Abstract

   Tumour cell plasticity is a major barrier to the efficacy of targeted
   cancer therapies but the mechanisms that mediate it are poorly
   understood. Here, we identify dysregulated RNA splicing as a key driver
   of tumour cell dedifferentiation in colorectal cancer (CRC). We find
   that Apc-deficient CRC cells have dysregulated RNA splicing machinery
   and exhibit global rewiring of RNA splicing. We show that the splicing
   factor SRSF1 controls the plasticity of tumour cells by controlling
   Kras splicing and is required for CRC invasion in a mouse model of
   carcinogenesis. SRSF1 expression maintains stemness in human CRC
   organoids and correlates with cancer stem cell marker expression in
   human tumours. Crucially, partial genetic downregulation of Srsf1 does
   not detrimentally affect normal tissue homeostasis, demonstrating that
   tumour cell plasticity can be differentially targeted. Thus, our
   findings link dysregulation of the RNA splicing machinery and control
   of tumour cell plasticity.

   Subject terms: Cancer models, Colon cancer, Oncogenes
     __________________________________________________________________

   The influence of mRNA splicing on colon cancer development and
   progression is unclear. In this study, the authors demonstrate that the
   SRSF1 splicing factor is essential to sustain the stem cell phenotype
   of WNT-activated colorectal cancers.

Introduction

   Colorectal cancer (CRC) is a leading cause of cancer related mortality
   responsible for around 900,000 deaths annually^[58]1. The emergence of
   novel targeted therapies for treatment has offered hope of improved
   outcome^[59]2, but these are generally beneficial to only small subsets
   of patients and emergence of resistance is common, even in initially
   responsive patients. For example, resistance to the highly specific
   EGFR inhibitor cetuximab, used for the treatment of some KRAS wild-type
   tumours, routinely emerges and survival benefit is limited^[60]3. In
   this case resistance emerges via multiple, distinct genetic and
   transcriptomic mechanisms that bypass the tumour’s requirement for
   signalling via EGFR, suggesting tumour heterogeneity may play a role in
   mediating therapy resistance^[61]3. Indeed, intra- and inter-tumoural
   mutational diversity within patients has been shown to result in CRC
   subpopulations having variable responses to therapeutic agents^[62]4,
   together suggesting that targeting single specific oncogenic driver
   genes may not deliver durable responses, even in highly stratified
   patient populations. Therefore, there is an urgent need to identify
   more general, clinically targetable biological mechanisms that exploit
   general cancer cell characteristics rather than the classically studied
   oncogenic driver genes that mediate them, such mutated APC, TP53 or
   RAS/MAPK^[63]5.

   Another key therapeutic challenge is overcoming the plasticity of stem
   cell fate, where stem cells can not only differentiate into more
   specialised cell types, but these differentiated cells have the ability
   to revert back to the stem cell phenotype following tissue
   damage^[64]6–[65]8. This is especially important when considering
   clinical treatment as this phenomenon has also been observed in tumours
   where differentiated tumour cells (non-Lgr5 + ) can be reactivated
   following cancer stem cell depletion and subsequently dedifferentiate
   and fuel tumour growth^[66]9,[67]10.

   Alternative RNA splicing increases proteome diversity and can induce
   radical effects on cellular phenotype including promoting
   carcinogenesis^[68]11–[69]13. For example, the alternatively spliced
   Rac1 isoform, Rac1b, is constitutively active and is upregulated in
   Wnt-driven CRC driving efficient tumourigenesis^[70]14–[71]16.
   Additionally, mRNA splicing is thought to be the rate-limiting step in
   generating functional transcripts and has been proposed as a potential
   therapeutic vulnerability in MYC driven breast cancer and
   lymphoma^[72]17–[73]19. Therefore, activity levels of the RNA
   spliceosome may be an important facilitator of cancer cell growth and
   its targeting may present a viable therapeutic strategy.

   Here, we aimed to identify universal therapeutic targets for CRC by
   determining the mechanisms important for cancer cell growth. We
   identified widespread dysregulation of RNA splicing factors and global
   reprogramming of RNA splice isoforms in a Wnt-driven animal model of
   CRC. Using 3D organoid cultures, we performed a synthetic lethal screen
   of splicing factors altered in CRC, identifying the splicing factor
   SRSF1 as a therapeutic vulnerability in CRC organoids. Moreover, we
   determined that SRSF1, by modulating the pro-proliferative Kras 4B
   isoform, alters cancer stem cell plasticity, the invasive potential of
   advanced tumours and the viability of primary CRC organoids derived
   from patients. Together, these data suggest that RNA splicing, and
   SRSF1 in particular, control the plasticity of tumour cells and is a
   viable therapeutic target in CRC.

Results

RNA splicing is dysregulated following Wnt hyperactivation

   To identify cellular processes activated at early stages of intestinal
   carcinogenesis, we deleted both copies of the CRC tumour suppressor
   gene Apc in mice using the intestinal specific villin-Cre^ERT2
   (villinCre^ERT2 Apc^fl/fl). Five days after tamoxifen induced-gene
   recombination, small intestinal tissue was dissected and activation of
   the Wnt hyperproliferative phenotype confirmed by immunohistochemical
   staining of BrdU and Wnt/β-catenin target gene upregulation
   (Fig. [74]S1a, [75]S1b and [76]S3a).

   Wild-type and Apc^fl/fl intestinal tissue was subjected to RNAseq
   (Supplementary Data [77]1) and biological pathway analysis (KEGG) used
   on differentially expressed transcripts to identify enriched biological
   processes. Compared to normal tissue, RNA processing events were the
   most significantly upregulated processes following Apc loss
   (Fig. [78]1a). Specifically, we found that 60 genes involved in RNA
   splicing were upregulated (Supplementary Data [79]2 and Fig. [80]S1c).
   This suggests there may be an increased requirement for RNA splicing
   following CRC initiation.

Fig. 1. RNA splicing is dysregulated following Wnt hyperactivation.

   [81]Fig. 1
   [82]Open in a new tab

   a RNAseq performed on wild-type and Apc^fl/fl mouse small intestines
   isolated 5 days post tamoxifen inductions and sequences were subjected
   to biological pathway enrichment analysis. b Alternative splicing
   events following Apc-deletion compared to wild-type. c RT-PCR
   validation of a selection of alternative splicing events. d
   Representative images of wild-type and Apc^fl/fl intestinal organoids
   treated with the splicing inhibitor pladienolide B. Scale bar 250 µm. e
   Half maximal inhibitory concentration (IC[50]) of pladienolide B in
   intestinal organoids, n = 3 independent experiments. f Representative
   images of organoids derived from normal human colon and 2 colonic
   tumours treated with 1 nM pladienolide B. Scale bar 1000 µm. g
   Quantification of viability assays following pladienolide B treatment,
   n = 3 independent experiments. h Schematic explanation of CRISPR screen
   design. i Guide RNA enrichment or depletion in surviving Apc^fl/fl
   organoids compared to wild-type following completion of CRISPR screen.
   Data represented as mean and error bars SD. Data analysed with
   two-tailed, unpaired t-test, p values are indicated in figure panels.
   See also Fig. [83]S1.

   To ascertain if splicing factor upregulation correlated with changes in
   RNA splicing events, we conducted differential splicing analysis using
   SUPPA2^[84]20 and identified 1,661 alternative splicing events
   following Apc loss (p < 0.05) (Fig. [85]1b and Supplementary
   Data [86]3) indicating a global rewiring of alternative RNA splicing
   upon Wnt activation. The largest category of splicing event detected
   using this method was alternative first exon (46.4%) which could be an
   indication of alternative transcriptional regulation. To investigate
   this, we compared these events to differentially expressed genes
   identified in Supplementary Data [87]1. 140/631 alternative first exon
   events were also differentially expressed indicating the majority of
   these events are not linked to transcriptional changes and likely
   represent alternative use of first exons and/or differential promoter
   usage (Fig. [88]S1d). SUPPA2 analysis detects annotated splicing events
   but is unable to detect unannotated events, limiting our analysis. To
   investigate whether loss of Apc also induces novel splicing events we
   carried out rMATS analysis, which can also detect novel splicing
   events^[89]21. Interestingly, this analysis identified a large number
   of splicing events utilising novel splice sites indicative of
   widespread splicing alterations (Supplementary Data [90]4 and
   Fig. [91]S1e). Splicing changes were detected and validated in numerous
   cancer-related genes, including pro-tumourigenic Cd44 variant
   isoforms^[92]22, Fgfr2 and eIF4A2 (Fig. [93]1c and S[94]1f–S[95]1h).
   Notably, we found a significant change in the splicing ratio of the
   oncogene Kras (Fig. [96]1c and S[97]1f) favouring the Kras4B isoform
   over the Kras4A isoform. These isoforms have been previously reported
   to have pro-proliferative and proapoptotic effects respectively
   (Figure [98]S1g)^[99]23,[100]24. We next analysed previously published
   RNA splicing analysis of The Cancer Genome Atlas (TCGA) data^[101]25 to
   determine whether the splicing events identified in our mouse model are
   found in human CRC samples. Of the top 100 identified skipped exon
   splicing events, 18 showed the same alteration in human CRC including
   KRAS, CD44 and EIF4A2 (Supplementary Data [102]5 and Fig. [103]S1i). In
   summary, we identified widespread alterations of alternative splicing
   events following Apc loss, several of which are associated with cancer
   progression and a proportion of which are found in human CRC samples.

   As RNA splicing activity is dysregulated in Apc^fl/fl intestine, we
   investigated the effects of global splicing inhibition on oncogenic
   growth compared to wild-type. 3D intestinal organoids from normal or
   Apc^fl/fl intestine were exposed to the highly potent splicing
   inhibitor pladienolide B^[104]26 (Fig. [105]1d). At low concentrations
   of the drug, Apc^fl/fl organoids showed signs of cell death
   (Fig. [106]1d) and a drop in cell viability (Fig. [107]1e and
   S[108]1j), whereas wild-type organoids were more resilient. We found a
   significant difference in the IC[50] of pladienolide B for wild-type
   and Apc^fl/fl intestinal organoids (Fig. [109]1e). We extended this
   analysis to organoids derived from human patient material. We found
   that CRC tumour organoids exhibited significant loss of viability upon
   pladienolide B treatment but organoids derived from normal tissue did
   not (Fig. [110]1f, g). These results support a proof-of-concept that
   the spliceosome may be a therapeutic vulnerability in Apc-deficient
   cells.

The splicing factor SRSF1 is critical for Apc-deficient growth

   To identify which splicing factors are required for Apc-deficient cell
   growth, we conducted a splicing factor-targeted, synthetic lethal
   CRISPR screen comparing wild-type to Apc^fl/fl intestinal organoids. We
   generated a guide RNA (gRNA) library targeting the 60 splicing factor
   genes upregulated following Apc deletion (Supplementary Data [111]2 and
   [112]6) and delivered them by lentiviral transduction into either
   wild-type-Cas9 or Apc^fl/fl-Cas9 intestinal organoids (Fig. [113]1h) at
   low multiplicity of infection (MOI). Surviving organoids were harvested
   after 7 days of antibiotic selection and gRNA sequences in surviving
   clones analysed via deep sequencing. A comparison of the gRNAs present
   in surviving wild-type-Cas9 and Apc^fl/fl-Cas9 organoids showed that
   guide RNAs targeting Ddx10 and Srsf1 were significantly
   underrepresented in Apc^fl/fl-Cas9 population (Fig. [114]1i and
   Supplementary Data [115]7). This implies deletion of these genes is
   less well tolerated in the context of Apc loss compared to normal
   growth conditions. To confirm these findings, we CRISPR-targeted Ddx10
   and Srsf1 individually in pooled gene-edited populations of both
   wild-type and Apc^fl/fl organoids. Targeting Ddx10 or Srsf1 reduced
   Apc^fl/fl organoid growth, but only Srsf1 showed a significant
   reduction in viability in Apc^fl/fl organoids compared to wild-type
   organoids where Srsf1 is deleted (Fig. [116]2a, [117]b and S[118]2a,
   S[119]2b). Therefore, we chose to investigate the function of Srsf1 in
   further detail.

Fig. 2. The splicing factor SRSF1 is critical for Apc-deficient growth.

   [120]Fig. 2
   [121]Open in a new tab

   a Images of clonogenicity assays in wild-type and Apc^fl/fl Cas9
   organoids, treated with the indicated gRNA. Scale bar 250 µm. b
   Clonogenicity quantification of organoids treated with indicated gRNA.
   c Images of doxycycline (dox)-induced (Tet-On) shRNAs against Renilla
   or Srsf1 in wild-type intestinal organoids. Red fluorescent protein
   (RFP) is co-expressed with the shRNA following addition of dox. White
   arrows indicate a tracked organoid over the time-course following dox
   treatment. d Organoid size of tracked wild-type intestinal organoids. e
   Images of tracked Apc^fl/fl organoids treated with the indicated Tet-On
   shRNA following addition of dox. f Size quantification of Apc^fl/fl
   organoids with the indicated shRNA, +/− addition of dox. Statistical
   test indicated for shSrsf1 + /− dox. g AKP organoids with the indicated
   Tet-On shRNA following addition of dox. h Quantification of AKP
   organoid size. For (c, e and g) scale bar is 250 µm. All images shown
   are representative. Data shown as mean and error bars are SD (b) and
   SEM (d, f, h). All data are n = 3 independent experiments and analysed
   with two-tailed, unpaired t-tests, with (d, f, h) using Bonferroni
   post-hoc correction (alpha = 0.05), p values are indicated in figure
   panels. See also Fig. [122]S2.

   Notably, despite showing significantly reduced viability in Apc^fl/fl
   organoids, deletion of Srsf1 by CRISPR also reduced viability in
   wild-type organoids. However, previous reports have demonstrated that
   modest reductions in Srsf1 expression levels are well tolerated in mice
   suggesting a potential therapeutic window for targeting Srsf1^[123]27.
   As therapeutic targeting does not lead to target inhibition equivalent
   to genetic knockout we aimed to phenocopy clinical intervention more
   accurately using doxycycline-inducible short hairpin RNA (shRNA)
   expression. This also allowed us to deplete Srsf1 levels in fully
   developed organoids of increasing genetic complexity and
   aggressiveness. Expression of the Srsf1 shRNA reduced SRSF1 levels in
   wild-type, Apc-deficient and AKP (Apc;^∆/∆ Kras; ^G12D/+ Trp53^∆/∆)
   organoids (Figures [124]S2c, [125]S2d). Validating the findings with
   CRISPR deletion, knockdown of Srsf1 significantly reduced the growth of
   fully mature Apc-deficient and AKP organoids (Fig. [126]2e–[127]h).
   However, we observed no impact of Srsf1 depletion on wild-type organoid
   growth (Fig. [128]2c, [129]d). Together, these data suggest a potential
   therapeutic vulnerability of CRC cells to depletion of Srsf1.

SRSF1 is required for epithelial cell hyperproliferation

   To investigate the effects of targeting Srsf1 in vivo, we combined Apc
   deletion in our mouse model, where SRSF1 was upregulated
   (Fig. [130]S3a), with a deletion of a single copy of Srsf1
   (villinCre^ERT2 Apc;^fl/fl Srsf1^fl/+) (Figs. [131]S3b–[132]S3c and
   [133]S3h). Homozygous deletion of Srsf1 led to gut toxicity with
   elevated levels of intestinal cell apoptosis (Fig. [134]S3k) so
   heterozygous deletion of Srsf1 was selected to phenocopy a more
   clinically relevant scenario where impaired SRSF1 activity might be
   achieved rather than complete deletion.

   A 30–50% reduction in Srsf1 levels (Figure [135]S3h) had no
   proliferative or morphological effect in wild-type intestinal tissue
   (Fig. [136]3a, [137]b and S[138]3d–S[139]3e). However, we observed a
   significant reduction in the number of BrdU-labeled (Fig. [140]3a,
   [141]b) and Ki-67-positive cells (Figs. [142]S3d, [143]S3e) upon
   deletion of one copy of Srsf1 in the context of Apc loss. Interestingly
   we found that the proliferative reduction seen in villinCre^ERT2
   Apc;^fl/fl Srsf1^fl/+ was observed predominantly in the intestinal
   villus zone (Fig. [144]3a–[145]b and S[146]3f) and this was independent
   of variations in crypt-villus size or levels of apoptosis
   (Figures [147]S3g and [148]S3i-[149]S3j). We investigated this
   proliferative reduction further by harvesting villi from either
   Apc^fl/fl or Apc;^fl/fl Srsf1^fl/+ tissue and generated villi-derived
   3D organoids in vitro. The clonogenic capacity of these purified
   Apc;^fl/fl Srsf1^fl/+ epithelial cells was strongly impaired
   (Fig. [150]3c–[151]e) demonstrating a persistence of this phenotype in
   a stromal-independent context. To determine in more detail whether
   Srsf1 depletion affected normal intestinal stem cell homeostasis we
   generated inducible vil-Cre-ERT2 WT and Srsf1^fl/+ mice carrying the
   Lgr5GFP-CRE^ERT2 transgene^[152]28. Following tamoxifen-induced gene
   deletion, we analysed the LGR5 + ISC population using GFP IHC to
   visualise LGR5-GFP expression and flow cytometry to determine the
   percentage of GFP positive cells. Both analyses showed that Srsf1
   depletion had no impact on the LGR5 ISC population (Figure [153]S3l–n).
   In addition, sorted single LGR5-GFP positive cells had the same
   capacity to form organoids when plated in vitro and showed no changes
   stem cell marker gene expression following Srsf1 depletion
   (Figure [154]S3o–q). Together, these data indicate Srsf1 depletion does
   not affect normal LGR5 + intestinal stem cell homeostasis but
   significantly impairs the growth of Apc deficient cells.

Fig. 3. SRSF1 is required for epithelial cell hyperproliferation and affects
cell-type plasticity.

   [155]Fig. 3
   [156]Open in a new tab

   a Representative images of BrdU-stained mouse small intestines with the
   indicated genotypes, 5-days after tamoxifen induction. Letters ‘c’ and
   ‘v’ indicate demarcated crypt and villus compartments respectively and
   the red line highlights the proliferative zone. Scale bar 250 µm. b
   Quantification of proliferative cells in the crypts and villi of the
   indicated genotypes, n = 4 (WT) vs 3 (Srsf1^fl/+) vs 5 (Apc^fl/fl) vs 4
   (Apc;^fl/fl Srsf1^fl/+) biologically independent mice. c Representative
   images of clonogenicity assays using villi-derived intestinal organoids
   of indicated genotypes. Scale bar 500 µm. d Quantification of
   villi-derived organoid clonogenicity, n = 10 vs 7 independent
   experiments. e qPCR gene expression of Srsf1 in organoids normalised to
   β-actin, n = 3 vs 4 independent experiments. f qPCR quantification of
   upregulated late transit-amplifying (differentiated) cell markers in
   Apc;^fl/fl Srsf1^fl/+ mouse intestinal tissue, relative to Apc^fl/fl
   intestinal tissue, using Gapdh loading control, n = 3 vs 3 biologically
   independent mice. g Prox1 stem cell marker qPCR quantification in mouse
   small intestines of indicated genotypes, normalised to Gapdh, n = 3 vs
   3 biologically independent mice. h qPCR-derived ratio of Kras splicing
   isoforms in the indicated genotypes, n = 7 (WT) vs 3 (Srsf1^fl/+) vs 8
   (Apc^fl/fl) vs 4 (Apc;^fl/fl Srsf1^fl/+) biologically independent mice.
   i Images of organoids of the indicated genotype, treated with control
   or Srsf1 shRNA. AKP (Apc;^∆/∆ Kras;^G12D/+ Trp53^∆/∆), KPN
   (Kras;^G12D/+ Trp53;^fl/fl Rosa26^N1icd/+). Scale bar 500 µm. j
   Relative clonogenicity and k size of surviving organoids following
   control or Srsf1 shRNA treatment, n = 3 vs 3 independent experiments.
   Data in bar charts are represented as mean and error bars are SD with
   data analysed with two-tailed, unpaired t-tests, p values are indicated
   in figure panels. All biological replicates are shown as individual
   value plots. See also Fig. [157]S3.

Oncogenic SRSF1 levels affect cell-type plasticity

   To investigate whether the reduced proliferative phenotype observed in
   the intestinal mucosa of Apc;^fl/fl Srsf1^fl/+ was due to changes in
   intestinal cell identity, we performed RNAseq analysis of Apc^fl/fl and
   Apc;^fl/fl Srsf1^fl/+ intestines. We compared the Apc;^fl/fl Srsf1^fl/+
   transcriptome (Supplementary Data [158]8) with previously defined
   signatures of intestinal cell types^[159]29,[160]30 using gene set
   enrichment analysis (GSEA). Interestingly, we found a significant
   overlap between genes associated with late transit-amplifying and
   differentiated enterocytes and gene overexpressed in Apc;^fl/fl
   Srsf1^fl/+ intestinal cells (Figs. [161]S3r–[162]S3u) (Supplementary
   Data [163]9). Numerous differentiated cell marker genes were
   overexpressed in Apc;^fl/fl Srsf1^fl/+ compared to Apc^fl/fl, including
   Slc13a2 and Apoc2 (Fig. [164]3f). We also found that expression of the
   Wnt target gene and cancer stem cell marker Prox1 was decreased in
   Apc;^fl/fl Srsf1^fl/+ (Fig. [165]3g). Interestingly, the increase in
   the splicing isoform ratio of Kras4B/Kras4A observed after Apc deletion
   (Fig. [166]1c and S[167]1f) was reversed upon simultaneous deletion of
   Srsf1 (Fig. [168]3h). This was not due to alterations in nuclear
   localisation of β-catenin, a previously described function of Srsf1, as
   demonstrated by IHC analysis (Fig. [169]S3v, w)^[170]31. These data
   suggest that Srsf1 promotes a less differentiated, stem cell-like
   phenotype following Apc deletion. To investigate the generality of
   these findings we depleted Srsf1 using shRNA in a number of
   tumourigenic organoid models (Fig. [171]S3x–z)^[172]32,[173]33 and
   determined stem cell function using clonogenicitiy assays. Validating
   our in vivo experiments, Srsf1 knockdown led to a significant reduction
   in clonogenicity, growth rate and viability of Apc^fl/fl organoids
   suggesting reduced stem cell function (Fig. [174]3i–[175]k and
   S[176]3x–z). Notably, the same results were observed following Srsf1
   knockdown in two mouse CRC organoid lines that model late-stage,
   metastatic disease (Fig. [177]3i–[178]k and S[179]3x–z). The expression
   of the stem cell marker Lgr5 was also reduced in Srsf1 depleted
   organoid lines (Fig. [180]S2d). Together, these data demonstrate that
   Srsf1 depletion reduces stem cell properties and results in colorectal
   cancer cells adopting a more differentiated cell transcriptional
   phenotype.

SRSF1 facilitates intestinal cell dedifferentiation

   Due to these findings we hypothesised that SRSF1 is required for
   differentiated enterocytes to dedifferentiate and acquire stem cell
   properties. Activation of KRAS or IκB kinase (IKK) combined with Apc
   loss has been shown to induce dedifferentiation of villus
   enterocytes^[181]34. Therefore, we generated mice carrying
   villinCre^ERT2 Apc;^fl/fl Kras^G12D or villinCre^ERT2 Apc;^fl/fl
   Kras;^G12D Srsf1^fl/+ alleles and induced gene recombination with
   tamoxifen. As previously described^[182]34,[183]35, a shorter 3-day
   post tamoxifen time point was employed so that dedifferentiating cells
   in the villus could be studied alone, without contamination from
   recombined cells migrating from the crypts. After 3-days, villi from
   the small intestines were dissected, digested to single cells and
   plated to determine dedifferentiation efficiency (Fig. [184]S4a). We
   compared enterocyte dedifferentiation between villinCre^ERT2 Apc;^fl/fl
   Kras^G12D and villinCre^ERT2 Apc;^fl/fl Kras;^G12D Srsf1^fl/+ mice and
   found that depletion of Srsf1 led to a significant reduction in the
   proportion of organoids formed (Fig. [185]4a, b). The clonogenic
   capacity of the organoid lines derived from these dedifferentiated
   cells was also reduced in the Srsf1^fl/+ genotype (Fig. [186]4c)
   suggesting that the acquisition of stem cell properties was impaired.
   Intriguingly, Kras isoform splicing was still modulated by SRSF1 even
   in the context of constitutively active KRAS signalling
   (Figure [187]S4b).

Fig. 4. SRSF1 facilitates intestinal cell dedifferentiation.

   [188]Fig. 4
   [189]Open in a new tab

   a Dedifferentiation assay using AK (Apc;^fl/fl Kras^G12D) cells
   directly purified from mouse intestinal villi from the indicated
   genotypes isolated 3 days postinduction, and quantification of their
   ability to form organoid clones, n = 4 vs 4 biologically independent
   mice. b Images from AK dedifferentiation assays. c Clonogenicity assays
   on organoids derived from and passaged from dedifferentiated
   villi-derived AK cells, n = 5 vs 5 independent experiments. d An
   alternative dedifferentiation assay using AI (Apc;^fl/fl IKK2^ca) cells
   directly purified from mouse intestinal villi from the indicated
   genotypes, and quantification of their ability to form organoid clones,
   n = 3 vs 3 biologically independent mice. e Images from AI
   dedifferentiation assays. f Clonogenicity assays on organoids derived
   from and passaged from dedifferentiated villi-derived AI cells, n = 5
   vs 6 independent experiments. g Schematic of experimental strategy to
   isolate differentiated villus cells 2 days post tamoxifen induction. h
   Images from EPHB2 –ve cell sorted AK dedifferentiation assays. i
   Quantification of colony-forming capacity of differentiated cells,
   n = 3 vs 3 biologically independent mice. j Overview of FACs plots used
   to gate for EPHB2^high and EPHB2^low cells. k Representative images of
   colonies formed from EPHB2^low cells. GFP indicates Apc knockdown and
   RFP indicates Renilla control or Srsf1 knockdown. l Quantification of
   colony formation, n = 6 vs 6 independent experiments. Scale bars in b,
   e and h are 500 µm and in K 1000 µm. All images are representative.
   Data in bar charts are represented as mean and error bars are SD with
   data analysed with two-tailed, unpaired t-tests, p values are indicated
   in figure panels. All biological replicates are shown as individual
   value plots. See also Fig. [190]S4.

   To confirm these observations, we employed a different
   dedifferentiation model driven by hyperactive NF-κB signalling^[191]34,
   (villinCre^ERT2 Apc^fl/fl IKK2^ca) and subjected these mice to the same
   experimental procedure. Again, we found a similar impairment in
   dedifferentiation capacity of villi cells and reduced clonogenicity in
   derived organoids following heterozygous Srsf1 deletion
   (Fig. [192]4d–[193]f).

   To rule out the possibility that these clonogenic cells derived from
   crypt cells migrating into the villi we sorted differentiated villus
   cells based on expression of EPHB2. EPHB2 marks a gradient of stem and
   progenitor crypt cells with expression absent in differentiated villus
   cells^[194]30. We sorted EPHB2 negative epithelial cells from wild-type
   crypts and confirmed that they did not have clonogenic capacity
   (Figure [195]S4c–e). We next utilised this sorting strategy to obtain
   differentiated villus cells from induced villinCre^ERT2 Apc;^fl/fl
   Kras^G12D and villinCre^ERT2 Apc;^fl/fl Kras;^G12D Srsf1^fl/+ mice
   (Fig. [196]4g). EPHB2 negative differentiated Apc;^fl/fl Kras^G12D
   cells were able to form colonies, indicative of cellular
   dedifferentiation. This ability was significantly impaired following
   Srsf1 depletion, demonstrating the requirement of Srsf1 for the
   acquisition of stem cell properties (Fig. [197]4h, i). To confirm these
   findings we utilised an analogous model where doxycycline-inducible
   shRNA depletion of Apc is used to activate Wnt signalling in Kras^G12D
   mutant colonic organoids^[198]36. In the absence of doxycycline,
   organoid growth is dependent on media supplementation with Wnt3a and
   R-spondin. In addition, EPHB2 negative cells have poor clonogenic
   capacity indicative of a differentiated cell phenotype
   (Figure [199]S4f). We used shRNA to deplete Srsf1 in these organoids
   and following doxycycline induction FACS sorted for EPHB2 negative,
   differentiated cells. Again, depletion of Srsf1 significantly impaired
   the ability of such cells to form colonies (Fig. [200]4j–l). In
   addition, organoids derived from Srsf1 depleted differentiated cells
   expressed significantly higher levels of the differentiated cell
   markers Muc2 and Krt20 (Figure [201]S4g, [202]S4h). Together, these
   data demonstrate a requirement for Srsf1 expression in mediating
   intestinal cell plasticity.

   To determine whether Srsf1 also drives increased stem cell properties,
   we overexpressed wild-type Srsf1 (Srsf1^WT) or Srsf1 with mutations in
   the second RNA recognition motif (RRM2) leading to impaired RNA
   binding^[203]37, (Srsf1^D136A,K138A) in Apc^fl/fl organoids
   (Figure [204]S4i). Consistent with our deletion studies, Srsf1^WT
   overexpression led to increased organoid growth, clonogenicity and
   increased expression of the stem cell marker Lgr5 (Figures [205]S4i–l).
   By contrast the Srsf1 mutant did not affect organoid growth,
   demonstrating that the RNA-binding ability of SRSF1 is necessary for
   its dedifferentiation activity. Based on these findings, we conclude
   that SRFS1 mediates intestinal cell dedifferentiation and the
   acquisition of stem cell properties.

Wnt-induced SRSF1 levels enforce splicing dysregulation

   To ascertain the extent to which Srsf1 contributes to Wnt-driven
   splicing dysregulation and identify potential splicing changes that
   mediate oncogenesis, we conducted differential splicing analysis on
   Apc^fl/fl and Apc;^fl/fl Srsf1^fl/+ intestinal tissue, in which we
   observed reduced proliferation and stem cell activity, using SUPPA2. We
   identified 577 alternative splicing events (p < 0.05) (Fig. [206]5a and
   Supplementary Data [207]10). The majority of these were exon skipping
   alternative splicing events (30.5%), supporting previous work using in
   vitro models showing that SR proteins promote splice site selection
   through exonic splicing enhancer recognition^[208]38. We then
   determined the proportion of unique dysregulated alternative splicing
   events in Apc^fl/fl that were SRSF1-dependent. We compared the
   alternative splicing events identified from our wild-type/Apc^fl/fl
   analysis (Fig. [209]1b and Supplementary Data [210]3) with the
   alternative splicing events found in Apc^fl/fl/Apc;^fl/fl Srsf1^fl/+
   (Supplementary Data [211]10). There was a significant enrichment of
   alternative splicing events from Apc^fl/fl/Apc;^fl/fl Srsf1^fl/+ that
   were also found in our wild-type/Apc^fl/fl dataset (p < 1e^−5)
   (Fig. [212]5b). In addition, a significant proportion of the
   alternative splicing events occurring after Apc deletion were reverted
   upon deletion of Srsf1^fl/+ (127 alternative splicing events with
   discordant dPSIs, p = 0.049) (Supplementary Data [213]11). Again, rMATS
   analysis identified a large proportion of potentially novel splicing
   events suggesting widespread changes in alternative splicing mediated
   by SRSF1 (Figure [214]S5a and Supplementary Data [215]12).

Fig. 5. Wnt-induced SRSF1 levels enforce splicing dysregulation.

   [216]Fig. 5
   [217]Open in a new tab

   a Alternative splicing events in Apc;^fl/fl Srsf1^fl/+ compared to
   Apc^fl/fl in small intestine isolated 5 days postinduction. b Gene set
   enrichment analysis showing the intersecting alternative splicing
   events discovered in the wild-type versus Apc;^fl/fl Srsf1^fl/+ dataset
   compared to the Apc^fl/fl versus Apc;^fl/fl Srsf1^fl/+ dataset. c
   RT-PCR validation of a selection of alternative splicing events and d
   quantification, n = 3 vs 3 biologically independent mice. e Images of
   wild-type, Apc^fl/fl and KPN organoids treated with deltarasin or
   control vehicle DMSO. Scale bar 500 µm. f Relative cell viability
   following deltarasin treatment in wild-type, Apc^fl/fl and KPN
   organoids, n = 3 vs 3 (Wt and Apc^fl/fl) and 3 vs 4 (KPN) independent
   experiments. g Images showing Apc^fl/fl organoids following treatment
   with a doxycycline (dox) inducible Srsrf1 shRNA + /− dox, and with GFP
   or Kras4b-GFP overexpression. RFP is co-expressed with the shRNA
   following addition of dox. Scale bar 500 µm. h Quantification of
   organoid size under the stated conditions, n = 3 vs 3 vs 3 vs 3
   independent experiments. All images are representative. Data in bar
   charts are represented as mean and error bars are SD with data analysed
   with two-tailed, unpaired t-tests, p values are indicated in figure
   panels. All biological replicates are shown as individual value plots.
   See also Fig. [218]S5.

   We validated several of the alternative splicing events discovered from
   our differential splicing analysis (Fig. [219]5c, [220]d).
   Additionally, isoform level changes in Kras splicing were detected and
   validated confirming that the Wnt-induced increase in the
   pro-proliferative Kras4b isoform over the pro-apoptotic Kras4a isoform
   was reversed upon Srsf1 depletion. SRSF1-dependent changes in Kras
   isoform splicing levels were also present in multiple other models
   (Figs. [221]3h, S[222]4b, S[223]6j and S[224]6m). To determine whether
   these transcripts were direct targets of SRSF1 we carried out RNA
   immunoprecipitation experiments. By pulling down SRSF1 from the CMT93
   mouse colorectal cancer cells line and carrying out qRT-PCR we
   identified significant binding of Srsf1 to several of the alternatively
   spliced transcripts identified by RNAseq (Fig. [225]S5b, c). In
   addition, alternative splicing events (events identified via SUPPA2
   analysis, not including AF events) in our Apc^fl/fl/Apc;^fl/fl
   Srsf1^fl/+ dataset, which were repressed following Srsf1 depletion,
   were enriched for SRSF1 binding motifs, together suggesting loss of
   direct, functional SRSF1 binding to these transcripts led to the
   splicing changes we observed (Fig. [226]S5d). Besides the
   well-characterised role of SRSF1 in pre-mRNA splicing in the nucleus,
   this shuttling SR protein also has post-splicing functions, which
   include mRNA export, mRNA translation and nonsense-mediated
   decay^[227]12,[228]39. We utilised a novel mouse model carrying a
   nucleo-cytoplasmic shuttling defective Srsf1 mutant to determine
   whether non-nuclear functions of SRSF1 could explain the phenotypes we
   observed^[229]40. Apc deficient organoids carrying a single allele of
   Srsf1-NRS (which is unable to shuttle to the cytoplasm) had no defect
   in clonogenic capacity (Fig. [230]S5e–g). This is unlike organoids
   carrying heterozygous Srsf1 deletion (Fig. [231]3c, e) demonstrating
   that the cytoplasmic function of Srsf1 does not explain the phenotypes
   we observe in this model. Together, these data suggest the primary
   cause of the effects we observed are due to nuclear (presumably
   splicing) functions of SRSF1.

   To functionally determine whether Kras4b plays a role in colorectal
   tumourigenesis we treated Apc^fl/fl and metastatic KPN (Kras;^G12D/+
   Trp53;^fl/fl Rosa26^N1icd/+) organoids with the KRAS4B specific
   inhibitor deltarasin (Fig. [232]S5h)^[233]41,[234]42. Deltarasin is a
   small molecule inhibitor of the KRAS4B -PDEδ interaction, and PDEδ has
   been shown to chaperone and recruit KRAS4B to the plasma membrane, but
   is not required for KRAS4A recruitment. Treatment with deltarasin led
   to a rapid organoid dissociation and a highly significant reduction in
   viability of both Apc^fl/fl and KPN organoid lines demonstrating an
   important role for KRAS4B in maintaining CRC organoid viability
   (Fig. [235]5e, [236]f). Treatment of wild-type organoids led to a
   smaller reduction in viability (Fig. [237]5e, [238]f). Treatment of
   human normal and CRC organoids showed a similar effect, with organoids
   derived from normal colon being resistant but those derived from
   tumours being highly sensitive to deltarasin treatment (Fig. [239]S5i,
   [240]S5j). To validate these findings we designed an antisense
   morpholino that inhibits the splicing of Kras4b. Treatment of organoids
   with this morpholino proved highly effective at reducing the
   Kras4b/Kras4a splice ratio (Fig. [241]S5m, [242]S5n) and significantly
   reduced the clonogenic capacity and viability of Apc^fl/fl and KPN
   organoids (Fig. [243]S5k–[244]S5p).

   To further this analysis, we next asked whether Kras4b expression is
   sufficient to rescue organoid viability in Srsf1 depleted organoids. We
   generated organoid lines overexpressing GFP alone, or Kras4b (also
   marked with GFP expression) both carrying a DOX inducible shSrsf1 RFP
   construct. In this model Kras4b expression (marked by GFP) can be
   maintained alongside Srsf1 depletion (marked by RFP expression)
   (Fig. [245]S5q). We allowed organoids to form, induced with DOX and
   tracked GFP/RFP double-positive organoids over 3 days. Similar to our
   previous results, depletion of Srsf1 in GFP expressing controls led to
   a significant reduction in organoid growth (Fig. [246]5g, [247]h).
   However, this growth reduction was significantly rescued to wild-type
   levels by ectopic expression of Kras4b indicative of a functional role
   for Kras4b in mediating the phenotypic effects of SRSF1 expression in
   this model (Fig. [248]5g, [249]h).

   To infer the biological function of the two Kras splicing isoforms in
   the context of CRC, we employed proximity-dependent biotin
   identification (BioID) to uncover interacting proteins of KRAS4A and
   KRAS4B (Fig. [250]S5r–[251]S5u). We found 83 proteins that
   significantly (p < 0.05) interacted with both KRAS4A and KRAS4B
   (Fig. [252]S5t and Supplementary Data [253]13). We also identified 23
   proteins that monogamously interacted with KRAS4A and 22 proteins that
   uniquely interacted with KRAS4B. Some of these uniquely interacting
   proteins were associated with contrasting cell signalling pathways and
   associated with unique cellular compartments (Supplementary
   Data [254]13). For example, IGF1R and RALA interacted only with KRAS4A
   whereas BRAF and RAP1A were exclusively associated with KRAS4B. These
   data show that KRAS4A and KRAS4B have specific subsets of protein
   interactors indicating SRSF1-controlled splicing changes can result in
   significant changes in the oncogenic protein interactome.

High Srsf1 expression mediates tumour cell plasticity and colorectal cancer
invasiveness

   To explore the role of SRSF1 in advanced stage colonic carcinogenesis
   we utilised a previously described model of carcinogen-induced
   tumourigenesis^[255]43. Cohorts of villinCre^ERT2 p53^fl/fl and
   villinCre^ERT2 p53;^fl/fl Srsf1^fl/+ mice were treated with repeated
   rounds of azoxymethane (AOM) and aged until signs of colonic
   tumourigenesis became apparent (Fig. [256]S6a). Mice treated with AOM
   presented predominantly with colonic tumours, with tumours of the small
   intestine rarely occurring (Fig. [257]S6b). Although we did not find
   evidence that targeting Srsf1 provided a survival advantage or a change
   in tumour number or burden (Figs. [258]S6c–[259]S6e) we found that
   deletion of one copy of Srsf1 (Fig. [260]S6f) significantly reduced the
   proportion of mice presenting with invasive tumours in this model
   (Fig. [261]6a, [262]b). Additionally, overall number and percentage of
   invasive tumours per mouse and the presence of extensive collagen
   deposition was significantly reduced in Srsf1 depleted mice
   (Figs. [263]6c and S[264]6g–S[265]6i).

Fig. 6. Srsf1 mediates tumour cell plasticity and colorectal cancer
invasiveness.

   [266]Fig. 6
   [267]Open in a new tab

   a Representative histological images of mouse colons and tumours
   stained with hematoxylin and eosin (H&E). Scale bars are 1 mm (2.5x),
   500 µm (5x) and 250 µm (10x). Tumours were isolated when mice showed
   clinical signs of intestinal tumourigenesis. b Quantification of the
   presence or absence of invasive tumours in mice from the indicated
   genotypes, with statistical difference calculated using a two-sided
   Chi-square test, n = 23 vs 34 biologically independent mice. c Box and
   whisker plot showing the number of invasive tumours as a percentage of
   the total number of tumours for each mouse for each genotype. The box
   extends from 25th to 75th centiles, the centre line is median and
   whiskers extend to minima and maxima, n = 23 vs 34 biologically
   independent mice. d Representative histological images of p53^fl/fl and
   p53;^fl/fl Srsf1^fl/+ mouse intestines and tumours stained for PROX1
   (stem cell marker) and SLC13A2 (differentiation marker). Scale bars are
   2.5 mm (1.25x), 500 µm (5x) and 50 µm (40x). e Histoscore and staining
   strength quantification for PROX1, n = 5 vs 6 biologically independent
   mice and f SLC13A2, n = 7 vs 5 biologically independent mice. g
   Schematic depiction of the experimental strategy used to investigate
   the invasive potential of tumour-derived intestinal organoids. h
   Representative images of calcein-stained organoids once they had
   invaded through Matrigel and through the porous membrane at the bottom
   of the assay well. Scale bar 500 µm. i Quantification of the invasive
   potential of p53^fl/fl tumour-derived organoids following control or
   Srsf1 shRNA manipulation, n = 9 vs 9 independent experiments. Data in
   bar charts are represented as mean and error bars are SD with data
   analysed with two-tailed, unpaired t-tests, p values are indicated in
   figure panels. All biological replicates are shown as individual value
   plots and n > 3. See also Fig. [268]S6.

   Upon examination of the invasive tumours, we found evidence of cell
   plasticity changes. Immunohistochemical analysis of the cancer stem
   cell marker PROX1 revealed a significant decrease in PROX1 expression
   in invasive p53;^fl/fl Srsf1^fl/+ tumour epithelial cells
   (Fig. [269]6d, [270]e) corroborating our RNAseq results in short-term
   Cre-lox recombined Apc;^fl/fl Srsf1^fl/+ tissue (Fig. [271]3g). We also
   found a significant corresponding increase in expression in the
   differentiation marker, SLC13A2 (Figs. [272]6d, [273]f), and altered
   Kras splicing in these tumours (Fig. [274]S6j). These data suggest that
   impaired Srsf1 expression reduces the invasive and stem cell behaviour
   of colonic tumours in vivo.

   To determine whether SRSF1 directly controls these phenotypes, we
   cultured 3D tumour organoids from invasive AOM p53^fl/fl tumours and
   depleted Srsf1 expression using shRNA (Fig. [275]S6k). We found that
   both PROX1 expression and Kras4b/Kras4a splicing ratio were impaired
   upon Srsf1 knockdown confirming a direct role for Srsf1 in mediating
   these effects (Figs. [276]S6l, [277]S6m). We then carried out an
   organoid invasion assay and observed a significant reduction in
   organoid cell invasion through matrigel upon Srsf1 depletion
   (Fig. [278]6g–[279]i).

   Together, these data suggest that SRSF1 directly controls tumour cell
   invasion and maintenance of stem cell properties in late-stage colon
   cancer.

SRSF1 depletion inhibits growth and stemness of human colorectal
cancer-derived organoids

   To ascertain how SRSF1 levels correlate with human CRC progression, we
   analysed colon cancer tissue microarrays (TMAs) containing varying
   severities of the disease. First, we investigated if the cancer stem
   cell marker PROX1, which has been show to promote metastatic outgrowth
   of cells^[280]44, correlated with SRSF1 expression. We found a positive
   correlation of SRSF1 and PROX1 expression (Fig. [281]7a, [282]b and
   S7a) similar to that seen in our laboratory models (Figs. [283]3g and
   [284]6d, [285]e). We next examined SRSF1 expression in different tumour
   stages and found that more invasive and metastatic tumours had
   significantly higher expression of SRSF1 (Fig. [286]7c) showing that
   SRSF1 levels correlate with advanced-stage colorectal cancer.

Fig. 7. Srsf1 depletion inhibits growth of human colorectal cancer-derived
organoids.

   [287]Fig. 7
   [288]Open in a new tab

   a A sample of cores taken from a human colon cancer tissue microarray
   (TMA), with the same respective core stained for SRSF1 or PROX1 on
   different sections. Scale bar 500 µm. b Linear regression analysis
   showing the correlation of SRSF1 and PROX1 staining (based on
   histoscore) on the TMA shown in a, with each datapoint representing a
   core taken from a patient. c Relationship between SRSF1
   immunohistochemistry staining and tumour stage in human patients, using
   TMA CO2081b, n = 54 vs 56 biologically independent tumour cores. d
   Representative images of patient-derived organoids (PDOs) treated with
   control or Srsf1 shRNAs. MD175 (polyp), MD20853 (CRC), MD20910 (CRC),
   MD19648 (FAP rectum Tumour), MD20043 (rectal carcinoma), C-002 (liver
   metastasis). Scale bar 1000 µm. e Number of surviving organoid clones
   after shRNA treatment and f size of indicated PDOs, n = 3 vs 3
   independent experiments. g Viability of PDO MD20043 after Srsf1 shRNA
   treatment and h mRNA expression (qPCR) of indicated genes, n = 3 vs 3
   independent experiments. i Model outlining the role of SRSF1 in
   modulating tumour cell plasticity and invasion in colorectal cancer.
   Data in bar charts are represented as mean and error bars are SD with
   data analysed with two-tailed, unpaired t-tests, p values are indicated
   in figure panels. All biological replicates are shown as individual
   value plots and n > 3. See also Figure [289]S7.

   To phenocopy clinical intervention by way of attenuating SRSF1 levels
   in human CRC, we manipulated SRSF1 in patient-derived organoids (PDOs)
   using shRNA. Consistent with our animal experiments the clonogenicity,
   size and viability of the PDOs was impaired by depleting SRSF1
   expression (Fig. [290]7d–[291]g). Reduced SRSF1 expression was
   associated with a corresponding reduction in LGR5 suggesting that in
   human cancer, as in our mouse models, SRSF1 mediates stem cell function
   (Figs. [292]7h and S[293]7b–S[294]7d). We also observed an impairment
   in the ratio of KRAS4B/KRAS4A (Fig. [295]7h) which we had earlier found
   to be critical in facilitating oncogenic growth in mouse (Fig. [296]5e,
   [297]f).

   Taken together these data establish that, not only can SRSF1
   attenuation mitigate the oncogenic potential of CRC in mouse models,
   but also in the human disease.

Discussion

   Tumour growth has been shown to increase the transcriptional output of
   cells leading to an elevated burden of pre-mRNAs requiring processing,
   including an increased requirement for RNA splicing^[298]45. The RNA
   addiction of hyperproliferative cells might therefore be exploited for
   clinical purposes. This idea was supported by earlier work showing that
   the spliceosome is a potential therapeutic vulnerability following MYC
   induction in 2D breast cancer cells and also during
   lymphomagenesis^[299]18,[300]19. Here we have shown that genes
   associated with RNA metabolic processing are upregulated in a mouse
   model of colorectal cancer initiation. These include genes involved in
   RNA splicing. 60 splicing related genes are upregulated following Apc
   loss, accounting for approximately a quarter of all known splicing
   genes. Based on this, we surmised that splicing impairment might
   effectively target rapidly proliferating intestinal organoids with
   hyperactive Wnt signalling. Indeed, treating organoids with
   pladienolide B (which targets the SF3B complex) successfully targeted
   Apc^fl/fl organoids without adversely affecting wild type organoids
   suggesting that RNA splicing generally may act as a therapeutic
   vulnerability in CRC. However, this drug has recently failed in
   clinical trials as it caused optic-nerve dysfunction and vision loss in
   patients^[301]46. On the other hand, a novel drug targeting the
   splicing factor SRSF1, ABX300, has been shown to impair SRSF1 splicing
   activity to treat diet-induced obesity in mice without any observed
   toxicity^[302]47. Furthermore, villinCre^ERT2 Srsf1^fl/+ mice had
   normal intestinal morphology without any change in cell proliferation
   or normal stem cell function, and an inducible shRNA against Srsf1 had
   no observable phenotype in wild-type organoids. Thus, there is a clear
   therapeutic window for targeting SRSF1 in colorectal cancer.

   In conjunction with an increased requirement for splicing factors, we
   identified dysregulation of splicing itself, as shown by the shift in
   RNA isoforms generated following Apc loss. We found that the Kras4a and
   4b isoforms were alternatively spliced depending on the level of Wnt
   activation, and this was mediated by SRSF1. SRSF1 has previously been
   proposed to be oncogenic and is a direct transcriptional target of
   Myc^[303]48,[304]49. In addition, SRSF1 expression has been shown to be
   dependent on Wnt signalling in colorectal cancer suggesting that Wnt
   driven dysregulation of RNA splicing is partially mediated by SRSF1
   following Apc loss^[305]50. The Kras4a isoform has been shown to
   promote cell death^[306]24 and, when mutated, has significantly less
   oncogenic potential in mice than Kras4b^[307]51. Not only did we see a
   shift in splicing of these isoforms following Apc deletion, but we
   found that when we impaired the function of Apc-upregulated KRAS4B
   using deltarasin or an antisense morpholino, we could hamper oncogenic
   growth in our organoid models. Furthermore, the expression of Kras4b
   could rescue the growth of cells with impaired Srsf1 levels. As Srsf1
   depletion leads to alterations in numerous splicing events it is
   unlikely that this single splicing event fully explains the role of
   SRSF1 in CRC. It is likely that numerous SRSF1 targets contribute to
   its phenotypic effect, making SRSF1 an attractive therapeutic target,
   depletion of which could impact on multiple different pathways.

   Targeting Srsf1 resulted in reduced proliferation and stem cell
   function consistent with similarly described functions of SRSF1 in
   breast cancer^[308]50,[309]52. Transcriptome analysis of Apc;^fl/fl
   Srsf1^fl/+ intestinal tissue revealed a role for SRSF1 in modulating
   cellular plasticity with the gene expression signature becoming
   distinctly enterocyte-like. Dedifferentiation of enterocytes has been
   shown to support a ‘top-down’ model of colorectal tumour morphogenesis
   where adenomas originate at the top of intestinal
   crypts^[310]34,[311]53. Our data support a role for SRSF1-driven
   dedifferentiation in enterocytes. Following early activation of
   constitutive KRAS or NF-kB signalling with Apc loss, a reduction in
   dedifferentiation-driven clonal events arising from villi-derived
   differentiated cells was observed with impaired Srsf1 levels. Thus,
   SRSF1 can mediate cellular plasticity. As SRSF1 is a splicing factor,
   such plasticity changes might be brought about indirectly as a result
   of a general splicing repression response, or due to a change in the
   RNA isoform repertoire. As well as showing increased expression of
   differentiation marker genes, Srsf1 reduction resulted in a decrease in
   the stem cell marker Prox1. PROX1 is a Wnt-regulated transcription
   factor that has been shown to advance colon cancer progression by
   promoting dysplasia in colonic adenomas^[312]54, as well as enhancing
   metastasis in Wnt-driven progenitor cells^[313]44. Not only were Prox1
   levels impaired in the intestines of Apc;^fl/fl Srsf1^fl/+ mice
   following acute Apc deletion, but this reduced level was observed in
   the advanced tumours of AOM treated p53;^fl/fl Srsf1^fl/+ mice. These
   tumours were also significantly less invasive and displayed evidence of
   the same cell differentiation phenotypes observed in our early-stage
   tumour model. Encouragingly, this demonstrates a potential to target
   SRSF1 mediated cellular plasticity even in the advanced stages of the
   disease.

   In conclusion, our investigation using mouse models, ex-vivo organoid
   systems and patient-derived samples demonstrates that intestinal
   Wnt-driven cancers are addicted to the spliceosome. We have shown that
   targeting an individual oncogenic splicing factor, SRSF1, impairs
   cancer progression through a variety of mechanisms (Fig. [314]7i).
   Modulation of cell-type plasticity in favour of a gene expression
   signature with reduced stemness, an impaired ability of cells to
   dedifferentiate, as well as tumours having a lower invasive potential,
   all make targeting SRSF1 a highly attractive option, and may complement
   current standard-of-care therapies.

Methods

Contact for reagent and resources sharing

   Requests for further information, reagents and resources should be
   directed to and will be fulfilled by the Lead Contact, Kevin B. Myant:
   (kevin.myant@igmm.ed.ac.uk).

Animals models

   Species used: Mus musculus. All animal experiments were performed in
   accordance with a UK Home Office licence (Project License 70/8885), and
   were subject to review by the animal welfare and ethics board of the
   University of Edinburgh. Both genders of mice were used for all
   experiments at an age of between 6 and 12 weeks once they had reached a
   minimum weight of 20 g. Mice were bred at the animal facilities of the
   University of Edinburgh and were kept in 12 h light–dark cycles and
   were given access to water and food ad libitum. Mice were maintained in
   a temperature- (20–26 °C) and humidity- (30–70%) controlled
   environment. Colonies had a mixed background (50% C57Bl6J, 50% S129).
   The genetic alleles used for this study were as follows:
   villinCreER^[315]55, Apc (floxed)^[316]56, ASF/SF2 (Srsf1
   floxed)^[317]27, Kras (G12D)^[318]57, IKK2ca^[319]58, P53
   (floxed)^[320]59. Mice were genotyped by Transnetyx (Cordoba, USA). At
   experiment endpoints, mice were humanely sacrificed by cervical
   dislocation (CD) in line with UK Home Office regulations.

Tumour models and treatments

   For short-term Cre-Lox recombination where animals were taken at a
   specific time point of between 2 and 5 days postinduction of gene
   recombination (depending on the experimental model), mice were induced
   with a single dose of tamoxifen (Sigma-Aldrich, T5648) by
   intraperitoneal injection of 120 mg/kg on the first day and then
   optionally with a further treatment of 80 mg/kg on the second and third
   days.

   For the long-term Cre-Lox recombination in tumour cohorts, mice were
   given a 300 µL, 10 mg/mL dose of tamoxifen on day 0 and a 200 µL,
   10 mg/mL dose of tamoxifen dose on day 1. Mice were then treated via
   intraperitoneal injection with azoxymethane (AOM) (Sigma-Aldrich,
   A5486) at a dose of 10 mg/kg once a week, every week, starting on day
   12, for 6 weeks.

   Mice were aged until symptomatic of disease (rectal bleeding, weight
   loss, hunching and/or pale feet). Mice were randomly distributed by sex
   and age and a minimum of 20 or 3 mice were used for long and short-term
   experiments respectively. In long-term tumour cohort mice, tumour
   number and burden were macroscopically quantified in situ after mouse
   termination.

   For animals which were used for BrdU analysis, 200 µL of cell
   proliferation labelling reagent (GE Healthcare, RPN201) was
   administered via intraperitoneal injection 2 h before culling.

Splicing factor-targeted synthetic lethal CRISPR screen & organoid
transductions

   A targeted guide RNA (gRNA) library for the 60 APC-induced splicing
   factors was generated using sequences from Mouse GeCKOv2 Library A
   (Zhang Lab/GeCKO website), using 3 guide RNAs per gene and 9
   non-targeting control guides (Supplementary Data [321]6). In order to
   avoid ligation bias, oligonucleotides (Sigma) corresponding to each
   guide RNA were cloned individually into the lentiGuide-Puro vector
   (Addgene, 52963), Sanger sequenced, and then pooled at equimolar
   concentrations. LentiGuide-Puro was a gift from Feng Zhang^[322]60.
   Large-scale plasmid prep was achieved by electroporation of Endura
   electrocompetent cells (Lucigen) as previously described (Zhang
   Lab/GeCKO website). Equal guide RNA coverage within the library was
   confirmed via next-generation sequencing (NGS). The guide RNA library
   was used to generate infectious lentiviral vector particles.

   The workflow of the synthetic lethal screen was as follows: Day 0 –
   Wild type-Cas9 and Apc^fl/fl-Cas9 expressing organoids (derived from
   tamoxifen-inducible villinCre^ERT2 LSL-Cas9 mice) were plated and grown
   (full 24-well plate each, 20 µl Matrigel/well) in organoid growth media
   supplemented with 6 µM CHIR-99021 (GSK-3 inhibitor, Abcam, ab120890).
   Day 3 – Organoids were further expanded to two full 24-well plates each
   and grown in ‘Organoid+ media’: ADF (Advanced DMEM/F12 + B27 + N2)
   (500 µL/well), Noggin, EGF, R-spondin (only for Wild type organoids),
   10 µM Y-27632 dihydrochloride (ROCK inhibitor, Tocris, 1254), 6 µM
   CHIR-99021, 1 µM Jagged-1 (188–204) (Notch Ligand, AnaSpec, AS-61298)
   and 1 mM valproic acid (Sigma, PHR1061). Day 5 – organoids were
   dissociated via mechanical disaggregation followed by treatment with
   StemPro Accutase Cell Dissociation Reagent (Gibco, A1110501) for 5 min
   at 37 °C and then neutralised in 1% BSA. Cells were counted using the
   Countess II Automated Cell Counter (Invitrogen) and 1×10^6 wild
   type-Cas9 or Apc^fl/fl-Cas9 organoid cells were transduced at a
   multiplicity of infection (MOI) of 0.3 for each of the three biological
   replicates. (The precise working titre of the virus was calculated in
   advance by viral titration experiments on non-Cas9 expressing Apc^fl/fl
   organoids and measuring survival following puromycin selection).
   Briefly, 250,000 cells were plated on an 80 µL bed of Matrigel in each
   of four wells of a 12-well plate (1×10^6 cells per genotype). Cells
   were incubated with 0.3 MOI lentivirus and 4 µg/mL of Hexadimethrine
   bromide/polybrene (Sigma, H9268) in a total volume of 500 µL of
   organoid+ media. Cells were allowed to transduce for 24 h. Day 6 –
   virus was removed and adhered organoid cells were overlaid and set with
   100 µL Matrigel, followed by standard organoid growth media with
   Y-27632 only. Day 7 – Puromycin was added at a concentration of 2 µg/mL
   and Y-27632 was maintained in the media. Day 8 – fresh puromycin and
   Y-27632 containing media was added to organoids. Day 11 - puromycin and
   Y-27632 treatment was stopped and cells were grown in standard organoid
   growth media, and this was also refreshed on Day 13. Organoids were
   harvested on Day 14 and genomic DNA was purified using the DNeasy Blood
   & Tissue Kit (Qiagen, 69504). 4 independent replicate experiments were
   performed.

Clonogenicity analysis

   Intestinal organoids were passaged as usual via mechanical
   disaggregation and DMEM/PBS washes. After the last wash step, organoid
   cell pellets were treated with 1 mL StemPro Accutase cell dissociation
   Reagent (Gibco, A1110501) and incubated at 37 °C for 5–10 min in order
   to generate single cells. An equal volume of 1% BSA was added to stop
   the digestion reaction and then diluted in 10 mL DMEM/F-12 media. Cells
   were then passed through a 40 µm cell strainer and centrifuged at 300 g
   for 3 min. Cells were counted using the Countess II Automated Cell
   Counter (Invitrogen). Between 1–10 × 10^3 single cells were plated per
   5 µL drop of Matrigel/BME and a minimum of 4 drops were plated for each
   genotype/condition. Organoid growth media was added and resultant
   spheres/clones were counted after 4 days. The clonogenic capacity was
   determined by calculating the average percentage of spheres formed in
   each drop per number of single cells plated.

Dedifferentiation assay

   Mice were administered 120 mg/kg tamoxifen on day 0 and 80 mg/kg
   tamoxifen on days 1 and 2. On day 3, mice were culled and the first
   10 cm of the small intestine following the duodenum was dissected and
   washed twice with PBS. The intestine was opened longitudinally with
   small scissors and the opened intestine was rinsed in PBS. Villi were
   removed by scraping using a glass coverslip, and were collected in
   DMEM/F12 in a 50 mL centrifuge tube. The tube was gently inverted 5–6
   times in order to dissolve the mucus and debris was allowed to settle
   for 30 s. After the larger aggregates settled down, the supernatant
   containing the villi was decanted into another centrifuge tube. This
   supernatant was then centrifuged at 100 g for 3 min and the collected
   villi pellet was resuspended gently in 10 mL DMEM/F12 so as not to
   fragment or disrupt the villi structure. Whole villi were observed
   under a light microscope and then counted. Equal numbers of villi were
   taken for each genotype/condition, centrifuged at 100 g for 3 min and
   then resuspended in 3–5 mL TryplE Express (Gibco, 12605010) and
   incubated at 37 °C for 30 min. During the incubation, the suspension
   was vigorous resuspended via pipetting every 10 min. After each
   resuspension, a droplet of the digestion medium was observed under the
   microscope to check digestion. After digestion to single cells, cells
   were resuspended in 10 mL DMEM/F12 and passed through a 40 µm cell
   strainer. Single cells were counted and then 12 droplets of
   Matrigel/BME of 10 µL containing 50,000 cells per droplet were plated
   in organoid growth media, with the addition of 10 µM Y-27632
   dihydrochloride (ROCK inhibitor, Tocris, 1254). Colonies/spheres
   resulting from dedifferentiated villi-derived single cells were scored
   after 7 days.

RNAseq

   For the wild type vs Apc^fl/fl and Apc^fl/fl vs Apc;^fl/fl Srsf1^fl/+
   RNAseq experiments, 3 mice from each genotype were given
   intraperitoneal injections of tamoxifen of 120 mg/kg on day 0 and
   80 mg/kg tamoxifen on day 1. The mice in the Apc^fl/fl vs Apc;^fl/fl
   Srsf1^fl/+ experiment received an additional 80 mg/kg tamoxifen
   induction on day 2. Mice were culled on day 5, and the small intestine
   was dissected and flushed with PBS. A 1 cm piece of small intestine
   following the duodenum was placed in RNAlater solution (Sigma). RNA was
   later extracted as previously described and was sent to Edinburgh
   Genomics for sequencing. Truseq (Illumina) mRNA-seq libraries were
   prepared from total RNA and these were then sequenced using the
   Illumina HiSeq 4000 using 150 base paired-end sequencing.

Human organoid culture conditions

   Human adenoma, carcinoma and liver met organoids were cultured in: 30%
   R-spondin conditioned media, 1% Noggin conditioned media, 1x B27,
   50 ng/mL EGF, 10 nM Gastrin, 10 nM PGE2, 10 mM Nicotinamide, 10 µM
   SB202190, 600 nM A83-01, 12.5 mM N-Acetylcysteine in ADF (with 1X
   PenStrep, HEPES, Glutamine). Human organoid transduction conditions
   were the same as for mouse organoid transductions except that
   single-cell suspension were generated in TrypLE with 10 µM Y27632 for
   10 min at 37 °C with mechanical dissociation every 3 min.

Human patient-derived organoids (PDOs)

   C-002 is a cetuximab-resistant PDO generated from a biopsy of a liver
   metastasis from a gastrointestinal cancer patient and was generated
   elsewhere^[323]61. The following PDOs were generated by Dr Farhat Din
   and Prof Malcolm Dunlop, University of Edinburgh: MD175 is a 50 yr old
   female with familial adenomatous polyposis who previously underwent a
   colectomy and ileorectal anastomosis who then developed a stage 2
   rectal cancer - individual polyp used in this study. MD20043 is an
   81 yr old male with stage 4 rectal cancer, TNM (T3, 2, 1). MD19648 is a
   45-year-old female with familial adenomatous polyposis complicated by a
   stage 3 rectal cancer - FAP rectum tumour, TNM (pT1, pN1a, 0). MD20853
   is a 71 yr old male with a tubulovillous adenoma with low-grade
   dysplasia. MD20910 is a 60 yr old male with rectal cancer, TNM (pT2,
   N1b, 0). Ethical approval for human CRC organoid derivation was carried
   out under NHS Lothian Ethical Approval Scottish Colorectal Cancer
   Genetic Susceptibility Study 3 (SOCCS3) (REC reference:
   [324]11/SS/0109). All patients provided fully informed consent for the
   use of their tissues.

Biological pathway enrichment analysis

   RNAseq data of the 2330 genes upregulated in Apc^fl/fl relative to wild
   type were analysed using g:Profiler version e98_eg45_p14_ce5b097,
   accessed on 24/01/2020 using the following data sources: biological
   pathways-KEGG.

Organoid Culture

   To generate organoids from wild type or genetically modified mice, the
   first 10 cm of small intestine following the duodenum was isolated,
   flushed with phosphate-buffered saline solution (PBS) and opened
   longitudinally using scissors. Villi were removed by scraping using a
   microscope coverslip. Remaining tissue was then washed several times
   with PBS and then incubated with 2 mM EDTA in PBS with gentle shaking
   at 4 °C for 30 min. Crypts were then removed from the tissue by
   vigorous pipetting and selecting through a 70 µm cell strainer. Crypts
   were scored and ~100 crypts were plated in 20 µL droplets of cold
   liquified Matrigel Growth Factor Reduced (GFR) Basement Membrane
   Matrix, Phenol Red-free, LDEV-free (Corning #356231/734-1101) or
   Cultrex Reduced Growth Factor Basement Membrane Extract Matrix, Type 2
   (BME 2) (Trevigen #3533-010-02). A single 20 µl matrix droplet was
   grown in each well of a 24-well plate. Matrix droplets with suspended
   cells were allowed to solidify at 37 °C for 10 min prior to the
   addition of growth media. All organoids were grown in a humidified
   incubator at 37 °C supplemented with 5% CO[2].

   Organoids were grown in ‘organoid growth media’: Advanced DMEM/F-12
   media (ADF, Gibco) supplemented with 100 units/mL penicillin, 100 µg/mL
   streptomycin, 2 mM L-Glutamine and 10 mM HEPES (Life technologies), N2
   (Gibco, 17502048), B27 (Gibco, 17504044), 50 ng/mL EGF (Peprotech,
   315-09-500) and 100 ng/mL Noggin (Peprotech, 250-38-500). Wild type
   organoids were grow with the addition of 500 ng/mL R-spondin-1
   (Peprotech, 120-38-500). We later used Noggin and R-spondin via way of
   conditioned media made by transformed HEK 293 cell lines. The
   transformed Noggin producing cell line was a gift from Hans Clevers’
   group (Hubrecht Institute) and the R-spondin producing cell line was
   purchased from Trevigen (#3710-001-01). Organoids were passaged via
   mechanical disaggregation using a serological pipette, where several
   rounds of ADF or PBS washing and centrifugation were used to remove the
   residual matrix, before resuspending organoid fragments in the fresh
   matrix.

Tumour cell isolation and organoid culture

   Tumours were dissected from the intestine and then cut into pieces.
   Tumour tissue fragments were then incubated in 2 mL organoid growth
   media supplemented with 1 mg/mL collagenase II (Sigma, C1764),
   0.5 mg/mL hyaluronidase (Sigma, H3506) and 10 µM Y-27632
   dihydrochloride (ROCK inhibitor, Tocris, 1254). Incubation was at 37 °C
   for 60–90 min with vigorous shaking until the tumour was completely
   disaggregated. The digested reaction was then neutralised using 100 µL
   1% BSA and then fragments were filtered through a 70 µm cell strainer.
   The filtered cells were then centrifuged at 600 g for 3 min and washed
   2–3 times in DMEM/F12, before being resuspended and plated in an 20 µl
   droplets of BME in 24-well plates and overlaid with organoid growth
   media.

Pladienolide B treatment

   Pladienolide B stock (Calbiochem, 5.30196.0001) was made up to 1 mM in
   DMSO. Wild type and Apc^fl/fl organoids per broken into fragments via
   mechanical disaggregation and were plated at a density of ~40
   fragments/5 µL drop of Matrigel in a 96-well plate, with at least 6
   wells per Pladienolide B concentration for each of the three biological
   replicates. Organoids were allowed 48 h to develop prior to the
   addition of Pladienolide B, and were treated for 48 h with
   concentrations of Pladienolide B ranging from 0.5 nM–200 nM. For human
   organoid experiments the same protocol was used but a single dose of
   1 nM Pladienolide B was used.

MTT/Resazurin cell viability analysis

   For Thiazolyl Blue Tetrazolium Blue (MTT) cell viability assay,
   organoids were plated and treated in 96-well plates as already
   described. MTT (Sigma, M2128) solution was added to the organoid
   culture media at a final concentration of 500 µg/mL and incubated for
   2–3 h at 37 °C, 5% CO[2]. The medium was then discarded and 20 µL of 2%
   SDS solution in H2O was added to solubilize the Matrigel (2 h, 37 °C).
   Then, 100 µL of DMSO were added for 1 h (37 °C) to solubilize the
   reduced MTT. The optical density was measured on a microplate
   absorbance reader (Wallac, 1420 Victor2 microplate reader) at 570 nm
   and a background read at 690 nm was subtracted from this value. Cell
   viability was normalised to either Wild type or Apc^fl/fl DMSO control
   vehicle-treated organoids.

   For Resazurin cell viability assay, Resazurin (R&D systems, AR002) was
   added at a volume equal to 10% of the cell culture media volume and
   incubated for 4 h at 37 °C. Fluorescence was read using 544 nm
   excitation and 590 nm emission wavelength. For CRISPR gene-targeting
   assays, cell viability was assessed by first removing the fluorescence
   of puromycin treated, non-transduced (and therefore dead) organoids,
   followed by normalisation relative to either wild type or Apc^fl/fl
   non-targeting guide RNA control transduced organoids.

Overexpression of SRSF1 wild-type/DK mutant

   Srsf1WT and Srsf1DK were generated as above and cloned into the
   pUltra-GFP vector (Addgene, 24129). The pUltra was a gift from Malcolm
   Moore^[325]62. PCR amplificaton used 5’-XbaI and 3’-BamHI restriction
   sites, with the addition of a C-terminal FLAG tag. Transduction was
   performed as already described and cells were expanded and subsequently
   subjected to cell sorting (FACS) for GFP + cells. For the clonogenicity
   assay – cells were subjected to FACS and seeded at 5000 cells/10 µl BME
   drop in Y27632, and cells were counted 5 days following FACS sorting.

Real-time quantitative polymerase chain reaction (qPCR)

   Gene expression analysis was achieved via qPCR of mRNA. Total RNA was
   extracted using the RNeasy Mini Kit (Qiagen, 74106) followed by DNase
   treatment (Invitrogen, AM1906) and first-strand cDNA synthesis using
   500 ng -1 µg RNA (Quantabio, 95048-100). qPCR was performed on 1/10
   diluted cDNA using SYBR Select Master Mix (Applied Biosystems, 4472920)
   according to the manufacturer’s instructions on the CFX Connect
   Real-Time System (Bio Rad). Gene-specific oligonucleotides used are
   listed in (Supplementary Data [326]14). Ct-values were normalized to
   β-Actin, GAPDH or 18srRNA Ct-values, or the geometric mean of a
   combination of these loading controls. The delta-delta Ct method was
   used in order to calculate the relative fold change in gene expression
   of samples.

Reverse transcription polymerase chain reaction (RT-PCR) analysis of splicing
isoforms

   RNA extraction and first-strand cDNA synthesis using 500 ng RNA were
   performed as previously described. A PCR reaction was then performed
   with gene-specific oligonucleotides (Supplementary Data [327]14) using
   Phusion High-Fidelity DNA Polymerase (NEB, M0530S) following the
   manufacturer’s instructions using 5 µL of 1/10 diluted cDNA. PCR
   products were then separated by electrophoresis on a 1% agarose gel and
   fragment sizes were analysed using ImageJ^[328]63 using GAPDH RT-PCR
   for loading normalisation. Splicing isoform changes were calculated by
   generating a ratio of expression of each respective isoform in each
   condition.

CRISPR screen target validation experiments

   For validation of individual genes arising from the CRISPR screen, the
   most effective guide RNA (based on Z-score) was selected for gene
   targeting. Guide RNAs used for the validation were: Non-targeting#2
   (5’-GCTTTCACGGAGGTTCGACG-3’), Srsf1 (5’-CGGGTCCTCGAACTCAACGA-3’), Sf3b4
   (5’-TTTAGATGCCACGGTGTACG-3’) and Ddx10 (5’-AATATCACCTACTCGAGAAC-3’).
   Guide RNAs were individually cloned into the lentiGuide-Puro vector
   (Addgene) and transformed into Stbl3 chemically competent E. coli
   (Invitrogen, C737303). Wild type-Cas9 and Apc^fl/fl-Cas9 expressing
   organoids, as well as non-CAS9 expressing wild type and Apc^fl/fl
   organoids, were transduced with lentivirus containing the specific
   guide RNA. This was done using 250,000 cells for each of the four cell
   genotypes. Two days after transduction, cells were selected with
   puromycin for 3 days. Surviving organoids were then digested to single
   cells using StemPro Accutase Cell Dissociation Reagent (Gibco,
   A1110501) and a clonogenicity assay was performed by plating 10,000
   cells per 5 µl drop of Matrigel, with at least 4 Matrigel droplets per
   condition. Organoid growth media was added, with the addition of
   R-spondin-1 and 10 µM Y-27632 dihydrochloride (ROCK inhibitor, Tocris,
   1254) to the Wild type-Cas9 and Wild type-Cas9-null organoids only.
   Four days after seeding, organoids resulting from single cells were
   scored and the clonogenic capacity was calculated as a percentage based
   on the number of cells seeded. Normalised clonogenicity values were
   then calculated by first normalising the Cas9-expressing cells with the
   Cas9-null cells for each genotype, followed by normalising these values
   to the nontargeting guide RNA expressing condition.

   For the cell viability assay in Apc^fl/fl organoids, 10,000 single
   cells of either Apc^fl/fl-Cas9 or Apc^fl/fl-Cas9-null single cells were
   plated onto a bed of 20 µl Matrigel in a well in a 96-well plate in
   triplicate and were transduced with lentivirus-containing media. The
   day after transduction (day 1), lentivirus was removed and cells were
   overlayed with 20 µl Matrigel and media. On day 2 puromycin selection
   was started. On day 8, resazurin (R&D systems, AR002) was added to the
   media for 4 h at 37 °C and fluorescence was read using 544 nm
   excitation and 590 nm emission wavelength. Cell normalisation was done
   as previously described, first using the Cas9-null organoids, and then
   using the non-targeting guide RNA condition as a reference.

Barcode PCRs and deep sequencing of integrated gRNAs

   Genomically integrated guide RNAs were PCR amplified using staggered,
   barcoded PCR primers as previously described by Zhang lab^[329]64.
   Sequencing of amplified PCR products was achieved by Illumina deep
   sequencing (MiSeq v2).

Immunohistochemistry and histology

   Intestinal tissue was dissected and washed with PBS, and fixed either
   as a Swiss-roll or a parcel in formaldehyde 4% for 24 h at 4 °C. For
   tumour scoring, intestinal tissue was fixed in Methacarn (60% Methanol,
   30% Chloroform and 10% Glacial acetic acid) for 5 min at room
   temperature. Tissue processing was done using the Tissue-TeK VIP
   infiltration Processor (Sakura) followed by embedding in paraffin.
   Tissue sections were prepared at 5 µm using a microtome (Leica).
   Standard immunohistochemistry & histology techniques were used. Harris
   Hematoxylin and Eosin Y were used for nuclei and cytoplasmic staining
   respectively (Thermo Fisher Scientific).

   The following antibodies were used: BrdU (BD Biosciences, 347580, pH6,
   1/500), PROX1 (R&D systems, AF2727, pH6, 1/100 for human tissue array,
   1/200 for mouse sections), SLC13A2 (Atlas antibodies, HPA014963, pH8,
   1/100), SRSF1 (Invitrogen/Thermo Fisher Scientific, 32-4600, pH8,
   1/10,000), Ki67 (Abcam, ab15580, pH6, 1/2000), Active Caspase-3 (R&D
   systems, AF835, pH6, 1/800). Secondary detection was achieved using
   Dako EnVision + /HRP rabbit/mouse System, neat (Agilent technologies,
   K400311-2), except for PROX1 where Rabbit anti-Goat IgG was used 1/200
   (Thermo Fisher Scientific, 81–1620). Staining was achieved by
   Diaminobenzidine (DAB) Quanto Chromogen and Substrate (Thermo Fisher
   Scientific).

   To stain collagen, Picro Sirius Red was used: de-waxed slides were
   submerged in Picro Sirius Red solution for 2 h. Picro Sirius Red
   Solution: 0.1% Direct red 80 (Sigma, 41496LH), 0.1% Fast green FCF
   (Abcam, ab146267) in Picric acid solution 1.3% in H2O (Sigma, P6744).
   Post staining, slides were washed twice in acidified water, dehydrated
   according to standard protocols and mounted.

   Images were taken by using either the BX53 Upright Microscope (Olypmus)
   with the CellSens imaging software (Olympus) or the Nanozoomer Digital
   slide scanner (Hamamatsu) with the NDP.view2 software (Hamamatsu).
   Image analysis and quantification was done using QuPATH^[330]65.

   The colon cancer human tissue arrays used were CO804b and CO2081b.
   (Biomax).

shApc;KrasG12D/+;shRNA de-differentiation assay

   shApc;KrasG12D/+ organoids were previously described^[331]36, and
   cultured in ADF + + media with 1X B27, 1X N2, EGF (50 ng/mL), 1% Noggin
   CM, 10 mM Nicotinamide, 1.25mM N-acetylcysteine. Organoids were
   lentivirally transduced with either pTripZ-minP-shRenilla, or
   pTripZ-minP-shSrsf1. Organoids were given either a 12 hr or 24 hr
   treatment with or without doxycycline (1 µg/mL) before being collected
   and digested to single cells in TrypLE containing 10 µM Y-27632 for
   30 min at 37 C. Cells were washed in ADF + + before being passed
   through a 40 µM cell strainer. Single cells were centrifuged at 500x g
   for 5 min at 4 C before being stained with 1:200 EphB2-APC conjugated
   antibody (BD Bioscience, Clone 2H9) for 30 min at room temperature in
   PBS with 0.1% BSA and 10 µM Y-27632. Cells were then washed twice in
   PBS at 500x g for 5 min at 4 C before being subjected to FACS (BD
   FACSARIA™ III) for EphB2low and EphB2high (no doxycycline) or
   GFP + /RFP + /EphB2^low and EphB2^high (doxycycline) populations.
   Single cells were then seeded into 20 µl BME in organoid media
   containing 10 µM Y-27632 for 48 h. Formed organoids were counted 7 days
   after FACS, and clonogenicity was calculated as
   (#GFP + /RFP + organoids/#single cells seeded x100).

EPHB2 negative in vivo dedifferentiation assay

   Mice were administered 120 mg/kg tamoxifen on day 0. On day 2, mice
   were culled and the first 10 cm of the small intestine following the
   duodenum was dissected and washed twice with PBS. The intestine was
   opened longitudinally with small scissors and the opened intestine was
   rinsed in PBS. Villi were removed by scraping using a glass coverslip,
   and were collected in DMEM/F12 in a 50 mL centrifuge tube. The tube was
   gently inverted 5–6 times in order to dissolve the mucus and debris was
   allowed to settle for 30 s. After the larger aggregates settled down,
   the supernatant containing the villi was decanted into another
   centrifuge tube. This supernatant was then centrifuged at 100 g for
   3 min and the collected villi pellet was resuspended gently in 10 mL
   DMEM/F12 so as not to fragment or disrupt the villi structure. Whole
   villi were observed under a light microscope and then counted. Equal
   numbers of villi were taken for each genotype/condition, centrifuged at
   100 g for 3 min and then resuspended in 2 mL TryplE Express (Gibco,
   12605010) and incubated at 37°C for 30 min. During the incubation, the
   suspension was vigorous resuspended via pipetting every 10 min. After
   each resuspension, a droplet of the digestion medium was observed under
   the microscope to check digestion. After digestion to single cells,
   cells were resuspended in 10 mL DMEM/F12 and passed through a 40 µm
   cell strainer. Single cells were washed once with 0.1% BSA in PBS and
   stained with 1:200 EphB2-APC conjugated antibody (BD Bioscience, Clone
   2H9) and 1:200 Epcam-PE-conjugated antibody (BD Biosciences, 563477)
   for 30 min at room temperature in PBS with 0.1% BSA. Cells were then
   washed twice in 0.1% BSA in PBS at 500 g for 5 min at 4 C before being
   subjected to FACS (BD FACSARIA™ III) for Epcam positive, EphB2 negative
   populations. Single cells were then seeded into 10 µl BME in organoid
   media containing 10 µM Y-27632. Formed organoids were counted 7 days
   after FACS, and clonogenicity was calculated as (#organoids/#single
   cells seeded x100).

LGR5-GFP clonogenicity

   vil-Cre-ERT2 WT and Srsf1^fl/+ mice carrying the
   Lgr5GFP-CRE^ERT2transgene were administered 120 mg/kg tamoxifen on day
   0 and 80 mg/kg tamoxifen on day 1. On day 7 mice were culled and the
   first 10 cm of the small intestine following the duodenum was dissected
   and washed twice with PBS. The intestine was opened longitudinally with
   small scissors and the opened intestine was rinsed in PBS. Villi were
   removed by scraping using a microscope coverslip. The remaining tissue
   was then washed several times with PBS and then incubated with 2 mM
   EDTA in PBS with gentle shaking at 4 °C for 30 min. Crypts were then
   removed from the tissue by vigorous pipetting and selecting through a
   70 µm cell strainer. Equal numbers of crypts were taken for each
   genotype/condition, centrifuged at 100 g for 3 min and then resuspended
   in 2 mL TryplE Express (Gibco, 12605010) and incubated at 37 °C for
   15 min. During the incubation, the suspension was vigorous resuspended
   via pipetting every 5 min. After each resuspension, a droplet of the
   digestion medium was observed under the microscope to check digestion.
   After digestion to single cells, cells were resuspended in 10 mL
   DMEM/F12 and passed through a 40 µm cell strainer. Cells were washed
   twice in 0.1% BSA in PBS at 500 g for 5 min at 4 C before being
   subjected to FACS (BD FACSARIA™ III) for the Lgr5-GFP positive
   population. Single cells were then seeded into 10 µl BME in organoid
   growth media containing 10 µM Y-27632. Formed organoids were counted 7
   days after FACS, and clonogenicity was calculated as
   (#organoids/#single cells seeded x100).

Tumour stage classification

   Staging of tumours was scored according to the TNM staging system; T0-
   no evidence of primary tumour, Tis- carcinoma in situ, T1- tumour
   invading submucosa, T2- tumour invading muscularis propria, T3- tumour
   invading through the muscularis propria into the pericolorectal tissue,
   T4a- tumour penetrating to the surface of the visceral peritoneum, T4b-
   tumour directly invading or adherent to other organs or structures.

BrdU cell counting

   Images of stained small intestinal parcels were analysed in ImageJ
   where villi and crypts were first divided into villi and crypt
   compartments. BrdU positive cells in each compartment were scored from
   several images analysing an average of 24 villi/mouse and 82
   corresponding crypts/mouse, using 3–5 mice per genotype.

cDNA constructs and SRSF1 mutagenesis

   Srsf1 cDNA was purchased from GeneCopoeia (EX-Mm13449-Lv122) and Prox1
   cDNA was from Origene (MR210370). Kras 4 A and Kras 4B were cloned from
   cDNA from Apc^fl/fl mice.

   The Srsf1 RNA recognition motif 2 (RRM2) mutant ‘DK’ was created by
   forcing amino acids 136D and 138 K both to mutate to alanine. This was
   achieved by conducting a mutagenesis PCR on Srsf1 within the Lv122
   vector using the following overlapping oligonucleotides:
   5’-CAAGTGGAAGTTGGCAGGCTTTAGCGGATCACATGCGTGAAG-3’ and
   5’-CTTCACGCATGTGATCCGCTAAAGCCTGCCAACTTCCACTTG-3’.

SRSF1 RNA immunoprecipitation Protocol

   10 million CMT93 cells were pelleted and washed twice with PBS. Cells
   were homogenised in 1 ml lysis buffer, pH 7.4 (100 mM NaCl, 10 mM
   MgCl2, 30 mM Tris-HCl, 1 mM DTT, Protease & phosphatase inhibitor, 40
   U/ml RNase OUT, 0.5% Triton X-100). Homogenate was centrifuged at
   10,000 g for 10 min and 100ul of supernatant removed for RNA and
   protein input samples. The remaining lysate was split in 2 with half
   used for IgG IP and half for SRSF1 IP. The lysate was precleared with
   25ul washed Dynabeads, 0.05% BSA and 0.1ug/ml yeast tRNA for 1 h, 4
   degC. Lysate was moved to a fresh tube and incubated with 2 ug SRSF1
   antibody (sc-73026) or 2 ug mouse IgG (ab37335) for 1 h 4 degC
   pre-incubation with rotation, followed by a further 2 h incubation in
   the presence of 50ul dynabeads. The lysate was discarded and beads
   washed 3 x with lysis buffer. 1/3rd of the beads were taken for Western
   blots and the remaining resuspended in 100ul lysis buffer, followed by
   the addition of 50ug Proteinase K. Beads were incubated for 15 min 37
   degC to elute RNA from beads. RNA was purified in Trizol and purified
   RNA reverse transcribed. cDNA was used for qRT-PCR reactions to
   determine relative binding of SRSF1 to various RNA transcripts.

BioID protein interaction screen

   The pUltra vector (Addgene, 24129)^[332]62 was first modified, swapping
   out the EGFP for a gene encoding puromycin resistance. DNA sequence
   encoding the BirA enzyme with an N-terminal Myc tag (Myc-BioID2-MC,
   Addgene, 74223) was then subcloned into the pUltra vector (Addgene,
   24129). Myc-BioID2-MCS was a gift from Kyle Roux^[333]66,[334]67.
   Finally, primers containing the sequences of EcoRV and EcoRI
   restriction enzymes were used for Kras 4 A and Kras 4B amplification.
   Primer sequences were Kras 4 A/4B forward: 5’-
   CTCTCTGATATCGACTGAGTATAAACTTGTGGTGGTTGGAGCTGGTGGCGTAG -3’, Kras 4 A
   reverse: 5’-AGAGAGGAATTCTTACATTATAACGCATTTTTTAATTTTCACACAGCCAGGA-3’and
   Kras 4B reverse:
   5’-AGAGAGGAATTCTCACATAACTGTACACCTTGTCCTTGACTTCTTCTTCTTC-3’. For the
   purpose of a BioID interaction control protein, the gene encoding GFP
   was cloned into the BirA-pUltra vector.

   The BioID streptavidin affinity purification experiment was based on a
   previously published protocol^[335]66. In the first instance, 10 cm^2
   dishes were seeded each with 2 × 10^6 CMT-93 cells (mouse rectal
   carcinoma), 24 h prior to transfection. CMT-93 cells were grown in
   Dulbecco’s Modified Eagle’s Medium (DMEM, Sigma) supplemented with 10%
   foetal bovine serum, 2 mM glutamine and 1% of penicillin-streptomycin.
   Four plates were used for each of the three conditions: BirA-GFP,
   BirA-Kras4A and BirA-Kras4B. The following day, 8 µg of plasmid DNA was
   used for transfection in each plate using Lipofectamine 2000
   (Invitrogen, 11668030) and Opti-MEM (Gibco, 51985026) according to the
   manufacturer’s instructions. 24 h later, growth media was replaced and
   supplemented with 50 µM biotin (Sigma, B4501). Cells were incubated for
   24 h, washed in PBS, and then scraped into 500 µL RIPA buffer.
   Following protein clarification, 25 µL Streptavidin Sepharose slurry
   (GE Healthcare, 17-5113-01) was washed in RIPA buffer before being
   added to 1 mL of protein lysate, which was incubated for 6 h at 4 °C
   with rotation. Beads were purified by centrifugation at 1000 g for
   5 min and washed four times with buffer (50 mM TrisCl and 8 M Urea, pH
   7.4). On-bead digestion was done with trypsin as previously
   described^[336]68, followed by mass spectrometry using the Fusion Lumos
   mass spectrometer (Thermo Fisher Scientific). Proteins analysis was
   done using the MaxQuant-Perseus software^[337]69, and proteins were
   mapped to the mouse Uniprot database. Proteins enriched in BirA-Kras4A
   and BirA-Kras4B vs BirA-GFP were determined by first generating a Log2
   of the LFQ values and then normalising by Z-score. A student’s t-test
   (p < 0.05) was performed versus the GFP control samples. Significant
   hits were also filtered by two additional parameters: >1.5 fold higher
   levels in Kras 4A/4B expressing cells and at least 2 peptides detected
   in each of the three biological replicates. Cluster analysis was done
   using STRING^[338]70.

Western blotting

   Cells were lysed using RIPA buffer (Sigma, R0278) supplemented with 1%
   of phosphatase and protease inhibitors (Sigma, P0044 and P8340).
   Protein concentration was calculated using the BCA Protein Assay kit
   (Pierce). 10 µg of denatured protein lysate was separated by
   electrophoresis on NuPAGE 4–12% Bis-Tris precast protein polyacrylamide
   gels (Invitrogen) and blotted onto 4 µm nitrocellulose membrane
   (Amersham). Membranes were immersed in blocking solution (5% milk, 0.1%
   PBS-tween) for 1 h at room temperature, before being incubated in
   primary antibody at 4 °C overnight. Following 3 washes in 0.1%
   PBS-tween, the membrane was incubated in secondary antibody for 1 h at
   room temperature, followed by 3 more PBS-tween washes. Antibody signal
   was detected by using ECL Plus Western blotting substrate (Pierce,
   32132) and visualised using the ImageQuant LAS 4000 (GE Healthcare).
   Primary antibodies used were: β-actin, 1/5000 (Cell Signalling
   Technology, 4970), SF2/SRSF1, 1/1000 (Abcam, ab133689), c-Myc, 1/1,000
   (Cell Signalling Technology, 9402), Prox1, 1 µg/mL (R&D systems,
   AF2727), Ras (27H5), 1/1000 (Cell Signalling Technology, 3339), Myc-Tag
   (9B11), 1/10,000 (Cell Signalling Technology, 2276), and Streptavidin
   HRP, 1/10,000 (Abcam, ab7403). Secondary antibodies used were:
   Anti-Rabbit IgG HRP-linked, 1/1000 (Cell Signalling Technology, 7074),
   Anti-Mouse IgG HRP-linked, 1/1000 (Cell Signalling Technology, 7076)
   and Rabbit anti-Goat IgG, 1/1000 (Thermo Fisher Scientific, 81–1620).
   Western blot signal densitometry analysis was done using
   ImageJ^[339]63.

Lentiviral particles

   Lentiviral particles were made using cloned plasmid vectors, either at
   the Shared University Research Facility (University of Edinburgh), or
   in our own laboratory. Briefly, 10 µg gene-specific lentiviral vector
   was mixed with 7.5 µg lentiviral packaging vector psPAX2 (Addgene) and
   2.5 µg envelope protein producing vector pCMV-VSV-G (Addgene) and
   transfected into HEK 293 T cells in a 10 cm^2 dish using
   polyethylenimine transfection reagent (Polysciences, 23966). After 48 h
   virus was purified by first filtering the supernatant media with a
   0.45 µm filter, followed by virus concentration using Lenti-X
   Concentrator (Takara Bio, 631232) and resuspension of viral particles
   in PBS.

Diagrams & illustrations

   Diagrams were created with Biorender.com (Figs. [340]4G, [341]6G,
   [342]7I, S[343]3R, S[344]5H, S[345]5Q, S[346]5R, S[347]6A and S[348]6K)
   or in Microsoft PowerPoint.

Deltarasin treatment

   Organoids were mechanically dissociated by vigorous pipetting. 100
   fragments were seeded in 10 µl BME in a 24-well plate. 24 h after
   seeding, media was removed and replaced with fresh media containing
   either Deltarasin (Tocris, 5424) (10 µM) or DMSO. Cell proliferation
   was assessed after 48 h adding 10% Resazurin (R&D systems, AR002).
   Fluorescence was measured using Victor2 Multilabel Plate Reader
   (PerkinElmer) as already described.

Kras 4b morpholino treatment

   Organoids were mechanically dissociated by vigorous pipetting. 20–100
   fragments were seeded in 10 µl BME in a 24-well plate. 24 h after
   seeding, media was removed and replaced with fresh media containing
   either 5 μM Kras4B vivo-morpholino (Genetools -
   GTATAGAAGGCATCGTCAACACCCT) or 5 μM control vivo-morpholino (Genetools -
   CCTCTTACCTCAGTTACAATTTATA). Cell proliferation was assessed after 6
   days later by adding 10% Resazurin (R&D systems, AR002). Fluorescence
   was measured using Victor2 Multilabel Plate Reader (PerkinElmer) as
   already described.

Organoid invasion assay

   The invasion assay was based on a previously described
   protocol^[349]71. FluoroBlok HTS 24 Multiwell Insert System with 8.0 µm
   Pore High Density PET Membrane (Scientific Laboratory Supplies Ltd.,
   351157) was used. Inserts were coated with 50 μg/mL Matrigel solution
   diluted in cooled DMEM/F12 media containing EGF and Noggin. The
   organoids were dissociated with TryplE (Gibco, 12605010), resuspended
   in DMEM/F12 media containing EGF, Noggin and 5 ng/ml TGF beta-1
   (PeproTech EC Ltd, 100-21-10) and 3 × 10^4 cells were seeded in each
   apical chamber of the Fluoroblock insert, while DMEM/F12 media
   containing EGF and Noggin was added in the basal chambers. Following
   72 h incubation at 37 °C, 5% CO[2] atmosphere, cells were stained for
   1 h with Calcein AM (Abcam, ab141420) and bottom and top fluorescence
   was read on a Victor2 Multilabel Plate Reader (PerkinElmer) at
   wavelengths of 485/535 nm (Ex/Em).

Bioinformatics

   Differential Gene Expression of Wild type vs Apc^fl/fl and Apc^fl/fl vs
   Apc;^fl/fl Srsf1:^fl/+ Adaptor sequences were removed from paired reads
   using TrimGalore and cutadapt an8d reads mapped to mouse genome GRCm38
   (mm10) using tophat 2.1.1^[350]72. Differential expression was called
   using cuffdiff 2.2.1^[351]73 making use of Ensembl gene annotation 84
   for genes (for Wild type vs Apc^fl/fl) and Ensembl gene annotation 87
   for genes (for Apc^fl/fl vs Apc;^fl/fl Srsf1^fl/+). Genes with an
   adjusted p value of less than 0.05 were considered significantly
   regulated.

   Splicing analysis Wild type vs Apc^fl/fl and Apc^fl/fl vs Apc;^fl/fl
   Srsf1:^fl/+ Paired reads were pseudoaligned to the GRCm38 (mm10)
   Ensembl 87 transcriptome using salmon v0.9.1^[352]74. Splicing changes
   were inferred from transcript TPMs using SUPPA2^[353]20 from gene
   definitions in the Ensembl 87. A second splicing assessment was made
   with rMATS^[354]21 using the —novelSS setting to obtain splicing events
   involving unannotated splice sites.

   Gene Set Enrichment analysis (GSEA): Enrichment analysis of
   differentially expressed genes in the Apc^fl/fl vs Apc;^fl/fl
   Srsf1^fl/+comparison in gene sets defining enterocyte, paneth, goblet,
   tuft and enteroendocrine cell types was performed using the GSEA and
   enricher functions of Clusterprofiler^[355]75. The gene sets for these
   cell types was compiled previously^[356]29. Enrichment of
   differentially expressed genes in ISC, proliferating, late TA and Lgr5
   gene sets, defined previously^[357]30, was performed similarly after
   mapping genes across species using homologs specified in biomart.

   Splicing factor targeted CRISPR screen wild type vs Apc:^fl/fl A
   library of 180 gRNAs targeting 60 genes (plus 9 non-targeting controls)
   was designed using Mouse GeCKOv2 Library A (Supplementary Data [358]6).
   Stagger sequences were removed from reads using cutadapt prior to read
   counting and statistical testing of counts in treatment vs control
   using MAGeCK v0.5.6^[359]76. Statistics for gRNAs were used to select
   genes with gRNA having increased counts in treatment.

   Enrichment of splicing events: An enrichment algorithm for the
   comparison of an ordered set (A) of splicing events with another (B)
   was implemented following the gene set enrichment method^[360]77.
   Splicing events in the reference set A were ordered by p value (for
   events with p < 0.1) and absolute dPSI used to associate a probability
   with each event. Events common to a test set B and A were considered
   hits and increased a cumulative enrichment score by the probability of
   each matching event, non-hit events decreased the score. The observed
   enrichment was the maximum of this score. The maximum enrichment
   possible was where the number of intersecting events occurred at the
   head of A, and a probability value for the actual enrichment was
   calculated from a null distribution of maximum scores found by randomly
   selecting 100,000 subsets of A with the size of the intersection of A
   and B.

Mouse shRNA constitutive & Tet-On system and organoid size experiments

   For constitutive knockdown, the control (scramble) shRNA was a gift
   from David Sabatini (Addgene, 1864)^[361]78.

   TRC lentiviral shRNAs (Mouse Srsf1 ENTREZGENE:110809) were purchased
   from Dharmacon. These were supplied in the pLKO.1 vector: shSrsf1-3
   Clone Id: TRCN0000109142, sequence: AACATCAGCGTAACATACATC.

   For inducible knockdown, shRenilla and shSrsf1–4 were cloned into
   Tet-ON inducible vector pTRIPZ, modified with a synthetic minimal
   promoter (minP) (synthesised by GenScript) to replace the minimum CMV
   promoter. shRNAs were cloned into pTripZ vector using 5’ XhoI and
   3’EcoRI restriction sites. Organoids were seeded into BME and treated
   with 1 µg/ml doxycycline (Alfa Aesar by Thermo Scientific – J60579.22)
   dissolved in water.

   shSrsf1–4 sequence (97mer):
   TGCTGTTGACAGTGAGCGATAGGCTTATGTTTGAACACTATAGTGAAGCCACAGATGTATAGTGTTCAAAC
   ATAAGCCTACTGCCTACTGCCTCGGA. shRenilla sequence: (97mer):
   TGCTGTTGACAGTGAGCGCAGGAATTATAATGCTTATCTATAGTGAAGCCACAGATGTATAGATAAGCATT
   ATAATTCCTATGCCTACTGCCTCGGA.

   Images were taken at the indicated time points and analysed using
   ImageJ 2.0. For organoid size (area) determination for the constitutive
   shRNA model, the minimum number of organoids used to calculate organoid
   area per biological replicate were: Apc:^fl/fl 15, AKP: 20 and KPN: 40.
   For tracking and size determination in the Tet-On models, the number of
   organoids measured & tracked per biological repeat were: Wild-type
   organoids: 5–17, Apc^fl/fl organoids: 8–20 and AKP organoids: 5. For
   the shRNA Srsf1/Kras 4B rescue experiments in Apc^fl/fl organoids,
   14–22 organoids were measured for size per biological replicate. For
   the human PDO organoids, the number of organoids measured were 33–250.
   All organoid size experiments had n = 3 biological replicates.

Human shRNAs

   TRC lentiviral shRNAs (Human SRSF1 ENTREZGENE: 6426.) were purchased
   from Dharmacon. These were supplied in pLKO.1 vectors. shSRSF1–2 Clone
   ID: TRC0000001094, Sequence: TTAACCCGGATGTAGGCAGT.

Quantification and statistical analysis

   Statistical analyses were performed using GraphPad Prism software (v8.3
   GraphPad software, La Jolla, CA, USA) and Microsoft Excel (v2016,
   Redmond, WA, USA) performing the tests as indicated in the figure
   legends or main text. Significance levels were calculated according:
   p < 0.05 (*), p < 0.01(**) and p < 0.001 (***).

Reporting summary

   Further information on research design is available in the [362]Nature
   Research Reporting Summary linked to this article.

Supplementary information

   [363]Supplementary Information^ (5.6MB, docx)
   [364]Peer Review File^ (458.1KB, pdf)
   [365]41467_2022_30489_MOESM3_ESM.pdf^ (132.3KB, pdf)

   Description of Additional Supplementary Files
   [366]Supplementary Data 1^ (347.4KB, xlsx)
   [367]Supplementary Data 2^ (9.7KB, xlsx)
   [368]Supplementary Data 3^ (701.9KB, xlsx)
   [369]Supplementary Data 4^ (791.1KB, xlsx)
   [370]Supplementary Data 5^ (36.3KB, xlsx)
   [371]Supplementary Data 6^ (16.2KB, xlsx)
   [372]Supplementary Data 7^ (49.9KB, xlsx)
   [373]Supplementary Data 8^ (33.7KB, xlsx)
   [374]Supplementary Data 9^ (52.9KB, xlsx)
   [375]Supplementary Data 10^ (389.3KB, xlsx)
   [376]Supplementary Data 11^ (26.2KB, xlsx)
   [377]Supplementary Data 12^ (561.3KB, xlsx)
   [378]Supplementary Data 13^ (2.5MB, xlsx)
   [379]Supplementary Data 14^ (14.8KB, xlsx)
   [380]Reporting Summary^ (3.7MB, pdf)

Acknowledgements