Integrative CRISPR Screening and RNA Analyses Discover an Essential Role for PUF60 Interactions with 3' Splice Sites in Cancer Progression.

Introduction
RNA-binding proteins (RBP) govern posttranscriptional gene regulation, a mechanism exploited by cancer cells to modulate protein expression levels and sustain oncogenic proliferation (1). RBP–RNA interactions control RNA splicing, polyadenylation, stability, subcellular localization, and translation (2). Dysregulated splicing is a significant driver of cancer progression as most tumor types show extensive alterations compared with healthy tissues (3). Aberrant expression of splicing-regulatory RBPs generates alternatively spliced isoforms that affect oncoprotein and tumor suppressor function, influencing apoptosis, angiogenesis, proliferation, and metastasis (4). Although RBP dysregulation in cancer is increasingly recognized, experimental and computational advances have identified thousands of novel RBPs for which functions in cancer remain poorly understood (5, 6). Therefore, large-scale functional discovery and characterization of RBPs in cancer is crucial for deepening our understanding of tumor biology and identifying new therapeutic targets.
Triple-negative breast cancer (TNBC) is the deadliest breast cancer subtype, lacking expression of the estrogen receptor, progesterone receptor, and HER2 (7). Therapies that block the effects of estrogen and progesterone are effective in treating hormone receptor–positive breast cancers (8); however, TNBCs lack such targeted therapies because of their absence of hormone receptor expression (9), emphasizing an urgent need for the discovery of effective targets. Accumulating evidence indicates that TNBCs display significant differences in the expression of mRNAs subject to RBP-dependent posttranscriptional control from those in hormone receptor–positive breast cancers and normal breast tissue (10). Moreover, RBPs involved in mRNA splicing and m6A-mediated RNA decay are selectively required for cell survival in TNBC but not for other cell types (11, 12). These findings emphasize the importance of systematically identifying RBPs that are essential for TNBC tumor growth yet are nonessential in healthy tissue–derived cells.
Here, we performed in vivo and in vitro RBP-focused CRISPR/Cas9 screens to identify RBPs that are specifically required to sustain the survival and proliferation of TNBC cells. We identify 50 RBPs that are required for the survival of TNBC tumors while being dispensable for normal cellular survival. Proteins associated with the U2 small nuclear ribonucleoprotein (U2 snRNP) complex were enriched among the candidates, with the poly(U)-binding splicing factor 60 (PUF60) emerging as a key modulator of TNBC proliferation and survival. We found that knockdown of PUF60 or inhibition of PUF60 binding at splice sites severely restricts cellular proliferation and induces tumor regression in multiple TNBC xenograft models. PUF60 drives exon inclusion within a network of proliferation-associated transcripts, including those involved in cell-cycle progression, chromatin organization, and the DNA damage response. mRNAs within these proliferation-associated pathways undergo widespread exon skipping and destabilization after disrupting PUF60-dependent splice site activity, resulting in TNBC cell apoptosis. Overall, our studies define the functional and molecular role of PUF60-mediated splicing regulation in posttranscriptional gene regulation of cell cycle and genomic stability pathways in TNBC and suggest that PUF60 inhibition is a promising therapeutic approach.

Materials and Methods

Cell culture
HEK293xT (female, Takara Bio, cat. #632180, RRID: CVCL_4401) immortalized cells were maintained in DMEM (Thermo Fisher Scientific, cat. #11965118) supplemented with 10% FBS (Thermo Fisher Scientific, cat. #26140079). MCF10A (female, ATCC, cat. #CRL-10317, RRID: CVCL_0598) immortalized cells were maintained in Human Mammary Epithelial Cell Basal Medium (Thermo Fisher Scientific, cat. #M171500) supplemented with 0.4% bovine pituitary extract, 0.01 μg/mL recombinant human insulin-like growth factor, 0.5 μg/mL hydrocortisone, and 3 ng/mL human EGF (Thermo Fisher Scientific, cat. #S0155). MDA-MB-231 (female, ATCC, cat. #HTB-26, RRID: CVCL_0062) immortalized cells were maintained in DMEM (Thermo Fisher Scientific, cat. #11965118) supplemented with 10% FBS (Thermo Fisher Scientific, cat. #26140079). MDA-MB-436 (female, ATCC, cat. #HTB-130) immortalized cells were maintained in DMEM (Thermo Fisher Scientific, cat. #11965118) supplemented with 10% FBS (Thermo Fisher Scientific, cat. #26140079) and 10 μg/mL recombinant human insulin (Thermo Fisher Scientific, cat. #12585014). SUM149 (female, Applied Biological Materials, cat. #T1250 RRID: CVCL_3422) immortalized cell lines were maintained in Ham’s F-12 Nutrient Mix (Thermo Fisher Scientific, cat. #11765054) supplemented with 5% FBS (Thermo Fisher Scientific, cat. #26140079), 1 μg/mL hydrocortisone (STEMCELL Technologies, cat. #07925), and 5 μg/mL recombinant human insulin (Thermo Fisher Scientific, cat. #12585014). IMR-90 (female, ATCC, cat. #CCL-186, RRID: CVCL_0347) and WI-38 (female, ATCC, cat. #CCL-75, RRID: CVCL_0579) immortalized cells were maintained in Eagles Minimum Essential Medium (ATCC, cat. #302003) supplemented with 10% FBS (Thermo Fisher Scientific, cat. #26140079). MCF12A (female, ATCC, cat. #CRL-3598, RRID: CVCL_3744) immortalized cells were maintained in a 1:1 mixture of DMEM and Ham’s F12 medium (Thermo Fisher Scientific, cat. #11320082), 20 ng/mL human EGF (Lonza, cat. #CC-4107), 100 ng/mL cholera toxin (MilliporeSigma, cat. #C8052), 0.01 mg/mL bovine insulin (MilliporeSigma, cat. #11070738), 500 ng/mL hydrocortisone (STEMCELL Technologies, cat. #07925), and 5% horse serum (Thermo Fisher Scientific, cat. #16050130). All cells were passaged every 2 or 3 days with TrypLE Express Enzyme (Thermo Fisher Scientific, cat. #12604021) and maintained in a humidified incubator at 37°C with 5% CO2. Cells were routinely tested for Mycoplasma contamination with a MycoAlert Mycoplasma Detection Kit (Lonza, cat. #LT07-318) and were found negative for Mycoplasma.

Animal studies
Animal protocols were approved by the University of California, San Diego, Institutional Animal Care and Use Committee under protocol S12099. Athymic Nude-Foxn1nu (Envigo, RRID: IMSR_ENV:HSD-069) were purchased at 6 weeks of age.

CRISPR plasmid library preparation
A subset of single-guide RNA (sgRNA) sequences from our previously developed CRISPR/Cas9 lentiviral library (13) was ordered as a pool of equal molar oligonucleotides (Supplementary Table S1). The lentiCRISPR RBP plasmid library was cloned using previously reported methods (14). To construct a doxycycline (DOX)-inducible library, the TLCV2 backbone (Addgene, cat. #87360, RRID: Addgene_87360; ref. 15) was digested with BsmBI, and PCR-amplified sgRNA oligonucleotide inserts were Gibson assembled. To maintain 300× library complexity, plasmid products were amplified using parallel electroporations. Electroporated bacteria were spread on carbenicillin agar plates (24.5 × 24.5 cm) and grown for 16 to 18 hours at 32°C. One day later, colonies were scraped, collected, and maxi prepped.

Lentivirus production and purification
HEK293xT cells were plated on 15-cm plates at 70% confluency the day before transfection. One hour before transfection, media were removed and replaced with 15 mL of prewarmed Opti-MEM (Thermo Fisher Scientific, cat. #31985062). Transfections were carried out using 62.5 μL Lipofectamine 3000 (Thermo Fisher Scientific, cat. #L3000001), 125 μL of P3000 Enhancer, 12.5 μg lentiCRISPR plasmid library, 6.25 μg pMD.2g (Addgene, cat. #12259, RRID: Addgene_12259), and 9.375 μg psPAX2 (Addgene, cat. #12260, RRID: Addgene_12260). Eight hours after transfection, media were changed to DMEM supplemented with 10% FBS. After 48 hours, the medium supernatant was filtered through a 0.45-μm nitrocellulose membrane. The virus was ultracentrifuged at 24,000 rpm for 2 hours at 4°C and resuspended overnight in PBS at 4°C.

Multiplicity of infection
To determine the volume of virus required for a multiplicity of infection (MOI) of 0.3 for each cell type, virus was titrated across a six-well plate with 1 million cells per well in medium containing 8 μg/mL polybrene (MilliporeSigma, cat. #TR-1003-G). After 48 hours, the medium was replaced with DMEM supplemented with 10% FBS and 2 μg/mL puromycin. Three days later, the number of surviving cells was counted to establish the MOI based on the virus volume that resulted in 30% cell survival.

In vitro RBP CRISPR screen
MCF10A, MDA-MB-231, MDA-MB-436, and SUM149 cells were infected with RBP CRISPR library lentivirus at 0.3 MOI in medium containing 8 μg/mL polybrene (MilliporeSigma, cat. #TR-1003-G). After 24 hours, media without polybrene were replaced. Following an additional 48 hours, media containing 2 μg/mL puromycin (Thermo Fisher Scientific, cat. #A1113803, CAS: 58-58-2) were replaced and maintained on for 4 days to select for infected cells. Cells expressing the RBP CRISPR library were then expanded by plating 4 million cells per 15-cm plate, with a total of two plates per replicate. A plate from each replicate was treated with 1 μg/mL DOX (MilliporeSigma, cat. #D9891, CAS: 24390-14-5) once. The cells were cultured for 14 days, with media and DOX replaced every 2 days and cell splitting every 4 days. For each condition and replicate, 4 million cells were harvested and snap frozen.

In vivo RBP CRISPR screen
MDA-MB-231s, MDA-MB-436s, and SUM149s expressing the RBP CRISPR library were expanded and subcutaneously transplanted (3 × 106 cells per mouse to maintain 200× library complexity) into the right flank of athymic nude mice (female, 6 weeks old). Following sufficient tumor engraftment (tumor volume of ∼300 mm3), mice were randomized and maintained on 5% sucrose water (five mice/cell line, –DOX) or 5% sucrose water with 2 mg/mL DOX (MilliporeSigma, cat. #D9891, CAS: 24390-14-5; five mice/cell line, +DOX) for 14 days. Tumors were washed with sterile PBS and minced with a sterilized blade into 2- to 3-mm cubes. Tumor pieces were dissociated in enzyme medium (∼0.1 cm3 tumor/∼2 mL), shaking for 30 minutes at 37°C. The enzyme medium consisted of 10× enzyme mix of collagenase IV [1 g/100 mL Hank’s Balanced Salt Solution (HBSS); Sigma, cat. #C-5138], hyaluronidase (100 mg/100 mL HBSS; Sigma, cat. #H-6254), and deoxyribonuclease (20,000 U/100 mL HBSS; Sigma, cat. #D-5025) in RPMI-1640 medium (Thermo Fisher Scientific, cat. #11875093) with 10% penicillin/streptomycin (Thermo Fisher Scientific, cat. #15070063). For each replicate, 4 × 106 cells were spun down and snap frozen.

Bulk sgRNA library preparation for in vitro and in vivo RBP CRISPR screens
DNA libraries were prepared using a targeted enrichment approach as previously described (13). Briefly, genomic DNA (gDNA) was extracted from pellets of 4 million cells using the DNeasy Blood and Tissue kit (Qiagen, cat. #69504) and then sonicated to ∼1000 bp by Biorupter. The average fragment size was assessed with gDNA ScreenTapes on the Agilent TapeStation (Agilent, cat. #5067-5365). Fragments containing sgRNAs were isolated with biotinylated RNA probes that target a conserved region on the TLCV2 backbone, and these were captured with streptavidin beads. Isolated gDNA fragments were purified using the DNA Clean and Concentrator-5 Kit (Zymo Research, 11-303C). PCR was then performed with following primers flanking the sgRNA:Forward (5′→3′): CCT​ACA​CGA​CGC​TCT​TCC​GAT​CTT​GTG​GAA​AGG​ACG​AAA​CAC​CG;

Reverse (5′→3′): GTT​CAG​ACG​TGT​GCT​CTT​CCG​ATC​TCC​ACT​TTT​TCA​AGT​TGA​TAA​CGG​ACT​AGC​C

This was followed by a second amplification with Illumina sequencing primers. Library quality was evaluated using an Agilent D1000 ScreenTape on the Agilent TapeStation (Agilent, cat. #5067-5582), and then the libraries were sequenced to 6M reads per library on the HiSeq 4000 sequencer in paired-end 55 bp mode.

Statistical analysis of in vivo and in vitro RBP CRISPR screens
Reads were aligned to the RBP library, and the MaGeCK version 0.5.4 software package (RRID: SCR_025016; ref. 16) was used to statistically identify RBP candidates. RBP candidates with a Wald P value <0.05 and β score <0 were considered candidate essential RBPs.

The Cancer Genome Atlas and Genotype-Tissue Expression data description
Publicly available datasets from The Cancer Genome Atlas Breast Invasive Carcinoma (TCGA-BRCA; RRID: SCR_003193; ref. 17) and the Genotype-Tissue Expression (GTEx; RRID: SCR_013042; ref. 18) were directly downloaded from the UCSC Xena portal at https://xena.ucsc.edu/ (RRID: SCR_018938; ref. 19). For gene expression data, we used mRNA expression z-scores calculated by RNA-seq by Expectation Maximum and categorized the data by samples from normal breast tissue (normal) and basal clinical tumor subtype (basal). For survival data, the Kaplan–Meier probability was determined using the UCSC Xena portal using clinical overall survival status among TCGA-BRCA patients by grouping the first and fourth quartiles of PUF60 copy-number variations (CNV) determined by the log2 (tumor/normal).

Knockdown experiments
Cells were transduced with the TRC lentiviral short hairpin RNA (shRNA) vector nontargeting control (NTC; MilliporeSigma, cat. #SHC002) and TRC lentiviral shRNA vector PUF60 (shPUF60-1; MilliporeSigma, cat. #TRCN0000322833; shPUF60-2; MilliporeSigma, cat. #TRCN0000017248; and shPUF60-3; MilliporeSigma, cat. #TRCN0000322893) or TRC lentiviral shRNA vector PSMA1 (shPSMA1; MilliporeSigma, cat. #TRCN0000003872) for 48 hours before treatment with 2 μg/mL puromycin (Thermo Fisher Scientific, cat. #A1113803, CAS: 58-58-2). Cells were analyzed 4 to 5 days after the addition of lentivirus for all assays unless noted otherwise.

Western blotting
Cells were lysed with RIPA buffer (Sigma-Aldrich, cat. #R0278) containing protease inhibitor (Thermo Fisher Scientific, cat. #A32953). Protein lysates were centrifuged to pellet and remove insoluble material and were then quantified using the Pierce BCA Kit (Thermo Fisher Scientific, cat. #A55863). Protein lysates were run on a 4% to 12% NuPAGE Bis-Tris gel (Thermo Fisher Scientific, cat. #NP0322) and transferred to a polyvinylidene fluoride membrane. Membranes were blocked in TBS containing Tween 20 (TBST) with 5% milk for 20 minutes and probed overnight at 4°C with the following primary antibodies: rabbit pAb anti-PUF60 (Fortis Life Sciences, cat. #A302-817, RRID: AB_10631036), rabbit pAb anti-PSMA1 (Invitrogen, cat. #PA1-963, RRID: AB_2171381), mouse mAb anti-GAPDH (Millipore, cat. #MAB374, RRID: AB_2107445), rabbit pAb anti-PARP (Cell Signaling Technology, cat. #9542, RRID: AB_2160739), rabbit mAb anti-APC6 (Cell Signaling Technology, cat. #9499, RRID: AB_10896848), and rabbit pAb anti–V5 tag (Bethyl Laboratories, cat. #A190-120A, RRID: AB_67586). Membranes were washed three times for 5 minutes with TBST and then probed for 1 hour at room temperature in TBST containing 5% milk with the following secondary antibodies diluted 1:5,000: anti–mouse IgG, HRP linked (Cell Signaling Technology, cat. #7076, RRID: AB_330924) and anti–rabbit IgG, HRP linked (Cell Signaling Technology, cat. #7074, RRID: AB_2099233). Membranes were washed three times for 5 minutes with TBST and developed using Pierce ECL detection kits (Thermo Fisher Scientific, cat. #32106) on an Azure Western Blot Imaging System (RRID: SCR_026671).

Generation of L140P base-edited MDA-MB-231 cells
MDA-MB-231 cells were co-transfected using polyethyleneimine (Fisher, 502559821) either with a plasmid encoding the ABE8e-SpRY base editor and GFP (Addgene, #226858) and an sgRNA targeting human PUF60 (5′-TCC​CCC​AGC​TCA​TAG​TAG​AT) to introduce the L140P mutation or the base editor plasmid alone as a control. The sgRNA sequence can be found in the key resources table. Forty-eight hours following transfection, GFP-positive cells were isolated by FACS and a subset was harvested for gDNA extraction and Sanger sequencing to determine A-to-G editing efficiency. Sorted cells were seeded into a 96-well plate and imaged using the CELLCYTE X imaging system (Cytena) 96 hours following transfection. Cells were monitored over a 5-day period at 10× magnification, with phase contrast images captured every 4 hours. Image analysis was performed using CELLCYTE Studio to quantify cell confluence over time.

Plasmid construction and generation of L140P-PUF60 and wild-type PUF60 stable cell lines
To generate L140P-PUF60 and wild-type (WT) PUF60 stable cell lines, an inducible lentiviral CRISPRi expression backbone (Addgene, cat. #167935, RRID: Addgene_167935) was digested at XHO1 and BAMH1 sites to remove the KRAB, dCas9, and DHFR domains positioned immediately downstream of the Tet operator sequence. gBlocks (Integrated DNA Technologies) containing codon-optimized PUF60 open reading frames with (L140P-PUF60) and without (WT-PUF60) the L140P mutation, a C-terminal V5 tag, and 25nt complimentary sequences were assembled with the digested lentiviral backbone (immediately downstream of the Tet operator sequence) in a two-fragment Gibson assembly reaction. To generate lentiviral particles for inducible L140P-PUF60, WT-PUF60, and SMARTvector-inducible PUF60 shRNAs (tet-shPUF60-1; Horizon Discovery, V3SH11252-228880222; tet-shPUF60-2; Horizon Discovery, V3SH11252-228079444; and tet-shPUF60-3; Horizon Discovery, V3SH11252-224918077), we seeded HEK293T cells on 10-cm plates at 70% confluency, 24 hours prior to transfection. Transfections were performed using 45 μL Lipofectamine 3000, 60 μL of P3000 Enhancer, 7 μg transgene, 3.6 μg pMD.2g, and 6.6 μg psPAX2. Eight hours following transfection, media were changed to DMEM + 10% FBS. Forty-eight hours following transfection, the medium supernatant was filtered through a 0.45-μm low protein-binding membrane. The virus was concentrated with Lenti-X Concentrator (Takara Bio) and centrifuged and resuspended according to manufacturer’s instructions. To make tet-shPUF60 cell lines, MDA-MB-231 cells were individually transduced with the inducible PUF60 shRNAs for 48 hours before treatment with 2 μg/mL puromycin (Thermo Fisher Scientific, cat. #A1113803, CAS: 58-58-2). To make L140P-PUF60 and WT-PUF60 cell lines, stable tet-shPUF60-1 and tet-shPUF60-2 MDA-MB-231 cell lines were then transduced with L140P-PUF60 and WT-PUF60 lentivirus and sorted for the top 10% of red fluorescent protein (RFP)–expressing cells on a BD Influx Cell Sorter.

Enhanced cross-linking and immunoprecipitation sequencing library preparation
Experiments were performed as previously outlined (20) with biological duplicates. In brief, 20M cells were UV cross-linked at 400 mJ/cm2 constant energy, lysed, and sonicated by Bioruptor. Lysates were treated with RNase 1 to fragment RNA, and then protein–RNA complexes were immunoprecipitated (IP) with PUF60 antibody (rabbit pAb anti-PUF60, Fortis Life Sciences, A302-818) or V5 antibody (rabbit pAb anti–V5 tag, Fortis Life Sciences, A190-120A) precoupled to Sheep Anti-Rabbit Dynabeads. Paired inputs (2% of lysate) were first removed from each sample. IP samples were stringently washed, and RNA from all samples was dephosphorylated with FastAP (Thermo Fisher Scientific) and T4 RNA ligase (NEB), followed by on-bead ligation of barcoded RNA adapters to the 3' end of RNA in samples (NEB). RNA–protein complexes were run on 4% to 12% NuPAGE Bis-Tris gels and transferred to nitrocellulose membranes in which the RNA in the region 60 to 140 kDa was excised from the membrane and treated with proteinase K (NEB) to release RNA. Input samples were dephosphorylated with FastAP (Thermo Fisher Scientific) and T4 PNK and a 3' RNA adapter was ligated with T4 RNA ligase (NEB) so synchronize input and IP samples. RNA was reverse transcribed using SuperScript III Reverse Transcriptase (Thermo Fisher Scientific), followed by ExoSAP-IT (Applied Biosystems) to remove excess primers. qPCR was used to determine the appropriate number of PCR cycles for library amplification, followed by amplification with Q5 (NEB, cat. #M0492S) and quality control using an Agilent D1000 ScreenTape (Agilent, cat. #5067-5582). Libraries were sequenced to 20M reads on the HiSeq 2000, 2500, or 4000 platform.

Computational analysis of enhanced cross-linking and immunoprecipitation data
Reproducible eCLIP peaks were called using the latest release of the Skipper pipeline, available on GitHub (https://github.com/YeoLab/skipper RRID: SCR_026260; ref. 21). In brief, Skipper trims and aligns reads, determines counts per window, and compares enrichment in IP samples against input samples. Enriched elements are then considered reproducible if they meet a 20% false discovery threshold in both replicates for each sample. Only those with a significant FDR less than 0.05 and enrichment odds ratio greater than 3 were included in further analysis. For V5-tagged enhanced cross-linking and immunoprecipitation (eCLIP), nonspecific V5 antibody binding in WT MDA-MB-231 cells was removed from the analysis of L140P-PUF60 and WT-PUF60 eCLIPS using BEDTools version 1.31.1 (RRID: SCR_006646; ref. 22). Motifs were analyzed using HOMER version 4.11 (RRID: SCR_010881; ref. 23), and metagene plots were generated using Metadensity version 0.0.1 (24). Overlapping windows among cell types were determined using BEDTools version 1.31. (22), and splicing maps were generated using RBP-Maps version 0.1.4 (25).

RNA sequencing library preparation
To assess the impact of PUF60 knockdown on both normal and TNBC cell lines, total mRNA was isolated from three biological replicates of nontargeting controls and three biological replicates for each PUF60-targeting shRNA (shPUF60-1, shPUF60-2, and shPUF60-3) in MCF10A, MDA-MB-231, MDA-MB-436, and SUM149 cell lines. To evaluate the effect of the L140P mutation on PUF60 splicing activity, total mRNA was extracted from three WT PUF60 and three L140P-mutant biological replicates in MDA-MB-231 cells using a Direct-zol RNA Miniprep Kit (Zymo Research, cat. #R2071). RNA quality was assessed using an Agilent RNA ScreenTape (Agilent). A total of 500 ng RNA was rRNA depleted, and library preparation was carried out with the Stranded mRNA Prep Ligation Kit (Illumina, cat. #20040534). Libraries were quality controlled using an Agilent D1000 ScreenTape (Agilent, cat. #5067-5582). Sequencing was performed on HiSeq 2000, 2500, or 4000 platforms, collecting 60 million 150 bp paired-end reads per sample.

Computational analysis of integrated RNA sequencing and eCLIP data
Adapter trimming and read mapping to the human genome build hg38 were performed using STAR version 2.4.0. Differential expression analysis for genes with a minimum transcripts per million (TPM) of 10 in any sample was conducted with DEseq2 version 1.22.1 (26). Differential alterative splicing (AS) events were analyzed using rMATS version 4.0.2 (RRID: SCR_023485; ref. 27), with only differential splicing events with a sum of 30 reads across all conditions considered for downstream analysis. Significant differentially splicing was defined by an absolute inclusion-level difference greater than 5% and an FDR of less than 5%. Exon splicing events containing an adjacent upstream 3′SS eCLIP window were determined using BEDtools version 1.31.1 (RRID: SCR_006646; ref. 22).

Gene Ontology analysis
Gene ontology enrichment was analyzed using Metascape version 3.5 (RRID: SCR_016620; ref. 28), with background genes consisting of those expressed with a TPM greater than 10 in each cell line for the integrated eCLIP and RNA sequencing (RNA-seq) analysis. For the CRISPR screen candidate ontology analysis, the entire cohort of RBPs from the Lenti CRISPR library was utilized as the background list.

siRNA reverse transfection
A total of 1.5 × 106 MDA-MB-231 cells were reverse transfected with Lipofectamine RNAiMAX (Thermo Fisher Scientific, 13778030) according to the manufacturer’s protocol. siRNAs targeting PUF60 (Horizon Discovery, L-012505-01), CDC16 (Horizon Discovery, L-012505-01), or a nontargeting control siRNA (Horizon Discovery, D-001810-10) were added at a final siRNA concentration of 35 nmol/L. Cells were then harvested 48 hours later for downstream analysis, including qRT-PCR and time-course imaging.

qRT-PCR analysis
RNA was extracted with a Quick-RNA Miniprep Kit (Zymo Research, cat. #R1054), and cDNA was synthesized from total RNA using a High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems, cat. #4368814). Real-time PCR was performed using Power SYBR Green PCR Master Mix (Applied Biosystems, cat. #4367659) and primers specific to human GAPDH (F 5′-GAA​GGT​GAA​GGT​CGG​AGT​C; R 5′-GAA​GAT​GGT​GAT​GGG​ATT​TC), human PUF60 (F 1 5′-GGA​AAC​CTC​CAC​AGG​GCA​CA; R 1 5′-CGC​GAT​GGT​CTG​CTT​CAC​CA; F 2 5′-TGG​AAA​CCT​CCA​CAG​GGC​AC; 2 5′-GCT​CTG​CTC​CAT​GGC​GTA​CT), and human CDC16 (F 1 5′-CAA​CAG​TAT​CAA​AGT​GCT​CTA​TTT​TGG​GCA​G; R 1 5′-TTG​TCC​AGT​TTT​CGT​GAC​CGA​AGT​G; F 2 5′-AGC​ATG​TCG​TTA​CCT​TGC​AGC​TAG​G; R 2 5’-TGA​AGC​CAC​TTT​CGT​CCT​TCA​AGT​AT). Values of gene expression were normalized to GAPDH expression using the ddCT method.

Cell cycle analysis
MDA-MB-231 and MCF10A cells were transduced with NTC, shPUF60-1, shPUF60-2, or shPUF60-3 virus and selected for 2 days with 2 μg/mL puromycin (Thermo Fisher Scientific, cat. #A1113803). Five days after transduction, cells were stained with Zombie Violet dye from the Zombie Violet Fixable Viability Kit (BioLegend, cat. #423113) according to the manufacturer’s instructions. Then, cells were fixed in 66% ice-cold ethanol and stained with propidium iodide (PI) using the Propidium Iodide Flow Cytometry Kit (Abcam, cat. #ab139418) according to the manufacturer’s instructions. Cells were analyzed by flow cytometry using the BD LSRFortessa flow cytometer under the Pacific Blue (Zombie Violet) and PE-Texas Red (PI) channels. Analysis and gating were performed using FlowJo (RRID: SCR_008520).

Comet assay
The comet assay was performed according to the manufacturer’s instructions (Abcam, cat. #ab238544). Briefly, MDA-MB-231 and MCF10A cells were seeded in 12-well plates 4 days after transduction with either nontargeting control NTC or shPUF60-1, shPUF60-2, or shPUF60-3 virus. Cells expressing shNTC were treated with either 40 µmol/L etoposide (Thermo Fisher Scientific, cat. #J63651.MC, CAS: 33419-42-0) or DMSO vehicle control for 4 hours before collection by gentle scraping. A total of 1,000 cells in 10 μL were mixed with 60 μL agarose and transferred onto comet slides. The agarose–cell suspension was allowed to solidify, followed by lysis at 4°C for 1 hour. Electrophoresis was performed with cold TBE buffer for 15 minutes at 60 volts. The comet slides were rinsed in H2O three times and then fixed with 70% cold ethanol for 5 minutes. Air-dried slides were then stained with Vista Green DNA Dye and viewed on a fluorescence Keyence microscope. The percentage of DNA in the tail, tail length, and tail moment were assessed using the ImageJ plug-in in OpenComet (RRID: SCR_021826), with 50 to 100 nuclei measured per condition.

PUF60 knockdown in vivo tumorigenicity assays
To determine the effect of PUF60 knockdown on MDA-MB-231 tumor growth, MDA-MB-231 cells stably expressing tet-shPUF60-1 (Horizon Discovery, cat. #V3SH11252-228880222) or tet-shPUF60-2 (Horizon Discovery, cat. #V3SH11252-228079444) were selected with 2 μg/mL puromycin (Thermo Fisher Scientific, cat. #A1113803, CAS: 58-58-2) and subcutaneously transplanted (3 × 106 cells in a single transplant site per mouse) into 20 athymic nude mice (female, 6 weeks old) per shRNA. Mice were randomized and half were maintained on 5% sucrose water (−DOX) or 5% sucrose water with 2 mg/mL DOX (MilliporeSigma, cat. #D9891, CAS: 24390-14-5; +DOX) 14 days after transplantation. To determine the effect of PUF0 knockdown in additional TNBC models, MDA-MB-436 and SUM149 cells stably expressing tet-shPUF60-1, tet-shPUF60-3 (Horizon Discovery, cat. #V3SH11252-224918077), or tet-shNTC (Horizon Discovery, cat. #VSC11657) were similarly selected with 2 μg/mL puromycin and then subcutaneously transplanted (3 × 106 cells in each of two transplant sites per mouse) into athymic nude mice (female, 6 weeks old). For each cell line, five mice were used per PUF60-targeting shRNA group (tet-shPUF60-1 and tet-shPUF60-3) and 10 mice for the tet-shNT control group. All mice in these cohorts were maintained on 5% sucrose water with 2 mg/mL DOX 14 days after transplantation. For all cohorts, tumor volume and body weight were measured twice weekly using calipers and a digital scale. RFP-positive, DOX-treated mice were imaged in an IVIS Spectrum in vivo imaging system (Revvity, RRID: SCR_027425) at the initial (days 3–6) and final (days 39–58) days of DOX treatment. Tumors were harvested after reaching an average volume of 800 to 1,000 mm3 and processed by either snap freezing for protein extraction or fixation in 4% paraformaldehyde for IHC analysis.

IHC
Tumor samples were fixed in 4% paraformaldehyde and then paraffin embedded. Sections of 5-μm thickness were prepared using a microtome by the Moores Cancer Center Histology Core. Sections were deparaffinized in Histoclear and rehydrated through a graded alcohol series. Antigen retrieval was performed with citrate buffer. Tissues were stained with primary antibody against rabbit mAB anti–cleaved caspase-3 (1:150; Cell Signaling Technology, cat. #9664, RRID: AB_2070042) and subsequently with a secondary horse anti–rabbit IgG using the ImmPRESS Kit (Vector Laboratories, cat. #MP-7801-15, RRID: AB_3712183). Slides were stained with diaminobenzidine chromogen (Vector Laboratories, cat. #SK-4105, RRID: AB_2336520), followed by hematoxylin QS (Vector Laboratories, cat. #H-3404-100) counterstaining. Images were acquired on the Revolve microscope (Echo Laboratories, RRID: SCR_026523) using 10X objectives for quantification and 17X for figure representation. Quantification of cleaved caspase-3–positive cells was performed with QuPath-0.5.1 (analyze, cell detection, and positive cell detection, RRID: SCR_018257), analyzing 20 to 35 images per tissue section from each of nine DOX-treated tumors and 10 vehicle control tumors. The mean percentage of cleaved caspase-3 (CC3)–positive cells per tumor was then used to assess significant differences in CC3 signal between DOX-treated and vehicle control tumors.

L140P-PUF60 and WT-PUF60 in vivo tumorigenicity assays
To determine the effect of L140P PUF60 mutation on tumor growth, RFP-sorted L140P-PUF60 and WT-PUF60 MDA-MB-231 cells were subcutaneously transplanted (3 × 106 cells per mouse) into athymic nude mice (female, 6 weeks old; 15 mice for L140P-PUF60 and 15 mice for WT-PUF60). All mice were maintained on 5% sucrose water with 2 mg/mL DOX (MilliporeSigma, cat. #D9891, CAS: 24390-14-5; +DOX) 14 days after transplantation. Twice each week, tumors were measured with calipers and mouse body weight was recorded. RFP-positive, DOX-treated mice were imaged in an IVIS Spectrum in vivo imaging system (Revvity, RRID: SCR_027425) on day 0 (initial) and day 58 (final) of DOX treatment. Tumors were harvested after reaching a volume of 1,000 mm3 on average, followed by snap freezing for protein extraction.

Time-lapse microscopy
To measure confluence over time, L140P-PUF60 and WT-PUF60 MDA-MB-231 cells were seeded at 10k cells in Incucyte ImageLock plates (Essen BioSciences, cat. #4379) 24 hours prior to imaging. Plates were loaded into the Incucyte and imaged at 10× magnification for 108 hours every 12 hours. Phase images were analyzed using the Incucyte ZOOM Basic Analyzer. To quantify apoptotic cells, L140P-PUF60 and WT-PUF60 MDA-MB-231 cells were seeded at a density of 900 cells per well in 384-well plates 24 hours before imaging. Fresh media supplemented with caspase-3/7 live-cell dye (Sartorius, cat. #4440) were added just prior to imaging. The plates were then loaded into the CELLCYTE X imaging system (Cytena, RRID: SCR_021911) and imaged at 10× magnification over a period of 120 hours, with images captured every 4 hours. Phase contrast and green fluorescent channel images were analyzed using the CELLCYTE Studio to assess cell confluence and measure fluorescent object count per field of view (FOV). Caspase-3/7 FOV values were then normalized to total cell confluence per FOV.

Quantification and statistical analysis
Researchers overseeing and measuring individual tumor xenografts were not blinded to the experimental conditions. Mice were assigned to experimental groups through simple randomization. All animal studies were conducted in accordance with institutional and national animal welfare regulations. Power analysis was employed to determine the sample size necessary for detecting significant changes in tumor size. Details of statistical tests, including the test used, exact sample sizes, and precision measures, are provided in the figure legends. A P value of less than 0.05 was considered significant, with specific P values reported in the figures or their legends.

Results

Dual CRISPR/Cas9 screens identify RBPs that are required for TNBC survival
To identify essential RBPs in TNBC tumors, we performed in vivo CRISPR/Cas9 screens in tumor xenografts derived from the TNBC cell lines MDA-MB-231, MDA-MB-436, and SUM149. These cell lines harbor alterations that are frequently found in TNBC, including mutations in TP53, BRCA1, and RB1, deletion of CDKN2A and CDKN2B, and MYC amplification (29, 30). Additionally, we conducted in vitro screens in all three TNBC lines and the nontransformed MCF10A cell line (“normal”), which is derived from normal human breast epithelial tissue and does not form tumors in mice. This dual-screening approach enabled detection of RBPs essential for TNBC survival while excluding those required in normal MCF10A cells.
The CRISPR RBP-targeting library used for both in vitro and in vivo studies was derived from RBPs prioritized through CRISPR/Cas9 drop-out screens conducted in isogenic cells engineered with TNBC-associated genetic alterations. We initially screened 1,078 RBPs in vitro using isogenic cells containing MYC amplification (11), KRASG13D gain-of-function (Supplementary Table S1), or heterozygous BRCA1 loss-of-function mutations (Supplementary Fig. S1A and S1B; Supplementary Table S2). From these, we compiled 206 RBPs essential in at least one model to generate a focused CRISPR/Cas9 lentiviral library, which contained five sgRNAs for each target, 140 sgRNAs targeting 37 essential genes as positive controls, and 225 nontargeting sgRNAs as negative controls (Supplementary Table S3).
MDA-MB-231, MDA-MB-436, SUM149, and MCF10A cells were transduced with our CRISPR library and puromycin selected, and then the TNBC cell lines were subcutaneously transplanted into athymic nude mice (Fig. 1A). The CRISPR library was constructed in the lenti-CRISPR V2 backbone (15), enabling DOX-inducible expression of Cas9 after tumor growth to at least a volume of 300 mm3. After 14 days of DOX induction, equatorial tumor cross-sections were harvested, dissociated, and processed through next-generation sequencing to quantify pooled sgRNA distribution using the MAGeCK package (16). Additionally, all four cell lines underwent in vitro screening to exclude RBPs that were essential in normal MCF10As from downstream analysis. As expected, following 14 days of DOX induction in vitro and in vivo, we observed a reduction in essential gene-targeting sgRNA read counts compared with nontargeting sgRNA read counts (Supplementary Fig. S1C).
Computational analysis revealed that 69 RBPs (Wald P < 0.05) were significantly depleted in DOX-induced TNBC tumors versus uninduced control tumors. Of these, 50 candidates were specifically depleted in TNBC tumor cells versus MCF10A cells, indicating unique roles in TNBC survival (Fig. 1B; Supplementary Table S4). This cohort included known TNBC-associated RBPs such as IGF2BP2 (31), SF3B1 (32), METTL3 (33), and TRA2B (34). We also identified 40 RBPs that have not yet been explored in TNBC, alongside several candidates with no known oncogenic function, including CARS2, RPUSD3, SRPR, INTS5, PARS2, and GTPBP3. Analysis of public data from TCGA pan-cancer clinical data resource (17) and the GTEx comprehensive tissue resource (18, 19) showed that 29 of 50 candidates are highly expressed in basal-like tumors (Fig. 1C), a subtype associated with TNBC characteristics (35). To define a high-confidence cohort of RBPs, we identified 15 TNBC-essential candidates significantly depleted in both in vitro and in vivo DOX-induced TNBC cells. Gene ontology analysis highlighted enrichment of U2 snRNP components (PUF60, DHX15, SRSF1, PRPF8, HNRNPH1, and SF3B1) among these high-confidence candidates (Fig. 1D; Supplementary Fig. S1D). Consistent with reports of U2 snRNP dysregulation across cancers (36, 37), our results suggest that TNBC tumors rely on RBPs that recruit U2 snRNP to 3′ splice sites (3′SS) or integrate directly into the complex to influence branchpoint usage.

Depleting PUF60 induces apoptosis of TNBC cells
Among the U2 snRNP–associated RBPs, transcript levels of the poly(U)-binding splicing factor PUF60 were the most upregulated in basal-like tumors versus normal breast tissue (Fig. 1E). PUF60 functions as a splicing factor by binding polypyrimidine tracts (PPT) to promote U2 snRNP recruitment to weak 3′SSs within pre-mRNA (38, 39). It is located within the 8q24 chromosomal region, alongside the MYC oncogene, which is one of the most frequently amplified regions in TNBC. This region contains risk loci for multiple epithelial malignancies, including breast cancer (40). Thus, previous work has reported copy-number gain or elevated expression of PUF60 in various tumor types, along with roles in cancer progression (41–45). However, whether these cancer-promoting functions necessitate MYC amplifications remains insufficiently explored. Although PUF60 is a nucleic acid–binding protein involved in nuclear splicing and transcriptional processes (38, 46), only one prior study has explored PUF60-mediated splicing activity in cancer (44). The majority of studies propose that its cancer-related functions primarily involve cytoplasmic interactions (43, 45) or RNA- and DNA-independent activity (41). Furthermore, the transcriptome-wide direct binding and splicing targets of PUF60 remain undefined in any cell type, emphasizing the need to investigate its direct, splicing-regulatory targets in TNBC.
To explore the clinical relevance of PUF60, we cross-referenced the TCGA-BRCA dataset and observed significantly improved overall survival of patients with breast cancer tumors in the lowest quartile of PUF60 CNV compared with patients with breast cancer in the highest quartile (Fig. 1F). Consistently, we confirmed significant upregulation of PUF60 transcript levels in TNBC cell lines MDA-MB-231, MDA-MB-436, and SUM149 compared with normal MCF10A cells (Supplementary Fig. S1E).
Using three independent shRNAs, we demonstrated that loss of PUF60 expression selectively induced apoptosis in human TNBC cells, evidenced by an increase in PARP1 cleavage, whereas nontransformed human epithelial mammary cells (MCF10A and MCF12A) and lung fibroblasts (IMR-90 and WI-38) showed no change (Fig. 1G; Supplementary Fig. S1F). In contrast, knockdown of the essential RBP PSMA1 induced PARP-1 cleavage in all cells. These results highlight that PUF60 is required for TNBC survival but is dispensable for the survival of normal epithelial breast and lung fibroblast cells.

PUF60 targets are enriched for genes regulating cell-cycle progression, the DNA damage response, and chromatin remodeling
To identify PUF60 splicing targets in TNBC, we performed eCLIP (20) in MDA-MB-231, MDA-MB-436, SUM149, and MCF10A cells (Supplementary Fig. S2A). Consistent with its role in U2 snRNP assembly (38), we observed enriched PUF60 binding adjacent to 3′SS (Fig. 2A; Supplementary S2B), confirmed by metagene analysis (Fig. 2B). There is no consensus motif known to influence PUF60’s selection of PPTs and, although they are typically U-rich without a strict consensus (47), motif analysis revealed a highly significant alternating UC residue, suggesting a strong preference at PPTs containing a UCUCUCUC motif (Fig. 2C).
To investigate PUF60’s role in TNBC survival, we compared high-confidence 3′SS binding sites among TNBC and normal cells. We found a significant overlap of 3'SS binding sites across each cell line (Supplementary Fig. S2C) and defined a common set of 4,991 3′SS eCLIP binding sites (black outline) occurring in at least two of four cell lines (Fig. 2D; Supplementary Table S5). This cohort was enriched for genes associated with cell-cycle progression, the DNA damage response, and chromatin organization (Fig. 2E). Unlike in healthy tissues, cancer cells undergo continuous rounds of division, increasing their reliance on cell cycle–dependent processes, including the DNA damage response and chromatin remodeling (48). Our findings indicate that PUF60 binds to the 3′SS of transcripts associated with these cancer-promoting processes, implicating a potential role for PUF60 in sustaining high rates of oncogenic proliferation.

Loss of PUF60 downregulates cell cycle– and DNA damage–associated transcripts
We investigated the impact of PUF60 knockdown on AS by performing RNA-seq on PUF60-depleted TNBC and normal cell lines (Supplementary Fig. S3A and S3B). Globally, we detected a large proportion of exon skipping events compared with all other AS events (Fig. 3A; Supplementary Fig. S3C), consistent with prior RNA-seq knockdown studies (49). Among PUF60 3′SS targets, we also observed decreased exon inclusion upon PUF60 knockdown, confirming that PUF60 binding upstream of an exon promotes inclusion of that exon (Fig. 3B).
Next, we assessed the functional consequences of PUF60 depletion and found that PUF60 knockdown downregulated the majority of PUF60 targets containing exon-skipping events (Supplementary Fig. S3D). We hypothesized that this decrease was caused by asymmetric exons (exon nucleotide length not divisible by 3), or exons that disrupt the downstream reading frame when removed, triggering RNA surveillance pathways such as nonsense-mediated decay (NMD). Indeed, across all four cell lines, more than 50% of differentially expressed PUF60 regulatory targets were asymmetric exons, and these transcripts were more frequently downregulated than those with symmetric exons (Fig. 3C). Our findings suggest that PUF60 knockdown destabilizes transcripts because they contain skipping events within asymmetrical exons.
To identify PUF60 targets with cancer-specific roles, we analyzed the expression of exons bound at the 3′SS in at least two cell lines (black outline in Fig. 2D). Hierarchical clustering of normalized TPMs revealed a large cluster (orange, 1,273) of genes with higher expression in TNBC cells compared with MCF10As (Fig. 3D; Supplementary Table S6). Exons within this cluster exhibited skipping events that were greater in magnitude in TNBC compared with MCF10A cells (Supplementary Fig. S3E; Fig. 3E), indicating that PUF60 depletion more strongly affects the splicing of transcripts that are highly expressed in TNBC. We then performed hierarchical clustering of the differentially expressed PUF60 targets within the orange cluster (Fig. 3D) to identify pathways uniquely contributing to apoptosis after PUF60 depletion (Fig. 3F; Supplementary Table S7). Two clusters (blue and gray) showed downregulation of PUF60 targets across all TNBC cell lines. These clusters were enriched for genes regulating cell cycle, chromosome organization, and collagen fibril assembly, all of which are dynamically regulated during cellular proliferation (Fig. 3G). Our findings indicate that depletion of PUF60 downregulates genes that encode and regulate cellular proliferation through exon-skipping events within overexpressed genes in TNBC. Importantly, only TNBC cells were highly sensitive to downregulation of these pathways, possibly because of their increased dependence on cell cycle control to maintain continuous proliferation.
To explore the clinical relevance of PUF60 targets downregulated upon knockdown (blue and gray clusters, Fig. 3H), we integrated TCGA (17) and GTEx (18, 19) RNA-seq data. Hierarchical clustering of Z scores revealed that most genes within this cohort were overexpressed in TNBC tumors compared with normal breast tissue (139; Supplementary Fig. S3F). Additionally, we found a negative correlation (r = −0.46), indicating that PUF60 targets highly expressed in tumor samples are generally lowly expressed in normal breast tissue (Fig. 3H). Among these were MCM4 (50), CHEK2 (51), NCAPD2 (52), RCC1 (53), SETDB1 (54), and CDC16 (55), all of which are implicated in the DNA damage response and/or cell-cycle progression in various cancers.
We next investigated how PUF60-mediated splicing of CDC16, a subunit of the anaphase-promoting complex involved in the metaphase-to-anaphase transition (56), affects TNBC survival. PUF60 bound the 3′SS upstream of exons 7 and 8 in both MCF10A and MDA-MB-231 cells (Supplementary Fig. S3G), but its knockdown caused selective exclusion of these exons in MDA-MB-231 cells, suggesting that compensatory factors may contribute to proper splicing in MCF10As after PUF60 knockdown. Skipping of one or both exons would induce a downstream frameshift, likely triggering NMD and reducing CDC16 mRNA and protein expression. Indeed, CDC16 protein expression was consistently reduced upon PUF60 depletion with three independent shRNAs in MDA-MB-231 but not in MCF10A cells (Supplementary Fig. S3H). To assess the functional impact of this reduction, we performed siRNA-mediated knockdown of CDC16 (Supplementary Fig. S3I), which impaired MDA-MB-231 proliferation (Supplementary Fig. S3J) and increased caspase-3/7–positive cells (Fig. 3I). These findings indicate that loss of CDC16, whether via direct silencing or because of PUF60 knockdown, induces apoptosis in TNBC cells; however, it remains to be determined whether CDC16 is the primary or sole mediator of PUF60-induced apoptosis.
To examine the impact of PUF60 knockdown on cell cycle, we used PI staining, which captures G1 and G2–M phases but has limited sensitivity for S phase detection (57). Notably, PUF60 depletion in MDA-MB-231 cells led to an increase in the proportion of cells in the G1 phase and a corresponding decrease in the G2–M phase, whereas MCF10A cells showed no significant changes (Fig. 3J; Supplementary Fig. S3K). Although PUF60 knockdown decreases cell cycle–associated transcript expression in both TNBC and normal cells, only TNBC cells, which are characterized by unchecked proliferation, were reliant on PUF60 for proper cell-cycle progression. To further characterize the importance of PUF60’s regulatory targets associated with DNA damage, we performed a comet assay to quantify DNA damage resulting from PUF60 depletion in both TNBC and normal cells. Indicators of DNA damage, including the percentage of DNA in the tail and tail moment, were elevated in TNBC cells following PUF60 knockdown but not in normal cells (Fig. 3K; Supplementary Fig. S3L). Our findings suggest that TNBC cells are much more susceptible to DNA damage after PUF60 knockdown, contributing to the essential role of PUF60 for TNBC cell survival. In summary, PUF60 binds to and regulates the splicing of key transcripts to prevent DNA damage and cell cycle arrest, processes which are typically evaded in highly proliferative cancer cells.

Depleting PUF60 induces TNBC tumor regression and triggers apoptosis in vivo
To assess whether PUF60 depletion affects tumor growth, we established stable MDA-MB-231 cells transduced with three DOX-inducible PUF60 shRNAs. DOX treatment effectively reduced PUF60 protein levels for all three shRNAs (Supplementary Fig. S4A, left). Following subcutaneous engraftment of MDA-MB-231 cells expressing two of our validated shRNAs and subsequent DOX administration (Fig. 4A, top), we observed an 86% reduction in tumor burden compared with vehicle-treated controls (Fig. 4B and C; Supplementary Fig. S4B). Protein lysates from these tumors confirmed sustained PUF60 depletion throughout the treatment period (Supplementary Fig. S4C). To determine whether this result extends to additional TNBC models, we established stable MDA-MB-436 and SUM149 cells transduced with two distinct DOX-inducible PUF60 shRNAs and a DOX-inducible nontargeting shRNA (Supplementary Fig. S4A, right). Following subcutaneous engraftment and DOX administration (Fig. 4A, bottom), PUF60 knockdown resulted in an 82% and 90% reduction in tumor burden in MDA-MB-436 and SUM149 tumors, respectively (Fig. 4D–G; Supplementary Fig. S4D and S4E). Protein lysates from these tumors confirmed sustained PUF60 depletion (Supplementary Fig. S4F and S4G). Fluorescent in vivo imaging of the RFP-tagged tumors further corroborated these findings, showing either a loss of RFP signal or minimal tumor growth at endpoint relative to initial imaging (Supplementary Fig. S4H and S4I).
To better mimic a clinical scenario, we assessed the therapeutic potential of PUF60 knockdown by initiating DOX treatment only after tumors reached a larger size (average volume of 132.8 mm3). Following 29 days of DOX treatment of MDA-MB-231 tumors expressing two distinct DOX-inducible PUF60 shRNAs, tumor volume decreased by 53% (Supplementary Fig. S4H–S4J; Supplementary Table S8), demonstrating that PUF60 knockdown not only inhibits tumor growth but also induces tumor regression. Protein lysates from these tumors confirmed PUF60 depletion (Supplementary Fig. S4K). Fluorescent in vivo imaging of the RFP-tagged tumors further corroborated these findings, showing loss of RFP signal by day 47 of DOX treatment (Supplementary Fig. S4L). Additionally, we detected an increase in CC3-positive cells in tumor sections from DOX-treated mice, indicating that PUF60 depletion leads to tumor regression via induction of apoptosis in TNBC tumors (Fig. 4J and K). Our findings collectively indicate that PUF60 is required to sustain growth and survival across multiple TNBC tumor models.

A residue within PUF60’s RNA recognition motif 1 is required for global recognition of 3′SS targets
Although previous studies have proposed that PUF60’s cancer-related functions involve cytoplasmic protein–protein interactions (43, 45) or nucleic acid–independent activities (41), we observed predominantly nuclear signal in MDA-MB-231 cells (Supplementary Fig. S5A). Thus, we investigated whether disrupting its ability to bind 3′SSs would replicate the effects observed in our knockdown studies. Using an ABE8e-SpRY A-to-G base editor, we introduced an L140P mutation into the RNA recognition motif 1 (RRM1) domain of PUF60 in TNBC cells, a substitution previously shown to impair binding to U-rich 3′SS substrates (Supplementary Fig. S5B; ref. 58). L140P markedly reduced cell proliferation (Supplementary Fig. S5C), suggesting that disruption of RNA binding phenocopies PUF60 knockdown. To circumvent the limited proliferative capacity of constitutively edited cells, we employed a DOX-inducible system to dynamically regulate L140P expression. MDA-MB-231 cells stably expressing an inducible shRNA (tet-shPUF60-1) targeting endogenous PUF60 were subsequently transduced with inducible codon-optimized, V5-tagged WT-PUF60 or L140P (L140P-PUF60) PUF60 constructs that were resistant to the shRNA-mediated knockdown (Supplementary Fig. S5D and S5E). Corroborating our base editing results, we observed that 48 hours after DOX treatment, TNBC cells expressing L140P-PUF60 ceased proliferation, whereas cells expressing WT-PUF60 continued to proliferate at rates similar to those of the untreated cells (Fig. 5A). Notably, L140P-PUF60 MDA-MB-231 cells also exhibited increased levels of caspase-3/7–positive cells (Fig. 5B; Supplementary Fig. S5F), underscoring the essential role of PUF60–RNA binding in sustaining TNBC cell survival.
To investigate RNAs bound by this mutant, we performed eCLIP on L140P-PUF60 and WT-PUF60 MDA-MB-231 cells using a V5 antibody (Supplementary Fig. S5G). De novo motif analysis of WT-PUF60–reproducible eCLIP windows identified the same UCUCUCUC motif (P < 10−131) detected in our endogenous PUF60 eCLIP assays. In contrast, enrichment for this motif was absent in the L140P-PUF60 eCLIP signal. Furthermore, we observed significant enrichment for uracil-rich sequences in WT-PUF60 eCLIP windows, aligning with PUF60’s established role in directly binding PPTs, whereas L140P-PUF60 windows lacked this enrichment (Fig. 5C). Despite these differences, canonical donor (AG′GURAGU) and acceptor (YAG) splice sites were enriched among L140P-PUF60 motifs (Fig. 5C). Although our analysis indicates that L140P-PUF60 is not interacting with PPTs, our findings suggest that the RRM1 mutant may be recruited to splice sites through protein–protein interactions with U2 snRNP components, facilitated via its intact U2AF homology motif (UHM) domain.
In line with this hypothesis, co-IP assays revealed that L140P-PUF60 interacts with the U2-associated splicing factor, U2AF65, although with lower affinity than WT-PUF60 (Supplementary Fig. S5H; ref. 39). Interestingly, RNase A treatment reduced only WT-PUF60-U2AF65 interactions, suggesting that PUF60-U2AF65 interactions are partially mediated by RNA. Our observations indicate that although L140P-PUF60 can bind U2 snRNP proteins at splice sites, its RNA-mediated interactions may be impaired by reduced binding capability at U-rich RNA substrates. Consistently, WT-PUF60 showed fourfold higher reproducible eCLIP signal at 3′SSs (Fig. 5D), and 66% of WT-PUF60–bound genes lost 3′SS signal upon L140P substitution (Supplementary Fig. S5I and S5J), confirming that L140P substantially reduces PUF60’s RNA binding affinity and presence at 3′SSs.
We investigated the impact of L140P on AS by performing RNA-seq on L140P- and WT-PUF60 MDA-MB-231 cells (Fig. 5A). Our data were in line with a previous study (58) that showed that the L140P substitution led to exclusion of the PUF60-regulated UBE2F exon 5 (rMATS FDR = 9.67 × 10−4; Supplementary Fig. S5K–S5M). We also found that ∼75% of differential exon skipping events observed after PUF60 knockdown were also detected upon L140P mutation (Fig. 5E and F), whereas the remaining 25% may reflect indirect effects of PUF60’s non–RNA-binding functions.
To determine whether L140P-PUF60 binding affects exon usage, we assessed the frequency of WT- and L140P-PUF60 3′SS eCLIP peaks proximal to exon splicing events. WT-PUF60 bound near substantially more exon-skipping events, which were also greater in magnitude [WT-PUF60 Δ percent spliced in (PSI) = −0.19; L140P-PUF60 ΔPSI = −0.11] than skipping events near L140P-PUF60 peaks (Fig. 5G), indicating that L140P-PUF60 does not direct AS of its targets. We further investigated the position-specific impact of L140P-PUF60 binding on neighboring exon usage by constructing a “splicing map” outlining eCLIP signal near skipped exons compared with all native cassette exons (25). WT-PUF60 3′SS binding showed global enrichment near skipped exons compared with native cassette exons (Fig. 5H), whereas L140P-PUF60 lacked this pattern, supporting that L140P disrupts 3′SS-dependent splicing regulation.

Disruption of PUF60’s 3′SS-dependent splicing activity downregulates cellular proliferation pathways
To assess whether PUF60 knockdown and L140P mutation produce consistent splicing changes in TNBC, we examined splicing events involving exons bound by endogenous PUF60 at their 3'SS. Hierarchical clustering of inclusion levels showed that 86% of PUF60’s target exons clustered together (gold) showing exon skipping under both conditions, indicating substantial similarity in splicing outcomes between both conditions (Fig. 5I; Supplementary Table S9). We next verified that L140P mutation also downregulated cellular proliferation pathways as hierarchical clustering of differentially expressed conserved targets (Fig. 5J; Supplementary Table S10) identified two downregulated clusters (gray and blue) after L140P mutation and PUF60 knockdown. Unlike upregulated clusters, the gray cluster was enriched for genes associated with cell cycle and DNA metabolism (Fig. 5K). These findings are consistent with our endogenous PUF60 eCLIP and knockdown RNA-seq data, which identify PUF60 as a regulator of transcripts associated with genomic integrity and cell-cycle progression (Fig. 3G). The blue cluster contained cellular matrix adhesion and junction assembly–associated targets that were downregulated in TNBC but upregulated in normal cells. The ability of cancer cells to communicate with the extracellular matrix through cellular adhesions is crucial for intracellular survival and proliferation signaling (59), which also supports our understanding of PUF60’s role in cancer cell division (Fig. 5K). We conclude that disrupting PUF60’s ability to bind the 3'SS replicated the molecular effects observed in our PUF60 knockdown studies, suggesting that PUF60-mediated splicing activity is vital for the regulation of transcripts encoding cellular proliferation regulators.

Silencing of PUF60 splicing activity by L140P mutation suppresses tumor growth
To assess the functional implications of PUF60-mediated 3'SS recognition, in vivo, we subcutaneously transplanted L140P-PUF60 and WT-PUF60 MDA-MB-231 cells into mice (Fig. 6A). Upon tumor engraftment and DOX treatment, we observed a 93% reduction in tumor burden compared with mice engrafted with WT-PUF60 cells (Fig. 6B and C; Supplementary Fig. S6A; Supplementary Table S11). Furthermore, we observed a 67% reduction in L140P-PUF60 tumor size after 58 days of treatment compared with tumor volume on the first day of treatment (Fig. 6D and E). Our observations indicate that the PUF60L140P mutation not only suppresses growth but also induces tumor regression. These results were corroborated by fluorescent in vivo imaging of RFP-tagged tumors, which revealed a loss of RFP signal by day 47 of DOX treatment (Supplementary Fig. S6B–S6D). In conclusion, our investigation into the molecular and functional consequences of a mutation within the PUF60 RRM1 domain strongly substantiates our findings that PUF60 binding at 3′SSs is required to sustain TNBC growth and survival in both cells and tumors. Moreover, our findings underscore that TNBC cells lacking PUF60-3′SS binding activity show similar gene expression changes to PUF60 knockdown cells, confirming that PUF60′s role in regulating cell cycle and DNA metabolism pathways is primarily through its splicing activity at key transcripts required for maintaining the survival and proliferation of TNBC cells (Fig. 6F).

Discussion
Our RBP-focused in vivo functional screen identified 50 candidates that contribute to TNBC development within a tumor xenograft context. Although these RBPs were implicated in various steps of RNA metabolism, our candidate cohort was significantly enriched for the U2 snRNP spliceosomal complex, which is responsible for intronic branch site identification, and includes RBPs frequently mutated or aberrantly expressed in various cancers, including breast cancer (60, 61). Consistent with these findings, we identified that the U2-associated splicing factor PUF60 is essential for sustaining TNBC cell and tumor growth. Through integrated eCLIP and knockdown RNA-seq analysis, we revealed that PUF60 drives exon inclusion within transcripts encoding proteins involved in cell cycle and DNA damage pathways. Upon PUF60 depletion, these mRNAs exhibited widespread exon skipping and subsequent destabilization, leading to selective activation of programmed cell death in TNBC. Our observations complement previous work showing that TNBC is especially vulnerable to the inhibition of cell cycle and genomic stability regulators (62, 63). As these pathways collaborate to protect TNBC cells from fatal DNA insults that accompany uncontrolled proliferation rates, our study implicates a role for PUF60 in regulating transcripts that are crucial for TNBC survival (64). Collectively, our findings reveal how PUF60 modulates the splicing of key cell cycle– and DNA repair–encoding transcripts to contribute to cellular survival–associated gene regulation in TNBC.
RBPs are multifunctional proteins and determining their disease-specific regulatory roles remains a significant challenge in cancer biology. Many studies suggest that PUF60’s cancer-promoting functions may be independent of RNA (41) or cytoplasmic (43, 45). Here, we establish an additional function for PUF60 within the nucleus by disrupting PUF60’s 3′SS direct RNA interactions using a previously characterized RRM1 domain mutation (58) and identify splicing activity as a key molecular mechanism through which PUF60 contributes to TNBC survival and growth. Our eCLIP analysis of transcripts directly bound by PUF60 reveals a network of candidate targets within proliferation-associated pathways, which safeguard genomic integrity during the mitotic cell cycle. Our multiscale transcriptomics further suggest that PUF60’s oncogenic functions depend on the regulation of several mRNA targets, positioning splicing activity as central to its role in TNBC. As such, our datasets serve as a valuable resource for identifying comprehensive signatures of splicing vulnerabilities in TNBC progression.
Aberrant genomic repair accompanied by chronic replication stress is a dependency shared across many tumors, suggesting that other cancers may require PUF60 for survival. In line with this, several studies have found that PUF60 depletion also induces apoptosis in lung adenocarcinoma (44) and ovarian cancer (43), both characterized by replication stress due to high proliferation rates (65). These findings underscore PUF60’s potential as a broader therapeutic target in cancers with genomic instability.
Our in vivo findings highlight PUF60 inhibition as a potential therapeutic strategy for inducing TNBC tumor regression. Although transcriptional and translational modulators exist, few spliceosomal inhibitors have been thoroughly characterized, and none target PUF60 (66, 67). However, compounds that inhibit the U2-associated SF3b1 complex have been shown to exhibit anticancer properties (67, 68), and cancers driven by spliceosome dysfunction are sensitive to SF3b inhibition (69), supporting the promise of selective splicing modulators for aggressive cancers like TNBC. Although PUF60-specific inhibitors are not yet developed, small molecules blocking UHM–UHM ligand motif (ULM) interactions, present in PUF60 and other U2-associated factors, could be leveraged (70, 71). However, the presence of UHM and ULM domains in nonspliceosomal proteins, such as kinases, raises concerns about potential off-target effects from these inhibitors (72). Future work should explore PUF60-specific inhibitors, such as aptamers or small molecules, as a therapeutic strategy for TNBC; however, future studies are essential to evaluate the safety of PUF60 inhibition, particularly its impact on normal tissues in vivo. Alternatively, splice-switching antisense oligonucleotides that bind to pre-mRNA and block splicing machinery interactions could reverse specific downstream splicing events and reduce potential toxicity from general PUF60 inhibition (73). Although we identified CDC16 as a potential downstream effector of PUF60, further studies are needed to validate these findings and to determine whether additional downstream AS events contribute to apoptosis in PUF60-depleted TNBC cells. Our comprehensive study provides a molecular framework for identifying critical splicing events that could be leveraged to disrupt genomic maintenance and cell cycle processes that safeguard TNBC against replication stress.

Supplementary Material
Supplementary Figure 1CRISPR screening in isogenic cell lines and TNBC-derived cell lines identifies PUF60 as an essential RBP in TNBC, related to Figure 1.

Supplementary Figure 2Quality control of PUF60 eCLIP and cell line comparisons of 3' splice site targets, related to Figure 2.

Supplementary Figure 3KD RNA-sequencing reveals exon skipping events modulated by PUF60 in TNBC and normal cells, related to Figure 3.

Supplementary Figure 4Depletion of PUF60 in TNBC cells significantly shrinks tumor volume, related to Figure 4.

Supplementary Figure 5PUF60-activated exons are excluded following L140P substitution, related to Figure 5.

Supplementary Figure 6Silencing of PUF60 splice site activity shrinks tumor volume, related to Figure 6.

Supplementary TablesSupplementary tables 1-11