The PRMT5-splicing axis is a critical oncogenic vulnerability that regulates detained intron splicing

Summary

Protein arginine methyltransferase 5 (PRMT5) is a promising cancer target, yet it is unclear which PRMT5 roles underlie this vulnerability. Here, we establish that PRMT5 inhibition induces a special class of unspliced introns, called detained introns (DIs). We used the depletion of CLNS1A, a PRMT5 cofactor that specifically enables Sm protein methylation, to interrogate the impact of DIs. We found that the disruption of Sm protein methylation is sufficient to induce DI upregulation, cell cycle defects, and loss of viability. Finally, we discovered that PRMT5-regulated DIs, and the impacted genes, are highly conserved across human, and also mouse, cell lines but display little interspecies conservation. Despite this, human and mouse DIs have convergent impacts on proliferation by affecting essential components of proliferation-regulating complexes. Together, these data argue that the PRMT5-splicing axis, and particularly appropriate DI splicing, underlie cancer’s vulnerability to PRMT5 inhibitors.

Graphical abstract

Highlights

• •PRMT5i promotes nuclear DI formation, decreasing coding isoform and protein levels
• •Sm protein methylation loss is sufficient to induce DIs and decrease viability
• •PRMT5-regulated DIs are conserved within, but not between, species
• •DIs from distinct species impact the same core pro-proliferation protein complexes

Gene network; Genomic analysis; Cancer; Transcriptomics

Introduction

PRMT5 is the major type II arginine methyltransferase and the primary enzyme responsible for catalyzing the symmetric dimethyl arginine mark (SDMA).¹ PRMT5 is highly expressed in stem cells, and its expression decreases as cells enter post-mitotic, terminally differentiated fates.² Importantly, PRMT5 is re-expressed in nearly all cancers, and chemical or genetic inhibition shows drastic anti-proliferative effects.^1,3 Many PRMT5 inhibitors (PRMT5i) show promising preclinical results in a wide variety of tumor types,^4,5,6,7 and several PRMT5i are in, or moving toward, clinical trials.⁸

PRMT5 writes the SDMA mark on proteins involved in cellular processes, including transcription, translation, and splicing. PRMT5 functions in a hetero-octameric complex comprised of four PRMT5 molecules and four cofactor molecules.^9,10 One of these cofactors, MEP50, is required for efficient PRMT5 activity. Additional co-factors, CLNS1A, RIOK1, and COPR5, augment PRMT5 activity by recruiting the complex to specific targets associated with splicing, translation, or transcription, respectively.^1,11 The PRMT5-MEP50-CLNS1A complex, referred to as the methylosome,^12,13 is the most studied of these complexes. It binds to and promotes the methylation of three Sm proteins: SmB/B′, SmD1, and SmD3.^12,14,15,16 These methylated Sm proteins are then incorporated into a larger, seven-member Sm protein ring, which is loaded around an snRNA to form an snRNP – a core component of the spliceosome.^17,18 PRMT5 is known to methylate other splicing factors,¹⁹ but does so without CLNS1A.

Which aspect(s) of PRMT5 biology underlie the vulnerability of cancer cells to PRMT5i remains a key question in the field. Because the methylosome is one of PRMT5’s most studied roles, much attention has been focused on assessing splicing problems upon PRMT5 inhibition. In a seminal study, Bezzi et al. examined PRMT5 deletion in neural stem and progenitor cells and showed that this was sufficient to cause neonatal inviability.²⁰ They showed that the PRMT5-deficient neural cells display an array of splicing defects, highlighting the inclusion of a cassette exon in MDM4 and consequent p53 activation, but also noting elevated levels of unspliced introns.²⁰ In a different study, we identified PRMT5 as the top hit in an in vivo screen for essential genes in glioblastoma, and our molecular analysis revealed that the predominant impact on splicing was increased levels of introns.² Notably, based on their relative stability and presence in poly-adenylated mRNAs, we concluded that these were likely a special class of unspliced introns, called detained introns (DIs).² DIs occur in fully processed transcripts that remain in the nucleus, making them unavailable for translation.²¹ This distinguishes them from retained introns (RIs), which exist in transcripts that reach the cytoplasm and typically undergo nonsense-mediated decay (NMD).²¹ It is now common to use splicing defects as evidence of PRMT5i action,^{2,4,5,6,7,19,22,23,24,25} reinforcing the prevailing view that PRMT5’s splicing role largely underlie the vulnerability of cells to PRMT5 inhibition. However, it is important to note that previous studies use experimental approaches that disrupt all PRMT5 functions. Therefore, while splicing defects are evident, it is an open question whether these are responsible for PRMT5i sensitivity versus other roles of PRMT5. Intriguingly, a recent study showed that the bulk depletion of CLNS1A, the methylosome co-factor, reduced the viability of pancreatic ductal adenocarcinoma KP4 cells.²² This observation is consistent with a key role for the PRMT5-splicing axis, but, unfortunately, this study did not look at the impact of CLNS1A loss on splicing.

In this current study, we conduct in-depth analyses of PRMT5-regulated splicing events and directly test their essentiality. We demonstrate that elevated intron levels are the predominant splicing defect and confirm that these are bona fide DIs. We specifically probe the functional importance of the methylosome through CLNS1A depletion, which leaves other PRMT5 functions intact. We find that CLNS1A is essential for viability in some, but not all, cell lines. Importantly, loss of viability directly reflects the loss of Sm protein methylation and upregulation of DIs. Finally, we find that the identities of PRMT5-regulated DIs are highly conserved across human cell lines, as well as mouse cell lines, irrespective of tissue/tumor type. In contrast, there is little conservation at the DI or gene level between human and mouse DIs, even within the same tissue type. Despite this, human and mouse DI-containing genes strongly converge on biological processes associated with cell proliferation. This reflects the fact that many DI-regulated genes encode for proteins that are essential, but largely non-overlapping, components of multi-protein complexes that play central roles in proliferation. Together, these data suggest that maintaining appropriate DI splicing in core proliferation-associated genes plays a critical role by which PRMT5 enables cell viability.

Results

PRMT5-regulated intronic splicing events are bona fide detained introns

We began our study by analyzing the splicing changes resulting from PRMT5 inhibition. We conducted a 3-day PRMT5i treatment using 10 nM JNJ-64619178⁶ in U87 glioblastoma cells, which we have previously shown to be PRMT5i-sensitive.² We enriched for poly-adenylated mRNAs in whole cell extracts, performed deep sequencing (> 40 million reads/sample), and conducted analyses to identify and quantify PRMT5-regulated splicing changes. The majority of significant changes (p_adj < 0.05) were elevated levels of introns, with the remainder being low levels of exon-based splicing events [cassette exons (CAS), mutually exclusive exons (MXE) and alternative 5’ (A5PSS) or 3’ (A3PSS) splice sites] (Figures 1A–1C). For each of these splicing events, the impact of JNJ-64619178 showed high concordance with findings from our prior study,² in which we treated U87 cells with a first-generation PRMT5i, EPZ015666⁴ for 3 days (Figure S1A). Notably, JNJ-64619178 and EPZ015666 have distinct mechanisms of action (targeting SAM versus substrate binding, respectively), arguing that the preferential impact on elevated intron levels is a general consequence of PRMT5 inhibition. Moving forward, we used JNJ-64619178 as our PRMT5i, unless otherwise stated.

Figure 1.

PRMT5 inhibition causes the formation of DIs specifically in the nucleus

(A–C) U87 cells were treated with vehicle or 10 nM JNJ-64619178 (PRMT5i) for 3 days and polyA + mRNA was deep sequenced and analyzed to identify and quantify exon-based alternative splicing (AS) events, including DIs, as shown in the schematic (A).

(B) Dots represent individual events for each AS type, with color denoting those significantly (p_adj < 0.05) or not significantly (gray) altered by treatment.

(C) Proportion of each AS event type that is significantly (p_adj < 0.05) altered by treatment. The total number of significant AS events is indicated above each bar.

(D–F) U87 cells were treated with vehicle or PRMT5i (as above). Nuclear and cytoplasmic fractions were isolated prior to sequencing (as above). (D) Heatmap shows intron-containing isoforms that are significantly up- or down-regulated by PRMT5i-treatment in either nuclear or cytoplasmic compartments (p_adj < 0.05). Columns show replicates, rows indicate individual introns, and the coloring indicates Z score normalized isoform counts where red denotes higher expression and blue denotes lower expression. (E) For all PRMT5-regulated DI isoforms (Left), and their coding isoforms (Right). The y axis values show the proportion of reads present in nuclear versus cytoplasmic fractions (1 – exclusively nuclear, 0 – equal distribution, −1 – exclusively cytoplasmic). The x axis values show the proportion of reads in vehicle versus PRMT5i-treated samples (1 – exclusively in PRMT5i samples, 0 – equal distribution, −1 – exclusively in vehicle samples). Each dot represents a specific isoform. (F) Representative sequencing tracks for a DI (red), and its adjacent exons (gray), in DNA2 in nuclear or cytoplasmic fractions and in whole cell lysates, with or without PRMT5i treatment.

(G) Comparison of the induction of PRMT5-regulated DIs (Left) and their coding isoforms (Right) detected in nuclear (DIs) or cytoplasmic (coding) fractions versus whole cell lysate. Inset shows the region displayed in the larger graphs. Correlation values reflect the Pearson correlation coefficient.

(H) Bar graphs (Top) show the quantification of DI (red) and coding (gray) isoform levels for six representative PRMT5-resposive genes (as indicated), determined by qRT-PCR of total RNA from whole cell lysates of U87 cells treated with vehicle or 10 nM JNJ-64619178 (PRMT5i) for 3 days. Data are mean ± SD of 3 technical replicates. ∗∗p < 0.01, ∗∗∗p < 0.001, and ∗∗∗∗p < 0.0001, Student’s t test. Protein levels (Bottom) were determined by western, using GAPDH as a loading control. Quantifications are shown below each blot. For this and all subsequent western blots, numbers indicate approximate molecular weight in kilodaltons.

Our next goal was to establish that the persisting introns were bona fide DIs, not RIs. A defining hallmark of DI-containing transcripts is their nuclear localization.²¹ Thus, we repeated the 3-day 10 nM PRMT5i treatment of U87 cells and performed cellular fractionation, confirming efficacy by western blotting (Figures S1B and S1C). We conducted deep sequencing and quantified the levels of intron-containing, versus coding (fully spliced), isoforms within the cytoplasmic and nuclear compartments (Figure S1B), as well as the impact of PRMT5i treatment (Figures 1D, 1E, and S1D). We detected 5,885 intron-containing isoforms and found that the vast majority (94%) showed profound nuclear enrichment (Figure S1D), which is entirely consistent with these being DIs. Most were lowly present in the absence of PRMT5 inhibition, befitting the known occurrence of DIs in normal cells.²¹ In response to PRMT5i, 44% of intron-containing isoforms (2,618/5,885) showed significant changes (Figure S1D). Importantly, the vast majority (94%) of these PRMT5i-regulated isoforms were DIs, as judged by their predominant nuclear localization (Figures 1D and 1E). For these PRMT5i-regulated DIs, treatment rarely impacted the cytoplasmic levels, but instead evoked profound changes in nuclear levels (Figures 1D and 1E). Moreover, in most cases (77%), the nuclear levels were dramatically increased by PRMT5i treatment (Figures 1D and 1E).

Our working model is that DI formation, and consequent nuclear trapping of mRNAs, leads to low levels of coding transcripts.² Thus, we examined the levels of coding transcripts, initially considering all detected species (Figures S1E and S1F). Most were observed in both compartments and were unaffected by PRMT5i treatment (Figures S1E and S1F). We then considered the coding isoforms of genes that contained PRMT5-regulated DIs (Figure 1E). Their behavior matched the reciprocal effect predicted by our hypothesis; all nuclear-enriched DIs showed concomitant decreases in the cytoplasmic representation of their coding isoforms (Figure 1E). Since labs in the PRMT5 field often use whole cell lysates to assess splicing defects, we asked whether these compartment-associated effects can still be discerned from whole cell extracts. Nuclear fractionation increases the detection sensitivity of sequencing reads in DI, compared to whole cell lysates, but the latter method is sufficient to detect basal levels and drug-induced DI upregulation with appropriate depth of sequencing (Figure 1F). Additionally, data from whole cell lysates correlated reasonably well with that from the nuclear fraction for DI isoforms (r = 0.70) or the cytoplasmic fraction for coding isoforms (r = 0.83; Figure 1G). Most importantly, we could directly validate the drug-induced changes on transcript and protein levels in whole cell lysates. Specifically, qRT-PCR confirmed significantly higher levels of DI-containing transcripts (2.5- to 24-fold over vehicle, p < 0.01) as well as significantly reduced levels of the coding transcripts (42–78% of vehicle, p < 0.01) in response to 3-day PRMT5i treatment (Figure 1H). In each case, this resulted in lower protein levels (21–57% of vehicle) detected by western blotting after 5 days of treatment, as the drug-induced splicing changes take effect on the proteome.

As a final proof that the PRMT5i-upregulated introns are DIs, we examined the impact of NMD, which would only affect RI levels. For this, we treated U87 cells with 10 nM PRMT5i for 3 days, and introduced siRNAs against UPF1, an NMD-initiating factor, or a non-targeting control, for the last 2 days of treatment. We confirmed significant siRNA-mediated UPF1 knockdown through qRT-PCR (p < 0.01) and western blotting (Figures S1G and S1H), and then established by qRT-PCR that PRMT5i treatment increased the intron levels within DNA2 (> 6-fold, p < 0.05) and EIF4E (> 20-fold, p < 0.01), but that the depletion of UPF1 did not further increase DI levels (Figure S1I). Taken together, our data show that PRMT5 inhibition specifically promotes the formation of bona fide DIs, thereby reducing the levels of cytoplasmic coding transcripts and protein product.

Detained introns are highly conserved across human cell contexts

Having demonstrated that DI formation is the major splicing change in response to PRMT5 inhibition in U87 cells, we wanted to determine if DIs are conserved across different human cell lines. For this analysis, we examined six additional lines, including four different cancers, a virally transformed cell line, and an immortalized normal cell line (Figure 2A). We treated these with 10 nM PRMT5i for 3 days, generated whole cell extracts, validated the inhibition of SDMA, and performed deep sequencing and AS analyses (Figures S2A and S2B). DIs accounted for the majority (62–75%) of AS events significantly altered (p_adj < 0.05) in response to PRMT5 inhibition in every context (Figures 2B and S2C). Collectively, we identified 2,794 unique DIs that were significantly altered (p_adj < 0.05) in response to PRMT5 inhibition in at least one cell line (Figure 2C). Like U87 cells, the vast majority (87–98%) of DIs were upregulated in response to drug treatment (Figures 2C and S2C). We then investigated the degree to which significant PRMT5i-regulated AS events (p_adj < 0.05) were conserved across human cells. Initially, we performed pairwise comparisons between all seven cell lines, plotting both the degree of pairwise conservation and absolute number of conserved events for each AS class (Figure 2D). This revealed remarkably little conservation for any of the exon-based AS events, but high conservation of DI identity, for all pairwise comparisons, affecting more than 1,000 individual DIs (Figure 2D). Given the high pairwise DI conservation, we then considered the overlap across all lines (Figures 2E and S2D). This revealed remarkable conservation of PRMT5i-regulated DIs; more than 500 were significantly upregulated (p_adj < 0.05, log₂FC > 0) in 7/7 cell lines, and more than half of the DIs in any one cell line were conserved in at least 5/7 lines (Figures 2E and S2D). These data indicate that PRMT5 inhibition upregulates a common set of DIs, irrespective of tissue type, driving mutation, or oncogenic status.

Figure 2.

PRMT5 regulates a conserved set of DIs across multiple human cell lines

(A) Human cell lines used for AS analysis, including cancer or tissue of origin, driver mutations or construct, and MTAP status.

(B) Proportion of each AS event type that show significantly altered levels in the indicated cell line in response to 3-day treatment with 10 nM JNJ-64619178 (PRMT5i) compared to vehicle control (p_adj < 0.05). The total number of significant AS events for that cell line is indicated to the right of each bar.

(C) Relative expression of individual DIs across all human cell lines after 3-day vehicle or PRMT5i treatment. Each column represents a specific DI that is significantly up- or down-regulated by PRMT5i-treatment in at least one cell line. Coloring represents Z score normalized intron counts, where red indicates higher relative expression and blue indicates lower relative expression.

(D) Similarity matrix of the indicated AS event significantly altered by PRMT5i treatment (p_adj < 0.05). Each square represents the similarity between the two intersecting cell lines, with coloring indicating percent similarity (dice similarity score) and square size indicating the number of similar events.

(E) Bar charts of DI conservation across 7 human cell lines. Bar length represents the number of significant PRMT5i-upregulated DIs (p_adj < 0.05, log₂FC > 0) for each cell line and the color denotes the number of cell lines in which a given DI is conserved across. The total number of significant DIs in each line is indicated to the right of each bar.

(F) Enriched GO terms from the KW Biological Process gene set for PRMT5i-upregulated DIs in each cell line (p_adj < 0.05, log₂FC > 0). Terms are displayed if they are significant in at least one cell line (FDR < 0.01). Color represents enrichment score, while size inversely correlates with significance value.

Given this high conservation, we next determined the biological processes that PRMT5-regulated DIs are associated with. Gene ontology (GO) analysis for all upregulated DIs (p_adj < 0.05, log₂FC > 0), showed strong enrichment of core proliferation-associated processes, including multiple cell cycle-associated gene sets, in all seven lines (Figure 2F), consistent with our prior analysis of U87 cells.² Together, these data support a model where PRMT5 enables efficient intron splicing of a conserved set of DIs in proliferation-associated genes.

Detained intron upregulation from methylosome impairment mirrors viability impairment

Our observations fit the prevailing view that the vulnerability of cancer cells to PRMT5i results from splicing defects that impact proliferation control. However, neither our, nor existing, data directly address the importance of the PRMT5-splicing axis because they disrupt all PRMT5 functions (Figure 3A). To address this limitation, we interrogated the necessity of the methylosome, which mediates PRMT5’s core splicing role, using the dTag degron system²⁶ to deplete CLNS1A, the PRMT5-cofactor that is essential for this function (Figure 3A). We selected three of our seven PRMT5i-responsive cell lines – CAL51, HEK293T, and HCT116 (Figure 2A) – and introduced the dTag degron domain, which also contains an HA-tag, into the endogenous CLNS1A locus. For each line, we identified four clones that have homozygous degron cassette insertion, as determined by a 12 kDa molecular weight increase for CLNS1A and HA positivity (Figures 3B and S3A). We challenged parental cell lines with the degron inducer, dTag-13, for 6 days and showed that doses up to 1 μM avoid off-target toxicity (Figure 3C). We then treated all of our clones for 3 days with 1 μM dTag-13 and showed that this achieved high levels of CLNS1A depletion for all cell lines (Figure 3D).

Figure 3.

Generation of endogenous dTag-CLNS1A in multiple human cell lines to specifically inhibit the PRMT5-splicing axis

(A) Schematic indicating how use of the dTag-CLNS1A degron system specifically separates the disruption of the methylosome from other PRMT5 functions.

(B) Western blot analysis of representative dTag-CLNS1A clones for CAL51, HEK293T, and HCT116 cells, alongside parental controls with antibodies against CLNS1A, the inserted HA tag, or loading control GAPDH. Asterisk (∗) marks a non-specific band.

(C) Dose-response curves for three parental cell lines treated with dTag-13 for 6 days. Data are mean ± SD of 3 technical replicates/line. Indicated p-values represent the most significant comparison between the respective drug concentration and vehicle-treated cells. Vertical dashed line represents the concentration above which dTag-13 has off-target effects. ∗∗∗∗p < 0.0001, Student’s t test.

(D) Western blot analyses of HA (CLNS1A) and loading control GAPDH in four CAL51, HEK293T, and HCT116 dTag-CLNS1A clones treated with either vehicle or 1 μM dTag-13 for 3 days.

Having validated our dTag-CLNS1A clones, we directly compared their sensitivity to CLNS1A depletion versus PRMT5 inhibition. Specifically, we treated each with vehicle, 1 μM dTag-13, or 10 nM PRMT5i for various timepoints. First, we conducted molecular analyses, using a 3-day timepoint to match the preceding studies. PRMT5i behaved as expected; it had no detectable effect on the levels of HA-CLNS1A (Figures 4A and 4B) but caused a dramatic and significant (p < 0.01) downregulation in the methylation of SmB (Figures 4A and 4B), as well as all other SDMA-modified proteins (Figure S4A). Consistent with our clone validation (Figure 3D), dTag-13 treatment induced robust and significant (p < 0.01) depletion of HA-CLNS1A in all cases (Figures 4A and 4B). Unexpectedly, we observed two distinct downstream molecular phenotypes, which depended on the cell line and were consistent across all clones. CAL51 and HEK293T displayed one set of phenotypes: dTag-13 treatment caused significant (p < 0.05) loss specifically of SDMA on SmB (Figures 4A and 4B), but not other PRMT5 targets (Figure S4A), and yielded robust DI induction (Figures 4C and S4C). Importantly, there was high conservation in the identity of DIs affected by CLNS1A-depletion in the clones versus PRMT5i treatment of their relevant parental CAL51 or HEK293T lines (Figure 4D), and less overlap for other AS events (Figures S4D–S4G). The HCT116 clones displayed starkly different downstream molecular phenotypes. Specifically, dTag-13 treatment yielded no significant loss of methylation on Sm proteins (Figures 4A and 4B), and almost no significant induction of DIs, or other AS changes (Figures 4C, 4D, and S4C–S4G).

Figure 4.

Loss of Sm protein methylation correlates with increased DI formation and decreased viability

(A–C) dTag-CLNS1A clones for CAL51, HEK293T, and HCT116 were treated with vehicle, 10 nM JNJ-64619178 (PRMT5i), or 1 μM dTag-13 for 3 days (A) Western blot analyses of HA (CLNS1A), SDMA (methyl-SmB), and loading controls (GAPDH for CAL51 and HEK293T, or HSP90 for HCT116). (B) Quantification of HA (Left) and methyl-SmB (Right) levels normalized to loading control. Data are mean ± SD of 4 biological replicates. Welch’s t-test. (C) Quantification of indicated classes of AS event in dTag-CLNS1A #1 for CAL51, HEK293T, and HCT116. Dots represent individual events, denoting ones that are significantly (p_adj < 0.05, color), or not significantly (gray) altered by dTag-13 treatment.

(D) Venn diagrams show overlap of DIs in CAL51, HEK293T, and HCT116 cells that were significantly altered by dTag-13 versus PRMT5i treatment, compared to their vehicle controls (p_adj < 0.05). Hypergeometric test.

(E) CAL51, HEK293T, and HCT116 dTag-CLNS1A #1 clones were treated with vehicle, 10 nM PRMT5i, or 1 μΜ dTag-13. These were analyzed by FACS for: (Left graphs) cell cycle profile by DAPI/EdU staining after 4 days, or (Right graphs) cell death by Annexin-V/PI staining after 6 days of treatment. Cell cycle data represent mean ± SD of 4 biological replicates, and cell death data represent mean ± SD of 3 biological replicates. Corresponding statistics, and additional timepoints, are shown in Figures S5A, S5B, S6A, and S6B.

(F) Relative viability of four dTag-CLNS1A clones versus parental lines for CAL51, HEK293T, and HCT116 cells after a 6-day dTag-13 treatment. Data represent mean ± SD of 3 technical replicates. Student’s t test.

For all panels, ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, and ∗∗∗∗p < 0.0001.

PRMT5 inhibition is known to yield proliferation defects, which can trigger cell death or senescence.² Thus, we screened the CAL51, HEK293T, and HCT116 dTag-CLNS1A #1 clones for these defects, directly comparing the effects of treatment with dTag-13 with those of PRMT5i. Exactly as expected, PRMT5i induces cell cycle defects, and then apoptosis, determined by Annexin-V positivity (Figures 4E, S5A, S5B, S6A, and S6B). In contrast, dTag-13 treatment of dTag-CLNS1A #1 clones yielded distinct outcomes in the CAL51 and HEK293T, versus HCT116, in agreement with their differential molecular response. The first observed difference was in cell cycle phasing, determined by EdU incorporation and DAPI staining via flow cytometry after 2, 4, or 6 days of treatment (Figures 4E, S5A, and S5B). For the 2-day timepoint, all three lines largely resembled the vehicle control (Figure S5A). By day 4, and continuing on day 6, the CAL51 and HEK293T clones both showed significant differences in the distribution of G1, S, and G2/M populations, indicating cell cycle progression defects (Figures 4E, S5A, and S5B). Moreover, by day 6, both lines showed a significantly higher level of cells with sub-G1 DNA content, indicating cell death (Figures S5A and S5B). To reinforce this finding, we conducted Annexin-V and propidium iodide (PI) staining on 6-day treatment samples and confirmed that dTag-13 significantly increased the level of apoptotic and dead cells in both CAL51 and HEK293T (Figures 4E, S6A, and S6B). Importantly, for both CAL51 and HEK293T clones, the nature, magnitude, and time of appearance of the cell cycle and apoptotic defects were strikingly similar for dTag-13 versus PRMT5i treatment. As noted above, we observed a different outcome with HCT116. Specifically, dTag-13 treatment of the HCT116 dTag-CLNS1A clone had minimal impact on the cell cycle distribution, the level of sub-G1 cells, or the degree of Annexin-V/PI staining at any time point, while PRMT5i induced cell cycle defects and apoptosis (Figures 4E, S5A, S5B, S6A, and S6B). These differential phenotypic effects between dTag-13 and PRMT5i treatment correlated perfectly with their differing abilities to impact SmB methylation and DI formation.

To further probe the differential cell cycle and apoptotic responses of CAL51, HEK293T, and HCT116 dTag-13, we compared the relative viability of all four dTag-CLNS1A clones to their parental controls across a range of dTag-13 concentrations (Figure 4F). The dTag-13 response of the four HCT116 dTag-CLNS1A clones was indistinguishable from that of their parental control. In contrast, the CAL51 and HEK293T dTag-CLNS1A clones all showed exceptionally heightened sensitivity (p < 0.0001). To confirm that these dTag-13 effects were on target, we generated dTag-CLNS1A #1 variants that ectopically expressed FLAG-tagged CLNS1A (Figure S7A). We treated these cells with dTag-13 and found that ectopic CLNS1A-FLAG rescued the drug induced loss of SmB protein methylation, induction of DIs, and reduced viability of CAL51 and HEK293T clones, essentially restoring them to the parental state (Figures S7A and S7C). Taken together, these data show that selective loss of SDMA on Sm proteins is sufficient to phenocopy central effects of PRMT5 inhibition, including the upregulation of DIs, and subsequent activation of cell cycle defects, and cell death.

CLNS1A is fully dispensable in some contexts

The finding that all four HCT116 dTag-CLNS1A clones were insensitive to CLNS1A depletion was highly unexpected, given that it is believed to be cell essential. We realized that dTag-13 treatment does not achieve complete CLNS1A loss, and the residual levels in HCT116 are typically higher than those observed in CAL51 and HEK293T clones (Figures 4A and 4B). To determine whether a more prolonged treatment would reveal an impairment in HCT116 cells, we treated two HCT116 dTag-CLNS1A clones with 1 μM dTag-13 for 50 days, harvesting and replating cells every 5 days. These cells showed no perceptible growth impairment and no noticeable loss of SmB methylation, even though CLNS1A was efficiently depleted throughout the time-course (Figure 5A). Thus, HCT116 can tolerate a dramatic, long-term CLNS1A depletion.

Figure 5.

HCT116 cells are CLNS1A-independent until PRMT5 activity is decreased

(A) Western blot analyses of HA (CLNSIA), SDMA (methyl-SmB), and HSP90 as loading control in HCT116 dTag-CLNS1A #1 (Left) and #2 (Right) maintained in 1 μM dTag-13 for the indicated number of days.

(B) Representative Western blot analysis of CLNS1A, SDMA (methyl-SmB), and loading control GAPDH levels in HCT116 sgCtrl and four sgCLNS1A clones (n = 4 biological replicates).

(C) Relative proliferation of HCT116 sgCtrl and four sgCLNS1A clones through analyses of cumulative population doubling over a 6-day period. Data represent mean ± SD of 10 technical replicates.

(D) Relative viability of HCT116 sgCtrl and indicated sgCLNS1A clones treated with the PRMT5i JNJ-64619178 for 6 days. Data represent mean ± SD of 3 technical replicates.

(E–G) HCT116 parental or two dTag-CLNS1A clones treated with or without 1μM dTag-13 in combination with the indicated concentrations of JNJ-64619178 (PRMT5i). (E) Relative viability was assessed after treatment for 6 days. Data represent mean ± SD of 3 technical replicates. Red markings correspond to the critical PRMT5i doses used in Figures 5F and 5G. (F) Western blot analyses of HA (CLNS1A), SDMA, and loading control HSP90 after treatment with the critical indicated PRMT5i doses for 3 days (G) Relative levels of a representative DI in EIF4E were assessed by qRT-PCR after treatment with the critical indicated PRMT5i doses for 3 days. Data represent mean ± SD of 3 technical replicates.

For all panels, ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, and ∗∗∗∗p < 0.0001, Student’s t test.

Given this finding, we wondered whether CLNS1A is fully dispensable in HCT116 cells. We deployed CRISPR-Cas9-mediated gene targeting with the goal of generating CLNS1A knock-out (KO) lines, and identified two independent KO lines (sgCLNS1A-1 and sgCLNS1A-2), carrying different homozygous deletions that each caused frameshifts and premature termination codons (PTCs), as well as two independent hypomorphic clones (sgCLNS1A-3 and sgCLNS1A-4), in which one allele bore a small in-frame deletion, Δ117-118, and the other was disrupted by different frameshift mutations and consequent PTCs (Figure S8A). Western blotting confirmed that the KO clones lacked detectable CLNS1A and revealed that the hypomorphic clones had effectively undetectable protein levels (Figures 5B and S8B), indicating that these in-frame deletions impair protein stability. We then analyzed the properties of our sgCLNS1A clones, and found that Sm protein methylation was retained at relatively normal levels in two clones and with only a modest reduction in two others (Figures 5B and S8B), and all four sgCLNS1A clones proliferated at the same rate as the sgCtrl cells (Figure 5C). These data argue that HCT116 cells thrive in the absence of CLNS1A.

Given how unexpected this finding was, we wondered whether the sgCLNS1A clones had altered PRMT5-associated processes to compensate. We realized that there is an annotated alternatively spliced CLNS1A isoform, which skips the exon targeted by our sgRNA,²⁷ and screened for full length (FL)- and truncated-CLNS1A transcripts in the KO clones. Both had significantly reduced levels of FL-CLNS1A mRNA and upregulated levels of truncated-CLNS1A, albeit from near absence in sgCtrl cells to low levels in the KOs (Figures S8C and S8D). Alphafold2^28,29 yielded a protein structure for truncated-CLNS1A, but the portion encoded by the skipped exons makes critical contacts to Sm protein substrates and also within the CLNS1A protein, likely supporting stability (Figure S8E). Accordingly, we introduced FL- or truncated-CLNS1A into sgCLNS1A-1 cells using retroviruses at a similar multiplicity of infection, and observed lower protein levels of the truncated versus FL-CLNS1A (Figure S8F). We also considered the possibility of altered expression for other PRMT5 pathway components. Specifically, we interrogated PRMT5, MEP50 (PRMT5’s essential co-factor), PRMT9 (an alternate SDMA-writing enzyme), ROIK1 (a co-factor that competes with CLNS1A), and SNRPB/SmB, using qRT-PCR and western blotting, but found no differences between the four sgCLNS1A clones and the sgCtrl controls (Figure S8G).

Since CLNS1A functions in a complex with PRMT5, we anticipated that CLNS1A deficiency, and subsequent ectopic expression, would modulate the sensitivity of the HCT116 cells to PRMT5i. Thus, we treated sgCLNS1A-1 and sgCLNS1A-4 with a range of PRMT5i concentrations for 6 days, and found that both showed significantly (p < 0.01) increased susceptibility to PRMT5i, compared to the parental control (Figure 5D). We then examined the effect of the ectopic expression of core PRMT5 pathway components on this phenotype. As expected, FL-CLNS1A made all sgCLNS1A clones significantly (p < 0.01) more resistant to PRMT5i (Figures S9A and S9B). In contrast, truncated-CLNS1A had no effect in the sgCLNS1A-1 cells (Figure S9C). Finally, we found that the ectopic expression of PRMT5, MEP50, or PRMT9 did not modulate the PRMT5i-sensitivity of sgCLNS1A-1 or sgCLNS1A-3 (Figures S9D and S9E). These data yield two major conclusions. First, they show that truncated-CLNS1A is non-functional, demonstrating that the sgCLNS1A clones are truly CLNS1A-deficient, and establishing that CLNS1A is dispensable in HCT116 cells. Second, they show that CLSN1A loss increases the vulnerability of HCT116 cells to PRMT5 inhibition, which cannot be rescued by the overexpression of single PRMT5 pathway players.

We hypothesized that we could exploit the collaborative effects of CLNS1A deficiency and PRMT5 inhibition to further explore the relationship between the PRMT5-splicing axis and cellular viability. For this, we returned to the HCT116 dTag-CLNS1A clones, and cultured two alongside the parental line, with either vehicle control or 1 μM dTag-13 across a range of JNJ-64619178 (Figure 5E) or EPZ015666 (Figure S9F) concentrations. As expected, dTag-13 treatment did not alter the responsiveness of the parental HCT116 to either PRMT5i, but it made the dTag-CLNS1A lines significantly more sensitive to both inhibitors (p < 0.001). The largest viability differences with or without dTag-13 treatment were observed at 0.5 and 1 nM of the PRMT5i JNJ-64619178 (Figure 5E Center, red triangle and star). Thus, we focused on these concentrations to determine whether differences in Sm protein methylation and DI levels correlated with the observed viability differences. At both 0.5 and 1 nM PRMT5i, HCT116 dTag-CLNS1A #1 exhibited a greater loss of Sm protein methylation when dTag-13 treatment was added (Figure 5F). Moreover, decreased Sm protein methylation was accompanied by a significant (p < 0.01) increase in the levels of a canonical DI in EIF4E (Figure 5G). Importantly, there was no difference in the levels of other (non-methylosome) PRMT5 targets between minus and plus dTag-13 treatment at 0, 0.5, and 1 nM PRMT5i. Thus, CLNS1A depletion significantly sensitizes cells to PRMT5 inhibition, and this, again, tracks with the magnitude of impairment of Sm protein methylation and increased DI formation.

MTAP deficiency increases dependence on methylosome activity

We also considered how the status of methylthioadenosine phosphorylase (MTAP) modulates the influence of the methylosome and DI regulation on cell viability. MTAP is frequently deleted in many human tumors due to its adjacency to the CDKN2A tumor suppressor locus.^{30,31,32,33,34,35} MTAP loss causes the accumulation of its substrate, MTA, which acts as an endogenous inhibitor of PRMT5.^36,37 If DIs represent the core vulnerability of cancer cells, MTAP loss should collaborate with CLNS1A deficiency in HCT116 dTag-CLNS1A clones. To address this, we used CRISPR-Cas9-mediated gene targeting to generate two independent MTAP-deficient variants of HCT116 dTag-CLNS1A #1 (Figure S10A). MTAP deficiency had no effect on the normal rate of proliferation (Figure S10B). However, it altered the response to treatment with 1 μM dTag-13, exactly as expected; compared to MTAP-proficient cells, MTAP-deficient cells showed lower levels of SmB methylation and significantly upregulated canonical DIs in EIF4E and SNRNP200 after 3 days of treatment, and significantly reduced viability after 6 days (Figures S10C–S10E). Finally, we asked whether the collaboration between CLNS1A depletion and PRMT5 inhibition in HCT116 cells was modulated by MTAP deficiency. We treated our sgCtrl and sgMTAP lines in the presence or absence of 1 μM dTag-13 for 6 days, together with JNJ-64619178 or two next-generation PRMT5 inhibitors that are known to be MTA-cooperative (AMG-193 and TNG908) over a range of concentrations (Figure S11A). First, we considered the no dTag-13 condition (i.e., CLNS1A proficient) and found that the PRMT5i behaved exactly as expected; JNJ-64619178 had similar IC50s for sgCtrl and sgMTAP cells, while the MTA-cooperative inhibitors had lower IC50s in sgMTAP cells. We then considered how dTag-13 addition (i.e., CLNS1A deficient) altered this response. For both sgCtrl and sgMTAP cells, dTag-13 increased sensitivity to all three PRMT5i, particularly at the lower drug doses. The viability difference between dTag-13 plus and minus cells was most pronounced in sgMTAP cells, beginning with the no PRMT5i point, suggesting that it largely reflects the collaborative effects of CLNS1A deficiency and MTAP loss. This was apparent when we replotted the data, setting the no drug concentration viability to 1 for both dTag-13 plus and minus cells (Figure S11B). Based on these experiments, PRMT5 inhibition appears to be largely additive with CLNS1A deficiency and MTAP loss, at least in the setting of this single cell line.

Detained intron identity is conserved across mouse cell lines to regulate proliferation-associated processes

Our data show that DIs are highly conserved across human cell types and are a key vulnerability to PRMT5 inhibition. Given this, we expanded our analysis to murine cells. Specifically, we treated murine lung and pancreatic ductal adenocarcinoma lines for 3 days with PRMT5i [JNJ-64619178 (100 nM) or EPZ015666 (10 μM)] (Figures 6A, S12A, and S12B). Similar to human cells, DIs accounted for the majority of significant (p_adj < 0.05) AS events irrespective of cell line or inhibitor (Figures 6B and S12C). Moreover, DI identity was highly conserved across murine cell lines (Figures 6C–6E), and much more so than exon-based AS events (Figure 6D). Indeed, nearly 1,000 individual DIs were common to all murine cell lines examined (Figure 6E). Thus, similar to our finding in human cells (Figures 2C–2E), PRMT5 regulates a conserved set of murine DIs. We performed GO analysis on all upregulated DIs for each murine line (p_adj < 0.05, log₂FC > 0) and observed strong enrichment for core proliferation-associated processes (Figure 6F). Collectively, our data show that PRMT5 inhibition drives the formation of a conserved set of DIs in murine cells that impact proliferation-associated processes.

Figure 6.

PRMT5 regulates a conserved set of DIs in multiple mouse cell lines

(A) Summary of mouse cell lines used for AS analysis, including cancer type, driver mutations, PRMT5i used, and MTAP status.

(B) Proportion of each AS event type that is significantly different in indicated cell line after 3-day PRMT5i treatment compared to vehicle control (p_adj < 0.05). Total number of significant AS events is indicated above the bar.

(C) Relative expression of individual DIs across all mouse cell lines after 3-day vehicle or PRMT5i treatment. Each column represents a specific DI that is significantly upregulated or downregulated by PRMT5i-treatment in at least one cell line. Colors represent column Z score normalized intron counts, where red indicates higher relative expression and blue indicates lower relative expression.

(D) Similarity matrix of the indicated class of AS event that was significantly altered by PRMT5i treatment. Each square represents the similarity between the two intersecting murine cell lines. Square coloring indicates percent similarity (dice similarity score), and square size is proportional to the number of similar events.

(E) Bar charts of DI conservation across all mouse cell lines. Each bar represents the number of PRMT5i-upregulated DIs (p_adj < 0.05, log₂FC > 0) in each cell line and the color corresponds to the number of cell lines in which a given DI is conserved across. Total number of significant DIs in each line is indicated to the right of the bar.

(F) Enriched GO terms from the KW Biological Process gene set for PRMT5i-upregualted DIs in each cell line (p_adj < 0.05, log₂FC > 0). Terms are displayed if they are significant in at least one cell line (FDR < 0.01). Color represents enrichment score, while size inversely correlates with significance value.

Detained introns are not conserved across species, but the biological complexes and processes they affect are highly conserved

Our analyses indicate that DIs impinge on similar biological processes in human and mouse cells (Figures 2F and 6F). To explore this overlap, we conducted pairwise comparisons of the processes affected by PRMT5i-induced DIs in the human H23 lung cancer line with those detected in each of the murine lung cancer lines, KP393T5 and KP1233, and observed 71% (Figures 7A and 7B) and 79% (Figure S13A) overlap, respectively. Again, there was a high representation of core proliferation-associated processes. Similarly high conservation (56%) was observed when comparing human PANC1 and murine MDM1402 pancreatic cancer lines (Figure S13B). This high inter-species conservation of DI-affected biological processes was also agnostic of tissue type, as comparison of the biological processes conserved across all human lines with those conserved across all murine lines showed 67% conservation (Figure S13C).

Figure 7.

PRMT5-regulated DI identity is not conserved across species, but it occurs in genes in similar complexes to regulate proliferation-associated processes

(A) Overlap of common DI-regulated (Left) processes, (Center) genes, and (Right) complexes for PRMT5-regulated DIs in human (H23) and mouse (KP393T5) lung cancer cell lines. Percentages represent the dice similarity score for each comparison.

(B) Enrichment of DI-regulated biological processes in human and mouse cells. Each pair of bars represents a specific biological process, with the blue bar representing the FDR of the human H23 lung cancer line and the red bar representing the FDR of the mouse KP393T5 lung cancer line. Vertical dashed line represents the significance cutoff (FDR < 0.05).

(C) Schematic of representative DI-regulated complexes, including the minichromosome maintenance (MCM) and mediator (MED) complexes. Each shape represents a specific complex member, and a color fill of blue, red, purple, or white indicates that the complex member contained a DI in human, mouse, both species, or neither, respectively.

(D) Sankey plot showing convergence of DI-regulated complexes and biological processes. Each line represents a specific DI moving from (Column 1) gene to (Column 2) complexes to (Column 3) biological processes. Line color signifies conservation at the (Left) gene or (Right) complex level, and thickness corresponds to the number of (Left) complexes or (Right) biological processes a given gene is a part of. Column color represents conservation at the level of (Column 1) gene, (Column 2) complexes, and (Column 3) biological processes. Blue indicates unique to human cells, red unique to mouse cells, and purple common to both species. For biological processes, white indicates DIs/complexes that map to non-significant biological processes (FDR > 0.05).

We anticipated that the inter-species overlap would reflect the high conservation of the DI-containing genes. However, we observed a striking lack of inter-species conservation in DI-containing genes, even when we considered a specific tissue type (21%–25%; Figures 7A, S13A, and S13B) or the genes that were commonly regulated in all human versus all mouse lines (25%; Figure S13C). Moreover, when regulated genes were conserved between human and mouse, the DIs typically affected distinct introns. Thus, our data show that PRMT5i-induced DIs are poorly conserved across species at the intron or gene level, yet impact the same biological processes. As a possible explanation, we realized that many DI-containing genes encode for components of multi-protein complexes. Consequently, we performed CORUM analysis³⁸ to identify complexes affected by DI-containing genes in mouse versus human cells and then quantified the degree of overlap. For both comparisons within specific tissue types (Figures 7A, S13A, and S13B) and comparison of DIs commonly regulated in all human versus all mouse lines (Figure S13C), we found a strong overlap in complex identity (53–59%), which was much higher than the DI-containing gene identity, and approached the levels of biological process conservation. Examples of affected complexes are shown in Figures 7C, S13D, and S13E. Moreover, the examination of representative complexes revealed many cases where the PRMT5-regulated DIs impacted more than one complex component within a single species (Figures 7C, S13D, and S13E), which would further exacerbate defects in the assembly and levels of functional complexes. Finally, analyses at the global level (Figures 7D and S13F) clearly illustrate how DI upregulation in species-specific genes converges on multi-protein complexes to impact common, fundamental biological processes.

Discussion

The primary goal of this study was to address the prevailing view that PRMT5’s role in splicing is largely responsible for its oncogenic activity and therefore the cellular vulnerability to PRMT5 inhibitors. We began by investigating the spectrum of splicing problems that result from PRMT5 inactivation. Consistent with our prior study,² we find that the predominant significant changes are alterations in intron inclusion. We further establish that the vast majority of these are bona fide DIs, as judged by nuclear localization and independence from NMD, and most are upregulated as a consequence of PRMT5 inhibition. This DI upregulation occurs irrespective of the mechanism of action of the PRMT5i and is remarkably well conserved. Importantly, DI upregulation has regulatory consequences. We see, via both genome-wide analyses and validation of identified DI-regulated genes, a corresponding decrease in the levels of coding transcript available in the cytoplasm for translation. Additionally, western blot analyses confirm consistent reduction in protein levels. Notably, GO analyses show that the PRMT5i-responsive DIs selectively impact genes that control cell cycle regulation and proliferation, the biological processes that are well known consequences of effective PRMT5 inhibition. Accordingly, we find that DI upregulation is followed by cell cycle defects, induction of apoptosis, and loss of cell viability.

In a “separation of function” approach, we used the depletion or deletion of CLNS1A to specifically inactive the methylosome, and thus the PRMT5-splicing axis, while maintaining other PRMT5 functions. We observed a strong depletion of CLSN1A in all three cell lines examined. In two cases (CAL51 and HEK293T), this resulted in effective and specific methylosome inhibition (as judged by loss of SDMA marks on Sm proteins, but not other PRMT5 targets) while the third cell line (HCT116) showed no methylosome inhibition. Importantly, when methylosome inhibition was observed, we saw robust DI formation and subsequent cell cycle defects, apoptosis, and loss of viability. Moreover, in CAL51 and HEK293T cells, the kinetics and magnitude of the cellular defects were strikingly similar between CLNS1A depletion and PRMT5 inhibition. Collectively, our data show that PRMT5’s splicing role is sufficient to account for PRMT5’s cell essentiality. We note that these findings do not rule out roles for other PRMT5 functions. It is possible that the selective inactivation of other functions could be lethal too, or could serve to augment the key role of splicing. We also note that our separation of function experiments do not dissect the various aspects of splicing. Obviously, we favor the notion that DI regulation is the critical driver, based on their prevalence, high conservation, and enrichment in core proliferation regulators. However, we are not ruling out a role for exon-based AS events. Indeed, there is compelling evidence that PRMT5 inhibition can drive alternative splicing of MDM4, which results in p53 activation.²⁰ However, given the relatively low levels and conservation of exon-based AS events, we hypothesize that they have contributing roles in a context-dependent manner.

The finding that HCT116 cells were unaffected, phenotypically, by CLNS1A depletion or complete knockout was completely unexpected, as it is widely believed to be pan-cell essential. The existing view is based primarily on large-scale depletion screens, which effectively test fitness rather than essentiality. It remains an open question how HCT116 cells can survive without CLNS1A. Interestingly, we did not observe any obvious compensation in our CLNS1A KO cells via the upregulation of PRMT5, MEP50 or PRMT9, or downregulation of the competing cofactor, RIOK1. Since PRMT5-MEP50 complexes are sufficient to bind and methylate Sm proteins, at least in vitro,^22,39 we speculate that there is enough PRMT5-MEP50 activity in HCT116 cells to survive without CLNS1A. Regardless of the underlying mechanism, our analyses of CLNS1A-deficient and KO HCT116 cells support the essentially of methylosome activity. Specifically, these cells show no detectable perturbation of Sm protein methylation, DI levels, cell cycle phasing, or cellular viability. However, they do display increased sensitivity to PRMT5i or genetic ablation of MTAP. Importantly, this increased sensitivity reflects diminished Sm protein methylation and upregulated DI formation. The observed collaboration of CLNS1A deficiency with low level PRMT5i or MTAP loss in HCT116, at both the molecular and cellular levels, argues that these perturbations are working via a shared mechanism that converges on robust DI induction as a trigger for cell cycle defects and loss of viability.

Our study also reveals fascinating similarities, as well as differences, in DI biology between humans and mice. With respect to similarities, we find that on the order of 1000–2000 genes are affected by PRMT5i-responsive DIs in either human or mouse cells. These DIs are remarkably well-conserved in an intra-species manner; for example, approximately half of the PRMT5i-responsive DIs were conserved in all 7 human cell lines analyzed. This conservation is agnostic to tissue type, whether the cells are transformed or normal, or the nature of the driving mutations within the cancer cells. As noted above, many of the affected genes are cell essential and significantly enriched for roles in cellular proliferation. We also found that the same core biological processes are affected in mouse and human cells. However, remarkably, there is minimal conservation between these two species with regard to the identity of the DIs, or even the affected genes. Instead, our analyses argue that these targets frequently represent distinct, rather than overlapping, components of multi-protein complexes involved in proliferative control. Interestingly, in many cases, we find that more than one component of a given complex are encoded by DI-containing genes. It is easy to envisage that the upregulation of DIs, occurring simultaneously in hundreds of core proliferation regulators and affecting numerous essential proliferation-associated complexes, creates profound challenges for cell cycle progression. It is fascinating to consider how two species have developed a way to maintain regulation at the complex level without conservation at the DI or gene level. Presumably, with so many genes in play, a species could afford to lose or gain individual DIs, without significant perturbation of overall cell cycle control, as long as impact on core proliferation-associated complexes is conserved. This would allow drift at the DI and gene level, while maintaining conservation at the complex and process level. Notably, we believe that control of DI levels by the methylosome is a general mechanism of proliferation control, used by normal cells to regulate proliferative capacity throughout development, which is co-opted by cancer cells through the near universally observed upregulation of PRMT5 in malignancies.

Numerous PRMT5i are currently in, or approaching, clinical trials without a clear understanding of the root cause of cancer vulnerability. Our study strongly validates the prevailing view that the effectiveness of PRMT5i reflects the induction of splicing defects in proliferating cells. To improve cancer cell selectivity, the field has already generated PRMT5i that specifically target MTAP-deficient cells. Our data argue that the perturbation of MTAP also impinges on splicing, and more specifically, DI biology. In furtherance of increased efficacy, we hypothesize that PRMT5i that more specifically target the PRMT5-splicing axis, alone or in combination with inhibitors that target DI-regulated complexes, might further increase potency, assuming a viable therapeutic window exists.

Limitations of the study

Our study establishes that DI regulation is a key driver of cellular vulnerability to PRMT5 inhibition, but some limitations should be noted. First, although we clearly show that the inactivation of the PRMT5-splicing axis is sufficient to cause lethality, we note that our study does not address potential contributions by other PRMT5 functions. Second, our analysis notes that DIs are by far the most prevalent and most conserved AS events, and their formation correlates with inviability, but we cannot rule out that rarer and less well-conserved exon-based AS events contribute in some settings. Third, the identification of a single cell line, HCT116, that was impervious to CLNS1A loss, showing no splicing defects, was completely unexpected. We did not observe any obvious compensatory mechanisms, and thus, the underlying molecular basis for this remains to be determined. Additionally, it remains unclear if this is the sole exception or whether other cell lines can also survive without CLNS1A. Finally, we hypothesize that widespread DI induction and the consequent impairment of numerous core proliferation-associated complexes provide redundancy to the system, thereby enabling the drift of DI identity across species. However, we have not specifically tested this model. We acknowledge these limitations and underscore the need for deeper mechanistic understanding to fully exploit the PRMT5-splicing axis as a therapeutic vulnerability.

Resource availability

Lead contact

Further information and requests should be directed to the lead contact, Jacqueline A. Lees (jalees@mit.edu).

Materials availability

All unique reagents generated in this study are available from the lead contact upon reasonable request.

Data and code availability

• •The raw data and processed data files from RNA sequencing experiments are available on GEO: GSE294247. Additional processed data files can be found on Mendeley Data: https://data.mendeley.com/datasets/23vhyymx49/1.
• •The original code, including the alternative splicing pipeline, can be found on Mendeley Data: https://data.mendeley.com/datasets/23vhyymx49/1.
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

We are especially grateful to our colleague, Phillip A. Sharp, for insightful discussions throughout this study, as well as the Koch Institute’s Robert A. Swanson (1969) Biotechnology Center for technical support, particularly the Flow Cytometry and Barbara K. Ostrom (1978) Integrated Genomics and Bioinformatics Core Facilities. This study was supported by the National Institutes of Health (NCI) Koch Institute Support (core) grant P30-CA014051 and awards to: C.E.F. (NIH NIGMS training grant award 5T32GM136540, and David H. Koch Graduate Fellowship), P.L.B. (NIH GM141544), and J.A.L. (NIH NCI-5P01CA042063-35).

Author contributions

C.E.F. and J.A.L. conceived this study, with input from P.L.B., and wrote the article. C.E.F conducted all of the experiments with assistance from N.A.O. for the phenotypic characterization of cell lines and G.J.S. for RNA-Seq sample preparation and analysis. P.L.B. generated the computational alternative splicing identification pipeline with assistance from A.S.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Rabbit anti-CLNS1A Polyclonal Antibody	Thermo-Fisher	Cat#PA5-109555; RRID: AB_2854966
Rabbit anti-HA Monoclonal Antibody	Cell Signaling Technology	Cat#3724; RRID: AB_1549585
Mouse anti-GAPDH Monoclonal Antibody	Thermo-Fisher	Cat#AM4300; RRID: AB_2536381
Rabbit anti-Symmetric Di-Methyl Arginine Motif Monoclonal Antibody Mix	Cell Signaling Technology	Cat#13222; RRID: AB_2714013
Mouse anti-HSP90 Monoclonal Antibody	BD Biosciences	Cat#610418; RRID: AB_397799
Rabbit anti-MTAP Polyclonal Antibody	Cell Signaling Technology	Cat#4158; RRID: AB_1904054
Mouse anti-FLAG M2 Monoclonal Antibody	Sigma-Aldrich	Cat#F3165; RRID: AB_259529
Rabbit anti-PRMT5 Polyclonal Antibody	Cell Signaling Technology	Cat#2252; RRID: AB_10694541
Rabbit anti-MEP50 Monoclonal Antibody	Cell Signaling Technology	Cat#2018; RRID: AB_2215723
Rabbit anti-PRMT9 Polyclonal Antibody	Sigma-Aldrich	Cat#HPA042537; RRID: AB_2678043
Mouse anti-SmB/B′/N (A-5) Monoclonal Antibody	Santa Cruz Biotechnology	Cat#sc-374009; RRID: AB_10945300
Rabbit anti-RIOK1 Polyclonal Antibody	Thermo Fisher	Cat#17222-1-AP; RRID: AB_2284990
Rabbit anti-eIF4E Monoclonal Antibody	Cell Signaling Technology	Cat#2067; RRID: AB_2097675
Rabbit anti-DNA2 Polyclonal Antibody	Thermo Fisher	Cat#PA5-77943; RRID: AB_2735727
Rabbit anti-RPS6KB1 (p70 S6 Kinase) Monoclonal Antibody	Cell Signaling Technology	Cat#34475; RRID: AB_2943679
Rabbit anti-TYK2 Monoclonal Antibody	Cell Signaling Technology	Cat#14193; RRID: AB_2798419
Rabbit anti-STAT5 beta Monoclonal Antibody	Thermo Fisher	Cat#MA5-57741; RRID: AB_3680766
Rabbit anti-SNRNP200 Polyclonal Antibody	Thermo Fisher	Cat#A303-453A; RRID: AB_10949362
Goat anti-Mouse IgG IRDye 680RD	LI-COR Biosciences	Cat#926-68070; RRID: AB_10956588
Goat anti-Mouse IgG IRDye 800CW	LI-COR Biosciences	Cat#926-32210; RRID: AB_621842
Goat anti-Rabbit IgG IRDye 680RD	LI-COR Biosciences	Cat#926-68071; RRID: AB_10956166
Goat anti-Rabbit IgG IRDye 800CW	LI-COR Biosciences	Cat#926-32211; RRID: AB_621843
Bacterial and virus strains
DH5α	Thermo-Fisher	Cat#18265017
Stbl3	Thermo-Fisher	Cat#C737303
Chemicals, peptides, and recombinant proteins
JNJ-64619178	MedChemExpress	Cat#HY-101564
EPZ015666	L. Garraway	N/A
AMG-193	Selleck Chemicals	Cat#E1600
TNG908	Selleck Chemicals	Cat#E1349
dTag-13	Tocris	Cat#6605
NU7441	Selleck Chemicals	Cat#S2638
Critical commercial assays
RNeasy Mini Kit	Qiagen	Cat#74106
NEBNext Ultra II Directional RNA Kit with Poly(A) Selection	New England Biolabs	Cat#E7760L
Pierce BCA Protein Assay Kit	Thermo-Fisher	Cat#23227
Pacific Blue™ Annexin V Apoptosis Detection Kit with PI	BioLegend	Cat#640928
CellTiter-Glo	Promega	Cat#G7570
Cyquant	Thermo Fisher	Cat#C7026
Deposited data
EPZ015666-treated U87 cells	Braun et al.2	GEO: GSE93813
Raw and analyzed RNA-Seq data	This study	GEO: GSE294247
Resource website for this publication	This study	Mendeley Data: https://data.mendeley.com/datasets/23vhyymx49/1
Experimental models: Cell lines
Human: U87	ATCC	Cat#HTB-14
Human: NCI-H23	ATCC	Cat#CRL-5800
Human: PANC-1	ATCC	Cat#CRL-1469
Human: CAL-51	DMZ	Cat#ACC 302
Human: HCT 116	ATCC	Cat#CCL-247
Human: HEK293T	ATCC	Cat#CRL-3216
Human: RPE-1	ATCC	Cat#CRL-4000
Mouse: KP1233	Dimitrova et al.40	N/A
Mouse: KP393T5	Winslow et al.41	N/A
Mouse: MDM1402	Li et al.	N/A
Oligonucleotides
See Table S1 for Custom Oligonucleotides	Sigma-Aldrich	N/A
siUPF1-1	Sigma-Aldrich	Cat#SASI_Hs01_00101017
siUPF1-2	Sigma-Aldrich	Cat#SASI_Hs01_00101018
Non-targeting siRNA control	Sigma-Aldrich	Cat#SIC001
Recombinant DNA
Plasmid: pCRIS-PITChv2-Puro-dTAG	Addgene	Cat#91793
Plasmid: pCRIS-PITChv2-BSD-dTAG	Addgene	Cat#91792
Plasmid: pX330A-1x2	Addgene	Cat#58766
Plasmid: pX330S-2-PITCh	Addgene	Cat#63670
Plasmid: lentiCRISPRv2 Puro	Addgene	Cat#98290
Plasmid: lentiCRISPRv2 Hygro	Addgene	Cat#98291
Plasmid: pcDNA3.1+	Thermo Fisher	Cat#V79020
Plasmid: pUltra Blast	This study	N/A
Plasmid: pUltra Hygro	This study	N/A
CLNS1A	IDT	N/A
CLNS1A Truncation	cDNA	N/A
PRMT5	cDNA	N/A
MEP50	cDNA	N/A
PRMT9	cDNA	N/A
Software and algorithms
STAR v2.5.3a	Dobin et al.42	https://github.com/alexdobin/STAR
Bowtie v1.2.3	Langmead et al.43	https://bowtie-bio.sourceforge.net
Bedtools v2.26.0	Quinlan and Hall44	https://bedtools.readthedocs.io
DEXSeq v1.48.0	Anders et al.45	https://bioconductor.org/packages/release/bioc/html/DEXSeq.html
RSEM v1.3.1	Li and Dewey46	https://github.com/deweylab/RSEM
DESeq2 v1.42.0	Love et al.47	https://bioconductor.org/packages/release/bioc/html/DESeq2.html
Samtools v1.3	Li et al.48	http://www.htslib.org
DESeq 1.22.1	Anders and Huber49	https://www.bioconductor.org/packages/2.10/bioc/html/DESeq.html
DAVID	Huang et al.50 & Sherman et al.51	https://davidbioinformatics.nih.gov
NicheNet v2.1.7	Browaeys et al.52	https://github.com/saeyslab/nichenetr
Comprehensive Resource of Mammalian Protein Complexes (CORUM) v4.1	Tsitsiridis et al.38	https://mips.helmholtz-muenchen.de/corum/
Image Lab v6.1	Bio-Rad	https://www.bio-rad.com/en-us/product/image-lab-software?ID=KRE6P5E8Z
FlowJo v10.10.0	BD Biosciences	https://www.flowjo.com
Alphafold2 ColabFold v1.5.5	Mirdita et al.53	https://colab.research.google.com/github/sokrypton/ColabFold/blob/main/batch/AlphaFold2_batch.ipynb
R v4.2.0 and v4.3.2		https://cran.r-project.org
PyMOL v2.5.2	Schrodinger	https://www.pymol.org
Prism v10.1.0	GraphPad	https://www.graphpad.com
Illustrator v28.2	Adobe

Experimental model and subject details

Cell lines

U87, H23, PANC1, CAL51, HCT116, HEK293T, and RPE-1 cells were obtained from ATCC. Murine KP LUAD lines KP1233 and KP393T5 cell lines are previously described.^40,41 Murine KP PDAC line MDM1402 is previously described.⁵⁴ All cells were cultured in Dulbecco’s modified Eagle’s media (DMEM) supplemented with 10% fetal bovine serum (FBS) and 1% penicillin-streptomycin, and cultured in 37°C at 5% CO₂. H23 cells were cultured in RPMI-1640 media supplemented with 10% fetal bovine serum (FBS) and 1% penicillin-streptomycin, and cultured in 37°C at 5% CO₂. Cells were routinely tested for mycoplasma and were found to be negative.

Method details

Chemical inhibitors

For PRMT5 inhibitors: JNJ-64619178 was obtained from MedChemExpress (HY-101564) EPZ015666 was obtained from L. Garraway, and AMG-193 (E1600) and TNG908 (E1349) were both obtained from Selleck Chemicals. The dTag inducer molecule dTag-13 was obtained from Tocris (6605).

Cell line generation

dTag-CLNS1A knock-in vector generation

dTag (pCRIS-PITChv2-Puro-dTAG and pCRIS-PITChv2-BSD-dTAG) and PITCh (pX330A-1x2 and pX330S-2-PITCh) vectors were obtained from Addgene (91793, 91792, 58766, 63670). Both Puro and BSD dTag plasmids were modified to harbor homology to the N-terminus of CLNS1A as previously described²⁶ using Gibson Assembly. The PITCh vector (pX330A-1x2) was modified to contain an sgRNA targeting the N-terminus of CLNS1A as previously described.⁵⁵ Finally, pX330A-CLNS1A/PITCh was generated using Golden Gate Assembly.

dTag-CLNS1A knock-in Cell line generation

Knock-in cell lines were generated as previously described.²⁶ Briefly, 10⁶ cells were plated in a 10-cm dish prior to transfection. The following day, cells were transfected with 2 μg pX330A-CLNS1A/PITCh and 1 μg pCRIS-PITChv2-Puro-dTAG-CLNS1A using Lipofectamine 2000 (ThermoFisher, 11668027). After 8 h, media was replaced and supplemented with 500 nM NU7441 (Selleck Chemicals, S2638) to inhibit NHEJ. CAL51 cells did not tolerate NU7441 treatment, so this addition was omitted for that cell line. After 48 h, cells were selected with puromycin. Following selection and outgrowth, cells underwent a second transfection with 2 μg pX330A-CLNS1A/PITCh and 1 μg pCRIS-PITChv2-BSD-dTAG-CLNS1A, using the same transfection workflow to enrich for homozygous knock-ins. Following blasticidin selection, cells were single cell cloned by dilution. CLNS1A knock-in was confirmed by gDNA PCR and western blot analysis.

Knockout cell line generation

LentiCRISPRv2 Puro and lentiCRISPRv2 Hygro were obtained from Addgene (98290 and 98291). Vectors were modified to contain an sgRNA targeting MTAP, CLNS1A or a non-targeting control. For cell line generation, cells were infected with lentivirus containing lentiCRISPRv2 or lentiCRISPRv2 Hygro with a sgRNA targeting the gene of interest, or a control sgRNA (sgCtrl). Virus was incubated with cells for 2 days, after which puromycin or hygromycin was added for selection. Following selection cells were single cell cloned by dilution. For MTAP cell lines, knockout was confirmed by western blot analysis. For CLNS1A knockout cell lines, the CLNS1A locus was amplified and subcloned by Gibson Assembly into pcDNA3.1+ (ThermoFisher, V79020). Following transformation into DH5α cells, colonies were sequenced by Sanger Sequencing. CLNS1A knockout was additionally confirmed by western blot analysis.

Overexpression cell line generation

Overexpression studies constructs utilized the pUltra system, which we modified to incorporate a blasticidin or hygromycin resistance cassette. Briefly, pUltra Hot was digested with XbaI and AgeI-HF, and DNA containing the antibiotic resistance cassette introduced via Gibson assembly, generating the pUltra Blast or pUltra Hygro vector. Full length human CLNS1A was ordered as a gBlock (IDT). CLNS1A Truncation, PRMT5, PRMT9, and MEP50 were cloned from cDNA. Genes were cloned into the pUltra backbone via Gibson Assembly, which also incorporated a 3xFLAG-tag or 3xHA-tag at the C-terminus of the gene. Cells were infected with lentivirus to overexpress the gene of interest, or an empty vector control. Virus was incubated with cells for 2 days, after which antibiotic was added to generate stable cell lines.

RNA extraction, sequencing and alternative splicing analysis

Whole cell lysate RNA extraction

Following 3-day treatment with either DMSO (vehicle), PRMT5i, or dTag-13, cells were harvested by scraping. RNA was purified using the RNeasy Mini Kit (Qiagen, 74106) according to manufacturer’s instructions, including the optional DNaseI addition to remove contaminating gDNA (Qiagen, 79254).

Cellular fractionation and RNA Isolation from cellular fractions

U87 cells were seeded in a 15 cm plate at a 1.5 × 10⁶ cell/plate. One day after plating, cells were treated with vehicle (DMSO) or 10 nM PRMT5i (JNJ-64619178). After 3 days in drug, cells were harvested via trypsinization, pelleted, and separated into nuclear and cytoplasmic fractions as previously described.⁵⁶ Briefly, cell pellets were resuspended in cytoplasmic lysis buffer [0.15% (v/v) NP40, 10 mM Tris (pH7), 150 mM NaCl] supplemented with 10 U SUPERase.IN (Thermo Fisher, AM2694) and EDTA-free protease inhibitor cocktail (MilliporeSigma, 11836170001). Cells were then incubated on ice for 5 min. Cell lysate was layered on top of sucrose buffer [10 mM Tris (pH 7), 150 mM NaCl, 25% (w/v) sucrose] supplemented with 10 U SUPERase.IN (Thermo Fisher, AM2694) and EDTA-free protease inhibitor cocktail (MilliporeSigma, 11836170001). Cells were then centrifuged at 16,000 × g for 10 min at 4°C. The supernatant contains the cytoplasmic fraction and was harvested for RNA extraction. The remaining pellet contains the nuclear fraction and was washed twice with 800 μL cold PBS, centrifuging at 16,000 × g for 1 min at 4°C after each wash. The pellet was resuspended in cold RIPA buffer [50 mM Tris (pH 8), 150 mM NaCl, 1% (v/v) NP-40, 0.5% sodium deoxycholate, and 0.1% (w/v) SDS] supplemented with 10 U SUPERase.IN (ThermoFisher, AM2694) and EDTA-free protease inhibitor cocktail (MilliporeSigma, 11836170001). Lysate was incubated for 30 min on ice, with vortexing every 10 min, and then pelleted at 16,000 × g for 10 min at 4°C. The supernatant contains the nuclear fraction and was harvested for RNA extraction.

The RNA from the nuclear and cytoplasmic fractions were extracted using TRIzol Reagent (Thermo Fisher, 15596026) according to manufacturer’s instructions. Briefly, to each aqueous fraction, 0.5 mL of isopropanol and 1 mL of TRIzol Reagent was added, followed by an incubation on ice for 10 min. Samples were then centrifuged for 10 min at 12,000 × g at 4°C, pellets were washed in 1 mL 75% ethanol, and centrifuged again for 5 min at 7,500xg at 4°C. Pellets were air dried for 10 min and resuspended in RNase-free water.

Library preparation and sequencing

For the human cell lines, RNA-Seq libraries were prepared using the NEBNext Ultra II Directional RNA Kit with Poly(A) Selection and sequenced using the NovaSeq sequencer, with 150 nucleotide paired-end reads. For the murine LUAD KP1233 cells, RNA-Seq libraries were prepared using the NEBNext Ultra II Directional RNA Kit with Poly(A) Selection and sequenced using the Element AVITI sequencer, with 75 nucleotide paired-end reads. For the murine PDAC and LUAD 393T5 cells, RNA-Seq libraries were prepared using the NEBNext Ultra II Directional RNA Kit with Poly(A) Selection and sequenced using the NextSeq500, with 75 nucleotide paired-end reads. Raw reads for the EPZ015666-treated U87 cells² were obtained from Gene Expression Omnibus (GSE93813).

Alternative splicing analysis

Raw RNA-seq reads were mapped using STAR aligner version 2.5.3a⁴² with the following parameters:

 STAR

 --runMode alignReads

 --runThreadN 16

 --sjdbOverhang 149

 --outReadsUnmapped Fastx

 --outSAMtype BAM SortedByCoordinate

 --outFilterMultimapNmax 20

 --outFilterMismatchNmax 999

 --outFilterMismatchNoverLmax 0.04

 --alignIntronMin 70

 --alignIntronMax 500000

 --alignMatesGapMax 500000

 --alignSJoverhangMin 8

 --alignSJDBoverhangMin 1

 --outSAMstrandField intronMotif

 --outFilterType BySJout

 --twopassMode Basic

Additionally, raw RNA-seq reads were mapped using Bowtie 1.2.3⁴³ using the following parameters:

 bowtie -n 1 -l 30 -e 70 -t -p 4 -m 1 -X 500 -S

Detained intron and alternative splicing event identification was performed as previously described,^2,21 with the following modifications: Mapped reads were filtered using Bedtools v2.26.0 to remove reads overlapping expressed repeats from the UCSC genome browser RNA repeats and Gencode hg38 or mm10 and coding exons annotated in the Gencode hg38 or mm10 annotations. Alternative and constitutive classifications were performed using custom Python scripts and are agnostic with regard to existing annotations other than known gene boundaries. Importantly, all cell lines were run in parallel to ensure that the annotations would be consistent between the cell lines.

Detained intron and alternative splicing event quantification (DEXSeq)

Detained intron and alternative splicing events were quantified using DEXSeq.⁴⁵ A genome annotation including all alternative splicing events and detained introns was generated using custom python scripts and python scripts that accompany DEXSeq. Reads were then quantified to this genome using the DEXSeq python script with the following parameters:

 -p yes -s reverse -f bam -r pos

Finally, differential splicing was quantified using the DEXSeq package within R. Alternative splicing events and detained introns were called as significant if p_adj < 0.05 and lowly expressed exons and introns were filtered out if their normalized counts were below 80.

Detained intron and coding isoform quantification (RSEM-DESeq2)

Transcript annotation was performed as previously described.² Briefly, using custom Python scripts, we derived a complete transcriptome annotation based on Gencode hg38 gene start and end boundaries and the location of polyA sites for data derived from the human cell lines, the genomic locations of detained introns (as determined above), and the genomic locations of all splice junctions (from the STAR alignment). For comparisons of isoforms containing detained introns to the consensus coding sequence, the annotation for each gene was set to identify the consensus isoform, i.e., the junctions defining each exon are the most frequently used junction detected within all of the combined samples. Each detained intron within a gene was then assigned to an additional transcript.

Gene and isoform expression

Isoform-level quantification was performed using RSEM v1.3.1⁴⁶ using the custom genomic annotation described above, as well as Gencode hg38 annotations with the parameters:

 rsem-calculate-expression --forward-prob 0 --no-bam-output --paired-end

Differential expression analysis of gene and isoform count tables was performed using DESeq2 v1.42.0.⁴⁷ Normalized count tables were obtained from DESeq2. Gene and isoform expression differences were considered significant if p_adj < 0.01.

Dice Similarity Score

Dice Similarity scores were calculated for each AS event type across all pairwise cell line combinations using the following equation:

$$\%\text{Similarity}=100\%\times \frac{2\times \text{Intersection}\text{of}\text{AS}\text{Events}}{{\#\text{AS}\text{Events}}_{\text{Cell}\text{Line}1}+{\#\text{AS}\text{Events}}_{\text{Cell}\text{Line}2}}$$

Gene ontology analysis

Gene Ontology (GO) analysis was conducted using DAVID.^50,51 Genes who contained a significant DI upon PRMT5i (p_adj < 0.05, log₂FC > 0) were run against the KW Biological Process category. A hypergeometric test was used to determine the probability of observing a certain overlap between the inputted gene list and a specific GO term, compared to the background gene list. For these comparisons, the background gene list was all genes in the human or mouse genome. To account for multiple hypothesis testing, we report the Benjamini-Hochberg-corrected FDR. Gene sets were considered significant if the corrected FDR < 0.01. Gene sets are displayed if they met the significance threshold in at least one line.

Inter-species gene name conversion and complex analysis

Lists of genes containing a significant DI upon PRMT5i (p_adj < 0.05, log₂FC > 0) for each cell line of interest was created. Mouse gene annotations were converted to human gene annotations using NicheNet v2.1.7.⁵² The Comprehensive Resource of Mammalian Protein Complexes (CORUM) v4.1³⁸ was used to match DI-containing genes to annotated human protein complex members. Gene Ontology (GO) analysis was conducted using DAVID as described above. The DI-containing genes for each protein complex were used to connect complexes to related biological processes. We directly compared PRMT5 DI-regulated genes, protein complexes, and biological processes between human and mouse data based on tissue of origin (Lung: H23 & T5, Pancreas: PANC1 & PDAC), as well as data derived from PRMT5 DI-containing genes shared between all human cell lines (n = 7) or all mouse cell lines (n = 3). Similarities were determined via a Dice Similarity Score as described above.

siRNA knockdown

U87 cells were seeded cells in a 6 well dish at a 10⁵ cells/well. One day after plating, cells were treated with vehicle (DMSO) or 10 nM PRMT5i (JNJ-64619178). After one day in drug, cells were transfected with siRNAs targeting UPF1 (Sigma-Aldrich, SASI_Hs01_00101017 and SASI_Hs01_00101018) or a non-targeting siRNA control (Sigma-Aldrich, SIC001) at a 10 nM concentration with 12 μL of MISSION siRNA Transfection Reagent (Sigma-Aldrich, S1452) for a final volume of 2.2 mL per well. Cells were harvested via trypsinization 48 h post transfection and RNA was extracted as described above.

Western Blot

Cells were harvested by scraping, washed twice with PBS, and then lysed in RIPA buffer [50 mM Tris pH 8.0, 150 mM NaCl, 1% NP-40, 0.5% sodium deoxycholate, and 0.1% sodium dodecyl sulfate (SDS)] supplemented with EDTA-free protease inhibitor cocktail (MilliporeSigma, 11836170001). For drug treatment experiments, cells were treated with DMSO, PRMT5i, or dTag-13 at the indicated concentration for the indicated duration, before harvesting. Protein concentration was determined using the Pierce BCA Protein Assay Kit (Thermo Fisher, 23227), then 30 μg of protein was mixed with 4× Laemmli buffer and loaded onto 12% SDS-polyacrylamide gels. Proteins were separated by SDS-PAGE, transferred onto nitrocellulose membranes, and probed overnight at 4°C in 5% non-fat milk in TBST (unless otherwise stated) with antibodies against the following: CLNS1A (Invitrogen, PA5-109555, 1:1000), HA (Cell Signaling Technology, 3724S, 1:5000), GAPDH (Thermo Fisher, AM4300, 1:5000), SDMA (Cell Signaling Technology, 13222S, 1:1000, 5% BSA), Hsp90 (BD, 610418, 1:1000), MTAP (Cell Signaling Technology, 4158S, 1:1000), FLAG (Sigma-Aldrich, F3165, 1:2000), PRMT5 (Cell Signaling Technology, 2252S, 1:1000), MEP50 (Cell Signaling Technology, 2018S, 1:1000), PRMT9 (Sigma-Aldrich, HPA042537, 1:1000), SmB/B’ (Santa Cruz Biotechnology, sc-374009, 1:1000), RIOK1 (Thermo Fisher, 17222-1-AP, 1:1000), eIF4E (Cell Signaling Technology, 2067S, 1:1000), DNA2 (Thermo Fisher, PA5-77943, 1:1000), RPS6KB1/p70 S6 Kinase (Cell Signaling Technologies, 34475T, 1:1000), TYK2 (Cell Signaling Technologies, 14193T, 1:1000), STAT5B (Thermo Fisher, MA5-57741, 1:1000), SNRNP200 (Thermo Fisher, A303-453A-T, 1:1000). Membranes were then incubated for 1 h at room temperature with anti-mouse (LI-COR Biosciences, 926–68070 and 926–32210, 1:10,000) or anti-rabbit (LI-COR Biosciences, 926–68071 and 926–32211, 1:10,000) conjugated to IRDye 800CW and fluorescence was measured using the ChemiDoc Imaging System (Bio-Rad). Blots were quantified using Image Lab 6.1 (Bio-Rad).

RT-PCR and Quantitative Real-Time PCR (qRT-PCR)

Total RNA was purified from cultured cells using the Rneasy Mini Kit (Qiagen, 74106) according to manufacturer’s instructions, including the optional DNaseI addition (Qiagen, 79254). RNA was converted to cDNA using SuperScriptIII following the manufacturer’s instructions (Thermo Fisher, 18080044). Quantitative Real-Time PCR (qRT-PCR) reactions were performed using FAST-SYBR Green (Thermo Fisher, 4385612) on a StepOnePlus Real-Time PCR System (Applied Biosystems). Data were analyzed using the ΔΔC_T method, and relative messenger RNA (mRNA) levels were normalized to GAPDH. RT-PCR reactions were performed using Phusion. Following amplification, PCR products were run on a 2% agarose gel, which was imaged using the ChemiDoc Imaging System (Bio-Rad).

Cell cycle analysis

Cells were plated in 10 cm dishes in quadruplicate, and then drugged with DMSO, PRMT5i, or dTag-13 at the indicated concentration for the indicated duration, before harvesting. Prior to collection, 5-ethynyl-2′-deoxyuridine (EdU, Life Technologies #E10187) was added to cells a final concentration of 10 μM for 30 min. Cells were collected by trypsinization, fixed in a 4% formaldehyde solution, and stored at 4° before processing. Cells were permeabilized in a 0.5% Triton X-100, 1% BSA solution for 15 min at room temperature, and then rinsed the permeabilization solution. Cells were stained in 500 μL of a modified Click reaction cocktail (1 mM CuSO₄, 100 mM Sodium Ascorbate, 1 μM Alexa Fluor 647 Azide [Thermo Fisher, A10277] in PBS) for 30 min. Cells were then rinsed with the permeabilization solution and then resuspended in the permeabilization solution supplemented with 1μg/mL DAPI (BD, 564907) and 100 μg/mL RNaseA (Qiagen, 1007885). Flow cytometry was conducted using the BD Celesta and the raw data was analyzed using FlowJo v10.10.0.

Apoptosis analysis

Cells were plated in 10 cm dishes in triplicate, and then drugged with DMSO, PRMT5i, or dTag-13 at the indicated concentration for the indicated duration, before harvesting. Cells were harvested by trypsinization and 0.25-1x10⁶ cells were stained with the Pacific Blue Annexin V Apoptosis Detection Kit with PI per the manufacturer’s instructions (BioLegend, 640928). Flow cytometry was conducted using the BD Celesta and the raw data was analyzed using FlowJo v10.10.0.

Dose-response curves

For each dose-response curve, cells were plated in a volume of 150 μL media into each well of a 96-well plate in the following densities: CAL51 – 250 cells/well, HEK293T – 100 cells/well, and HCT116 – 150 cells/well. Perimeter wells were filled with 200 μL of media to avoid edge effects. The following day, 50 μL of 4× the final drug concentration in media was added to each well in triplicate. For the double-drug dose-response curves, cells were plated in 100 μL media, and then were drugged. The following day, for each drug, 50 μL of 4× the final drug concentration in media was added to each well. After 6 days of treatment, CellTiter-Glo (Promega, G7570) was used per manufacturer’s instructions. Luminescence was measured with a Tecan M200 Pro plate reader. Background signal was determined by averaging the luminescence of cell-free wells, and was subtracted from all other viability measurements. Average signal for each dose triplicate was computed, and drug-treated cells were normalized to their DMSO control by division. To generate dose-response curves, normalized viability values for each sample were fit to the following nonlinear model:

$$\text{Viability}={\mathrm{A}}_{\inf }+\frac{{\mathrm{A}}_{0-}-{\mathrm{A}}_{\inf }}{{1+\left(\frac{\left[\text{drug}\right]}{\text{EC}50}\right)}^{\mathrm{n}}}$$

where A_inf and A₀ are the lowest and highest drug doses, respectively. Area under the curve was computed and used for statistical testing.

Proliferation assay

For each cell line, 150 cells were plated in each well. Two replicate plates were plated. The following day, a plate was washed and frozen to serve as counting control. After 6 days, the other plate was washed and frozen. Cyquant was used according to the manufacturer’s instructions (ThermoFisher, C7026). Fluorescence was measured with a Tecan M200 Pro plate reader. Background signal was determined by averaging the fluorescence of cell-free wells and was subtracted from all other measurements. Population doublings were calculated via log-transformation of normalized fluorescence.

Alphafold2 prediction

Full length or Truncated CLNS1A protein sequences were inputted into the Alphafold2 ColabFold v1.5.5 software,⁵³ using the default settings. Structure predictions were visualized using PyMOL v2.5.2.

Statistical analysis and plotting software

R v4.3.2, GraphPad Prism v10.1.0, and Adobe Illustrator v28.2 were used for generating and plotting figures.

Quantification and statistical analysis

R v4.3.2 and GraphPad Prism v10.1.0 were used to perform statistical analyses. Data are shown as mean with SD (error bars), unless otherwise stated. All statistical tests used are indicated in the figure legend and/or STAR Methods section. All statistical analyses were performed using Student’s t test unless the variance was significantly different between the samples (calculated with the F-test) in which case it is then specified as Welch’s t test. For statistical significance, p < 0.05 was considered statistically significant, and p = n.s. refers to not statistically significant (p > 0.05).

Published: June 20, 2025

Supplemental information

Table S1. Custom oligonucleotides used in this study, related to STAR Methods

(A) Oligonucleotides used for qPCR analyses.

(B) Oligonucleotides used for cloning.