Genome-wide CRISPR screen for human factors involved in alternative polyadenylation based on differential localization of CD47

Abstract

At least 70% of the human protein-coding genes contain multiple polyadenylation sites (PAS) and undergo alternative polyadenylation (APA), generating distinct transcripts from a single gene. While APA has been implicated in various physiological and pathological processes, its regulatory factors and cellular mechanisms remain incompletely understood. A previous study demonstrated that APA influences the localization of the cell surface marker CD47. Here, we present the results of a genome-wide CRISPR screen aimed at identifying APA regulators using CD47 as a reporter. Given that isoform-specific knockdown of CD47, as well as knockdown of core 3′ end processing factors, alters CD47 localization, we developed an immunofluorescence-based method that simultaneously detects cell surface and intracellular CD47 protein, enabling the visualization of APA-dependent changes at the single-cell level. Leveraging this approach, we conducted a CRISPR screen and identified multiple genes affecting CD47 cell-surface expression. In addition to known membrane trafficking factors, we uncovered several nuclear factors, among which POLDIP2 emerged as a potential novel APA regulator with a global impact on APA. This study provides a foundation for further investigations into the molecular mechanisms governing APA.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-14782-7.

Introduction

Cleavage and polyadenylation is a fundamental process in mRNA processing, consisting of two coupled reactions: The endonucleolytic cleavage of pre-mRNA at the polyadenylation site (PAS) followed by the synthesis of a poly(A) tail at the 3′ end¹. Notably, a single gene can generate multiple transcript isoforms through the usage of different PASs, a phenomenon known as alternative polyadenylation (APA)². APA is widespread, with at least 70% of human protein-coding genes expressing APA isoforms³. APA is regulated by cis-acting elements within nascent mRNA and trans-acting factors, including four core protein complexes: Cleavage and polyadenylation specificity factor (CPSF), cleavage stimulation factor (CSTF), cleavage factor I (CFI), and cleavage factor II (CFII). These complexes, along with over 80 additional accessory proteins, interact with pre-mRNA through protein–protein and protein–RNA interactions⁴. In response to dynamic physiological and pathological conditions, 3′ end processing factors modulate PAS selection, thereby directing APA regulation⁵. For example, CFIm plays a critical role in maintaining distal PAS usage through the recognition of UGUA motifs⁶. Downregulation of CFIm typically shifts PAS usage toward proximal sites, leading to widespread 3′ UTR shortening⁷. In contrast, CSTF has been shown to enhance cleavage at proximal PASs in certain cellular contexts, further emphasizing the importance of identifying novel APA regulators⁸. Although several studies have successfully identified core 3′ end processing factors using conventional methods such as cellular fractionation, proteomic analysis, and UV cross-linking^4,9–11, these approaches have limitations in detecting the full spectrum of regulatory proteins. How to systematically screen and identify previously unrecognized APA regulators within the vast human proteome remains a key challenge. Addressing this challenge requires a robust and versatile genome-wide platform.

Genome-wide CRISPR knockout screening enables the simultaneous targeting of numerous genes, followed by functional selection to identify cells exhibiting phenotypic changes of interest. After applying selective pressure based on a research question, cells with the desired phenotype are enriched and analyzed using next-generation sequencing (NGS) to identify single guide RNAs (sgRNAs) that are enriched or depleted, thereby pinpointing candidate genes involved in the process under investigation¹². However, a major limitation of applying this approach to APA regulation is that APA-related changes are not easily detectable through conventional phenotypic readouts, making it challenging to identify APA-regulating genes using CRISPR screens.

A recent study demonstrated that APA regulates the expression and localization of CD47 by generating mRNA isoforms with different 3′ untranslated region (UTR) lengths¹³. When the distal PAS of CD47 is utilized, the resulting mRNA, termed CD47-LU (LU for long UTR), contains an extended 3′ UTR that serves as a scaffold for recruiting a protein complex, including the RNA-binding protein HuR and SET, to the site of translation. This interaction facilitates the association of SET with newly synthesized CD47 protein, promoting its translocation to the plasma membrane via activated RAC1. In contrast, when the proximal PAS is used, the resulting mRNA, termed CD47-SU (SU for short UTR), has a shorter 3′ UTR that lacks the sequence necessary for HuR and SET recruitment. Consequently, the CD47 protein produced from this transcript predominantly localizes to the endoplasmic reticulum (ER). Notably, CD47-SU has also been shown to play a specific role in stress-induced apoptosis, suggesting that APA-induced changes in CD47 localization contribute to its functional diversification¹³. These findings suggest that CD47 protein localization serves as a readout for APA and could be leveraged as a reporter system to study APA regulation.

In this study, we developed an immunofluorescence-based method that simultaneously detects both cell surface and intracellular CD47 protein, providing a direct visualization of APA-dependent changes. By integrating this method with pooled genome-wide CRISPR screening, we identified several candidate APA regulators. Notably, our findings suggest that POLDIP2 may play a pivotal role in global APA regulation.

Materials and methods

Cell culture

HeLa cells (ATCC) and HEK293FT cells (ATCC) were cultured in Dulbecco’s modified Eagle’s medium (DMEM) (Gibco) supplemented with 10% fetal bovine serum (FBS) (Biowest), 100 U/ml penicillin, and 100 μg/ml streptomycin (Sigma-Aldrich) at 37 °C in a humidified incubator with 5% CO₂. To generate a HeLa cell line stably expressing humanized Streptococcus pyogenes Cas9 (hSpCas9), HeLa cells were transduced with a pLenti-Cas9-BSD-based expression vector using lentivirus produced with the ViraPower Lentiviral Expression Systems (Thermo Fisher Scientific), following the manufacturer’s protocol. After 24 h, transduced cells were selected with 4 μg/ml blasticidin (Invivogen) for 2 weeks. Monoclonal cell lines were isolated using cloning cylinders, expanded in 6-well plates, and hSpCas9 expression was assessed by immunoblot analysis.

Gene knockdown using short hairpin RNA (shRNA)

shRNA sequences were cloned into pRSI9-U6-sh-UbiC-TagRFP-2A-Puro. For transient knockdown of core 3′ end processing factors, HeLa cells were transfected with either a control plasmid (U6 promoter alone) or a specific shRNA-containing plasmid using Lipofectamine 2000 (Life Technologies). Transfected cells were harvested 48 h post-transfection. For stable knockdown, recombinant lentiviruses were produced using the ViraPower Lentiviral Expression Systems and used to infect HeLa cells, following the manufacturer’s recommendations. Infected cells were selected in growth medium containing 0.5 μg/ml puromycin and harvested 4 days post-infection for confocal microscopy and fluorescence-activated cell sorting (FACS) analysis. The shRNA sequences are provided in Supplementary Table 1.

Quantitative PCR

Total RNA was extracted using Sepasol‐RNA I Super G (Nacalai Tesque) following the manufacturer’s instructions. qRT-PCR was performed using the One Step TB Green PrimeScript PLUS RT-PCR Kit (Perfect Real Time) (Takara Bio) on a StepOnePlus Real-Time PCR System (Applied Biosystems). Each experiment was conducted at least three times to obtain a minimum of three biological replicates. The distal PAS usage index was calculated as ΔCT = CTlong—CTtotal, where CT represents the threshold cycle. Data were presented as log2 fold changes by normalizing test samples to controls and calculating log base 2 values. A negative value indicates that the mRNA has a shortened 3′ UTR compared to the control. To precisely quantify the relative expression of the two CD47 3′ UTR isoforms, a standard CD47 cDNA was used to calibrate differences in amplification efficiency between primer sets. The fraction of the short 3′ UTR isoform was determined by subtracting the CT value of the long isoform from the CT value of total CD47 expression. The primers used for qRT-PCR are listed in Supplementary Table 2.

Immunofluorescence staining

For the simultaneous detection of cell surface and intracellular CD47, cells were fixed with 2% paraformaldehyde in phosphate buffered saline (PBS) for 15 min at room temperature (RT), followed by PBS washes. Cells were then blocked with 10% FBS for 15 min at RT and incubated with Alexa Fluor 488-conjugated mouse anti-human CD47 antibody (Thermo Fisher Scientific) for 1 h at RT in 10% FBS. After PBS washes, cells were permeabilized with 0.7% Tween-20 in PBS for 15 min at RT, washed again, and re-blocked with 10% FBS for another 15 min. Cells were then incubated with eFluor 450-conjugated mouse anti-human CD47 antibody (Thermo Fisher Scientific) for 1 h at RT in 10% FBS. Following additional PBS washes, cells were stained with 50 μg/ml 7-aminoactinomycin D (Invitrogen) for 30 min at 4 °C, followed by three PBS washes. Imaging was performed using a Zeiss LSM 780 inverted confocal microscope equipped with a 63×, 1.4 numerical aperture oil objective.

FACS analysis

For FACS analysis of cell surface and intracellular CD47, cells were fixed with 2% paraformaldehyde in PBS for 15 min at RT, washed with 1% FBS in PBS, and blocked with 10% FBS for 15 min at RT. Cells were then incubated with Alexa Fluor 488-conjugated mouse anti-human CD47 antibody for 1 h at RT in 10% FBS. After washing with 1% FBS in PBS, cells were permeabilized with 0.7% Tween-20 in PBS for 15 min at RT, washed again, and blocked with 10% FBS for another 15 min at RT. Subsequently, cells were stained with PE-Cyanine5-conjugated mouse anti-human CD47 antibody (Thermo Fisher Scientific) for 1 h at RT in 10% FBS. After final washes with 1% FBS in PBS, cells were resuspended in 1% FBS. Flow cytometry was performed on a BD FACSMelody cell sorter (BD Biosciences) with 3-laser, 8-color (2-2-4) configuration, with at least 10,000 cells analyzed per sample. Data were processed using the FlowJo v10.7 software (BD Biosciences).

Immunoblot analysis

HeLa cells were resuspended in high salt buffer, sonicated for 10 min, and cooled on ice. Lysates were resolved on 8% sodium dodecyl sulfate polyacrylamide gels and transferred onto PVDF membranes (Merck Millipore). Membranes were blocked with 5% skim milk or Blocking One (Nacalai Tesque) for 1 h at RT. Primary antibodies, including anti-FLAG M2 (Sigma) and anti-actin (Abcam), were diluted in 5% skim milk or Can Get Signal (Toyobo) and incubated with the membranes for 1 h at RT. After washing with Tris-buffered saline containing 0.1% Tween-20, the blots were incubated with HRP-linked horse anti-mouse IgG antibody (Cell Signaling Technology) for 1 h at RT. The blots were washed three times with the same buffer, and signal detection was performed using ECL Western Blotting Detection Reagents (GE Healthcare).

T7 endonuclease I (T7EI) assay

Genomic DNA was extracted by using the DNeasy Blood & Tissue Kits (Qiagen), following the manufacturer’s protocol. Target regions were amplified by PCR using KOD FX Neo (Toyobo). PCR products were denatured at 95 °C for 5 min, then re-annealed using a temperature ramp of− 2 °C per second to 85 °C, followed by a− 0.1 °C per second ramp to 25 °C. The re-annealed PCR products were incubated with T7EI (New England Biolabs) at 37 °C for 15 min. Digested products were analyzed by electrophoresis using a MutiNA system (Shimadzu). The primers used for the T7EI assay are listed in Supplementary Table 2.

FACS-based genome-wide CRISPR screens

For pooled genome-wide CRISPR screening, the Guide-it CRISPR Genome-Wide sgRNA Library (Takara), comprising 76,612 sgRNAs targeting 19,114 human genes, was packaged into lentivirus following the manufacturer’s protocol. Viral supernatants were used to transduce HeLa cells stably expressing hSpCas9 at a multiplicity of infection of < 0.3, ensuring a 1000-fold library representation. Selection of transduced cells with 200 μg/ml hygromycin B was initiated 24 h after infection. On day 7, transduced cells were harvested, washed with PBS, and fixed with 2% paraformaldehyde for 15 min at RT. Cells were blocked with PBS containing 10% FBS and incubated with Alexa Fluor 488-conjugated anti-CD47 antibody in PBS containing 10% FBS for 1 h at RT. Cells were then resuspended in permeabilization buffer containing 0.2% Tween-20 for 15 min at RT and subsequently stained with eFluor 450-conjugated anti-CD47 antibody in PBS containing 10% FBS for 1 h at RT. After three washes with PBS, cells were filtered through a 35-μm nylon mesh and sorted in PBS containing 1% BSA using a BD FACSMelody (BD Biosciences). Untransduced or sgRNA library-negative (mCherry) cells were excluded, and the 10% of the cells with the lowest or highest eFluor 450/Alexa Fluor 488 signal ratios were sorted into PBS containing 1% BSA. The gating strategy for flow cytometric sorting is shown in Fig. 3b. At least 1 × 10⁶ CD47 surface high and 1 × 10⁶ CD47 surface low cells were collected per replicate.

Fig. 3.

Genome-wide CRISPR screening identifies candidate APA regulators. (a) Schematic of the genome-wide CRISPR screen based on differential CD47 localization. (b) FACS analysis of cell surface and intracellular CD47 expression in HeLa cells stably expressing Cas9. The 10% of cells with the lowest (CD47 surface high) and 10% with the highest (CD47 surface low) intracellular-to-cell surface CD47 expression ratios were collected. (c), (d) Gene-level enrichment of sgRNAs in CD47 surface high (c) and CD47 surface low (d) cells, with the x-axis representing log₂ fold changes and the y-axis showing MAGeCK P-values. Dashed lines indicate P < 0.05 and log₂ fold changes > 1, with significantly enriched genes highlighted. (e) (f) GO enrichment analysis of top 100 ranked genes whose depletion resulted in cell surface CD47 retention (CD47 surface high) (e) or intracellular CD47 accumulation (CD47 surface low) (f).

Genomic DNA from sorted CD47 surface low, CD47 surface high, and unsorted control populations was extracted using the DNeasy Blood & Tissue Kit (Qiagen). PCR amplification was performed using a NEBNext Ultra II Q5 Master Mix (New England Biolabs) according to the manufacturer’s instructions. PCR products were gel-purified using the NucleoSpin Gel and PCR Clean-Up kit (Takara). Barcoded NGS libraries were pooled and sequenced using the HiSeq X platform (Illumina). The primers used for library amplification are listed in Supplementary Table 2.

Data analysis of pooled CRISPR screens

To quantify raw sequencing reads, adapter sequences in the 5′ and 3′ region of sgRNA sequences were trimmed using cutadapt. The trimmed FASTQ files were then mapped and counted using MAGeCK (version 0.5.9). To assess sgRNA enrichment, the MAGeCK-VISPR maximum likelihood estimation algorithm was implemented. First, count tables were filtered to exclude sgRNAs with fewer than 200 counts in both control and sorted samples before proceeding with downstream analyses. Read counts were normalized by total read counts, and average log₂ fold changes, P values, and false discovery rates (FDRs) were calculated using MAGeCK (version 0.5.9). Enrichment of sgRNAs in CD47 surface low and CD47 surface high populations was determined by comparing sgRNA counts in sorted populations to those in unsorted control populations. Gene Ontology (GO) terms were annotated using the GO.db and org. Hs.eg.db (both version 3.11.4) Bioconductor packages in R.

Secondary screen

To validate candidate genes identified in the primary screen, a secondary screen was performed. The top 100 hit genes, including the 25 most significant genes localized inside or outside the nucleus from CD47 surface low and CD47 surface high populations, were selected as candidates. The most effective sgRNAs targeting these genes were individually cloned into pLVXS-sgRNA-hyg, and HeLa cells stably expressing hSpCas9 were transduced with pooled sgRNAs targeting selected genes. FACS-based CRISPR screening and subsequent analysis were conducted following the same protocol as the primary screen. The sgRNA sequences used for the secondary screen are listed in Supplementary Table 1.

RNA sequencing (RNA-seq) and APA analysis

Total RNA was extracted using Sepasol‐RNA I Super G (Nacalai Tesque) according to the manufacturer’s protocol. RNA sequencing libraries were prepared and sequenced using the NovaSeq X Plus platform (Illumina), generating 2 × 150 base-pair paired-end reads. A total of 212 million reads were processed by trimming adapter sequences using cutadapt and mapped to the Homo sapiens genome (hg38) using HISAT2 (version 2.0.2). On average, 96% of total reads were mapped to the genome or transcriptome. Gene counts were derived based on human Ensembl gene IDs by using htseq-count. Differential expression analysis was conducted on raw gene counts using DESeq2. Differentially expressed genes (DEGs) were defined as those with |log₂ fold changes|> 1, where fold change was calculated as the expression in POLDIP2 knockdown samples relative to control samples. Gene expression changes were visualized as a volcano plot using the EnhancedVolcano package in R Studio. For heatmap analysis, DEGs were visualized using the ComplexHeatmap package in R Studio. GO terms were annotated using GO.db and org. Hs.eg.db (both version 3.11.4) Bioconductor packages in R.

APA was analyzed using the Quantification of APA (QAPA) bioinformatics algorithm from the RNA-seq data. QAPA quantifies PAS usage within genes by calculating poly(A) usage (PAU), which represents the percentage of PAS usage per transcript. Higher PAU values indicate greater utilization of a specific PAS, whereas lower PAU values indicate reduced usage. To integrate PAU values across all PASs within a gene, the weighted 3′ UTR Length Index (WULI) was calculated using the following formula:

where represents the 3′ UTR length up to each PAS, and denotes the poly(A) usage of the corresponding PAS. To identify significant shifts in 3′ UTR usage, the differences in mean WULI values (∆WULI) were calculated between three POLDIP2 knockdown and three control samples. Genes with |∆WULI/100|≥ 0.1 were considered significant. Using this approach, 1056 genes with significant changes in 3′ UTR usage in response to POLDIP2 knockdown were identified.

Statistical analysis

Statistical analyses were performed using GraphPad Prism 10. The specific statistical tests used are detailed in the corresponding figure legends. All data are presented as means ± s.d. Statistical significance was determined as follows: Not significant (ns), P ≥ 0.05; *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001.

Results

CD47 mRNA 3′ UTR length affects its protein localization

Previous studies have demonstrated that global alterations in mRNA polyadenylation, resulting from the utilization of different PASs, play a crucial role in post-transcriptional gene regulation under various physiological and pathological conditions^5,14. Several studies have identified core 3′ end processing factors as key regulators of PAS selection^15–18; however, a complete characterization of all potential 3′ end processing factors remains elusive. Therefore, we aimed to identify novel APA regulators using a genome-wide CRISPR screen. To facilitate the detection of APA events, we developed a novel method to visualize APA-mediated outcomes at the single-cell level. A recent report indicates that CD47 protein localization is regulated by APA¹³. Specifically, when the distal PAS of CD47 is utilized, the CD47 protein is transported to the cell surface, where it functions as a transmembrane protein. Conversely, when the proximal PAS is utilized, CD47 is retained intracellularly, primarily within the ER. Based on these findings, we hypothesized that CD47 protein localization could serve as a readout for APA regulation. To test this hypothesis, we developed an immunofluorescence staining protocol capable of differentiating between cell surface and intracellular CD47 (Fig. 1a). In the first step, cell surface CD47 was labeled using Alexa Fluor 488-conjugated anti-CD47 antibody. After permeabilization, intracellular CD47 was stained using eFluor 450-conjugated anti-CD47 antibody. To ensure that only intracellular CD47 was stained in the second step, we used identical monoclonal anti-CD47 antibodies conjugated with Alexa Fluor 488 or eFluor 450, preventing re-staining of cell surface CD47. Confocal microscopy confirmed that cell surface CD47 was exclusively labeled with Alexa Fluor 488, while intracellular CD47 was specifically stained with eFluor 450 (Fig. 1b).

Fig. 1.

Immunofluorescence staining method simultaneously detecting cell surface and intracellular CD47, enabling visualization of CD47 APA. (a) Schematic representation of the immunofluorescence staining method that simultaneously detects CD47 protein localization. (b) Fluorescence confocal microscopy images obtained using the double staining approach for cell surface and intracellular CD47 in HeLa cells stably expressing control shRNA, shRNA targeting total CD47 (shCD47-SU&LU), or shRNA targeting the long 3′ UTR isoform (shCD47-LU). Scale bars, 20 µm. (c) qRT-PCR analysis of total CD47 mRNA and the long 3′ UTR isoform from HeLa cells transfected with shCD47-SU&LU, shCD47-LU, or shControl. Data are presented as mean ± s.d. ****, P < 0.0001; **, P < 0.01 (two-tailed t-test for independent samples, n = 3). (d) Quantification of median fluorescence intensity (MFI) values from panel c. ****, P < 0.0001; ***, P < 0.001; **, P < 0.01; *, P < 0.05 (two-tailed t-test for independent samples, n = 3). (e) FACS analysis of cell surface and intracellular CD47 expression in HeLa cells from panel c. Representative histograms display cell surface and intracellular CD47 expression, with unstained cells shown in gray.

To validate the specificity of the antibodies used, we designed an shRNA targeting a common region present in both long and short isoforms of CD47 mRNA. qRT-PCR analysis revealed that shCD47-SU&LU overexpression significantly reduced the expression of both CD47-LU and CD47-SU (Fig. 1c). Confocal microscopy and FACS analysis demonstrated that knockdown of both isoforms led to a reduction in both cell surface and intracellular CD47 staining (Fig. 1b, d, e). These results validate the double staining approach as an effective method for simultaneously detecting cell surface and intracellular CD47.

To confirm the previous report showing that CD47 protein localization is regulated by 3′ UTR length¹³, we designed an shRNA specifically targeting the extended 3′ UTR present only in the long isoform (CD47-LU). qRT-PCR analysis confirmed the selective knockdown of the long isoform by shCD47-LU (Fig. 1c). As expected, confocal microscopy and FACS analysis demonstrated that CD47-LU knockdown selectively reduced cell surface CD47 expression, while intracellular CD47 expression was only slightly affected (Fig. 1b, d, e). Thus, these findings establish that CD47 protein localization is regulated by 3′ UTR length, and that our immunofluorescence staining method provides a reliable approach for visualizing APA-mediated outcomes.

Depletion of core 3′ end processing factors alters PAS usage in CD47 and affects CD47 protein localization

To assess whether the immunofluorescence staining method could detect APA alterations, we investigated how the knockdown of known APA regulators affects CD47 PAS usage and protein localization. shRNAs targeting core 3′ end processing factors were designed, and their knockdown efficiency was validated via qRT-PCR, confirming effective depletion of the corresponding mRNA levels (Supplementary Fig. 1). To determine the relative abundance of the two CD47 mRNA isoforms, isoform-specific qRT-PCR was performed to measure the distal PAS usage index before and after knockdown of core 3′ end processing factors. PCR primers were designed to target a common region present in all CD47 mRNA isoforms to quantify total CD47 mRNA expression. Additional primers were designed to specifically amplify the long CD47 isoform, allowing for the quantification of distal PAS usage. Changes in distal PAS usage were quantified using the 2^−ΔΔCT method , where the difference in CT values between distal and common amplicons was calculated and normalized to control shRNA samples, with log₂-normalized 2^−ΔΔCT values indicating a decrease (negative value) or an increase (positive value) in distal PAS usage upon knockdown. Knockdown of specific subunits of CPSF, CSTF, and CFIIm resulted in subtle but detectable shifts in the relative expression of CD47 isoforms. Notably, CPSF73 knockdown caused a significant reduction in distal PAS usage. In addition, knockdown of CFIm subunits, with the exception of CFIm59, led to a significant decrease in distal PAS usage (Fig. 2a). These findings are consistent with previous studies indicating that depletion of CFIm25 and CFIm68, but not CFIm59, leads to an overall increase proximal PAS usage^19,20. The lack of effect from CFIm59 knockdown may be attributed to functional redundancy between CFIm59 and CFIm68²⁰. Another possibility is that CFIm59 functions antagonistically with CFIm68, thereby downregulating CFIm activity²⁰. In contrast, PCF11 knockdown led to an increase in distal PAS usage (Fig. 2a), which aligns with previous findings demonstrating that PCF11 knockdown favors distal PAS selection²¹. These results indicate that core 3′ end processing factors play a crucial role in modulating CD47 PAS usage.

Fig. 2.

Core 3′ end processing factors influence CD47 APA, and CFIm25 depletion alters CD47 protein localization. (a) qRT-PCR analysis of total CD47 mRNA and the long 3′ UTR isoform (CD47-LU) in HeLa cells. Changes in distal PAS usage were calculated using the 2^−ΔΔCT method, where CT represents the threshold cycle. The relative abundance of CD47-LU was determined by calculating the difference in CT values between the distal and common amplicons (ΔCT = CT(distal) − CT(common)). ΔCT values obtained under knockdown conditions were then normalized to control shRNA samples, and log₂-transformed values are shown. Results are presented as mean ± s.d. **, P < 0.01; *, P < 0.05 (two tailed t-test for independent samples, n = 3). (b) Absolute quantification by qRT-PCR of total CD47 mRNA and its APA isoforms in HeLa cells transfected with shCFIm25 or shControl. Results are presented as mean ± s.d. **, P < 0.01; *P < 0.05 (two-tailed t-test for independent samples, n = 3). (c) Fluorescence confocal microscopy images obtained using the double staining approach in HeLa cells stably expressing control shRNA or shRNA targeting CFIm25. Scale bars, 20 µm. (d) Quantification of MFI values from panel c. ****, P < 0.0001; ***, P < 0.001 (two-tailed t-test for independent samples, n = 3). (e) FACS analysis of cell surface and intracellular CD47 expression in HeLa cells processed as described in panel (c) Representative histograms display cell surface and intracellular CD47 expression, with unstained cells shown in gray.

Among the core 3′ end processing factors, CFIm25, encoded by NUDT21, has been previously identified as a key factor in APA regulation under various physiological and pathological conditions^7,14,22. Consistent with these reports, absolute quantification by qRT-PCR, with CD47-SU expression determined as the difference between total CD47 and CD47-LU, demonstrated an increase in CD47-SU expression and a concomitant decrease in CD47-LU expression upon CFIm25 knockdown (Fig. 2b). To determine whether CFIm25 depletion affects CD47 protein localization, confocal microscopy and FACS analysis were performed. A shift in the ratio of cell surface to intracellular CD47 fluorescence was observed, with a significant increase in intracellular CD47 upon CFIm25 knockdown (Fig. 2c–e). These findings suggest that the double staining approach provides a reliable method for detecting APA-associated changes in PAS usage at the single-cell level, and that CD47 protein localization can serve as an functional indicator of APA process in a CRISPR-based genetic screen.

Identification of candidate genes affecting CD47 surface expression through a genome-wide CRISPR screen

We first established a HeLa cell line stably expressing humanized hSpCas9 and validated its expression and knockout efficiency by analyzing CD47 protein localization following NUDT21 knockout (Supplementary Fig. 2a–c). Cells were transduced with lentivirus expressing either a control sgRNA (sgCTRL) or sgRNA targeting NUDT21 (sgCFIm25). Seven days post-transduction, cell surface and intracellular CD47 expression levels were assessed by flow cytometry, revealing that NUDT21 knockout increased intracellular CD47 expression with a marginal effect on cell surface CD47 expression, consistent with the phenotype observed in CFIm25 knockdown experiments (Supplementary Fig. 2d). These results confirm that the CRISPR-based system provides a robust platform for studying APA regulation.

To systematically identify novel APA regulators, we performed a genome-wide CRISPR screen by transducing hSpCas9-expressing HeLa cells with a genome-wide sgRNA library comprising 76,612 sgRNAs targeting 19,114 human genes (four sgRNAs per gene) at a low multiplicity of infection and selected with hygromycin B (Fig. 3a). Seven days post-transduction, FACS was used to isolate two distinct populations based on intracellular-to-surface CD47 expression ratios, namely CD47 surface low and CD47 surface high cells (Fig. 3b). Genomic DNA was extracted from sorted and unsorted populations, and deep sequencing was performed to quantify sgRNA abundance. MAGeCK analysis identified numerous genes whose depletion altered CD47 protein localization (Fig. 3c, d; Supplementary Table 3). GO enrichment analysis of the top 100 ranked genes enriched in the CD47 surface high population revealed overrepresentation of genes involved in clathrin adaptor activity and cargo adaptor activity, including AP2 complex subunits, COPE, and CIDEB (Fig. 3e). Given their roles in endocytosis, depletion of these genes likely impaired CD47 internalization, leading to its retention at the plasma membrane. In the CD47 surface low population, genes associated with protein localization, vesicular organization, and membrane trafficking were enriched (Fig. 3f). Among them, several vacuolar ATPase components, the Sec61 complex subunit SEC61B, and subunits of the TRAPP complex were identified. Additionally, genes encoding Rab geranylgeranyltransferase subunits, RABGGTA and RABGGTB, and exocyst complex components were detected (Fig. 3f). Knockout of these genes resulted in reduced CD47 cell surface localization, suggesting their role in CD47 transport from the ER to the plasma membrane.

While most of the top-ranked hits were implicated in CD47 protein localization, the screen also identified several genes previously implicated in APA, including NUDT21, SSU72, CDK9, METTL3, and LSM11 (Fig. 3c–f)^23–25. However, some core 3′ end processing factors, such as those belonging to CPSF and CSTF complexes, were not among the top-ranked genes. This suggest that these factors have a limited impact on CD47 APA or that functional redundancy within these complexes compensates for the loss of individual components. This hypothesis is supported by previous studies demonstrating redundancy among subunits of 3′ end processing complexes, which could explain why depletion of these components did not significantly alter CD47 APA^{14,15,19,20,26}. Additionally, the screen identified genes involved in mRNA homeostasis and protein translation, including genes encoding signal recognition particle components. The presence of these genes among the top hits suggests that their knockout may indirectly impact CD47 protein localization, possibly by affecting mRNA stability or translation efficiency²⁷. Together, these results demonstrate that, while the majority of genes identified in our CRISPR screen are involved in protein localization and membrane trafficking, a significant subset is linked to APA regulation. The presence of NUDT21 and other APA-related genes among the top hits validates our approach, suggesting that the list may also include novel APA regulators.

Secondary screen identifies POLDIP2 as a potential APA regulator

Although the genome-wide screen successfully identified candidate genes, the experiment inherently favors sensitivity over specificity, leading to the inclusion of false positives among the identified hits. To address this limitation, we performed a secondary screen targeting the highest-ranked genes from the primary screen. Specifically, we selected the top 25 genes from both CD47 surface low and CD47 surface high populations, prioritizing those localized inside or outside the nucleus (Fig. 4a). Using similar experimental design as in the primary screen, we cloned the most effective sgRNAs from the primary screen into pLVXS-sgRNA-mCherry-hyg. To serve as an internal control, non-targeting sgRNAs were also included and cloned into pLVXS-sgRNA-hyg, which lacks mCherry, allowing for clear distinction from other sgRNAs. The secondary screen successfully identified several genes reproducible from the primary screen, such as CD47, components of the AP2 adaptor complex, and exocyst complex subunits. Notably, we also identified genes known to play a role in APA regulation, including SSU72, LSM11, and METTL3, whose knockdown altered CD47 APA (Fig. 4a, Supplementary Table 4)^23–25.

Fig. 4.

Secondary screen identifies candidate genes potentially involved in APA regulation. (a) Schematic of the secondary CRISPR screen based on differential CD47 localization. (b) Quantification of MFI values from FACS analysis of CD47. The double staining approach was applied to HeLa cells stably expressing control shRNA or shRNA targeting candidate genes. Results are presented as mean ± s.d. ****, P < 0.0001; ***, P < 0.001; **, P < 0.01 (two-tailed t-test for independent samples, n = 3). (c) qRT-PCR analysis of total CD47 mRNA and the long 3′ UTR isoform in HeLa cells stably expressing control shRNA or shRNA targeting candidate genes. Results are presented as mean ± s.d. ****, P < 0.0001; ***, P < 0.001; **, P < 0.01 (two-tailed t-test for independent samples, n = 3).

Since genes directly involved in APA regulation are likely to be localized in the nucleus²⁸, we selected seven genes encoding nuclear-localized proteins for further analysis, namely RBBP4, SLBP, RBM15, POLDIP2, HDAC3, BRD4, and ZBTB25. To assess their effects on CD47 protein localization, HeLa cells were transduced with lentivirus expressing shRNAs targeting these candidate genes, and knockdown efficiency was validated by qRT-PCR (Supplementary Fig. 3). Depletion of several candidate genes resulted in a significant increase in CD47 fluorescence intensity at the cell surface. Specifically, knockdown of CFIm25, RBBP4, SLBP, and POLDIP2 led to a marked increase in surface CD47 expression, whereas knockdown of RBM15, HDAC3, BRD4, and ZBTB25 had no significant effect. For intracellular CD47, CFIm25 and particularly POLDIP2 knockdown led to a substantial increase in fluorescence intensity (Fig. 4b). These FACS results confirmed that a subset of candidate genes affects CD47 protein localization.

To determine whether the observed CD47 localization changes were caused by APA alterations, we performed qRT-PCR analysis to quantify CD47-SU and CD47-LU isoform abundance. The results demonstrated that knockdown of the candidate genes affected CD47 APA, with changes in isoform expression patterns generally corresponding to the CD47 protein localization shifts observed in FACS analysis. Notably, POLDIP2 knockdown resulted in a substantial increase of CD47-SU, comparable to the changes observed with CFIm25 knockdown. Additionally, CD47-LU expression also increased, though to a lesser extent, upon POLDIP2 depletion (Fig. 4c). These results demonstrate that a subset of secondary screen candidate genes, particularly POLDIP2, significantly affects CD47 protein localization and its APA.

Depletion of POLDIP2 has a global impact on APA

In recent years, several bioinformatics tools have been developed to analyze APA using RNA-seq data, including Dynamic analyses of Alternative PolyAdenylation from RNA-seq (DaPars) and Tool for Alternative Polyadenylation Site Analysis (TAPAS)^29,30. To identify global targets of POLDIP2, we performed RNA-seq following POLDIP2 knockdown alongside a parallel control knockdown (Supplementary Fig. 4a). Differential expression analysis revealed 8124 genes with significantly altered expression (FDR < 0.01) in response to POLDIP2 depletion (Fig. 5a, Supplementary Table 5). Among them, 1613 genes exhibited significantly increased expression (log₂ fold change > 1), while 2213 genes showed significantly decreased expression (log₂ fold change <  − 1). Heatmap analysis confirmed that POLDIP2 depletion strongly affected most Differentially expressed genes (DEGs) (Supplementary Fig. 4b). Given that POLDIP2 is a multifunctional protein with context-dependent roles^31,32, it is unsurprising that its depletion broadly affects gene expression. GO enrichment analysis of the DEGs revealed biological regulation, cellular response to stimulus, developmental process, and cell differentiation as the most enriched biological processes. The lack of strong enrichment for specific pathways suggests that POLDIP2 knockdown disrupts gene expression globally rather than targeting a single pathway (Supplementary Fig. 4c, d). We next analyzed CD47 and observed reduced read density within the 3′ UTR in response to POLDIP2 depletion (Supplementary Fig. 4e). These results confirm that qRT-PCR findings reliably detect APA alterations and demonstrate that POLDIP2 knockdown induces 3′ UTR shortening, which can be visualized via RNA-seq read density analysis.

Fig. 5.

The QAPA algorithm identifies POLDIP2 as a novel APA regulator. (a) Volcano plot of gene expression changes in POLDIP2-depleted cells (POLDIP2KD) compared to control cells. Genes with FDRs < 0.01 and |log₂ fold changes|> 1 were considered significant. (b) Scatter plot of WULI values in control and POLDIP2-depleted cells. Genes exhibiting significant 3′ UTR shortening (n = 584) and lengthening (n = 472) upon POLDIP2 knockdown (|ΔWULI|≥ 0.1) are highlighted. (c), (d) Representative RNA-seq density plots and corresponding ΔWULI values for genes exhibiting 3′ UTR shortening (c) or lengthening (d) upon POLDIP2 knockdown (POLDIP2KD) versus control (CTRL) cells. The y-axis represents RNA-seq read coverage. (e) Expression levels of core 3′ end processing factors upon POLDIP2 depletion, derived from the RNA-seq data. Results are shown as log₂ fold change ± standard error. ****, FDR < 0.0001; ***, FDR < 0.001; **, FDR < 0.01; *, FDR < 0.05 (n = 3). (f) Correlation between ΔWULI values and gene expression changes. The x-axis represents ΔWULI, where negative values indicate a shift toward proximal PAS usage in POLDIP2 knockdown cells. The y-axis represents the log₂-transformed gene expression changes upon POLDIP2 knockdown.

To further assess POLDIP2’s role in global APA regulation, we applied QAPA, a bioinformatics algorithm that quantifies 3′ UTR differences³³. QAPA first identifies PASs in 3′ UTRs using non-redundant annotations from PolyAsite, a repository of PAS coordinates from published 3′-end sequencing datasets, and GENCODE PolyA annotation track, which contains manually annotated PASs. These annotations are used to refine existing proximal 3′ UTR coordinates and define alternative 3′ UTR isoforms. The difference in 3′ UTR usage between two samples is quantified as a change in PAU values, reflecting shifts in PAS selection. To enhance the robustness of APA quantification, we also incorporated the Weighted 3′ UTR Length Index (WULI), which accounts for all mRNA isoforms derived from distinct PASs. A positive ΔWULI value indicates 3′ UTR lengthening, whereas a negative ΔWULI value indicates 3′ UTR shortening in POLDIP2 knockdown cells compared to control cells³⁴. QAPA identified 1056 genes with a significant and reproducible shift in 3′ UTR usage upon POLDIP2 depletion (Fig. 5b, Supplementary Table 5). Among them, 584 genes, including DIAPH2 and GALNT2, exhibited significant 3′ UTR shortening, while 472 genes, including GOLGA8B and TFPI, showed significant 3′ UTR lengthening. Representative examples of POLDIP2-regulated APA genes are shown in Fig. 5c, d, along with their respective ΔWULI values. These results suggest that POLDIP2 influences PAS selection, leading to both 3′ UTR shortening or lengthening in affected genes.

Such bidirectional APA changes have rarely been observed. One possible explanation for the POLDIP2 knockdown phenotype is that POLDIP2 directly or indirectly influences both core 3′ end processing factors that favor proximal PAS usage and those that favor distal PAS usage. To explore this possibility, we analyzed RNA-seq data to examine how POLDIP2 knockdown affects the expression of core 3′ end processing factors. The results showed that CPSF160, WDR33, and CSTF77 were affected by at most a twofold change in expression (Fig. 5e). Given the impacts that knockdown of these factors had on CD47 APA (Fig. 2a), it is unlikely that these expression changes alone could account for the global APA alterations observed.

It is widely believed that mRNA isoforms with long 3′ UTRs are generally less stable than their shorter counterparts due to the presence of destabilizing elements and increased susceptibility to degradation pathways. However, this notion has been challenged by a genome-wide APA study in mouse cells. Using the transcription inhibitor actinomycin D to assess mRNA stability³⁵, the study found that long isoforms are only marginally less stable than short isoforms. Furthermore, additional regulatory elements such as stabilizing sequences within the UTR can significantly influence mRNA decay rates^35–37. To explore this complexity, we examined whether changes in 3′ UTR length upon POLDIP2 depletion correlated with altered gene expression levels. By plotting ΔWULI against gene expression changes, we assessed the relationship between APA-induced 3′ UTR length changes and transcript abundance. The results showed that transcripts exhibited significantly increased or decreased expression levels, independent of whether their 3′ UTRs were shortened or lengthened (Fig. 5f). These findings suggest that APA isoform stability is influenced by additional cis-regulatory elements beyond 3′ UTR length alone.

Discussion

In this study, we employed a pooled genome-wide CRISPR screen to systemically identify novel APA regulators. Since CRISPR-based screening requires detectable phenotypic changes, we integrated an immunofluorescence staining approach using CD47 protein localization as a proxy for APA changes. Additionally, we leveraged CD47’s differential localization to establish a cell selection system based on the ratio of cell surface to intracellular CD47 fluorescent signals. This double staining method facilitated FACS-based cell sorting for cell populations with altered CD47 PAS usage, which was then used for CRISPR-based screening. The strength of this approach lies in its ability to recapitulate endogenous APA regulation, allowing APA-related changes to be detected within their natural cellular context. This was demonstrated by CFIm25 knockdown, where we successfully detected CD47 protein localization changes. Our study identified numerous factors whose knockout either inhibited or promoted the localization of both cell surface and intracellular CD47. GO enrichment analysis revealed numerous genes associated with clathrin, protein localization, and vesicle organization, further supporting that the double staining method accurately reflects CD47 protein localization. While most of the top-ranked genes were expected to be involved in CD47 protein localization, we also identified genes implicated in APA regulation in both CD47 surface high and CD47 surface low populations, including NUDT21, SSU72, CDK9, METTL3, and LSM11. Notably, we also identified several proteins with functions seemingly unrelated to either protein localization or APA, suggesting potential unexplored regulatory mechanisms.

Our screening identified several interesting potential APA regulators. One such candidate, RBBP4, is associated with histone acetylation and chromatin assembly^38,39, and previous studies have suggested that histone modifications and chromatin remodeling influence APA^40–42. However, our results indicate that RBBP4 depletion did not specifically affect CD47 APA but instead led to a general increase in both CD47-SU and CD47-LU isoforms, suggesting that RBBP4 is involved in the transcriptional regulation of CD47 (Fig. 4c). Similarly, we identified BRD4 and HDAC3 as potential APA regulators. A previous study by Arnold et al. demonstrated that BRD4 depletion disrupts the recruitment of core 3′ end processing factors to the 5′ control regions, resulting in RNA cleavage and termination defects⁴³. While this strongly suggests that BRD4 is essential for optimal RNA cleavage and termination, our results showed that BRD4 depletion, much like RBBP4 depletion, led to a general increase in both CD47-SU and CD47-LU isoforms. Meanwhile, Xiang et al. performed a bioinformatics analysis of RNA-seq data following HDAC inhibition and HDAC3 depletion, revealing a global 3′ UTR shortening in HDAC inhibitor–treated samples⁴⁴. Their study also found that HDAC3 depletion produced a mixed outcome, with 175 genes exhibiting 3′ UTR shortening and 132 genes exhibiting 3′ UTR lengthening. However, in our study, the APA changes in CD47 upon HDAC3 depletion were relatively modest compared to those observed for another candidate, POLDIP2. Given this, we prioritized POLDIP2 as the primary candidate for further investigation (Fig. 4c).

POLDIP2 has been characterized as a multifunctional protein with distinct roles depending on its cellular localization^31,32. In the nucleus, POLDIP2 was initially identified as an interacting partner of the p50 subunit of DNA polymerase δ⁴⁵. It also associates with PCNA, suggesting a role in coordinating DNA replication and repair⁴⁵. Additionally, POLDIP2 has been implicated in DNA damage response through its interaction with translesion synthesis DNA polymerase η⁴⁶. However, conflicting evidence suggests that POLDIP2 does not directly participate in translesion synthesis, as it is not recruited to DNA damage sites following UV irradiation⁴⁷. Instead, POLDIP2 appears to translocate to spliceosomes, where it contributes to UV-induced alternative splicing of MDM2⁴⁷. Since POLDIP2 lacks a canonical RNA-binding domain, it has been hypothesized that it functions as a protein scaffold, facilitating protein–protein interactions involved in mRNA splicing⁴⁷. Beyond its nuclear role, POLDIP2 has also been identified as a mitochondrial protein, where it interacts with the C-terminal tail of p22phox, a NADPH oxidase subunit. This interaction contributes to Nox4 enzymatic activity and the production of reactive oxygen species⁴⁸.

This study further characterized POLDIP2, exploring its potential role as a previously unrecognized APA regulator. While knockdown of POLDIP2 led to an overall increase in CD47 mRNA and protein expression, it uniquely caused a disproportionate upregulation of the short 3′ UTR isoform relative to the long isoform (Fig. 4). This isoform imbalance could not be explained by general transcriptional upregulation alone and suggested a role for POLDIP2 in APA regulation, which we further validated on a transcriptome-wide scale in Fig. 5. Unlike well-established APA regulators, such as CFIm25, POLDIP2 appears to regulate APA in a bidirectional manner, leading to both 3′ UTR shortening and lengthening across different genes. This suggests that POLDIP2 may influence APA through multiple distinct mechanisms. Several possible explanations may account for the observed effects of POLDIP2 depletion on APA. For example, POLDIP2 may directly interact with RNA to regulate PAS selection. Several core 3′ end processing factors function as RNA-binding proteins with sequence specificity^4,6,15,18. For example, CFIm binds to two UGUA elements, one upstream and one downstream of a proximal PAS to suppress proximal PAS usage and favor distal PAS selection⁴⁹. CFIm25 depletion has consistently leads to global 3′ UTR shortening in human genes⁷. If POLDIP2 possesses sequence-specific RNA-binding capabilities, it may similarly bind to regions flanking either proximal or distal PASs in a gene-dependent manner. Consequently, POLDIP2 knockdown could result in bidirectional APA changes. Alternatively, POLDIP2 may act as a mediator that facilitates interactions among APA regulatory proteins. If POLDIP2 functions as a protein scaffold, it may coordinate protein–protein interactions involved in PAS selection and processing. Finally, POLDIP2 may indirectly modulate APA by regulating the expression of core 3′ end processing factors. Our results indicate that POLDIP2 depletion modestly alters the expression of several 3′ end processing factors (Fig. 5e). Given the impact of these factors’ knockdown on CD47 APA (Fig. 2a), it is unlikely that these expression changes alone account for the global APA alterations observed. However, their combined effects may have a more pronounced impact on APA. While further studies are necessary to fully elucidate its role in APA regulation, our findings demonstrate that POLDIP2 knockdown results in 3′ UTR alterations in a substantial number of genes. We recognize, however, that these effects may reflect both direct and indirect consequences of its loss. Additional work will be required to determine whether POLDIP2 directly regulates specific components of the APA machinery.

POLDIP2 has been implicated in various diseases, including atherosclerosis, ischemic injuries, Alzheimer’s disease, and cancers^50–55. Interestingly, APA has also been linked to multiple diseases^5,14,56, suggesting a potential connection between POLDIP2 function and APA regulation in pathological contexts. Notably, MAPT, the gene encoding tau, was found to exhibit 3′ UTR shortening upon POLDIP2 depletion (Supplementary Fig. 4f.). MAPT is known to undergo APA, generating two distinct 3′ UTR isoforms, with the long 3′ UTR isoform containing multiple cis-elements that repress its expression^57,58. The long mRNA isoform plays a critical role in neuronal cells, as it is transported to axons via an axonal localization signal within the 3′ UTR and locally translated. In contrast, when this signal is absent, both the mRNA and the protein remain confined to the cell body⁵⁹. Although the precise role of POLDIP2 in Alzheimer’s disease remains unclear due to conflicting reports^52,60, our findings suggest that POLDIP2 may contribute to its pathogenesis by influencing MAPT APA and causing tau mislocalization. Further studies are needed to determine whether POLDIP2-mediated APA alterations directly impact tau distribution and neurodegenerative processes.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

C.W. and Y.Y. conceptualized the project and developed the methodology. C.W. performed the investigations, conducted bioinformatics analyses, and wrote the original draft. J.Y., S.S., and Y.Y. reviewed and edited the manuscript. Y.Y. supervised the project.

Data availability

RNA-seq data have been deposited in GEO under accession code GSE288919. Raw and processed datasets generated during this study are available from the corresponding author on reasonable request.

Declarations

Competing interests

The authors declare no competing interests.