Matrin3 regulates mitotic spindle dynamics by controlling alternative splicing of CDC14B

SUMMARY Matrin3 is an RNA-binding protein that regulates diverse RNA-related processes, including mRNA splicing. Although Matrin3 has been intensively studied in neurodegenerative diseases, its function in cancer remains unclear. Here, we report Matrin3-mediated regulation of mitotic spindle dynamics in colorectal cancer (CRC) cells. We comprehensively identified RNAs bound and regulated by Matrin3 in CRC cells and focused on CDC14B, one of the top Matrin3 targets. Matrin3 knockdown results in increased inclusion of an exon containing a premature termination codon in the CDC14B transcript and simultaneous down-regulation of the standard CDC14B transcript. Knockdown of CDC14B phenocopies the defects in mitotic spindle dynamics upon Matrin3 knockdown, and the elongated and misoriented mitotic spindle observed upon Matrin3 knockdown are rescued upon overexpression of CDC14B, suggesting that CDC14B is a key downstream effector of Matrin3. Collectively, these data reveal a role for the Matrin3/CDC14B axis in control of mitotic spindle dynamics.

Matrin3 knockdown results in increased inclusion of an exon containing a premature termination codon in the CDC14B transcript and simultaneous down-regulation of the standard CDC14B transcript. Knockdown of CDC14B phenocopies the defects in mitotic spindle dynamics upon Matrin3 knockdown, and the elongated and misoriented mitotic spindle observed upon Matrin3 knockdown are rescued upon overexpression of CDC14B, suggesting that CDC14B is a key downstream effector of Matrin3. Collectively, these data reveal a role for the Matrin3/CDC14B axis in control of mitotic spindle dynamics.

Graphical abstract
In brief INTRODUCTION Matrin3 (MATR3) is a nucleic acid-binding protein 1,2 that is conserved in vertebrates and is a major component of the nuclear matrix, a highly structured residual framework composed of lamins and ribonucleoproteins (RNPs). 3,4 In vitro, Matrin3 can bind to DNA with its (RRMs). 5 By binding to DNA, Matrin3 can attach to specific chromatin structural elements termed matrix or scaffold attachment region (MAR or SAR, respectively). 6 Consistently, Matrin3 can regulate chromatin organization. 7,8 Nevertheless, most studies have focused on Matrin3 as an RNA-binding protein (RBP) regulating diverse-not necessarily related -RNA processes, including transcription, 9,10 splicing, [11][12][13] RNA stability, 14 nuclear export, 15 and nuclear retention of hyper-edited RNAs. 16 Moreover, we previously reported that after DNA damage in colorectal cancer (CRC) cells, the long non-coding RNA (lncRNA) PINCR directly binds and recruits Matrin3 to enhancers of p53 target genes that depend on PINCR for p53-mediated induction of a subset of p53 targets. 17 Additionally, during myogenesis, Matrin3 is required for paraspeckle formation, likely by controlling adenosine to inosine (A-to-I) RNA editing of the Ctn RNA encoded by SLC7A2. 18 Matrin3 knockout mice are embryonic lethal, indicating that it is required for normal development. 19 Conditional deletion of Matrin3 in the neuronal lineage showed that it is essential for maintaining neuronal survival. 20 Even though Matrin3 appears to be essential in vivo, disease-associated mutations were primarily found in neurodegenerative disorders, e.g., in familial amyotrophic lateral sclerosis (ALS) and frontotemporal dementia (FTD). 21 The impaired formation of dynamic nuclear condensates due to ALS-associated mutations in Matrin3 possibly contributes to its pathogenic mechanism. 22 There is conflicting evidence for a role of Matrin3 in cancer, indicating a potential role of Matrin3 as an oncogene or a tumor suppressor. [23][24][25] While little is known about the role of Matrin3 in CRC, the cancer type our research focuses on, this nevertheless places Matrin3 among the many RBPs with potential oncogenic or tumor-suppressive roles. Typically, RBPs can bind multiple transcripts, and aberrant expression of a single RBP can affect the expression of a vast array of genes and abnormal phenotypes implicated in diseases including cancer. 26 For instance, in CRC, the RBPs MSI1 and MSI2 27 and SRSF1 28 have oncogenic functions. In addition, the p53-induced RBP ZMAT3 has tumor-suppressive effects by functioning as a key splicing factor. 29,30 It remains unclear which of the many Matrin3 RNA targets [11][12][13] would mediate the function of Matrin3 in CRC or other types of cancer.
To understand the molecular mechanism(s) by which Matrin3 mediates its effects, we performed PAR-CLIP (photoactivatable ribonucleoside-enhanced crosslinking and immunoprecipitation), a UV crosslinking-based technique that identifies RBP target RNAs at nucleotide resolution on a transcriptome-wide scale. 31,32 In HCT116 cells (CRC), Matrin3 binds to thousands of pre-mRNAs at pyrimidine-rich sequence elements, repressing inclusion of nearby exons. For functional analysis, we focused on one of the top misspliced targets upon Matrin3 knockdown, CDC14B, a key regulator of mitotic spindle assembly. 33 We show that overexpression of CDC14B rescues the defects in mitotic spindles observed upon Matrin3 knockdown, suggesting that the effects of Matrin3 on mitotic spindle are mediated by CDC14B. Collectively, these data reveal a growth-promoting function of Matrin3 in CRC cells that is-at least partially-mediated through splicing changes of CDC14B.

Matrin3 binds to pyrimidine-rich intronic sites in nascent transcripts encoding proteins that regulate the mitotic spindle
We first determined Matrin3 mRNA expression in a panel of human tumors compared with normal tissues. Matrin3 mRNA was significantly overexpressed in tumors, most frequently in CRC (Figures 1A, S1A, and S1B). Consistent with these data, Matrin3 knockdown resulted in decreased proliferation and clonogenicity in multiple cell lines ( Figures 1B,  1C, and S1C-S1J), normal human diploid lung fibroblasts (IMR90 and WI38), and normal colon epithelial cells (HCEC-1CT) ( Figures S1K-S1P). Interestingly, Matrin3 knockdown in HCT116 did not result in cell death or changes in the cell cycle (Figures S1S and S1T). Furthermore, in the DepMap database (Cancer Dependency Map: https://depmap.org/portal), comparing the effect of specific gene loss on growth across more than 1,000 cancer cell lines on proliferation, Matrin3 scored negative on average, compared with the expected negative and positive scores, respectively, for an essential control gene (PLK1) and a tumorsuppressor gene (p53) (Figures S1Q and S1R). These data suggest that Matrin3 promotes growth in diverse cell types.
Next, we mapped Matrin3 binding sites on RNAs on a transcriptome-wide scale at nucleotide resolution in HCT116 cells using 4-thiouridine (4SU) PAR-CLIP. 31,32 For PAR-CLIP, we IPed endogenous, UV-crosslinked Matrin3 RNP. Autoradiography of crosslinked, ribonuclease-treated Matrin3 IP fractionated on denaturing protein gels revealed a prominent band migrating at the expected size of ~125 kDa, corresponding to the Matrin3 RNP ( Figure 1D). We recovered the RNA fragments of Matrin3 RNPs and converted them into small RNA cDNA libraries for next-generation sequencing. Next, we determined clusters of overlapping reads that harbor characteristic T-to-C conversions diagnostic of 4SU crosslinking events at higher frequencies than expected by chance. 34 For the three replicates, we found between 99,087 and 168,992 binding sites (Table S1). These binding sites were distributed on a set of >8,000 shared genes, suggesting that a significant proportion of expressed transcripts were bound by Matrin3 ( Figure 1E). Overall, the biological replicates showed excellent correlation with a R 2 of ~0.9 ( Figure 1F). Approximately 70% of Matrin3 binding sites were found on intronic regions ( Figure 1G), indicating that Matrin3 was interacting with nascent transcripts, consistent with its nuclear localization and previous CLIP-based studies. [11][12][13] Some of these previous studies found a role for Matrin3 in alternative splicing, and indeed, in our data, Matrin3 bound broadly across nascent transcripts with a pronounced enrichment near the 3′ splice site ( Figure 1H), most likely at the polypyrimidine tract required for splicing, considering that Matrin3 binding sites were enriched for 5-mer-containing pyrimidines ( Figures 1I and S2A).
To investigate the gene regulatory roles of Matrin3, we next performed RNA sequencing (RNA-seq) from HCT116 transfected with Matrin3 small interfering RNAs (siRNAs) or a negative control siRNA. Knockdown of endogenous Matrin3 led to a modest but significant decrease in target mRNA levels. The magnitude of this effect was dependent on the overall strength of binding, i.e., the number of Matrin3 binding sites per target mRNA, or the number of crosslinked reads per target mRNA normalized by overall mRNA abundance ( Figure 1J; Table S1). We previously found that both metrics correlated well with the occupancy of an RBP on its target. 35 We next asked which basic architectural features 36 differentiated Matrin3 targets from non-targets. Matrin3 targets were typically encoded on longer loci than non-targets. As one would expect from an RBP binding nascent transcripts, Matrin3 PAR-CLIP binding site numbers positively correlated with transcript length (Figure S2B), number of exons ( Figure S2C), exon length ( Figure S2D), or intron length ( Figure S2E). Top Matrin3 targets had lower GC content compared with weaker targets or non-targets ( Figure S2F), consistent with the preference of RRM domains binding unstructured RNA. Comparison of our Matrin3 PAR-CLIP (from HCT116) with Matrin3 enhanced CLIP (eCLIP) from HepG2 cells 37 showed that they were reasonably similar, with a Spearman's correlation coefficient of ~0.5 ( Figures S2H and S2I), reflecting the different transcriptome of the cell lines, different CLIP-seq methodology, and our deeper dataset ( Figure S2J). Pathway analysis showed that the top Matrin3 targets (>50 binding sites) were enriched for genes related to the mitotic spindle ( Figure 1K), a structure important for mitosis. Thus, the PAR-CLIP interactome complemented our cell-based findings that Matrin3 knockdown resulted in reduced proliferation and clonogenicity.

Matrin3 knockdown leads to inclusion of exons proximal to its binding sites
Matrin3 has been previously characterized as a regulator of alternative splicing. [11][12][13] Therefore, we analyzed alternative splicing (AS) patterns in our polyA-selected RNA-seq from HCT116 after Matrin3 knockdown. The majority of the AS events (false discovery rate [FDR] < 0.05, ΔPSI ≥ 10%) involved cassette exons or skipped exons (SEs) (Figures 2A and  S2G). Of those events, >70% were found in Matrin3-bound RNAs. As reported previously in other cell types, 11,13 Matrin3 acted mainly as a splicing repressor with ~66% of the SE events in Matrin3 targets resulting from exon inclusion after Matrin3 knockdown ( Figure  2B). We next validated some of these AS events on transcripts with high ΔPSI in HCT116 and three other CRC cell lines (LS174T, LS180, and LOVO) by qRT-PCR and in HCT116 cells also by semi-qRT-PCR. Specific exon inclusion events in the CDC14B, CD44, SETD5, ST7, and HP1BP3 pre-mRNAs mirrored the RNA-seq results in all cases ( Figures 2C,   2D, and S3). Taken together, our results indicate that Matrin3 binding close to 3′ splice sites suppresses AS and that its loss results in mRNA misprocessing that manifests in the inclusion of otherwise excluded exons.

The growth-promoting function of Matrin3 is mediated, in part, by CDC14B
Next, we chose to focus on CDC14B for functional analysis, as it was among the genes with most significantly changed ΔPSI upon Matrin3 knockdown. We hypothesized that the significant reduction in cell proliferation after Matrin3 knockdown is mediated, at least in part, by missplicing of CDC14B. CDC14B encodes a protein tyrosine phosphatase that is similar to Saccharomyces cerevisiae Cdc14 and regulates interphase nuclear architecture, (CDC14B-003 or ENST00000375241, only isoform detectable by long-read sequencing; Table S1), which we refer to as a "standard variant" or "CDC14B-s" ( Figure 2C). The decrease in CDC14B-s levels after Matrin3 knockdown was accompanied by increased inclusion of two exons, which are only annotated in CDC14B-008 (ENST00000481149), a potential processed transcript variant that does not encode a protein. We refer to the longer transcript variant as the "PTC variant" or "CDC14B-PTC" ( Figure 3A) because exon 13 introduces a premature termination codon (PTC), which makes this transcript a potential substrate for non-sense-mediated decay (NMD). Immunoblotting confirmed a decrease in CDC14B protein after Matrin3 knockdown ( Figure 3B). We were also able to specifically knockdown the CDC14B protein and CDC14B-s transcript with an siRNA specifically targeting the CDC14B-s transcript ( Figures 3B and 3C). Next, we validated that CDC14B-PTC is subject to NMD by knocking down UPF1, the core component of the NMD machinery, which indeed resulted in increased CDC14B-PTC abundance ( Figures 3D  and 3E).
In CRC, we observed a modest but significant positive correlation between Matrin3 and CDC14B-s mRNA levels ( Figure 3F). Knockdown of Matrin3 or CDC14B-s resulted in a similar decrease of proliferation, and concurrent knockdown of Matrin3 and CDC14B-s had an additive effect on proliferation ( Figure 3G). In colony-formation assays, knockdown of CDC14B-s alone had a stronger effect than knockdown of Matrin3, which was further augmented when both Matrin3 and CDC14B-s were depleted ( Figures 3H and 3I), suggesting that additional Matrin3 targets affecting clonogenicity could mask the effect of CDC14B alone. Colony-formation assays from cells overexpressing CDC14B-s showed a modest but significant (****p < 0.0001) rescue in clonogenicity of Matrin3 knockdown cells ( Figures 3J-3L). These data suggest that CDC14B splicing regulation contributes to the decreased growth observed upon Matrin3 knockdown.

Matrin3 regulates microtubule dynamics, mitotic spindle orientation and bipolarity via CDC14B
CDC14 is a conserved protein in eukaryotes and its crucial role in mitosis in budding yeast 38 prompted us to examine if Matrin3 affects mitotic spindle architecture via regulation of CDC14B. To visualize the architecture of the mitotic spindle, cells were immunostained with alpha-tubulin for spindle microtubules and Nuf2 as an outer kinetochore marker. We found that in metaphase, mitotic spindles were slender and significantly elongated upon knockdown of Matrin3 or CDC14B-s ( Figure 4A). CDC14B-s overexpression significantly rescued the elongated and misoriented mitotic spindles observed in Matrin3 knockdown cells ( Figure 4B), suggesting that the observed defects in spindle length upon Matrin3 knockdown are largely mediated by defects in CDC14B splicing.
We reasoned that the elongated spindles could be caused by altered microtubule dynamics with more polymerizing microtubules present, as seen in cells with more stabilized microtubules. 39 Therefore, we examined the density of EB1 comets that mark the plus ends of polymerizing microtubules in interphase. Qualitative analysis showed that in interphase cells, the density of EB1 comets was increased upon Matrin3 or CDC14B-s knockdown ( Figure 4C), suggesting that the density of polymerizing microtubules is higher.
Additionally, EB1 intensity measured by imaging and the pseudocoloring EB1 signal showed brighter comets in Matrin3 or CDC14B-s knockdown cells ( Figure 4D), confirming the presence of highly polymerized microtubules.
During symmetric cell division in epithelial cells, microtubule dynamics help orient mitotic spindles parallel to the plane of attachment. 40 Matrin3-or CDC14B-s-depleted cells showed misorientation of mitotic spindles, with higher z stack counts and presence of two poles of a mitotic spindle on different optical planes ( Figure 4E). 3D-reconstructed images clearly visualized this misorientation of mitotic spindles (Videos S1, S2, and S3). CDC14B-s overexpression significantly rescued the misoriented mitotic spindles observed in Matrin3 knockdown cells ( Figures 4F and 4G), suggesting that the observed defects in mitotic spindle orientation upon Matrin3 knockdown are largely mediated by defects in CDC14B splicing.
A previous study showed that after depletion of the microtubule depolymerizer MCAK, cells fail to form bipolar spindles. 41 Therefore, we transiently changed the mitotic spindle geometry by treating cells with monastrol, an Eg5 inhibitor that inhibits mitotic spindle bipolarity, 42 followed by monastrol washout and arresting cells in MG132 to capture cells in bipolar metaphase. 43 We found that proportions of monopolar mitotic cells were significantly increased in Matrin3 or CDC14B-s knockdown cells ( Figures S4A and S4B), suggesting a severe delay in mitotic spindle bipolarity. A delay in spindle bipolarity could lead to defects in chromosome congression, 43 and indeed, the proportion of cells with uncongressed chromosomes was increased in asynchronous Matrin3 or CDC14B-s knockdown cells ( Figure S4C). Chromosome congression defects often lead to chromosome segregation defects, especially if the chromosome uncongression was due to defects in kinetochore-microtubule attachments 44 or inefficient spindle assembly checkpoint. 45 Nevertheless, we did not observe significantly increased chromosome segregation defects in Matrin3 or CDC14B knockdown cells ( Figure S4D), suggesting that in Matrin3 or CDC14B-s knockdown cells, chromosomes eventually congress in metaphase and segregate normally, albeit with a delay. We propose that the delay in mitotic spindle bipolarity and chromosome congression could contribute to the decreased proliferation observed upon Matrin3 or CDC14B knockdown.
We arrive at the mechanistic model that Matrin3 contributes to maintenance of microtubule dynamics, spindle morphology, and proper mitotic spindle orientation by regulating CDC14B-s abundance and suggest that defects in these processes may contribute to the reduced proliferation of cells after knockdown of Matrin3.

DISCUSSION
Matrin3 is typically studied in the context of neurodegenerative disorders, like ALS. 46 Here, we discovered that Matrin3 promotes CRC cell growth by suppressing CDC14B splicing, resulting in altered microtubule dynamics.
Although several Matrin3 CLIP experiments were reported previously, 11-13 most of these reports focused on the relationship of the binding sites of Matrin3 and PTBP1, another splicing regulator. By integrating RNA-seq and PAR-CLIP data, we corroborated that Matrin3 acts mainly as a splicing repressor. 11, 13 We do not necessarily interpret this as evidence for Matrin3 being a regulatory molecule but rather as it being required for proper mRNA processing. Consistently, we found that Matrin3 coats its top targets across the entire pre-mRNA (exons and introns) and that loss of Matrin3 results in reduced target mRNA abundance, likely by cytoplasmic degradation pathways dedicated to sensing misprocessed and PTC-bearing mRNA, like NMD. 47 Nevertheless, many of the exon inclusion events upon Matrin3 knockdown will still result in alternatively spliced transcripts.
Matrin3 top targets were enriched for genes related to mitotic spindle formation and regulation and included CDC14B, which is important for mitosis. 48,49 Our data demonstrate that Matrin3-dependent down-regulation of the CDC14B-s variant affected proliferation and phenocopied Matrin3 knockdown. Because we found that CDC14B-s overexpression could rescue the effect of Matrin3 knockdown on clonogenicity, mitotic spindle length, and orientation, we concluded that CDC14B is one of the critical targets responsible for mediating these phenotypes.
We observed that Matrin3 represses the inclusion of exons 13 and 14 of the CDC14B mRNA, likely by coating splice sites on the nascent mRNA and preventing cotranscriptional splicing inclusion 50,51 of these exons and promoting the formation of the canonical CDC14B-s isoform. When included, exon 13 includes a PTC, marking the longer CDC14B-PTC transcript for degradation via NMD. We focused on CDC14B-s because the CDC14B-PTC is of low abundance in our RNA-seq data, consistent with it being a potential NMD target. Nevertheless, a noncoding, shorter transcript including exons 13 and 14 is found in databases (ENST00000481149), even though we do not find evidence of its existence by analysis of long-read sequencing (Isoseq) data from HCT116 cells.
Other studies also observed that Matrin3 loss results in reduced cell proliferation 24,25,52 and suggested increased apoptosis as a cause. Our data indicate that at least in CRC cells, Matrin3 knockdown results in increased polymerizing spindle microtubules, delayed mitotic spindle bipolarity, and delayed chromosome congression, possibly leading to reduced cell proliferation, as studies have shown that the stabilization of microtubules inhibit cell proliferation. [53][54][55] Functional studies of human CDC14B protein had conflicting results regarding its role in mitosis. For instance, Berdougo et al. 33 generated cell clones with non-functional CDC14B protein and showed that it did not alter the proliferation or cause mitotic spindle defects. On the other hand, Tumurbaatar et al. 49 found that CDC14B knockdown reduced cell proliferation, increased the amount of bi-and multinucleated cells, and resulted in impaired metaphase-anaphase transition and caused mitotic delay that resulted in cell death. We observed an increased proportion of cells with monopolar spindles following monastrol washout and uncongressed chromosomes upon knockdown of Matrin3 or CDC14B-s. We rule out that these phenotypes are due to defects per se in mitotic spindle bipolarity or chromosome congression, as we observe a mitotic delay rather than chromosome segregation defects. Other reports demonstrated that CDC14B can have an oncogenic effect through distinct mechanisms. Examples of these mechanisms include the activation of the Ras-Raf-Mek pathway, which mediates the oncogenic effect of CDC14B, 56 and also the degradation of p53, which is promoted by CDC14B phosphatase activity. 57,58 Matrin3 may be one of several RBPs regulating microtubule dynamics. Previously, the RNA-binding properties of the classical end-binding protein EB1, 59 as well as of the adenomatous polyposis coli (APC) 60 scaffolding proteins, were found to be required for their function, suggesting an intimate relationship between RNA regulation and microtubule dynamics. In conclusion, we found that Matrin3 regulates splicing of CDC14B, leading to increased expression of CDC14B-s, which promotes destabilization, shorter length, and proper orientation of microtubules, that culminates in more events of mitosis and, consequently, CRC cell proliferation.

Limitations of the study
We found that the Matrin3 regulates microtubule dynamics, mitotic spindle orientation, and bipolarity by repressing CDC14B pre-mRNA splicing. Although we validated the regulation of CDC14B pre-mRNA splicing by Matrin3 and decreased growth upon Matrin3 knockdown in multiple cell lines, the effects on the microtubule dynamics and the mitotic spindle upon Matrin3 knockdown were not validated in multiple cell types. We also found that the decreased clonogenicity observed upon Matrin3 knockdown was only partially rescued upon CDC14B overexpression in Matrin3 knockdown cells. This suggests that regulation of CDC14B splicing by Matrin3 is not the only mechanism by which Matrin3 functions to promote growth in CRC cells. In addition to CDC14B, several Matrin3 targets involved in regulation of cell cycle were identified in our PAR-CLIP and RNA-seq data. These targets remain to be validated at the pre-mRNA and protein levels and to be mechanistically dissected. Given that Matrin3 is essential for normal development in mice, 19 future studies could determine the key functional targets of Matrin3 in mouse cells and tissues. Finally, it would be important to determine if specific Matrin3-bound RNAs identified in our study also play a role in diseases other than cancer such as neurodegenerative disorders, where Matrin3 mutations have been primarily found. 21

STAR★METHODS RESOURCE AVAILABILITY
Lead contact-Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Ashish Lal (ashish.lal@nih.gov).
Materials availability-Cell lines and plasmids generated in this study are available from the lead contact upon request. All unique/stable reagents generated in this study are available from the lead contact with a completed materials transfer agreement.

•
The RNA-seq, Iso-Seq and PAR-CLIP data are available at Gene Expression Omnibus (GEO) and are publicly available as of the date of publication. Accession numbers are listed in the key resources table.
• This paper does not report original code.
• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. Cells prepared for RNA-seq, immunoblotting or colony formation assays were reverse transfected in 6 well plates. Briefly, 2 x 10 5 to 3.5 x 10 5 cells/per well were seeded with a solution containing 5 μl lipofectamine and 2.5 μl of a 20 μM siRNA in 500 μl of Opti-MEM ™ . The mixture was added to 2 ml of complete medium and cells.
Cells prepared for immunoblotting in Figures 3B and 3D were transfected twice. The second round of transfection was conducted 72 hours after the first transfection and harvested 72 hours after the second transfection.
Allstars Negative Control siRNA (Qiagen, Catalog no. 1027281) were used as control siRNAs. We used SMARTPool siRNAs (Horizon Discovery, Catalog no. L-017382-00-0005) against Matrin3 or (Horizon Discovery, Catalog no. L-011763-00-0005) UPF1. In order to target CDC14B standard variant, we designed an siRNA specific for the junction of the exons 12 and 15 of it (according to the model in the Figure 3A). The abovementioned siRNA, siCDC14B-s, was purchased from Integrated DNA Technologies (IDT). For proliferation and colony formation assays using those siRNAs, cells were transfected using siRNAs against more than one target (e.g., siCTRL and siMatrin3). Consequently, the final concentration of siRNAs was 40 nM Table S3 contains the sequences of custom siRNAs used.
Plasmid construction and generation of stable cell lines-cDNA sequence for the ENST00000375241 isoform of CDC14B-s was downloaded from the UCSC Genome Browser (GRCh38/hg38). The corresponding DNA sequence with linker and 3X-FLAG were synthesized using Twist Biosciences. The DNA fragments were PCR amplified and assembled into doxycycline inducible lentiviral vector using Gibson assembly. 73 The doxycycline inducible vectors with blasticidin selection marker (pCW-Cas9-Blast_empty vector and pCW-Cas9-Blast_CDC14B-FLAG) were obtained from Addgene (Addgene Catalog. no. 83481) and subsequently digested with NheI/BamHI for cloning purposes. The final constructs were verified using Sanger sequencing.
The plasmids were expanded in DH5α cells (Invitrogen, Catalog no. 18265017) and purified with Monarch plasmid miniprep kit (NEB, Catalog no. T1010L). They were cotransfected into 293T cells together with a third generation of lentivirus packaging system using Lipofectamine 2000 (Thermo Fisher, Catalog no. 11668027). HCT116 cells were transduced and after 2 days the cells were selected with 10 μg/mL of blasticidin (Thermo Fisher, Catalog no. A1113903) for 1 week at a MOI of ~0.5. Semi-quantitative RT-PCRs were performed using 1μl of non-diluted cDNAs, 12.5 μl of Phusion ® High-Fidelity PCR Master Mix with HF Buffer (NEB, Catalog no. M0531S) 0.8 μM of each primer (final concentration) and nuclease-free H 2 O enough for a 25 μl of reaction. The PCR cycle used was as follows: 94°C for 5 minutes for 1 X, 98°C for 10 seconds, 50 to 60°C (CDC14B: 50°C; ST7: 56°C; CD44 and SETD5: 57°C; HP1BP3: 60°C) for 30 seconds and 72°C for 0.5 or 2 minutes (2 minutes for CD44, 1 minute for CDC14B and 0.5 minute for the rest of primers) for X28. The PCR products were run on 1 to 2% agarose gels. Table S3 contains the sequences of primers used.

RNA extraction, RT-qPCR and RT-PCR-Cells
Immunostaining-For immunostaining, HCT116 parental or with inducible overexpression of CDC14B-S-FLAG or empty vector cells were transfected with siRNAs as described above, using 1.5 x 10 5 cells/per well that were seeded on coverslips (Corning, Catalog no. 2850-18) inside 6-well plates. To the cells with inducible overexpression of CDC14B-s-FLAG or empty vector was added doxycycline at final concentration of 1 μg/ml the day before the transfection and at the day of transfection. After 48 hours of transfection, cells were fixed with ice-cold methanol for 1 minute. In case we wanted to observe cells in metaphase, we treated the cells with MG132 (Sigma Aldrich, Catalog no. 474790-10MG) at 10 μM for 90 minutes prior to fixing them.
Fixation was followed by blocking with 1% BSA in PBS/0.1% Tween (PBST) for 45 minutes at room temperature. Cells were incubated in primary antibodies for 3 hours at room temperature, washed 3 times in PBST and incubated with secondary antibodies and DAPI for 1 hour at room temperature.  figure 5D were imaged with multiple 10 μm z-stacks and 3D rendering was performed with Imaris version 9.5.0.

3D-reconstructed images-3D reconstructed images derived from cells treated with MG132 immunostained cells with anti-CENP-A, anti-gamma-tubulin and DAPI showed in
Immunoblotting-Cells used for immunoblotting were lysed using 200 μL of RIPA buffer (Life Technologies, Cat no. 89901) and sonicated for 5 seconds three times at power set of 50% (VirTis VIRSONIC 100). The lysates were centrifuged for 10 minutes at 4°C at 16,000 x g, and the supernatant was collected. Total protein was quantified using Pierce ™ BCA Protein Assay Kit (Thermo Fischer, Catalog no. 23225) according to the manufacturer's protocol. 10 to 20 μg of the lysate was loaded into (6% or 10%) SDS-polyacrylamide gels or precast gels Novex ™ WedgeWell ™ 4 to 20% (Thermo Fischer, Catalog no. XP04205BOX).
Monastrol wash out assays-Following siRNA transfections for 48 hours, cells were treated with 10 μM monastrol for 3 hours, followed by washing off monastrol and growing them in monastrol free fresh media containg 10 μM MG132 for 45 minutes to capture cells in bipolar metaphase. Cells were then fixed and immunstained with antibodies against alpha-tubulin to visualize mitotic spindles and CENP-A to visualize centromeres. Cells were imaged to quantify cells with monopolar (failed to form bipolar) and bipolar spindles.
Immunoprecipitation-Cells prepared for immunoprecipitation (IP) were lysed as described above. 25 μl of Pierce ™ Protein A/G Magnetic Beads (Thermo Fischer, Catalog no. 88803) were prepared by washing them twice with PBS 1X and RIPA buffer once. 2 μg of anti-Matrin3 or normal rabbit IgG (Cell Signaling, Catalog no. 2729S) were coupled with the beads overnight. The next day the beads were washed again with RIPA buffer and 500 μg of lysate was used for IP for 4 hours. After that, the beads were washed five times with RIPA buffer, added to SDS-PAGE sample buffer and boiled at 95°C for 5 minutes. The same was done for the input sample. Samples were loaded into a 7.5% SDS-polyacrylamide gels for Immunoblotting using anti-Matrin3 antibody.
Colony formation assays-For this assay, cells were transfected with siRNAs twice. The second transfection was done 48 hours after the first one. Cells were seeded after 24 hours of the second transfection. 2 x10 3 to 3 x 10 3 cells/per well were seeded in 6-well plates for HCT116, LS180, U2OS and HCEC-1CT. In the case of HCT116 cells overexpressing CDC14B-s-FLAG or empty vector in an inducible system, doxycycline enough for 1 μg/ml was added 24 hours after seeding them. After 7 to 8 days, cells were fixed with ice cold methanol for 15 minutes and stained with crystal violet 0.5% in methanol (10%) for 15 minutes. The ImageJ (version 2.0.0-rc-43/1.52n) software package was used to analyze images of the area of colonies.
Proliferation assays-For proliferation assays, 5 x 10 2 to 3 x 10 3 cells/per well were seeded for HCT116, LS180, LS174T, IMR90 and WI38, in 96 well plates after 24 hours of transfection. Cells were incubated on Incucyte ® S3 Live-Cell Analysis Instrument and photographed each 6 hours for at least 4 days. The pictures were analyzed by measuring the occupied area (% confluence) of cell images over time with the software from the manufacturer's device. Sequence reads were aligned with the reference genome (Human -hg19) and the annotated transcripts using STAR (version 2.5.4a) 64 or TopHat (version 2.1.1) 63 for poly (A) capture or ribosomal knockdown sequencing, respectively. The gene expression quantification analysis was performed for all samples using RSEM (version 1.2.31) 65 . Differential gene expression was quantified using DESeq2 (version 1.26.0). 66 For transcript analysis, fastq files from poly (A) sequencing were trimmed using Trimmomatic (version 0.36) 67 and Trim Galore (version 0.4.5) (https:// www.bioinformatics.babraham.ac.uk/projects/trim_galore/). Gencode v19 version transcripts from were quantified using Salmon (version 0.14.1). 68 The normalized TPM expression values were obtained from the Sleuth package (version 0.30.0). 69

RNA-seq-RNA-Seq
Isoform-sequencing (Iso-Seq)-RNA from HCT116 cells was purified using the RNeasy Plus Mini Kit (Qiagen, Catalog number 74134) following the manufacturer's instructions. The library was prepared using Iso-Seq ™ Express Template Preparation protocol (Pacific Biosciences, CA, USA) for transcripts <2 kb. Raw subreads generated with PacBio SMRTlink (smrtlink-release_9.0.0.92188) were converted into HiFi circular consensus sequences (CCS). Demultiplexing was done using PacBio IsoSeq v3. The FLNC reads (Full-Length Non-Concatemer) were mapped to hg38 using Minimap2 software. 71 Isoforms classification was done using SQANTI3. 72 PAR-CLIP-Matrin-3 PAR-CLIP method was performed in three biological replicates as described previously. 31,61 Below is a summary of the protocol used: A. Cell culture and UV crosslinking: HCT116 cells were seeded in 15 mm dishes and were treated with 100 μM final concentration of 4-Thiouridine (Sigma-Aldrich, Catalog no. T4509) when they reached ~80% of confluency. For each sample, the cells were grown in 10 dishes. After 16 hours of treatment, the medium was removed, and cells were crosslinked for 5 min with 312 nm UV light using a Spectrolinker XL-1500 (Spectronics Corporation). Cells were scraped on ice using a rubber policeman and cold DPBS 1X, after which they were centrifuged at 500 xg for 5 min at 4°C. The cells were washed with DPBS 1X, centrifuged as above mentioned and the pellet was frozen at −80°C until further use.
The cell pellets were thawed on ice and cells were lysed using 1.5 volumes of RIPA buffer (150 mM NaCl, 1.0% NP-40, 0.5% sodium deoxycholate, 0.1% SDS (sodium dodecyl sulfate) and 50 mM Tris, pH 8.0 and 1 tablet of complete ™ , EDTA-free Protease Inhibitor Cocktail (Millipore Sigma, Catalog no. 11836170001) per 10 ml of solution). Subsequently, the cells were sonicated (VirTis VIRSONIC 100) 3 times for 30 seconds each (power set to 60%) and left on ice for 15 min. The lysates were cleared by centrifugation at 16,000 x g for 15 min at 4°C.

B. Immunoprecipitation, RNase T1 treatment and dephosphorylation:
The lysates were treated with RNase T1 (1 U/μl) (Thermo Fisher, Catalog no. EN0541) for 15 min at room temperature, and then cooled on ice for 5 min 80 μL of Pierce Protein A/G Magnetic Beads (Thermo Fisher, Catalog no. 88803) coupled with 20 μg of anti-Matrin3 antibody (Bethyl Laboratories, Catalog No. A300-591A) were added to the lysates and they were incubated at 4°C for 4 hours with rotation.

C. Phosphorylation with γ-32 P-ATP of crosslinked RNA, SDS-PAGE and Proteinase K digestion:
The resuspended beads were added to γ-32 P-ATP (0.5 μCi) and T4 PNK kinase (NEB, Catalog no. M0201S) (1 U/μl) and the beads were incubated at 37°C for 30 min with shaking. Next, non-radioactive ATP (100 μM) was added to the reaction and the beads were incubated for more 5 min at 37°C with shaking. Next, the beads were washed 5 times with 1 ml of PNK buffer without DTT, resuspended in 70 μl of 2x SDS-PAGE loading buffer, boiled for 5 min at 95°C, and vortexed.
The recovered material was loaded onto an SDS-Polyacrylamide gel NuPAGE ™ 4 to 12%, Bis-Tris, 1.0 mm, Midi Protein Gel, 12+2-well (Thermo Fisher, Catalog no. WG1401BOX) and after the run the gel was transferred to a nitrocellulose membrane using a semi-dry transfer apparatus. Next, the membrane was wrapped with plastic film and exposed to a blank phosphorimager screen overnight at −20°C.
The next day, the screen was scanned in a phosphorimager imager and the band correspondent to 125 kDa was cut. The membrane pieces were added to 1.5 ml tubes and ribonucleoproteins were digested for 30 minutes initially in 200 ml of Proteinase K buffer (50 mM Tris-HCl pH 7.5, 75 mM NaCl, 6.25 mM EDTA and 1% (w/v) SDS) containing 1.2 mg/ml of Proteinase K (Roche, Catalog no. 03450376103) and then 150 μl of the same buffer including 0.75 mg/ml of Proteinase K was added twice at each 30 minutes, completing 1 hour and 30 minutes of reaction. The recovered RNA was purified with phenol-chloroform extraction and ethanol precipitation. The RNA pellet was resuspended in 9 μl of water. The next day the samples were added to 20 μl of formamide gel loading solution (50 mM EDTA, 0.05% (w/v) bromophenol blue and formamide at 100%) and the samples were loaded onto a 15% denaturing Urea-PAGE, the same was done with the 19/35 nt RNA marker, at both sides of the gel. After the run, the gel was wrapped in plastic film and exposed to a blank phosphorimager screen for at least 1 hour at −20°C. The markers and samples in between 19/35 nt RNA marker were cut from the gel. The recovered RNA was purified with phenol-chloroform extraction and ethanol precipitation. The RNA pellet was resuspended in 9 μl of water.

E. 5' adapter ligation and cDNA synthesis:
For the 5' adapter ligation, the recovered RNA from last step (samples and RNA marker, in separate tubes) were mixed to 2 μl RNA ligase buffer containing ATP (500 mM Tris-HCl, pH 7.6, 100 mM MgCl 2 , 100 mM DTT and 10 mM ATP), 6 μl 50% PEG-8000 and 1 μl of 100 μM 5' adapter. The resultant solutions were boiled for 1 min at 90°C, cooled in ice for 2 min and added to 2 μl of Rnl1 (NEB, Catalog no. M0204S). The reactions were incubated at 37°C for 1 hour with shaking. The samples were added to formamide gel loading solution, loaded onto a 12% denaturing Urea-PAGE and the RNA was recovered as described in item D after 3' adapter ligation.
The ligated RNA was reverse transcribed with SuperScript ™ III Reverse Transcriptase (Invitrogen ™ , Catalog no. 18080044), using the 3' RT primer according to the manufacturer's protocol. Products of a size around 160 bp were isolated using a 3% agarose PippinPrep cassette (Sage Science, Catalog no. CSD3010) on a BluePippin device. After that, if the samples still presented a contaminant band below the expected size (linker-linker byproduct), a new round of PCR and size selection was made. The samples were quantified by TapeStation and sequenced on an Illumina HiSeq 3000 machine as single reads with 50 cycles. Analysis was performed as described previously using PARalyzer 34 built into the PARpipe pipeline 34 pipeline mapping the reads to human genome hg19.
Adapters and primers sequences used for PAR-CLIP are listed in Table S3.

Microscopy and image analysis-Immunostained cells were imaged on Delta Vision
Core system (Applied Precision/GE Healthcare, Issaquah, WA) consisting of Olympus . Z-stacks were acquired with the thickness interval of 10 μm each. Length of mitotic spindles were analyzed by measuring distance between two poles in a mitotic spindle and length across cellular cortex to cortex aligned with the spindle poles using "measure distance" tool in softWoRx. To prepare the figures, images were deconvolved, unless otherwise mentioned, with softWoRx and scaled manually to 8-bit using linear LUT and the same range of scaling for all the images.
Alternative splicing analysis-Alternative splicing analysis was done using fastq files from RNA-Seq made by poly (A) capture using rMATs. 70 The splicing events were filtered for the ones that appeared in at least 5

K-mer or motif analysis-
We used an in-house script to make the K-mer or motif analysis. It was implemented by counting all 5-mer occurrences within the binding site sequences and comparing these values with 5-mer abundances within a background file. The background file was built by matching each binding site with random continuous sequences of equal length taken from the region of the gene where that binding site was found, like intron, 3'UTR, CDS, etc.
Average coverage of binding sites-The visualization of binding site coverage ( Figure  1H) was performed using the RCAS tool (version 1.12.0) 62 using R software (version 4.0.4).

QUANTIFICATION AND STATISTICAL ANALYSIS
The statistical analysis for all data was performed using at least three replicates. The significance of statical analyses were tested using two-tailed Student's t-test when comparing two groups or one-way ANOVA when comparing more than four groups. Twoway ANOVA was performed for proliferation assays. Spearman's coefficient correlation was used for correlation analysis between different combinations of PAR-CLIP replicates samples (Scatterplots from T to C conversions). Two-tailed Kolmogorov-Smirnov test was used for comparisons of cumulative distribution. Data were considered significant when P < 0.05. Prism software (version 9) was used to make the analysis. Details of experiments can be found in figure legends.

Supplementary Material
Refer to Web version on PubMed Central for supplementary material.  Error bars: SD, and *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001, and ns, nonsignificant.  (B) Representative immunofluorescence images of MG132-treated HCT116 cells in metaphase immunostained for CDC1B-s-FLAG using anti-FLAG or α-tubulin (spindle microtubules) (left panel) after induction of CDC1B-s-FLAG with doxycycline and transfection with Matrin3 siRNAs or control siRNA. Nuclei were stained with DAPI, Scale bar, 5 μm; red bars indicate length of mitotic spindles. (Right panel) Bargraph of ratio between spindle length and cell size was calculated. "N", number of cells. Unpaired, two-sided t test.
(C) Representative immunofluorescence images showing EB1 comet density in interphase of HCT116 cells transfected with siRNAs as indicated and immunostained for EB1 and stained with DAPI for nuclei. Scale bar, 5 μm. n = 2. (D) Representative immunofluorescence images of MG132-treated cells in metaphase showing EB1 comets in cells treated with siRNAs as indicated and immunostained for EB1 and images pseudocolored to show highly intensified comets as yellow and less intensified comets as red. Scale bars, main image, 5 μm; insets, 1 μm. n = 2. (E) Representative immunofluorescence images of MG132-treated cells in metaphase showing mitotic spindle orientation as depicted by planes of two spindle poles denoted with red asterisks in HCT116 cells transfected with the indicated siRNAs and stained for γ-tubulin (left panel). Cells were imaged with multiple 10 μm z stacks, and the numbers of z stacks between planes of two poles were measured in each siRNA condition and plotted as a bar plot (right panel). "N" denotes number of cells analyzed from two independent experiments. Unpaired, two-sided t test. (F) Representative immunofluorescence images of MG132-treated HCT116 cells in metaphase showing mitotic spindle orientation as depicted by planes of two spindle poles in HCT116 cells transfected with the indicated siRNAs and immunostained for CDC1B-s-FLAG using anti-FLAG or γ-tubulin (spindle poles). Cells were imaged with multiple 10 μm z stacks, and the numbers of z-stacks between planes of two poles were measured in each siRNA condition and plotted as a bar plot (right panel). "N", number of cells from two independent experiments. (G) Quantification of the difference between two poles depicted in (F). Unpaired, two-sided t test. Error bars: SD, and *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001, and ns, non-significant.