMD via the deposition of the NMD-enhancing EJC, some spliceosome components are expected to be enriched in screens for NMD factors. For example, CWC22, an integral spliceosome component with a direct role in the recruitment and positioning of EIF4A3 on the 5’-exon within the catalytic spliceosome [14–16], was among or close to the top 200 hits in three of the screens (ranked 7, 233, and 379). Similarly, AQR, an NTC-related core spliceosome component with a documented role in EJC deposition, albeit via an unknown mechanism [18], was detected in the top 200 hits in three out of the four screens. However, the identification of a surprisingly large number of catalytic spliceosome components in NMD screens suggests that the connection between splicing and NMD is likely to extend far beyond spliceosome components previously known to be involved in EJC deposition. Interestingly, no such enrichment was observed for components of early spliceosomal complexes that are involved in splice site recognition but leave before the catalytic steps (Figure 1(C)). Thus, the catalytic spliceosome appears to have a broader and yet not fully appreciated impact on the proper functioning of NMD.

Depletion of several catalytic spliceosome components increases abundance of endogenous NMD targets

To investigate the impact of spliceosome components on NMD, we analysed data available from the ENCODE consortium as part of the ENCORE project [34–36], where individual proteins were knocked down or knocked out in K562 or HepG2 cell lines using siRNA or CRISPR-Cas9, respectively, followed by total RNA-Seq. Such RNA-Seq data are available from ENCODE for 16 of the 43 spliceosome proteins in the top 200 hits from at least two of the four screens. For the two snRNPs, U1 and U5, only one component was present among the top hits. Therefore, in our investigation, we included additional members from these snRNPs (SNRNPC for U1; PRPF6, PRPF8, and EFTUD2 for U5), even though they were not among the 43 proteins shared among the NMD screens. For all spliceosome component depletion datasets, we quantified gene expression at the transcript level (using kallisto [37] and hg38 transcriptome build 109) and performed differential expression analysis (using DESeq2 [38]) between the depletion and wild-type (WT) datasets (Table S4). Any datasets where the depletion and WT replicates did not segregate in a principal component analysis or where the primary protein-coding mRNA (as per the MANE select definition [39]) of the protein being depleted was less than 2-fold downregulated were not analysed further (Fig S1). We investigated 18 spliceosome components in total that met these criteria and represented among them are all snRNPs of the major spliceosome, as well as the nineteen complex (NTC) and NTC-related proteins (NTR) (Figure 2(A)). As controls, we used ENCODE RNA-Seq datasets from the same cell lines that were depleted for either the key EJC proteins MAGOH and EIF4A3 or the central NMD factor UPF1.

Figure 2.

Depletion of many catalytic spliceosome components upregulates endogenous NMD targeted mRNAs in K562 cells.

A) A heatmap on left displaying the -log10(p-value) of the Wilcoxon test comparing log2(fold change) of NMD-targeted as compared to MANE isoforms for all spliceosome component depletions tested. In the table to the right, spliceosome components are coloured according to when they leave the spliceosome. Stable complexes the shown proteins are part of are also given. GNB2L1 and SNRNP70 knockdown datasets are from HepG2 cell lines while all others are from K562 cells.

B) Boxplots displaying the log2(fold change) on the y-axis of MANE transcripts (purple) and NMD-targeted transcripts (orange) in the knockdowns indicated on x-axis. The number of transcripts in each group is indicated below the boxplot, the median of the boxplot is indicated on the boxplot, and the p-value of the Wilcoxon text comparing the two groups is above. Spliceosome component names are coloured according to when they leave the spliceosome as in A.

C) Boxplot showing the log2(fold change) of MANE transcripts (blue) and NMD-targeted transcripts (red) from genes with conserved poison exons. Boxplot and depletion annotations are as in B.

D) Genome browser view showing distribution of reads mapping to EIF4A2 in a representative wildtype (green) and knockdown (purple) replicates. The poison exon in EIF4A2 along with mapping reads are indicated by the dashed box, and the scale of each pair of tracks is indicated below the knockdown name on the left. Shown at the top is the exon structure of MANE select and PTC+ transcripts (thin lines: introns, thick lines: exons). Significance for all boxplots *: padj < 0.05. **: padj < 0.005. n.s = not significant (padj > 0.05).

To determine the effect of depletion of individual spliceosome components on NMD, we focused on endogenous transcripts that contain a termination codon at least 50 nucleotides upstream of an exon-exon junction, a set that we previously defined in Yi et al. [40]. Such transcripts are targeted to NMD due to presence of an EJC downstream of a termination codon, which is thus regarded as a premature termination codon (PTC) [40]. As not all PTC-containing transcripts undergo efficient NMD, we limited our analysis to PTC-containing transcripts (PTC+) that were previously shown to be upregulated upon combined depletion of SMG6 and SMG7 [12]. These PTC+ mRNAs are produced from ~ 2,000 genes and therefore provide a broad measure of NMD activity. As a control, we examined the effect of depletion on the MANE select transcripts produced from the same genes. We found that, like the depletion of known NMD/EJC factors, reduced levels of all 11 catalytic spliceosome components tested led to a higher median fold-change for PTC+ transcripts as compared to their corresponding MANE-select isoforms (Figures 2(A,B) and S2A). Among these, the strongest effect on the PTC+ group was observed for AQR, an NTC-related protein previously reported to aid EJC assembly [18]. Notably, depletion of two other NTC-related proteins, RBM22 and CDC5L, significantly increased PTC+ transcripts, suggesting that NTC-related components beyond AQR may play a role in EJC-dependent NMD (Figures 2(A,B) and S2A). Interestingly, components of the SF3B (SF3B1, SF3B3) and SF3A (SF3A1) subcomplexes of the U2 snRNP also showed a highly significant increase in median fold change for PTC+ transcripts as compared to the control group [(median fold-change for PTC+ versus MANE select transcripts – SF3B1:0.3 vs −0.36; SF3B3:0.231 vs –0.661)] (Figures 2(A,B) and S2A). Notably, a mutation in SF3B1 that causes myelodysplastic syndrome was previously shown to upregulate endogenous NMD targets as well as an NMD reporter [28]. Our results suggest that the link between NMD target abundance and SF3 subcomplexes of U2 snRNPs extends beyond SF3B1 (Figures 2(A,B) and S2A). Of the four U5 components tested, depletion of all but PRPF6 resulted in upregulation of PTC+ transcripts compared to MANE transcripts. Notably, PRPF6 is the only U5 component tested that leaves before spliceosome activation [32], further suggesting that the effect on NMD as a result of spliceosome disruption could be tied to the two catalytic steps of splicing. Of the other U5 components, EFTUD2, which is adjacent to EIF4A3 in the catalytic spliceosome and also engages with CWC22 [19], SNRNP200, and PRPF8 showed a significant effect on the abundance of PTC+ transcripts (Figures 2(A,B) and S2A). Among the factors that leave before spliceosome activation, depletion of only U2AF1, a U2 auxiliary factor, and PRPF4, a U4 component, had a significant effect on PTC+ transcripts, whereas both U1 snRNP components tested (SNRPC and SNRNP70), PRPF3, another U4 subunit, and GNB2L1, a C2 complex protein, had no or only a mild effect on NMD-targeted transcripts (Figure 2(A,B) and S2A). We observed overall very similar trends in the effects of spliceosome component depletion when we limited our analysis to a specific class of PTC+ transcripts, where splicing in (or inclusion) of PTC-containing poison-exons targets these mRNAs to NMD (Figure 2(C) and S2B). Notably, many poison exons are highly conserved and their inclusion is tightly linked to NMD-dependent transcript regulation [5], which can be critical for the overall regulation of these genes and their functions [6]. A visual examination of read coverage of a well-documented specific poison exon, exon 11 of EIF4A2, showed that more reads mapped to the poison exons in the knockdown samples than in the WT cells (Figure 2(D)). Overall, we conclude that upon loss of many spliceosome components, particularly those of the catalytic spliceosome, endogenous NMD-targeted transcripts accumulate at a higher abundance in K562 cells.

Compromised splicing activity upon spliceosome component depletion leads to decreased abundance of canonical isoforms

The depletion of spliceosome components is expected to cause widespread changes in splicing, which in turn can impact multiple steps in gene expression. Hence, the RNA-Seq datasets from spliceosome component depletion encompass changes in mRNA biogenesis (e.g. splicing), mRNA degradation (e.g. NMD), and all intermediate steps. To test whether the increase in the abundance of PTC+ transcripts is due to an effect on NMD or due to preferential generation of these isoforms via splicing, we first quantified global splicing changes in all depletion datasets using rMATS turbo v4.3.0 [41]. As expected, depletion of spliceosome components altered global splicing patterns, resulting in significant changes in annotated splice site usage and production of novel splicing events (Figure 3(A)). Notably, in all knockdowns, the number of novel splicing events observed was greater than the number of annotated splicing events that changed significantly (Figure 3(A)). Furthermore, all the spliceosome depletions tested produced a similar number of novel splicing events, with the exception of AQR, which produced an even greater number of novel splicing changes. Among the annotated events, the depletion of the catalytic spliceosome members, as compared to the early spliceosome components, tended to cause a larger change in significantly altered annotated splice events (compare red circles to green circles in Figure 3(A)). The most common annotated splicing event that changes when a spliceosome component is depleted is exon skipping, accounting for approximately half of the altered annotated splice events in most samples tested (Figure 3(B), top). Interestingly, among the novel splicing events, we observed a dramatic increase in alternative 3’- and 5’-splice-site usage upon spliceosome component depletion (Figure 3(B), compare yellow and green sections of the bars between annotated (top) and novel (bottom) events). This change could have resulted from the increased usage of weaker splice sites by the compromised spliceosome. When the combined effect of altered annotated and novel splicing events was considered on well-expressed genes (at least one transcript with transcripts per million (TPM) count > 5 in wild-type cells), we found that a large fraction of genes (0.21 to 0.55) were subjected to alternative splicing upon spliceosome component knockdown (Figure 3(C)).

Figure 3.

Widespread changes in annotated and novel splicing events upon spliceosome factor knockdowns reduce expression of the affected genes.

A) A dot plot showing the count of splicing events (log10 transformed after normalizing to per million mapped reads) for annotated (circle points) and novel splicing events (triangle points). Points are coloured according to when the components last leave the spliceosome.

B) The proportion of significantly changing annotated (top) and novel splicing events (bottom) that are each splice type: alternate 3’ splice site (yellow), alternate 5’ splice site (light green), mutually exclusive exons (dark green), retained introns (light blue), and skipped exons (dark blue).

C) The proportion of genes with length-scaled TPM > 5 in WT cells that have altered splicing patterns following the indicated KD.

D) The log2(fold change) of genes that are (green) and are not (pink) undergoing altered splicing following spliceosome component knockdown. Comparisons are made at the gene level.

E) The log2(fold change) of genes that are (green) and are not (pink) undergoing altered splicing following spliceosome component knockdown. Comparisons are made at the MANE transcript level. Significance for all boxplots *: padj < 0.05. **: padj < 0.005. n.s = not significant (padj > 0.05).

Normal gene expression is expected to be affected when spliceosome factor depletion causes widespread splicing alterations. Consistently, we observed that upon depletion of a majority of catalytic spliceosome factors, genes that are subjected to alternative splicing (≥1 novel or significantly changing annotated AS event) show an overall downregulation compared to genes with no detectable change in splicing (Figure 3(D) and S3A). In comparison, these effects are milder upon the depletion of early spliceosome factors. We further argued that the impact of altered splicing on gene expression under these conditions will be more apparent at the transcript level, as considerable changes in isoforms produced from a gene can still be masked in gene level estimates. Therefore, we performed transcript-level comparisons focusing on canonical (MANE-select) transcripts. It is conceivable that alternative splicing under such conditions could direct the splicing of pre-mRNA away from the MANE isoform towards a different, potentially novel, isoform, thereby reducing the pool of MANE isoforms from that gene. Consistently, in almost all catalytic spliceosome factor depletion conditions, the MANE select isoforms from genes that show evidence of alternative splicing are reduced in their levels compared to the MANE isoforms from genes where no significant or novel splicing changes were detected (Figure 3(E) and S3B). In contrast, the effects on MANE isoform abundance of AS genes remained mild for early spliceosome component depletion. These results suggest that the depletion of catalytic spliceosome components has a profound effect on mRNA isoform expression, which we further characterize below.

Altered gene expression upon spliceosome component depletion also affects levels of non-NMD-targeted isoforms

The reduction in gene-level and MANE isoform expression upon catalytic spliceosome component depletion suggests that these conditions are also likely to affect the abundance of other isoforms produced from a gene, including the NMD-targeted isoforms. Thus, we first compared the change in levels of the MANE-select isoform versus other non-canonical isoforms for genes that are either subjected to alternative splicing or not under depletion conditions. We found that for genes with no evidence of alternative splicing, the distributions of fold-change values for MANE select and other non-canonical isoforms were comparable (i.e. not significantly different, with a few exceptions among the catalytic spliceosome components) (Fig S4A). However, for the genes that undergo alternative splicing, the fold-change values of the non-canonical isoforms as compared to the canonical MANE select isoforms is significantly higher for all the catalytic spliceosome components tested (Figure 4(A) and S4B). Notably, knockdown of UPF1 or EJC factors and early spliceosome components showed no or only a small difference in fold-change distributions of MANE versus non-canonical isoforms from genes with or without alternative splicing. To separate the effects of alternative splicing and NMD on transcript levels, we attempted to identify genes that produce an NMD-targeted isoform, but did not show evidence for alternative splicing upon spliceosome component depletion. However, we found only a handful of such genes in most knockdowns, and thus could not perform any conclusive analysis.

Figure 4.

Transcript re-quantification after including novel isoforms reveals that all non-canonical isoforms are upregulated following spliceosome component depletion.

A) Boxplots of the log2(fold change) of the MANE (teal) and non-canonical isoforms (dark green) of genes that undergo significant alternative splicing following depletion of the indicated proteins.

B) Comparison of log2 (fold change) of MANE (purple), NMD-targeted (orange), and stable non-canonical isoforms (green) following spliceosome component knockdown. The fold-changes were recalculated using kallisto and DESeq2 after including novel isoforms in the reference transcriptome. Median and number of observations in each group is noted as in Figure 2. P-value is the result of a Wilcoxon test comparing the MANE and stable non-canonical isoforms to the PTC+ isoforms, with the alternative hypothesis being that the PTC+ isoforms will be more abundant. Significance for all boxplots *: padj < 0.05. **: padj < 0.005. n.s = not significant (padj > 0.05).

It is possible that our ability to accurately quantify the levels of annotated transcripts upon spliceosome component knockdown is affected by the generation of novel misspliced transcripts. These mis-spliced transcripts are not included in the reference sequences used by kallisto for transcript quantification in our workflow and can affect the assignment of sequencing reads to annotated transcripts, thereby altering transcript quantification. To address this issue, we used Stringtie to identify novel transcripts and include them in the reference list for kallisto-based transcript quantification (Fig S1). We found that after accounting for novel transcripts produced upon spliceosome disruption, both non-canonical and NMD isoforms showed an upregulation as compared to MANE-select isoforms in several catalytic spliceosome component depletion datasets (Figure 4(B) and S6A). Notably, for a subset of spliceosome component knockdowns (RBM22, EFTUD2, SNRNP200), there was no or only a minor difference in the median fold-change values for PTC+ versus other non-canonical transcripts. Thus, we conclude that for this set of spliceosome component knockdowns, a simple comparison of the levels of canonical MANE transcripts versus non-canonical transcripts is not sufficient to distinguish the effects of altered splicing from the impact of compromised NMD on the increased abundance of PTC+ transcripts. However, in the case of AQR, SF3B1, SF3B3, and CDC40 depletion, the median fold changes of other non-canonical isoforms were significantly lower than those of the PTC+ isoforms (Figure 4(B)). These conditions are at least somewhat comparable to UPF1, EIF4A3, and MAGOH knockdowns, where the levels of canonical and non-canonical isoforms are comparable and significantly lower than those of the PTC+ isoforms. Therefore, AQR, SF3B1, SF3B3, and CDC40 may have a more direct effect on the suppression of PTC+ transcripts by NMD. However, the increased abundance of non-canonical isoforms compared to canonical MANE isoforms in these knockdowns suggests that altered splicing could also contribute to the increased levels of PTC+ transcripts.

Deficiency of some spliceosome components may dampen EJC-dependent NMD

Although our analysis shows that the increased abundance of NMD targets upon depletion of many catalytic spliceosome components cannot be completely attributed to disruption of NMD, the clear increase in the levels of PTC+ transcripts under these conditions warrants an investigation of the possible underlying causes. One hypothesis is that spliceosome knockdowns hamper NMD by impacting EJC deposition. The four factors above that appear to specifically influence PTC+ transcripts are more likely to exert such an effect. A prediction of this hypothesis is that a reduction in EJC deposition will more profoundly affect NMD of PTC+ mRNAs with fewer downstream EJCs whereas NMD substrates with multiple exon junctions in 3’UTRs will be more refractory to reduction in EJC deposition. To test this prediction, we binned the PTC+ mRNAs by number of introns in their 3’UTRs and assessed fold-change of these binned transcripts in UPF1, EIF4A3, or a subset of spliceosome factor knockdown datasets. As predicted, in the EIF4A3 knockdown dataset, the highest upregulation is observed for transcripts with only one 3’UTR intron whereas the degree of upregulation decreases with more 3’UTR introns (Figure 5(A)). The more introns there are in the 3’UTR, the more chances the EJC will still be able to bind to the transcript when EIF4A3 amounts are lowered. A similar pattern of reduced upregulation of PTC+ transcripts with increasing number of 3’UTR introns is observed in AQR, CDC40 and to a lesser degree in SF3B3 knockdowns (Figure 5(A)). In these knockdowns, four or more 3’UTR introns lead to a significantly lower upregulation of PTC+ transcripts than those with only one 3’UTR intron. No 3’UTR intron count-dependent change in PTC+ transcript levels is observed upon UPF1 or SF3B1 knockdown whereas PTC+ transcripts show no significant change as compared to MANE transcripts upon SNRNPC depletion. These findings suggest that a reduction in levels of AQR, CDC40 and, to a lesser extent, SF3B3 may lower EJC deposition and thereby directly impact NMD. Notably, these spliceosome proteins do not affect the levels of transcripts that are regulated by NMD independently of 3’UTR introns and hence are EJC-independent (Figure 5(B)). Thus, the effects of AQR, CDC40 and SF3B3 appear to be specific to EJC-dependent NMD.

Figure 5.

EJC dependence of gene expression changes upon spliceosome component depletion.

A) Boxplots of the log2(fold change) of the MANE (purple) and NMD-targeted PTC+ isoforms binned by number of 3’UTR introns (shades of orange, as indicated in the legend on the right) in datasets where proteins indicated at the bottom were depleted.

B) Boxplots of the log2(fold change) of the transcripts without any 3’UTR introns classified into NMD sensitive (red, upregulated > 1.5-fold in two out of three UPF factor depletion datasets from [40]) or insensitive (teal, all other transcripts) as indicated in the legend on the right. Fold changes are from datasets where proteins indicated on the x-axis were depleted. Significance for all boxplots in A and B *: padj < 0.05. **: padj < 0.005. n.s = not significant (padj > 0.05).

C) Bar graphs depicting mean fold changes in transcript levels for MANE (unshaded bars) or PTC+ (filled bars) transcripts from genes indicated at the bottom upon spliceosome factor knockdown relative to control knockdown (siNC) across four biological replicates. Knockdown conditions are indicated in the legend on the right. AQR and EFTUD2 knockdowns were for 65 hours, while siSF3B1 and siSF3B3 knockdowns included two replicates at 65 hours and two at 48 hours that were combined as the trends were consistent across both time points. Transcript levels were normalized to GAPDH, and data are presented as mean ± standard error of means. Statistical significance versus the control group (siNC) was determined using two-tailed unpaired Student’s t-test and is indicated by asterisks (*p < 0.05). The dashed horizontal line at y = 1 denotes baseline expression in control samples.

We chose a subset of PTC+ transcripts and their MANE counterparts to validate their altered expression upon knockdown of AQR, EFTUD2, SF3B1 and SF3B3 using small interfering RNAs in HCT116 cells (Fig S5A), a colorectal cancer cell line we have used previously for investigating the human NMD pathway. In this quantitative RT-PCR-based analysis, we included EFTUD2 knockdown as an additional condition because this U5 snRNP component is adjacent to EIF4A3 in the catalytic spliceosome and could influence EJC deposition. The four PTC+ transcripts chosen for this analysis, but not the MANE isoforms from the corresponding genes, are robustly upregulated upon 24-hour treatment of the cells with a chemical inhibitor of NMD, SMG1i [42] (Fig S5B). All four PTC+ mRNAs are upregulated upon AQR knockdown although SRSF3 PTC+ transcript shows only a modest increase (Figure 5(C)). For the other three knockdowns, only 2/4 tested NMD substrates show upregulation. While these results support the role of spliceosome proteins in EJC dependent NMD, we noticed that some of these knockdowns also reduced the levels of the MANE transcripts from the tested genes with the strongest downregulation observed in the case of FTH1 MANE transcript. The NMD isoform of RSRC2 also shows a dramatic decrease under SF3B1 and SF3B3 knockdown conditions. These results are consistent with our bioinformatic analyses above in Figures 3 and 4, and further emphasize that compromised activity of the catalytic spliceosome can skew transcript abundance. An RT-PCR-based analysis further confirmed the varied effects of spliceosome factor knockdown on transcript expression from different genes (Fig S5C), with some genes showing minimal changes (e.g. FTH1) while others generating novel, unannotated isoforms (e.g. COX7A2). Overall, these experimental validations further underscore the direct as well as indirect effects of spliceosome inhibition on NMD.

Depletion of some catalytic spliceosome components leads to production of novel NMD targeted isoforms in excess of the endogenous NMD targets

We considered additional hypotheses to investigate other possible causes of reduced NMD activity upon spliceosome disruption. Other groups have previously speculated that disruption of the spliceosome may result in an overabundance of NMD targets, much more than the pathway can handle, thereby overwhelming the NMD machinery and lowering its ability to suppress natural NMD-targeted transcripts [43,44]. To test this possibility, we compared the overall levels of annotated and novel transcripts targeted to NMD in the spliceosome component versus the control knockdown conditions. It is expected that in conditions where NMD is overwhelmed, the concentration of novel NMD targets would surpass that of natural NMD targets. We used the Isoform Switch Analyzer algorithm to classify the novel transcripts in the knockdown samples as NMD targets if they contained a PTC more than 50 nt upstream of an exon–exon junction; novel transcripts without a PTC or with a PTC in the last exon were classified as stable transcripts [45]. To compare the relative concentrations of annotated and novel transcripts in each sample, we summed the TPM values of all transcripts in each of the following groups: canonical MANE, annotated NMD, novel NMD, and novel stable. As expected, MANE isoforms had the highest cumulative TPM estimates in all samples, whereas the amounts of annotated NMD isoforms were only fractional (Figure 6(A), left). Intriguingly, the cumulative amounts of novel NMD transcripts were higher than those of novel stable transcripts in the case of AQR, SF3B1, SF3B3, and CDC40 knockdown (Figure 6(A), right). Curiously, in these four depletion conditions, novel NMD transcripts were almost two-fold or more abundant than the annotated NMD transcripts (Figure 6(A), right inset and S6B). It is noteworthy that these four conditions also showed a significant increase in PTC+ transcripts compared to other non-canonical transcripts and MANE transcripts (Figures 4(B) and 5(A,C)). Indeed, there was a strong correlation between how significantly the knockdown upregulated the PTC+ targets and the ratio of the total abundance of the novel versus annotated NMD targets (Figure 6(B)). While these results hint at the possibility that excessive production of novel NMD targets could interfere with the suppression of endogenous NMD substrates, it remains to be seen whether a mere doubling of substrate concentration, as is observed in some spliceosome depletion conditions, would be sufficient to suppress the pathway.

Figure 6.

Effect of spliceosome component depletion on relative levels of novel and annotated NMD targeted transcripts and of NMD factor mRNAs.

A) Left: Cumulative length scaled TPM of MANE (purple) and NMD (orange) transcripts from genes that produce annotated NMD-targeted isoforms in the indicated samples. Right: Cumulative TPMs of stable (blue) and predicted NMD-targeted (red) novel isoforms produced from all genes in the indicated samples. Inset: the cumulative length scaled TPM of annotated NMD-targeted (orange) and novel NMD-targeted transcripts (red), from the left and right plots, respectively, are re-plotted for comparison. In all cases, TPM for each transcript was averaged across replicates before summation.

B) A scatterplot comparing the ratio of cumulative TPMs of novel:annotated NMD-targeted transcripts (y-axis) to the -log10(p-value) of the upregulation of annotated NMD-targeted transcripts when compared to their MANE counterparts. Dashed line is the linear regression fit with Pearson’s R and p-value shown in upper left corner.

C) A heatmap clustered along the x-axis showing the log2(fold change) of the MANE isoform of NMD factor genes (x-axis) in spliceosome component or NMD/EJC factor depletion datasets (y-axis). Column labelled ‘stage’ on the left indicates the stage where a spliceosome component leaves the spliceosome as indicated in the legend on the right. Changes less than 1.5-fold in either direction are white, upregulated transcripts are teal, and downregulated transcripts are purple.

Levels of some NMD factor mRNAs are reduced upon spliceosome disruption

Another possible explanation for the increase in NMD-targeted transcripts upon spliceosome component disruption would be a decrease in the abundance of NMD proteins themselves due to altered splicing of their mRNAs, thereby causing partial inhibition of the pathway. To test this possibility, we examined changes in the abundance of MANE isoforms in a set of genes that contribute to the NMD pathway [46]. As shown in Figure 6C, although MANE isoforms for many NMD factors did not change dramatically upon spliceosome knockdown, there were small clusters of NMD factors whose MANE isoforms were down- or upregulated in these datasets. The most prominently downregulated set contained two UPF3 paralogs, UPF3A and UPF3B, which were downregulated > 2-fold in multiple catalytic spliceosome component knockdowns tested (4/10 for UPF3A and 6/10 for UPF3B). Interestingly, this is the same group of spliceosome components, whose knockdown leads to an increase in the abundance of NMD-targeted transcripts (Figure 2). Similarly downregulated in 6/10 catalytic spliceosome knockdowns (or 8/18 spliceosome factor knockdowns) is the mRNA encoding MOV10. As UPF3A and UPF3B activate both EJC-dependent and EJC-independent NMD [40,47] and MOV10 has been suggested to assist UPF1-dependent steps of NMD on some targets [48], downregulation of protein-coding mRNAs of these NMD factors could contribute to reduced NMD upon spliceosome disruption. Interestingly, the MANE isoform of SMG8, a regulator of SMG1 kinase, showed a > 2-fold increase in abundance after 4/10 catalytic spliceosome factor depletion. It is notable that while many NMD factor-encoding mRNAs are known to be autoregulated by NMD [49,50], we did not observe an upregulation of transcripts encoding NMD factors, with the exception of SMG8 and a few other isolated cases. Even upon strong NMD inhibition upon dual depletion of SMG6 and SMG7 proteins in HEK293 cells, only SMG1 and UPF2 encoding MANE isoforms showed > 1.5-fold upregulation. Overall, we conclude that NMD inhibition in K562 cells depleted of spliceosome components could partially result from reduced levels of key NMD activators, such as UPF3 paralogs and MOV10.

Spliceosome inhibitor treatment also leads to increased relative abundance of PTC+ and other non-canonical isoforms

The increased abundance of NMD (and in some cases other non-canonical) isoforms in a wide range of spliceosome depletion conditions raises the question of whether the altered abundance is due to a general effect of spliceosome inhibition rather than a compromised NMD-specific function of an individual factor. We argued that this idea can be tested by examining changes in the levels of PTC+ and canonical and non-canonical transcripts in human cells treated with spliceosome inhibitors. We performed isoform-level quantification of the RNA-Seq data from Naro et al. [50], in which the prostate cancer 22Rv1 cell line was treated with pladienolide B (an SF3B1 inhibitor [51]), indisulam (targets RBM39 for proteasomal degradation [52]), and THZ531 (CDK12/13 inhibitor [53]). Interestingly, similar to the trends observed with several catalytic spliceosome factor depletion, including that of SF3B1, pladienolide B treatment led to a significant increase in the relative abundance of PTC+ and other non-canonical isoforms compared to the levels of MANE transcripts from the corresponding genes (Figure 7A). Similar trends were observed for indisulam and THZ531, but the median fold changes for PTC+ and non-canonical isoforms were modest. Notably, RBM39 is among the top 200 targets in one of the four NMD factor genetic screens [26] and its indisulam-mediated degradation likely resembles the knockdown of individual spliceosome components. We also compared the levels of MANE, PTC+, and other non-canonical transcripts in lymphoblastoid cells treated with high doses of risdiplam [54], which does not inhibit the spliceosome but leads to widespread alterations in pre-mRNA splicing by promoting weak splice-site usage [55]. We found that risdiplam treatment also caused an upregulation of NMD-targeted transcripts as compared to MANE transcripts, although the effect on other non-canonical isoforms was only modest (Figure 7(A)). Notably, treatment with these spliceosome inhibitors and drugs that alter spliceosome activity also affected the expression of MANE transcripts of various spliceosome components (Figure 7B, left) and NMD factors (Figure 7B, right). In particular, pladienolide B treatment, which showed the highest increase in the fold-change of PTC+ and non-canonical isoforms, also exhibited the strongest downregulation of mRNAs encoding numerous splicing and NMD factors tested (Figure 7B). These observations further indicate that upregulation of NMD transcripts upon spliceosome inhibition, either via chemical inhibitors or individual factor depletion, is likely due to multiple contributing factors that may include reduction in levels of NMD and splicing factors, overproduction of novel NMD substrates, and/or dramatically altered gene expression.

Figure 7.

Effects of chemical inhibitors and modulators of spliceosome activity on NMD and other transcripts.

A) Boxplots showing log2 (fold change) of MANE (purple), NMD-targeted (orange), and stable non-canonical isoforms (green) following treatment with splice altering drugs indicated on the x-axis. Medians, numbers of observations and p-values of comparisons shown are as in Figure 4. Significance values *: padj < 0.05; **: padj < 0.005; n.s = not significant (padj > 0.05)

B) Left: A heatmap of the MANE isoform of spliceosome components under investigation (x-axis) following treatment with splice altering drugs (y-axis). Heatmap colours are as in Figure 5. Right: A heatmap of log2 (fold change) of MANE isoforms of NMD factors (x-axis) following treatment of splice altering drugs (y-axis). Colours are as in Figure 5.

Altered gene expression due to disease causing mutations in spliceosome components also includes increased abundance of NMD targeted isoforms

Based on our findings, we predicted that spliceosome mutations that cause human disorders will result in an increased abundance of NMD targets, in addition to altering pre-mRNA splicing. We investigated this possibility by examining available RNA-Seq datasets from retinitis pigmentosa (RP) patient-derived fibroblasts with a mutation in PRPF8 and induced pluripotent stem cells (iPSCs) made from patient-derived fibroblasts with a mutation in PRPF31. The PRPF8 deficient fibroblasts were derived from a patient with a deletion (c6974-6994del) that disrupts the region required for interaction with EFTUD2 and SNRNP200 [56] whereas iPSCs with PRPF31 deficiency were derived from a patient with very severe RP caused by deletion (c1115-1125del11) that causes a frameshift leading to a truncated protein [57]. When compared with cells derived from the control normal individuals, cells with either PRPF8 or PRPF31 mutations showed an upregulation of PTC+ (Figure 8(A)) or poison exon-containing transcripts (Figure 8(B)) compared to their MANE counterparts. In the case of PRPF31 mutant cells, the upregulation of PTC+ appears to be somewhat more specific and significantly higher than that of the non-canonical isoforms. Notably, the effect of PRPF8 mutations was similar to that observe in PRPF8 knockout K562 cells from ENCODE (Fig S2A). As a further confirmation of the increased abundance of NMD targets in these cells, read distribution across EIF4A2 gene locus showed increased inclusion of an NMD-targeting poison exon in the cells with mutant PRPF31 mutant cells, although in cells with mutant PRPF8 there was little difference from the wild-type sample (Figure 8(C)).

Figure 8.

Effects of retinitis pigmentosa causing mutations in PRPF8 and PRPF31 on NMD targeted and other isoforms in patient-derived cells.

A) Boxplots showing log2 (fold change) distributions of MANE (purple), NMD-targeted (orange), and stable non-canonical isoforms (green) in cells with retinitis pigmentosa causing mutations in PRPF31 or PRPF8 (x-axis). Medians, numbers of observations and p-values of comparisons shown are as in Figure 4.

B) Boxplots as in A showing log2(fold change) distributions of MANE (blue) and NMD-targeted isoforms (red) of genes containing poison exons in cells with retinitis pigmentosa causing mutations in PRPF31 or PRPF8.

C) RNA-Seq read distribution on EIF4A2 gene locus in representative replicates of cells derived from retinitis pigmentosa patients (purple) or normal individuals (green). The poison exon in EIF4A2 is indicated by the dashed box. The scale of each pair of tracks is indicated under the sample name. Significance for all boxplots *: padj < 0.05. **: padj < 0.005. n.s = not significant (padj > 0.05)

Similar to several catalytic spliceosome component knockdowns, iPSCs with the PRPF31 mutation showed a reduced abundance of MANE isoforms from genes that undergo alternative splicing (Fig S7). However, the same pattern was not observed in fibroblasts from patients with mutations in PRPF8. Thus, PRPF31 and PRPF8 mutations could also alter the expression of key NMD-regulated genes that may contribute to disease phenotypes. One example is NRG1, which plays a role in motor and sensory neuron development. In RP patient-derived PRPF8 mutant cells, the NMD-targeted isoform of NRG1 has a log2 fold change of 1.34 (padj = 0.001, Table S5) while the MANE isoform only changes mildly. Although RP is not a craniofacial disorder, NMD-targeted isoforms of genes that are involved in craniofacial development are also upregulated in PRPF31 mutant cells, hinting at shared gene expression mechanisms underlying different types of spliceosomopathies. One example is PDLIM7, which encodes a scaffold protein that localizes LIM-binding proteins to actin filaments and is involved in bone formation, including flat bones in the mandible and cranium [58]. In RP patient-derived PRPF31 mutant cells, a poison-exon-containing NMD isoform of PDLIM7 was upregulated (log2FC 1.14, padj = 0.04; Table S5), whereas its protein-coding MANE isoform was mildly downregulated. Notably, similar changes in PDLIM7 PTC+ and MANE isoforms were observed in PRPF8 mutant cells. Other NMD-regulated isoforms upregulated in PRPF8 and PRPF31 mutant cells include well-known NMD targets, such as SRSF2 and SRSF3 as well as isoforms of genes with functions that may contribute to disease progression [for example, OSTM1 (regulates chloride channels in osteoclasts and melanocytes)]. These observations indicate that even if the effect of spliceosome factor mutations on NMD target abundance is pleiotropic in nature, disruption of some key NMD-regulated activities may contribute to disease progression in spliceosomopathies.

Discussion

Spliceosome disruption increases abundance of NMD substrates

The identification of a surprisingly large number of spliceosome components in several genetic screens for NMD factors in human cell lines indicates that our understanding of the effect of spliceosome function on NMD remains incomplete. In particular, proteins present in the spliceosome when performing the two catalytic steps of splicing appear to have a greater influence on NMD (Figure 1). Motivated by these observations, our further investigation revealed that when spliceosome function is disrupted due to the depletion of one of its many core components, transcripts normally suppressed by the NMD pathway are upregulated (Figure 2). Our analysis confirmed the broad impact of the spliceosome core on NMD, as the upregulation of PTC+ transcripts was observed in 15/18 spliceosome proteins tested. Moreover, an increased abundance of NMD-targeted isoforms was also observed in cells treated with spliceosome inhibitors (Figure 7) and in the transcriptomes of cells derived from human patients with spliceosome component mutations (Figure 8). As expected, depletion of spliceosome core proteins led to widespread changes in pre-mRNA splicing (Figure 3), which altered overall gene expression (Figures 3 and 4). Even after we tuned transcript quantification to account for these alterations in the transcriptome, depletion of the spliceosome components showed an increased abundance of EJC-dependent NMD targets as a group as compared to the canonical transcripts (Figure 4B and S6A), indicating that NMD-dependent gene regulation is compromised when the spliceosome is dysfunctional.

The effect on NMD substrate abundance is more pronounced upon reduction of catalytic spliceosome components compared to U1 and U4 components that function during early steps. (Figure 2, S2 and S6A). Depletion of all 11 catalytic spliceosome components caused a highly significant upregulation of PTC-containing transcripts, whereas in the case of early components, either the effect on NMD substrates was insignificant (SNRNPC and PRPF6) or only moderately significant (SNRNP70 and PRPF3) (Figure 2 and S2). Among the factors that are not part of the catalytic spliceosome, the depletion of only U2AF1 and PRPF4 has a strong effect on NMD targets. A more consequential effect of catalytic spliceosome components on NMD could be the compromised EJC assembly (see below). We noticed that, compared to the catalytic components, early spliceosome factor depletion affected fewer annotated splicing events (Figure 3(A)) and had a weaker effect on the levels of NMD factor mRNAs (Figure 6(C)). Intriguingly, a recent study showed that in HEK293 cell lines, depletion of U1 components resulted in more splicing changes than that of catalytic spliceosome components [59]. One possibility is that the increase in altered splicing events due to early spliceosome component depletion could hinder K562 cell survival. If this were the case, the early spliceosome components would be less amenable to acute depletion experiments, possibly explaining their under-representation in the ENCODE database and in screens for NMD factors (Figure 1). If the depletion of many early spliceosome components is lethal to cells, it stands to reason that the only early components that survive acute protein depletion are those with a milder effect on gene expression, including NMD.

Possible causes of NMD inhibition upon spliceosome disruption

NMD defects upon spliceosome disruption could result from either direct interference of the pathway or indirect effects. The direct effects could result from compromised EJC assembly, which is initiated upon recruitment of EIF4A3 by the CWC22/CWC27 heterodimer to the B^act spliceosomal complex [14–17,60]. EIF4A3 and/or the assembled pre-EJC bound to a contiguous stretch of 6 nt in the 5’-exon was observed in the C complex [61,62]. While CWC22 and CWC27 depletion datasets were not available in the ENCODE database to test their effects on global NMD, knockdown of CWC22 in HeLa cells has been previously shown to upregulate the levels of NMD targets [14]. Thus, the reduced ability to recruit and deposit EJC proteins on exonic RNA within the spliceosome can directly affect NMD. Another candidate for EIF4A3 recruitment is EFTUD2, which is adjacent to EIF4A3 in the C complex where its C-terminus engages with the RecA domain of EIF4A3 [61]. Intriguingly, similar to EIF4A3, mutations in EFTUD2 cause a disorder characterized by craniofacial defects and intellectual disabilities [22,23,63,64]. Although EFTUD2 knockdown upregulated PTC+ transcripts (Figures 4B and 5C), similar effects were observed for non-NMD targets (Figure 4B). Thus, in this case, we cannot completely separate the direct effects of EFTUD2 on EJC/NMD from the possible indirect effects (see below).

We identified four components of the catalytic spliceosome that had a specific effect on PTC+ isoforms (Figures 2 and 4B). First among these, AQR has been shown to contribute to EJC deposition and NMD [18]. Indeed, we observed that AQR depletion in K562 cells caused strong and specific upregulation of PTC+ isoforms compared to both MANE and other non-canonical isoforms (Figure 2 and 4B). A similar effect on NMD-targeted isoforms was observed in the case of the two components of the SF3B complex of the U2 snRNP, SF3B1, and SF3B3 (Figures 2 and 4B). The specific effect of AQR and, to a lesser degree, of SF3B1 and SF3B3 on PTC+ transcripts in HCT116 cells (Figure 5C), underscores the likely direct impact of these factors on NMD. The higher sensitivity of transcripts with one (or fewer) downstream EJCs to the levels of these spliceosome components (Figure 5A) suggests that their direct impact on NMD could potentially stem from a role in EJC deposition, an important avenue for future investigations.

Notably, Cheruiyot et al. recently showed that SF3B1 and U2AF1 variants carrying myelodysplastic syndrome-causing mutations cause upregulation of NMD-targeted endogenous and reporter mRNAs in K562 cells [28]. The effects of SF3B1 inhibition on NMD are also supported by the upregulation of NMD substrates by SF3B1 targeting the spliceosome inhibitor pladienolide B (Figure 7). It is possible that SF3 complexes within the U2 snRNP also play a role in determining the potential of a spliced RNA to undergo NMD, as several other components of SF3 complexes are strongly enriched in NMD factor screens (Table S3). Our results from SF3B3 knockdown K562 cells support this hypothesis (Figures 2 and 4B). Interestingly, in the catalytic spliceosome SF3A and SF3B complexes, AQR binds adjacent to the intron in a region close to the branch point. Moreover, the intron-binding complex nucleated by AQR can be chemically cross-linked with SF3A and SF3B proteins [65]. Another protein that has a specific effect on PTC+ isoforms is CDC40 (also known as Prp17), a step II factor (Figures 2 and 4B) [33]. CDC40 interacts with multiple proteins and RNA components within the catalytic spliceosome, including the U2-branch site helix, U6 snRNA, and U5 proteins. Through these interactions, it plays a crucial role in stabilizing the second-step conformation of the spliceosome [61,62]. A hypothesis that emerges from these observations is that protein components that properly position the intron within the catalytic spliceosome also directly impact the recruitment/deposition of EJC subunits, thereby reducing the potential of spliced mRNA to be regulated by NMD.

In addition to the possible direct effects on NMD through EJC deposition, the disruption of spliceosome function by reduced levels or mutations in its core components also causes pleiotropic effects that contribute to impaired NMD. One indirect effect could be the misregulation of genes encoding NMD factors, such as the two UPF3 paralogs, which enhance both EJC-dependent and EJC-independent NMD [66,67]. Although mRNAs encoding UPF3 factors are particularly sensitive to spliceosome disruption (Figure 6C), it remains to be seen whether their protein levels are also reduced. We also examined a previously proposed hypothesis that overproduction of novel NMD substrates due to missplicing could indirectly affect the ability of the pathway to regulate its normal targets [66,67]. Interestingly, in cells depleted of the four factors that produce a significant and specific effect on NMD-targeted isoforms, we observed that the overall concentration of novel NMD substrates was at least 2-fold higher than that of annotated NMD targets (Figure 6(A)). In the various knockdowns tested, we observed a significant direct correlation between the relative amounts of novel NMD transcripts detected and the significance of NMD target upregulation in each knockdown (Figure 6(B)). Thus, the amount of novel NMD targets generated could influence the degree of NMD inhibition. It remains to be seen if a mere two-fold increase in the NMD target concentration is sufficient to overburden the pathway. Such an outcome seems less likely, considering a previous report that NMD activity is stable across tissues [68], where the concentration of NMD substrates is expected to be highly variable. Indirect effects on NMD targets could also stem from departure from normal splicing patterns and a consequential increase in the production of non-canonical transcripts, including NMD isoforms. This possibility is supported by the reduced abundance of canonical isoforms and increased levels of non-canonical transcripts that are not targeted by NMD but are transcribed from the same set of genes in several knockdowns (Figure 4(B)), spliceosome inhibitor-treated cells (Figure 7), and RP patient-derived cell lines (Figure 8). Experimental strategies that can differentiate RNA production and degradation rates are necessary to parse out contributions of such indirect effects on NMD target abundance. Finally, due to cross-regulation between the splicing machinery [59], inhibition of splicing also alters spliceosome factor abundance to further compound these indirect effects. Most likely, the increased abundance of endogenous NMD targets observed in the analysis presented here and of NMD reporter RNAs in recent genetic screens [26–29] resulted from a combination of direct and indirect effects. Direct effects may be limited to a smaller set of spliceosome components, whereas the reduced/lost functions of most spliceosome factors are expected to result in at least some pleiotropic effects. Regardless of the exact mechanism, our observations support the conclusion that compromised NMD-based regulation is another hallmark of cells with impaired spliceosomal function.

Consequences of NMD substrate misregulation in spliceosomopathies

Many genes that are important for developmental processes and cell differentiation pathways rely on NMD for regulation [69–71]. A prominent group comprises genes containing poison exons that are enriched in developmental functions [6]. Our results show that NMD-targeted isoforms generated from this set of genes are upregulated upon spliceosome disruption, including in cells derived from two patients with retinitis pigmentosa mutations in PRPF8 and PRPF31 (Figure 8). Among these, developmentally important genes are related to neuronal growth (e.g. NRG1, Table S5). Interestingly, in these RP datasets, we also observed the upregulation of some genes with functions in craniofacial development (e.g. PDLIM7 and OSTM1, Table S5). These findings, in conjunction with the increased abundance of NMD-targeted mRNAs upon depletion of other spliceosome components that cause human disorders when mutated (e.g. EFTUD2, SNRNP200, PRPF4, PRP17, and SF3B1), suggest that impaired NMD is likely to be a contributing factor in these disorders. In conclusion, although the exact cause of NMD target upregulation upon spliceosome disruption remains to be determined, we recommend that future investigations into spliceosompathies should consider NMD disruption as a likely contributor to molecular phenotypes.

Methods

Identification of splicing factors in NMD screens

Extended data tables of hits in NMD screens were downloaded from previous studies [26–29]. To identify high-confidence hits, we restricted each list to the top 200 factors based on rankings in the original studies. Using custom R scripts, we determined the overlap between all the four lists. A complete list of human spliceosome components was downloaded from the spliceosome database [32], and a list of potential NMD factors was annotated based on the presence of a gene in the spliceosome database. The last complex of the spliceosome a protein is associated with was determined by selecting the last formed complex listed for that protein in the spliceosome database. STRING and GO biological process analyses, with filtering for redundant terms set at 0.75, were conducted using Cytoscape version 3.10.1 [72].

RNA-Seq datasets and their processing

RNA-seq datasets from siRNA knockdown or CRIPSR-Cas-mediated knockout of spliceosome components identified in two or more of the NMD screens, as well as other spliceosome components of the U1, U2, U4, U5, and U6 components, were identified from the ENCODE consortium’s ENCORE project [34–36] and retrieved from the SRA archive (Table S1). Reads were trimmed using trimmomatic version 0.36 [73] to remove adaptors, bases with a quality of less than 3, and reads shorter than 30nt. Further quality control and analyses performed on these datasets are shown in Figure S1.

Generation of NMD target lists

Transcripts containing predicted PTC have been described previously [40]. To obtain a list of NMD targets, we analysed the SMG6 knockdown and SMG7 combined knockout from Boehm et al. and identified transcripts that were upregulated by more than 1.5-fold upregulated [12]. Any PTC-containing transcripts that were upregulated by > 1.5-fold in SMG6 and SMG7 double depletion were considered NMD-targeted PTC+ transcripts. BioMart was used to retrieve isoform characteristics from the Ensembl GRch38 build 109 of the human genome, including MANE select transcript status and transcript biotype from each isoform on the PTC+ list [74]. Genes where the MANE selected transcripts were on either the PTC-transcript or the NMD-transcript list were removed from the dataset. MANE transcripts from this set of genes were used as the MANE isoform group and non-MANE non-NMD biotype isoforms were used as the stable non-NMD transcript group. The number of 3’UTR introns were determined by using a custom awk script to count the number of 3’UTR entries for each transcript in the Ensembl build 109 genome annotation.

The poison exon target list was created by identifying previously reported genes that contain conserved poison exons that introduce PTC [5]. All transcripts from these genes were obtained using BioMart. Transcripts annotated with the nonsense-mediated decay biotype were included as NMD targets, and MANE-select transcripts were included as non-NMD targets. Genes with a MANE select transcripts with an NMD biotype or on the NMD target list were excluded.

To create the list of NMD targets with no 3’ UTR introns, any transcripts with known introns (as defined by Kovalak et al, [4]) were removed. The remaining were classified as NMD targets if the transcript was ≥ 1.5-fold upregulated in at least two of the following conditions we previously tested: UPF1 KD, UPF2 KD, or UPF3A and UPF3B KD [40]. The remaining set of transcripts without any 3’UTR introns were used as ‘non-NMD’ control set.

Differential expression analysis

Reads were mapped and quantified using kallisto version 0.43.1, based on an index generated from cDNA from Ensembl GRch38 build 109 [74]. Kallisto was run on paired-end reads using default settings [37]. Transcript quantification was imported to R using tximport, which calculates the length-scaled TPM. Transcripts with a mean length-scaled TPM of less than 1 in wild-type or knockdown samples were filtered out. Principal component analysis (PCA) was conducted in R using the DEseq2 package on the raw counts with default settings, comparing the WT and depletion conditions. ENCODE datasets without clear segregation between the treatment conditions were removed from the study (Figure S1). Differential expression analysis was conducted by comparing KD to WT transcript levels with DESeq2, using default settings [38]. ENCODE datasets in which the MANE isoform of the spliceosome component was not more than 1.5-fold downregulated were also removed from the study (Figure S1). Gene level differential expression analysis was conducted as described above; however, the tx2gene option was specified, and a table of transcripts and corresponding gene IDs was provided to convert transcript level counts to gene level.

The results of differential expression analysis were annotated based on the characteristics under investigation for each plot, and the log2fold change of transcripts on those lists was plotted using ggplot2 [75]. The Wilcoxon test was used to determine the statistical significance between the two groups. Adjusted p-values were calculated for each figure using the Bonferroni method. Adjusted p-values are listed in Table S6.

Splicing analysis

Splicing analysis was performed with rMATS turbo version 4.2.0 [41], using a binary index and GTF from the ensemble GRch38 build 109. A read length of 100 was specified for datasets from ENCODE, whereas the average read length from all samples was used for the other datasets. The variable-read-length and novelSS flags were used to identify novel splicing events created following spliceosome knockdown. Splicing changes were calculated by read mapping to exon–exon junctions and exons (labelled by rMATS as JCEC). Splicing changes were considered significant if padj < 0.05. Genes were classified as undergoing alternate splicing in the dataset if there were one or more significant splicing changes or novel splicing events. The total number of significant or novel splicing events was normalized according to the average number of reads mapped by HISAT2 across all samples from an experiment. The results of the splicing analysis were used to annotate the output of the differential expression analysis to compare genes that did or did not experience altered splicing.

Novel isoform analysis

Following IsoformSwitchAnalyzeR documentation, novel isoforms were identified using HISAT2 v2.1.0 for mapping, and Stringtie v1.3.3b for isoform quantification [76,77]. HISAT2 uses an index with annotated splice sites and exons built from Ensembl GRch38 build 109 and was run using the -dta options. Following HISAT2, the reads were aligned and merged by StringTie using a reference annotation from Ensembl GRch38 build 109. Finally, the reads were quantified using StringTie with the -e option specified, using the merged transcriptome as the reference annotation. Transcripts and quantifications were imported into R and analysed via IsoformSwitchAnalyzeR v2.2.0 [45]. During the importRdata step, the merged transcriptome from StringTie was used for exon annotation, and sequences from the same file extracted via gffread v0.12.7 [78] were used as the fasta file. Default settings were used for pre-filtering and switch testing using DEXSeq. Open reading frame (ORF) analysis was added using the same GTF file used for StringTie, and a novel ORF analysis was conducted using default settings. The consequence of isoform switching was determined based on the NMD status. Transcripts were classified as NMD targets by IsoformSwitchAnalyzeR, using the programme’s default settings based on the 50nt rule.

The sequences of novel isoforms were extracted with IsoformSwitchAnalyzeR and used along with the transcriptome from Ensembl GRch38 build 109 to create a kallisto index for each dataset. Kallisto quantification was conducted as described above, using an index containing novel transcripts. Transcripts with a length-scaled TPM of less than 1 in both the WT and KD datasets were filtered out before the DESeq2 step, as described above. The length-scaled TPM for transcripts calculated using tximport was used to compute and compare the cumulative TPM for different classes of transcripts.

Analysis of non-ENCODE datasets

Datasets from cells treated with Pladienolide B, THZ531, and Induslam [79] were retrieved from the SRA and processed as described above. Differential expression analysis was performed as described above, with treatment datasets compared to DMSO-treated data. RNA-seq datasets from cells treated with risdiplam [54] were retrieved from the SRA database, and technical replicates were merged into one dataset. Reads were trimmed and quantified as described above. We considered the RNA-seq datasets treated with the two highest concentrations of risdiplam (3160 and 10,000 nM) as two replicates of the treatment conditions and compared them to cells treated with DMSO in differential expression analysis. Datasets from iPSC from healthy patients and patients with PRPF31 mutations [57] and from fibroblast cells from healthy patients and patients with PRPF8 mutations [56] were retrieved from the SRA and quantified as described above. Differential expression analysis for all datasets was conducted using DESeq2 as described above. The SRA IDs of all non-ENCODE datasets are listed in Table S1.

Cell culture

Human colorectal cancer (HCT116) cell lines were cultured at 37°C and 5% carbon dioxide in a humidified chamber. McCoy’s 5A (Modified) Medium (Gibco) for HCT116 cells were supplemented with 10% Fetal Bovine Serum (Sigma) and 1% Penicillin‐Streptomycin (Fisher).

siRNA transfection and NMD inhibition

For siRNA-mediated knockdown, 1.6 μL of RNAiMAX, 60 pmol of siRNA, and 200 μL of Opti-MEM were incubated for 20 minutes according to the manufacturer’s protocol. HCT116 cells (1–2 × 10⁵) were then added to the transfection mixture in a 6-well plate. The medium was replaced the following day. Total RNA was harvested 48–65 hours post-transfection in 1 mL of TRIzol. siRNA sequences are provided in supplementary table 7. To inhibit the NMD pathway, cells were treated with 1 μM SMG1i (Cystic Fibrosis Foundation Therapeutics lab, chemical compound N1, dissolved in DMSO) for 24 hours before harvest. DMSO alone was used as a negative control. After 24 hours, cells were harvested in 1 mL of TRIzol.

Western blotting

To validate siRNA-mediated knockdown, cell extracts from siRNA treated HCT116 cells were separated on SDS‐PAGE, and transferred to a nitrocellulose membrane using Trans‐Blot Turbo system. After incubating membrane with primary antibodies overnight at 4°C and followed by 1 hour incubation with infrared fluorophore conjugated secondary antibodies, membranes were imaged with LI‐COR Odyssey CLx imager. Table S7 lists all the primary antibodies used.

RNA harvest, reverse transcription, RT-qPCR and RT-PCR

TRIzol extracted RNA was treated with RNase-free DNaseI (NEB) and Phenol-Chloroform extracted. 800 to 1000 ng of DNase-treated total RNA was used for reverse transcription using Maxima H minus RT (Thermo Fisher scientific). Reactions were set up as per manufacturer guidelines except for the amount of reverse transcriptase used (only half the recommended amount of RTase was used per unit RNA). cDNA thus obtained was diluted to the equivalent of 10 ng/μL of RNA and 1 μL was used per qPCR reaction. qPCRs were set up in 10 μL total volume using the BioRad SYBR green master mix. Each reaction was set up in triplicate, and average Ct value was used for analysis. Relative fold changes of transcript isoforms were calculated using the ΔΔCt method. At least 3 biological replicates (samples transfected and collected on different days) were used for estimating errors. Primers used are listed in table S7.

For RT-PCR, 1 μL of cDNA from representative knockdown conditions (siNC, siAQR, siEFTUD2, siSF3B1, and siSF3B3) was used in a 25 μL PCR reaction using GoTaq Green Premix (Promega) to assess isoform distribution of a particular transcript by agarose gel electrophoresis. PCR was performed for 25–35 cycles. Primers were used at a final concentration of 0.3 μM each. Primer sequences used in RT-PCR are listed in table S7.

Funding Statement

This work was supported by grants from the NIH (R01-GM120209 and R35-GM149298) to G.S. C.M.E. was partially supported by an NIH T32 training grant (T32-GM141955). Funding agencies made no contribution to setting goals, performing the described work, or preparing the manuscript.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Acknowledgments

Conceptualization: C.M.E. and G.S.; Formal analysis: C.M.E., D.P., and A.S.; Investigation: C.M.E. D.P., and A.S.; Writing – original draft and preparation: C.M.E. and G.S.; Writing – review and editing: C.M.E., D.P., and G.S.; Funding acquisition, Project Administration and Supervision: G.S. All authors have read and approved the final version of the manuscript.

Data availability statement

All codes written in support of this publication are publicly available at https://github.com/ceOSU/Embree-et-al-2025-Direct-and-indirect-effects-of-spliceosome-disruption. All datasets examined in this study are listed in Table S1. The GitHub repository includes all the supplemental tables. Further underlying data will be made available upon reasonable request from the corresponding author.

Supplementary Information

Supplemental data for this article can be accessed online at https://doi.org/10.1080/15476286.2025.2552517