Introduction

Most eukaryotic mRNAs undergo polyadenylation, a mechanism determining the cleavage and the addition of a poly(A)-tail at the 3’UTR extremity. Such process is mediated by the CPA machinery, a multiprotein assembly of four core complexes: the CPA specificity factor (CPSF), the cleavage stimulation factor (CSTF), and the cleavage factors I and II (CFI and CFII) [1].

Up to 70% of genes are characterized by the presence of multiple polyadenylation signals (PAS), which allow the generation of mRNA isoforms with different 3’UTRs [2], through the process called alternative polyadenylation (APA). APA alterations are frequent in oncological, immunological, hematological, and neurological disorders [3]. A general shortening of 3’UTRs has been observed in cancers compared to normal tissues, as it activates oncogenes by escaping from the repressive effect of miRNAs, thus leading to increased mRNA stability and protein level in tumor cells [4]. For example, Survivin and AURKA oncogenes preferentially use proximal PAS in ovarian and triple-negative breast cancers, thereby escaping from the repression mediated by miR-34c and hsa-let-7a [5, 6]. Moreover, in breast and lung cancers, a correlation between short 3’UTR expression and increased aggressiveness or lower patient survival was observed [7]. APA dysregulation in cancer may be partially explained by a global dysregulation of 3’UTR processing factors. For example, CFIm25 (NUDT21), whose knockdown is associated to 3’UTR shortening, was found downregulated in many different cancer types [8]. In contrast, CSTF2 was reported as upregulated in tumors (reviewed in Ding et al. [9]). Similarly, CPSF3 and CPSF4 were found upregulated in lung cancer, where their increased expression levels correlate with reduced overall and recurrence-free survival [10, 11]. In addition, it has been recently shown that genetic eQTLs could contribute to APA regulation by poly(A)-motif alteration, thus playing an important role in cancer risk determination [12].

In this view, deregulated CPA can be seen as a new hallmark of cancer and a potential reservoir of novel therapeutic targets. However, the phenotypic effects arising from the inhibition of CPA (CPAi) in cancer cells has been tested only for a handful of factors in selected tumor types. Suppression of CPSF4 by siRNA was shown to inhibit lung cancer cell proliferation, colony formation, and induce apoptosis [11]. Recently, the CRISPR/Cas9-mediated genetic abrogation of multiple CPSF subunits was reported to hamper viability of ovarian cancer cells, by lengthening 3’UTR and enhancing transcriptional readthrough, and consequently suppressing oncogenic pathways. In contrast, knockdown of NUDT21 produced controversial results in different tumor types [13, 14].

The proof-of-concept that CPA factors can be targeted pharmacologically has been provided only in a handful of very recent studies. Cui et al. showed the possibility of using JTE-607, the first known pharmacological inhibitor of CPSF73 (CPSF3), as a novel therapy to suppress human cancers [15]. They reported a similar impact of JTE-607 treatment on gene expression in HeLa and HepG2 cells, likely due to perturbations of both transcriptional readthrough and changes in APA. They also showed similarities between JTE-607-mediated CPAi and PCF11 knockdown, suggesting that different CPA factors may be good therapeutic candidates. Interestingly, CPAi effects appeared stronger in highly proliferating cells as well as in cells displaying high CPA activities; in this regard, FIP1 (FIP1L1) overexpression was shown to cause a global activation of CPA, thus sensitizing cells to CPAi. JTE-607 was also shown to induce apoptosis in a subset of acute myeloid leukemia and Ewing’s sarcoma cancer cell lines [16, 17]. Interestingly, pre-mRNA 3′ processing was found to be inhibited by JTE-607 in a sequence-specific manner, highlighting potential determinants of vulnerability [18]. In another work, Shen et al. developed additional CPSF3 inhibitors building upon the benzoxaborole scaffold, which were shown to exert potent antitumor activity against ovarian cancer cells in vitro and in vivo [19].

To extend our view on the potential of CPAi approaches as anticancer therapies, we systematically interrogated cellular dependencies inferred from large-scale genetic perturbation platforms (CRISPR and RNAi screenings [20, 21]) of the DepMap project (https://depmap.org/portal) to test the vulnerability of about one thousand cell lines of different cancer types from the Cancer Cell Line Encyclopedia (CCLE) [22] to the depletion of all known CPA factors.

Results

We first interrogated CRISPR- and RNAi-dependency scores of subunits belonging to the four core CPA complexes and observed that their inhibition is overall detrimental for cancer cells (Fig. 1A, B). While most core CPA factors are commonly essential according to CRISPR screening, only FIP1L1, is commonly essential for both screenings (Fig. 1C). In contrast, CSTF2 is strongly selective for embryonal tumors as from CRISPR screening, whereas CSTF2T is strongly selective for different breast cancer cells, in both CRISPR and RNAi screenings (Fig. 1C). Overall, CPSF subunits show the lowest average scores (Fig. 1D), meaning that most cell lines (from different tumor histotypes) are extremely vulnerable to their inhibition. This may be explained by the fact that CPSF is the complex implied in PAS recognition [2]. It is important to note that RNAi scores are generally higher than CRISPR ones (Fig. 1A, B, D), probably depending on the different method of gene perturbation [20, 21]. Nevertheless, this result highlights CPA factors, especially CPSF subunits, as possible actionable vulnerabilities. Our analysis corroborates the results recently obtained with CPSF3 pharmacological inhibitors [15, 19], as this factor shows extremely low CRISPR and RNAi scores, and thus could be considered as an optimal therapeutic candidate (Fig. 1A, B). In accord with Cui et al. observation that FIP1L1 overexpression sensitizes to CPAi [15], FIP1L1 resulted to be commonly essential (Fig. 1C). We found a trend for correlation between the mRNA expression levels of CPA factors (used as proxy of CPA activity) and the corresponding dependency scores (Fig. 1E) in the RNAi screening, a result that was not recapitulated in the CRISPR screening (Fig. 1E), overall indicating that vulnerability cannot be simply predicted by expression. Similarly, we did not appreciate any correlation between dependency scores and the proliferation status of cells, assessed by the expression of a cell-cycle proliferation (CCP) signature [23], though core factors were more expressed in highly proliferating cells (Supplementary Fig. S1). The only exception was CPSF1, the inhibition of which resulted more deleterious in high vs low proliferating cells (according to both CRISPR and RNAi studies) (Fig. 1F).

Fig. 1: Analysis of cancer cell line dependencies on core CPA factors.
figure 1

Boxplots of the (A) CRISPR (CRISPR DepMap Public 23Q2 + Score, Chronos) and (B) RNAi (RNAi Achilles + DRIVE + Marcotte, DEMETER2) dependency scores from the DepMap project for all the core CPA factors: CPSF complex (CPSF1, CPSF2, CPSF3, CPSF4, FIP1L1, WDR33, and SYMPK subunits) in blue, CSTF complex (CSTF1, CSTF2, CSTF2Τ, and CSTF3 subunits) in green, CFI complex (NUDT21, CPSF6, and CPSF7 subunits) in yellow, and CFII complex (CLP1 and PCF11 subunits) in red. The lower the CRISPR or RNAi score is for a certain factor, the more sensitive is a cell line to the inhibition of such factor; −0.5 was chosen as cutoff. C Table reporting the percentage of cells that are sensitive to the inhibition of a specific core CPA factors, as from DepMap project. Factors defined commonly essential are highlighted in red and the ones defined strongly selective are highlighted in purple. D Boxplots of the average CRISPR and RNAi dependency scores from the DepMap project for all the core CPA factors. E Barplot showing the average correlation values (Spearman) of the different core CPA complexes. The correlation is calculated between the CRISPR or RNAi scores and the expression levels of the different factors (as from CCLE version 23Q2). F Boxplots of the CRISPR and RNAi dependency scores for CPSF1 in high vs low proliferating cells. Cell lines were divided in high vs low proliferating cells according to a cell-cycle proliferation (CCP) gene signature [23], i.e. cells belonging to the first quartile of the distribution of the average expression are low proliferating cells, instead, cells belonging to the fourth quartile of the distribution of the average expression are high proliferating cells. RNA-seq data available for 1569 cell lines. CRISPR and RNAi scores available for 1096 and 712 cell lines, respectively. Boxplots report median values and barplots report mean values and relative standard deviation. Analyses and boxplots were performed using R software and t-test p-values are reported; *p < 0.05.

Full size image

To extend our analysis, we divided CPA factors in those that are predicted to promote the usage of proximal or distal PASs, according to the classification by Pereira-Castro & Moreira [24]. We reported a more variable vulnerability for both proximal and distal accessory CPA factors with respect core CPA factors (Fig. 2A, B). While most accessory factors are commonly essential according to CRISPR screening, only FIP1L1, SRSF7 (factors promoting proximal PASs), SNRNP70, SRSF3, and PAF1 (factors promoting distal PASs) resulted to be commonly essential for both screenings (Fig. 2C). Moreover, almost all proximal and distal CPA factors are invariably overexpressed in high vs low proliferating cells (Supplementary Fig. S2). Although we expected that cancer cells were more vulnerable to the inhibition of CPA factors promoting proximal PASs, distal CPA factors resulted to be more deleterious when inhibited in the cells, as compared with proximal factors, according to both CRISPR and RNAi scores (Fig. 2D).

Fig. 2: Analysis of cancer cell line dependencies on accessory CPA factors promoting proximal or distal PASs.
figure 2

Boxplots of the (A) CRISPR and (B) RNAi scores from the DepMap project for all the CPA factors that promote the usage of proximal PAS (orange) and distal PAS (blue). C Table reporting the percentage of cells that are sensitive to the inhibition of a specific CPA factors promoting proximal or distal PAS, as from DepMap project. Factors defined commonly essential are highlighted in red and the ones defined strongly selective are highlighted in purple. D Boxplots of the average CRISPR and RNAi score for CPA factors promoting proximal and distal PAS. The lower the CRISPR or RNAi score is for a certain factor, the more sensitive is a cell line to the inhibition of such factor, −0.5 was chosen as cutoff. CRISPR and RNAi scores available for 1096 and 712 cell lines, respectively. Boxplots report median values. Analyses and boxplots were performed using R software and t-test p-values are reported; ****p < 0.0001.

Full size image

Perspectives and limitations

In this study, we confirmed core and accessory CPA factors as novel vulnerabilities for human cancer and as potential new targets for anticancer therapy, overall valorizing and expanding the recent findings obtained using CPSF3 pharmacological inhibitors [15, 19]. Partially in contrast Cui et al. results and with the common view of APA role in cancer, we found that sensitivity to CPAi apparently does not correlate with cell proliferation or CPA factor expression, or with the intrinsic role of the factor in promoting proximal vs distal PAS usage, as could be expected for tumor cells that preferentially have shorter 3’UTRs. In this regard, recent studies are challenging the conventional paradigm for which 3’UTR shortening is associated with the proliferative status of cells and only shortening events are relevant to cancer. In fact, the advent of novel techniques, such as single cell RNA-seq or 3’-seq evolutions, has allowed to reveal that 3’UTR shortening can also occur during hematopoiesis and trophoblast differentiation, where it is uncoupled from proliferation and is rather associated with protein secretion [25]. In addition, several reports show that lengthening events also occur with a variable frequency in human cancers, where they can be even associated with worse prognosis [26,27,28]. While not extensively explored so far, it is plausible that 3’UTR lengthening of tumor-suppressor genes can equally contribute to cancer initiation/progression. In light of these recent results, our findings indicate that the role of 3’UTR lengthening factors (and vulnerability of cells to their inhibition) is far from being fully elucidated both in the context of proliferation/differentiation processes as well as in cancer.

We acknowledge some limitations of our study. First of all, dependency scores derive from in vitro screenings and lack of an in vivo validation. One important bottleneck is also the unavailability of dependency data for non-immortalized cells. Indeed, testing the differential vulnerability of normal and tumor cells to CPAi is a fundamental step for the translation into therapy, as it would allow to define the therapeutic window as well as to better evaluate the distinct therapeutic potential of “proximal” and “distal” CPA factors. In this regard, we found that almost all CPA factors are significantly upregulated in cancer cells compared to normal cells (Supplementary Fig. S3), which raises optimism regarding CPAi selectivity. Similarly, CPA protein level may represent a more reliable proxy to ultimately determine whether CPA activity may predict CPAi efficacy. Though these aspects still need to be comprehensively addressed and experimentally validated, our analysis confirms CPAi as a novel therapeutic approach for human cancer and suggests extending pharmacological studies to factors other than CPSF3. In this regard, CPSF1 appeared as an optimal candidate, even for aggressive tumors, as it showed low dependency scores overall and particularly in highly proliferating cells in both CRISPR and RNAi screenings. Notably, CPSF1 is upregulated pancancer (as from TCGA data [29], Supplementary Fig. S4) and its siRNA-mediated silencing was shown to suppress cell viability, impair colony formation ability, induce cell cycle arrest, and promote apoptosis in ovarian cancer cells [30]. In addition, canSAR predictions [31] indicate that CPSF1 protein has druggable structural domains (i.e. cavities that are available for ligand interaction, as from protein structures obtained by cryo-EM or X-ray diffraction) and is ligandable by chemistry-based assessments (i.e. small molecules have been already identified to interact with it or its homologs), suggesting it as a pharmacologically exploitable vulnerability. If CPSF1 appeared as pancancer essential, essentiality of other CPA factors (e.g. CSTF2 and CSTF2T) seemed to be context dependent. In a personalized medicine perspective, the observed differential vulnerability of cancer cell lines to selected CPA factors may be used to build up signatures to predict response of individual human tumors to CPAi approaches.