Alternative polyadenylation is a determinant of oncogenic Ras function

Shortening ends of mRNAs cooperate with oncogenes in development and cancer.

Abstract

Alternative polyadenylation of mRNA has important but poorly understood roles in development and cancer. Activating mutations in the Ras oncogene are common drivers of many human cancers. From a screen for enhancers of activated Ras (let-60) in Caenorhabditis elegans, we identified cfim-1, a subunit of the alternative polyadenylation machinery. Ablation of cfim-1 increased penetrance of the multivulva phenotype in let-60/Ras gain-of-function (gf) mutants. Depletion of the human cfim-1 ortholog CFIm25/NUDT21 in cancer cells with KRAS mutations increased their migration and stimulated an epithelial-to-mesenchymal transition. CFIm25-depleted cells and cfim-1 mutants displayed biased placement of poly(A) tails to more proximal sites in many conserved transcripts. Functional analysis of these transcripts identified the multidrug resistance protein mrp-5/ABCC1 as a previously unidentified regulator of C. elegans vulva development and cell migration in human cells through alternative 3′UTR usage. Our observations demonstrate a conserved functional role for alternative polyadenylation in oncogenic Ras function.

INTRODUCTION

The transition of normal cells to the neoplastic state requires genetic alterations that include activation of oncogenes and loss of tumor suppressors. Genome-wide analysis of numerous cancers has yielded a dizzying array of secondary mutations in genes that can affect the evolution of the cancer cell and its tumorigenic potential (1). The Ras oncogene is conserved from yeast to human and is one of the most frequently mutated genes in cancer, with up to 30% of all tumors having gain-of-function (gf) mutations in one or more of the three Ras paralogs: KRAS, HRAS, and NRAS (2–4). Ras proteins are critical regulators of signaling pathways that control proliferation, cell survival, and tissue differentiation. Point mutations in the guanosine triphosphatase domain of Ras lead to constitutive activation of prosurvival pathways that contribute to tumor development and progression (3, 5). Ras and other oncogenes, such as MYC and NOTCH1, require specific levels of expression, pathway activation, or cross-talk with other signaling axes to sustain tumor phenotypes in mammalian systems (6, 7). Studying the mechanisms by which these signaling pathways interact is challenging using established cell lines, which are often highly mutated and aneuploid (8). Model organisms provide robust complementary systems to study genetic interactions within the context of living animals.

The powerful genetic tools of the nematode worm Caenorhabditis elegans have provided much insight into the in vivo functions of oncogenes such as Ras. Mutational activation of the C. elegans Ras homolog let-60 occurs by a glycine to glutamic acid substitution at amino acid 13 (the n1046 allele), similar to the glycine to valine at amino acid 12 found in numerous human cancers. The hallmark phenotype of this let-60 gf allele is the multivulva (Muv) phenotype, in which ectopic vulvae develop along the ventral body of the worm (9). Forward genetic screens in C. elegans have helped establish the hierarchical order of Ras/mitogen-activated protein kinase (MAPK) signaling components and uncovered numerous genes that regulate Ras/LET-60 signaling that are conserved from worm to human (10).

Although substantial efforts to target and directly inhibit Ras in cancer have been undertaken, these ventures have shown limited success (11). Likewise, strategies for targeting downstream Ras effectors have had limited clinical success due to dose-limiting toxicity of the compounds (12). In addition, genetic compensation and signaling cross-talk within the Ras signaling network contribute to drug resistance in many tumors (13). Hence, novel strategies to inhibit Ras signaling for cancer treatment are needed to overcome these obstacles. While ambitious genome sequencing efforts have uncovered variants in genes that can affect the tumorigenic state, much less is known about how epigenetic influences on gene expression and function cooperate with oncogenic pathways.

An important epigenetic mechanism of gene regulation involves an essential step in mRNA maturation, which determines the length of the 3′ untranslated region (3′UTR). This occurs by endonucleolytic cleavage of the pre-mRNA in the 3′UTR, followed by the addition of the polyadenylate [poly(A)] tail. This cleavage reaction is directed by poly(A) site (PAS) site consensus sequences composed of multiple cis elements, including (i) an AAUAAA hexamer recognized by the cleavage and polyadenylation specificity factor (CPSF) complex, (ii) downstream U/GU-rich elements that are recognized by the cleavage stimulation factor (CstF), and (iii) upstream U-rich elements recognized by cleavage factor Im (CFIm). These proteins establish a platform for subsequent recruitment of cleavage factor IIm (CFIIm) and its cofactors to form the functional 3′UTR processing holoenzyme that cleaves and polyadenylates pre-mRNA (14). Many genes contain multiple polyadenylation sites in their 3′UTRs, enabling cleavage and polyadenylation at different PA locations, a process termed as alternative polyadenylation (APA). APA serves to direct the placement of the poly(A) tail such that isoforms of the same gene are produced with different 3′UTR lengths. The 3′UTR contains numerous regulatory sites for the binding of microRNAs and RNA binding proteins (RBPs), which regulate mRNA stability, localization, and translational efficiency (15).

Shorter 3′UTRs of specific genes have been associated with higher rates of proliferation, and these transcripts undergo 3′UTR lengthening as the cells reach a more differentiated state (16). About 30 to 40% of genes in C. elegans and about 70% of genes in mammals undergo APA, respectively (17, 18). Alterations in APA have also been associated with tumorigenesis, where transformed cells tend to have shorter 3′UTRs of key proliferative and survival genes. This often results in higher translation rates and increased levels of proteins that promote tumorigenesis (19). Apart from regulating transcript stability and translation, recent studies have demonstrated how changes in APA of specific transcripts can also influence the localization and functions of resulting proteins (20, 21). These observations have provided insight into the complexity of APA in diseases such as cancer; however, the precise mechanisms by which APA factors cooperate with activated oncogenes remain unknown.

In this study, we uncover evolutionarily conserved roles of the APA factor, cfim-1/CFIm25, in oncogenic Ras signaling using C. elegans and human cancer cell models. We found that cfim-1/CFIm25 functions to buffer oncogenic Ras output by regulating the ratio of 3′UTR lengths for conserved transcripts in both systems. These include the multidrug resistance protein (mrp-5/ABCC1), where shortening of its 3′UTR cooperates with oncogenic Ras to enhance its functional output in development and cancer cell behavior.

RESULTS

RNAi screen uncovers cfim-1 as a novel regulator of activated let-60/Ras

We previously demonstrated that the hypersensitivity to DNA damage–induced germline apoptosis of the Ras/let-60(ga89) gf allele, which has a more penetrant phenotype in the germ line, is dependent on a functional insulin-like growth factor 1 receptor (IGF1R) homolog, daf-2. Loss-of-function (lf) mutations in daf-2 completely suppress sensitivity to DNA damage–induced germline apoptosis in let-60(ga89) mutants (22). To determine whether daf-2 ablation also buffers activated LET-60/Ras in the soma, we turned to the vulva, the egg-laying organ of C. elegans that is extremely sensitive to perturbations in Ras/MAPK signaling. Loss of Ras/MAPK signaling prevents development of the vulva, resulting in the vulvaless (Vul) phenotype, whereas increased activity causes the multivulva (Muv) phenotype (23). The Muv phenotype used to measure hyperactive LET-60/Ras was quantified in two ways: (i) by determining the percentage of worms with at least one ectopic vulva (% Muv) and (ii) by generating a vulva induction score. Vulva induction was measured by counting the number of cells that have adopted vulval fates at the fourth larval (L4) stage (described in Materials and Methods). In normal development, only three vulval precursor cells (VPCs) adopt vulval cell fates; thus, the induction score for wild-type (WT) worms is 3.00, and the % Muv is 0 (Fig. 1, A and B). Under conditions of hyperactive LET-60/Ras, more VPCs adopted vulval cell fates, so induction scores above 3 represent Muv worms. The let-60(n1046) gf allele causes strong hyperactivation of Ras in somatic tissue, with an average vulval induction score of 4.09 and a penetrance of 71% Muv (Fig. 1, A to C).

Fig. 1. cfim-1 regulates oncogenic Ras activity in C. elegans.

(A) Schematic outline of vulval development in C. elegans. (B) VPC induction scores for WT (N2) and combinations of let-60(gf), daf-2(lf), and daf-16(lf) mutants. % Muv indicates the percentage of animals with more than three VPCs induced. N indicates the number of animals measured. Y-axis origin is set to 3 to represent the normal number of VPCs that are induced to divide. ****P < 0.0001 and *P < 0.05, one-way analysis of variance (ANOVA) with Tukey’s post hoc test. Detailed descriptions for all strains used in the study are described in Materials and Methods. (C) Representative images of larval 4 stage (L4) worms expressing the hyperactive let-60(n1046) allele; let-60(gf); daf-2(lf) double mutant, in which the Muv phenotype was completely suppressed; and let-60(gf); daf-2(lf); daf-16(lf) triple mutant, in which the Muv phenotype was partially restored. Red arrows denote ectopic pseudovulvae. (D) Schematic representation of screen performed by feeding let-60(gf); daf-2(lf) worms (suppressed Muv) RNAi against DAF-16 target genes. (E) VPC induction scores for let-60(gf); daf-2(lf) on HT115 control and cfim-1 RNAi, let-60(gf); daf-2(lf); cfim-1(lf) mutants, and let-60(gf); cfim-1(lf) mutants. % Muv indicates the percentage of animals with more than three VPCs induced. N indicates the number of animals measured. Y-axis origin is set to 3 to represent the normal number of VPCs that are induced to divide. *P < 0.05, one-way ANOVA with Tukey’s post hoc test. (F) Percentage (%) Muv scored for the let-60(ga89gf) and let-60(ga89gf); cfim-1(lf) mutants. Each data point represents quantification from three independent experiments. ****P < 0.0001, one-way ANOVA with Tukey’s post hoc test compared to cfim-1(lf) or let-60(ga89). (G) Schematic representation of the effect of cfim-1 loss in the context of different levels of Ras activation.

The dependency on daf-2/IGF1R on activated LET-60/Ras was also observed in the soma, where daf-2(lf) suppressed the Muv phenotype of let-60(n1046gf) worms to 5% (Fig. 1, B and C) (24). This suppression was partially dependent on the transcription factor daf-16/FOXO3a, which functions downstream of the daf-2/IGF1R pathway. daf-2(lf); let-60(gf); daf-16(null) triple mutants had a restoration of the Muv phenotype to a 3.65 induction score and 54% Muv (Fig. 1, B and C). Details of all C. elegans strains are provided in Table 3. Taking advantage of the let-60(gf); daf-2(lf) double mutant in which the Muv phenotype was suppressed, we conducted a screen to identify novel genes that restore the Muv phenotype similar to what was observed by ablating daf-16 (Fig. 1D).

Table 3. List of C. elegans strains.

N2	WT
let-60(n1046)	Strong somatic Ras gain-of- function allele, (G13E substitution)
let-60(ga89)	Weaker gain-of-function Ras allele (L19F substitution)
daf-2(e1370)	daf-2 loss of function
daf-16(mgDf47)	daf-16 loss of function
cfim-1(on9)	cfim-1(lf) generated by CRISPR
daf-2(e1370); cfim-1(on9)	daf-2(e1370) crossed with cfim-1(lf)
let-60(n1046); daf-2(e1370)	let-60(n1046) crossed with daf-2(e1370)
let-60(n1046); daf-2(e1370); daf-16(mgDf47)	let-60(n1046);daf-2(e1370) crossed with daf-16(mgDf47)
let-60(n1046); daf-2(e1370); cfim-1(on9)	let-60(n1046); daf-2(e1370) crossed with cfim-1(lf)
let-60(ga89); cfim-1(on9)	let-60(ga89) crossed with cfim-1(lf)

We focused on genes that are either known or predicted to be DAF-16 transcriptional targets, reasoning that restoring the Muv phenotype in let-60(gf); daf-2(lf) mutants would identify transcriptional targets of DAF-16 that regulate LET-60/Ras function. Therefore, a list of DAF-16 transcriptional target genes was generated from several unbiased studies, including chromatin immunoprecipitation and DNA adenine methyltransferase identification (DamID). Genes in the top 5% of methylation peaks from the DamID study and at least one other study were considered potential DAF-16 targets and prioritized for the screen (25–31). On the basis of this, we screened 202 candidate genes by RNA interference (RNAi) and found that the ablation of cfim-1 increased the Muv phenotype of let-60(gf); daf-2(lf) double mutants to an induction score of 3.27 and 23% Muv (Fig. 1E). This effect was even more pronounced after we generated a CRISPR deletion allele of cfim-1 and built a let-60(gf); daf-2(lf); cfim-1(lf) triple mutant, which was 45% Muv and had a vulva induction score of 3.55 (Fig. 1E). cfim-1 loss in the context of the let-60(n1046) single mutant (high Ras activation) did not show any additional enhancement of Muv (Fig. 1E). Since the n1046 allele is a strong let-60/Ras(gf) mutant, we wondered whether cfim-1 might enhance a weaker allele. Therefore, we crossed the cfim-1(lf) allele into the let-60(ga89) mutant and observed an increase in Muv from 2 to 40%, indicating that cooperation between cfim-1 and let-60 was independent of the daf-2/IGF1R pathway (Fig. 1F). This indicates that cfim-1(lf) can enhance the Muv output with low levels of oncogenic Ras/LET-60 in C. elegans (Fig. 1G).

cfim-1 regulates APA of transcripts that mediate oncogenic let-60/Ras output

cfim-1 encodes the C. elegans homolog of mammalian NUDT21/CFIm25, a subunit of the 3′ polyadenylation cleavage CFIm complex that directs poly(A) tails to distal cleavage sites to produce transcripts with longer 3′UTRs. Depletion of CFIm25 causes a global shortening in 3′UTRs, resulting in transcripts that escape negative regulation by microRNAs and/or RBPs (Fig. 2A). Therefore, we asked whether cfim-1 modulates let-60/Ras signaling output through biased APA of specific transcripts. We reasoned that if cfim-1 functions to attenuate let-60/Ras signaling output, then changes in the 3′UTRs of genes in cfim-1(lf) mutants would uncover proteins that either feed back to the Ras/MAPK pathway or amplify downstream signaling. Worms were subjected to whole-genome 3′UTR sequencing, and downstream analysis was performed by calculating the proximal poly(A) site usage (PPAU). cfim-1(lf) single and daf-2(lf); cfim-1(lf) double mutants exhibited substantial 3′UTR shortening, where a significant portion of genes had a higher number of reads mapping to the most proximal PA site (P < 2.2 × 10⁻¹⁶) (Fig. 2, B and C, and fig. S1B). In contrast, the daf-2(lf) single and daf-2(lf); daf-16(lf) double mutants did not exhibit notable alterations in their 3′UTR landscapes compared with WT controls (fig. S1, A and B). Given that cfim-1 was initially screened as a predicted DAF-16 target, we investigated the transcript levels of cfim-1 and found that it was not significantly altered in daf-2(lf) and daf-2(lf); daf-16(lf) strains compared to the known DAF-16 targets, indicating that cfim-1 is not likely a transcriptional target of DAF-16 (fig. S1C). These data, together with cfim-1(lf) restoring Muv in the let-60(ga89) strain, independent of daf-2/IGF1R, suggest that cfim-1 functions downstream, or in a parallel pathway, to attenuate activated LET-60/Ras output.

Fig. 2. cfim-1 loss promotes 3′UTR shortening of let-60/Ras pathway–related transcripts.

(A) Schematic outline of CFIM-1/CFIm25 function in the context of APA. CFIm25 is a component of the CFIm complex (denoted in yellow), which, together with other protein complexes, dictates polyadenylation of transcripts. CFIm25 biases toward the usage of distal polyadenylation sites, resulting in a longer 3′UTR. The longer 3′UTR often contains regulatory sequences such as microRNA (miRNA) and RBP sites, which regulates transcript stability and translation. Loss of CFIm25 is associated with shortening of 3′UTRs and subsequent loss of regulatory sequences. (B) Proximal polyadenylation site usage (PPAU) calculated in cfim-1(lf) and daf-2(lf); cfim-1(lf) strains. ****P < 2.2 × 10⁻¹⁶ compared with N2 controls. Data are the average from three independent replicates for each strain. (C) Scatter diagram showing the distribution of APA-regulated genes based on delta-PPAU of respective mutants versus expression. A higher/positive PPAU indicates that a proximal PA site is selected for that gene and vice versa. A significant proportion of genes show positive delta-PPAU, indicating increased usage of the proximal polyadenylation site in cfim-1(lf) and daf-2(lf); cfim-1(lf) mutants. Red dots represent genes with significant PPAU, and gray is nonsignificant. Violin plot shows the distribution of genes based on PPAU calculation in each strain. CPM, counts per million. (D) DEXSeq analysis and quantification of the number of genes altered by APA. “Shorter” and “longer” denote the increased expression of this isoform in the respective strains as described in fig. S2. (E) RNAi screen of the Ras pathway–associated genes in the let-60(gf); daf-2(lf); cfim-1(lf) mutant. Each data point represents an independent experiment. Colors/shapes of data points correlate with the indicated colors/shapes of genes (representing functional categories) in network analysis (fig. S3).

We next asked how the altered 3′UTR landscape in the absence of cfim-1 might modulate Muv in the let-60(gf) mutant. In addition to measuring PPAU, we also used the DEXSeq method to classify APA events. PA sites within 3′UTR of a gene were denoted as “exons,” and the number of reads mapping to these PA sites was quantified. This enabled us to classify genes into different APA classes, such as isoform switching (increase in short and decrease in long or vice versa), or changes in just one isoform (short/long), between cfim-1(lf) mutants compared to WT controls (fig. S2). Consistent with the UTR index results, loss of cfim-1 significantly altered the APA landscape (Fig. 2D). We identified 505 genes with increased expression of the short isoform in both cfim-1(lf) single and daf-2(lf); cfim-1(lf) double mutants (denoted as “shorter in mutants”) and 181 genes shortened specifically in the daf-2(lf); cfim-1(lf) double mutant (Fig. 2D). We also found that 49 genes showed a decrease in expression of the long isoform in cfim-1(lf) mutants. These observations are consistent with the established roles of CFIm25 in biasing toward distal PAS usage efficiency, as described in other systems (32, 33). Gene Ontology analysis of the 686 genes with increased proximal PAS usage identified 119 that were enriched in Ras signaling or processes that regulate oncogenesis in mammalian cells, such as cell migration, cell cycle, and cytoskeletal dynamics, as well as genes involved in vulva development and kinase activity (fig. S3). Therefore, we screened these 119 genes by RNAi in let-60(gf); daf-2(lf); cfim-1(lf) triple mutants (45% Muv) and found 12 that modulated the Muv phenotype, of which 3 function in cell migration (kgb-1/MAPK10, sem-5/GRB2, and pak-1/PAK1) and 1 in vulva development (usp-46/USP46). Ablation of capg-1, a condensin complex subunit, increased the percentage of Muv (Fig. 2E). We also investigated these 12 genes in the let-60(n1046) single mutant and found no significant effect on the Muv phenotype, suggesting that these genes likely had cfim-1(lf) dependency (fig. S4A).

We further assessed the correlation between 3′UTR shortening and gene expression of the 12 cfim-1–dependent genes. Only egl-20 showed increased expression with 3′UTR shortening in daf-2(lf); cfim-1(lf) double mutants, whereas expression of the other 11 genes did not exhibit a significant correlation (fig. S4B). This suggests that 3′UTR shortening of those genes could have other consequences, such as altered stability of specific isoforms, altered translation and localization, and function (34). Recently, PAK1, the human ortholog of pak-1, was reported to undergo APA and served as a predictive marker of glioblastoma progression (35). Thus, our results indicate that cfim-1 regulates APA of common transcripts that affect oncogenic let-60/RAS output in C. elegans and human with established roles in cancer biology (Table 1).

Table 1. Disease-relevant C. elegans genes.

C. elegans gene	Human ortholog	Function
egl-20	WNT16	Cell fate regulator, paracrine signaling to promote tumor cell survival (66)
top-1	TOP1, TOP1MT	DNA topoisomerase, regulates DNA topology during transcription
C05D9.3	ITGB4	Transmembrane glycoprotein receptors that regulate cell-cell adhesion, up-regulated in many cancers (67)
usp-46	USP12, USP46	Deubiquitinating enzyme, has both oncogenic and tumor suppressor roles (68)
kgb-1	MAPK10/JNK3	Kinase involved in proliferation, differentiation, and cell death, activated by inflammatory cytokines (69)
T20H4.5	NDUFS8	Complex I of respiratory chain, responsible for NADH oxidation
pyp-1	PPA1, PPA2	Inorganic pyrophosphatase that plays a role in energy metabolism, up-regulated in numerous cancers (70)
rab-6.2	RAB6A	Involved in the transport of proteins from Golgi body to endoplasmic reticulum (71)
sem-5	GRB2	Activates RAS/MAPK downstream of the EGFR receptor (72)
R119.3	DHRS2	Member of the short-chain dehydrogenases family, metabolizes steroid hormones and lipids
pak-1	PAK1	Kinase involved in cytoskeletal remodeling, contributes to metastasis of many cancers (73)
capg-1	NCAPG	Part of the condensin complex, which condenses chromosomes during mitosis and meiosis

CFIm25 regulates migration and invasion of KRAS-mutant cancer cells

To determine whether our observations in C. elegans were functionally conserved, we assessed CFIm25 function in human KRAS-mutant cancer cells. We used two HCT116 isogenic cancer cell lines, where the parental line had a heterozygous KRAS G13D mutation (KRAS G13D), and its isogenic counterpart had the mutant KRAS allele knocked out (KRAS WT) (36). Using these cell lines, we explored the mechanism by which CFIm25 functions in the context of mutant Ras in human cancer cells.

Ablation of CFIm25 by small interfering RNAs (siRNAs) resulted in spindle-like morphology and cytoplasmic protrusions in cells (fig. S5A). This was reminiscent of cells undergoing an epithelial-to-mesenchymal transition (EMT), which is associated with metastatic characteristics such as increased migration and invasion. Depletion of CFIm25 significantly increased the migratory and invasive properties of KRAS G13D cells (Fig. 3A). Compared to KRAS G13D cells, KRAS WT cells showed reduced migration and invasion, and depletion of CFIm25 in KRAS WT cells did not significantly alter their migratory or invasive potential (Fig. 3A). We corroborated the EMT phenotype using markers and observed a decrease in the epithelial marker E-cadherin and an increase in expression of the mesenchymal marker Vimentin under CFIm25 knockdown (KD) conditions for KRAS G13D cells, with a less pronounced effect in WT cells (Fig. 3, B and C). To assess whether CFIm25 depletion can affect the MAPK pathway downstream of Ras, we assessed expression levels of phospho–extracellular signal–regulated kinase (ERK) and observed that CFIm25 depletion increased phospho-ERK levels in G13D cells (Fig. 3B). We observed higher levels of basal phospho-ERK levels in WT cells compared to G13D, and we saw slight increases with CFIm25 depletion. This observation could be attributed to feedback mechanisms as seen in prior studies (37). Nevertheless, CFIm25 depletion enhances RAS/MAPK signaling in cells having oncogenic Ras.

Fig. 3. CFIm25 depletion increases migration and invasion of KRAS-mutant cells.

(A) Quantification of migration and invasion of KRAS G13D and KRAS WT HCT116 cell lines with/without CFIm25 KD. Each data point represents a field of view captured across three independent (migration) or two (invasion) independent experiments. ****P < 0.0001; ns, not significant; two-way ANOVA with Tukey’s post hoc test. Representative images show a single field of view of cell migration following knockdowns with indicated siRNAs. (B) Representative Western blot of E-cadherin and phospho-ERK expression in KRAS WT and KRAS G13D cell lines 72 hours after transfection with the indicated siRNAs and quantification of E-cadherin protein levels normalized to tubulin across three independent experiments. (C) Representative immunofluorescent images (maximum projection) of Vimentin (green) in KRAS G13D and KRAS WT 72 hours after transfection with the indicated siRNAs. Nuclei were stained with DAPI (blue). Scale bars, 10 μm. (D) Time course growth assay measuring the growth of KRAS G13D and KRAS WT with or without CFIm25 KD using Alamar Blue. n = 5 independent experiments, means ± SD. ***P < 0.001, unpaired Student’s t test. (E) Cell cycle analysis of KRAS G13D and KRAS WT with or without CFIm25 after 72 hours of KD. Table depicts the proportion of cells in each phase ± SD across two independent experiments. (F) Representative images (maximum projection) and quantification of γ-H2AX intensities in G13D and WT cells transfected with indicated siRNAs. ****P < 0.0001 and **P < 0.01, two-way ANOVA with Tukey’s post hoc test. Scale bars, 14 μm.

Activation of the EMT program, apart from being associated with increased cell migration and invasion, has also been associated with decreased cell proliferation (38). Cell growth and cell cycle analysis revealed that depletion of CFIm25 in KRAS G13D cells led to decreased cell growth and an increased fraction of cells in the S and G₂-M phases, with similar changes occurring in KRAS WT cells (Fig. 3, D and E). Oncogenic Ras activation has also been associated with increased genome instability through the accumulation of DNA damage and subsequent cell growth arrest (39). Since the depletion of CFIm25 induced arrest in S phase, we wondered whether this was associated with increased DNA damage. Therefore, we quantified the DNA damage marker γ-H2AX and found that CFIm25-depleted KRAS G13D cells had significantly increased levels (Fig. 3F). WT cells displayed significantly lower levels of basal γ-H2AX compared to G13D cells. CFIm25 depletion in WT cells increased γ-H2AX levels to basal levels observed in G13D cells (Fig. 3F). Thus, our observations cumulatively suggest that cells depleted of CFIm25 can undergo numerous alterations in behavior associated with the EMT program, but the presence of mutant KRAS is required for their increased migration and invasion.

We extended our studies to other KRAS-mutated cell lines to determine whether CFIm25 depletion can enhance Ras-dependent cell migration in different cell models. We made use of the DLD-1 isogenic paired cell line (KRAS G13D mutation and its WT counterpart) and three pancreatic cancer lines [BXPC3 (KRAS WT), CFPAC-I (KRAS G12V), and HPAF-II (KRAS G12D)] (40). Consistent with our observations in HCT116 cells, CFIm25 KD increased migration of KRAS-mutant DLD-1 with no significant effects on the WT line, as well as in the CFPAC-I and HPAF-II lines, with no significant effects observed in the KRAS WT BXPC3 cell line (fig. S5, B and C). Together, these data indicate that CFIm25 depletion has a common role in regulating EMT and cell migration in Ras-driven cancer cells.

CFIm25 regulates 3′UTR shortening of conserved genes

To identify transcripts regulated by CFIm25 in cancer cells, we subjected both KRAS-mutant and WT HCT116 cell lines, with and without CFIm25 KD, to RNA sequencing (RNA-seq) (Fig. 4A). Sequencing was performed at high depth, and we used the quantification of alternative polyadenylation (QAPA) pipeline to quantify the relative abundance of short and long 3′UTR isoforms (41). CFIm25 depletion significantly increased the overall PPAU in both G13D and WT cells (Fig. 4, B and C). APA event classification was performed similar to our analysis of the C. elegans dataset (fig. S2). From this analysis, 415 genes had increased expression of shortened isoforms, and 1369 genes also showed a decrease in long isoform expression in CFIm25 KD cells compared to controls. The APA landscape of KRAS WT cells revealed that CFIm25 depletion promoted increased proximal isoform expression of 421 genes and decreased expression of long isoforms for 1508 genes (Fig. 4D). This analysis also uncovered previously identified CFIm25 targets such as cyclin D1 that are subject to 3′UTR shortening after CFIm25 depletion (fig. S6A), thereby validating our pipeline (33).

Fig. 4. cfim-1/CFIm25 regulates 3′UTR shortening of conserved genes in C. elegans and human.

(A) Schematic outline of RNA-seq performed in KRAS G13D and KRAS WT HCT116 cells. Sequencing was performed with three independent replicates for each condition. (B) Proximal PA site usage (PPAU) calculated for G13D and WT cells transfected with control or CFIm25 siRNA. ****P < 0.0001, unpaired Student’s t test compared to respective controls for each cell line. Data are the average from three independent replicates for each condition. (C) Scatter diagram showing the distribution of APA-regulated genes based on delta-PPAU of CFIm25 KD versus control. A significant proportion of genes with high Z scores (quantification is described in Materials and Methods) show positive delta-PPAU, indicating increased usage of the proximal polyadenylation site. (D) Classification of APA-regulated genes into categories based on changes in isoform usage. (E) Venn diagram showing overlap in the number of genes with increased expression of proximal isoform in C. elegans (genes have been converted into their respective human orthologs) and in H. sapiens G13D cells. (F) RNAi screen of commonly regulated C. elegans genes in the let-60(gf); daf-2(lf); cfim-1(lf) triple mutants. Each data point represents quantification from an independent experiment. ****P < 0.001; **P < 0.01, unpaired Student’s t test. (G) Bar diagrams depicting expression data for the proximal and distal isoforms of mrp-5 and its human ortholog ABCC1 in the indicated strains/cell lines. PPAU values across the three independent experiments are also shown. ****P < 0.0001, unpaired Student’s t test.

Ingenuity Pathway Analysis (IPA) of shortened transcripts in KRAS G13D cells revealed enrichment in cancer-related signaling pathways, such as protein kinase A, WNT/β-catenin, and ERK/MAPK axes, supporting our observations that CFIm25 depletion biases APA to shorter isoforms of genes involved in cancer cell migration and invasion-related pathways (fig. S6B). Mining our RNA-seq dataset also revealed expression changes in EMT-related genes after CFIm25 depletion, especially in the KRAS-mutant cells, including reduced Occludin, and up-regulation of multiple EMT-associated chemokines (fig. S6C), providing more evidence that CFIm25 regulates EMT (42).

Having identified hundreds of genes that are subject to 3′UTR shortening, we referred to the C. elegans APA dataset to identify novel APA-regulated genes that are important for oncogenic Ras function in human cancer cells. Since there is a high degree of pathway conservation, we reasoned that comparing the C. elegans and human datasets would allow us to identify key pathways subject to APA regulation. Our initial screen of genes with shortened 3′UTRs in C. elegans uncovered 11 that were required for enhanced Ras output (Muv) in cfim-1(lf) mutants. The 686 C. elegans genes that showed increased expression of the short/proximal isoform were converted to their corresponding human orthologs using Ortholist 2 (Fig. 4E) (43). This generated a list of 1215 human genes (a single C. elegans gene can have multiple orthologs in humans). IPA analysis of these genes showed enrichment of similar cancer-related pathways as the human shortened gene set (fig. S7A). Comparing the C. elegans list to the human dataset revealed 33 C. elegans genes that had 45 orthologous human counterparts (Fig. 4E and Table 2), including the genes that suppressed Muv such as pak-1/PAK1, as well as several genes that did not fall into the functional categories we initially focused on. We also compared the genes that showed decreased distal/long isoform usage and found less overlap in this case (17 genes commonly regulated) between the two organisms (fig. S7B). We therefore focused on the genes that showed increased proximal isoform expression due to the greater degree of overlap.

Table 2. Genes that are commonly regulated by 3′UTR shortening in C. elegans and human cancer cells.

C. elegans genes	H. sapiens genes
mrp-5	ABCC1
mrp-5	ABCC2
mrp-5	ABCC5
F59C12.3	AMOTL2
ensa-1	ARPP19
chd-3	CHD4
mup-4	COL12A1
let-2	COL5A1
scpl-1	CTDSP2
scpl-1	CTDSPL
dyn-1	DNM1
ensa-1	ENSA
epc-1	EPC2
epn-1	EPN3
D1069.3	FAM126B
fln-2	FLNB
act-2	GLIS3
hil-1	H1F0
lsy-13	ING4
wwp-1	ITCH
C28H8.4	KDELR1
ldb-1	LDB2
F58A3.1	LEMD3
oga-1	MGEA5
ZK1240.3	MID1
mml-1	MLXIP
hst-1	NDST1
mup-4	NOTCH2
pak-1	PAK1
ceh-18	POU2F1
ceh-18	POU2F2
kin-1	PRKACB
pkc-3	PRKCI
daf-18	PTEN
F35G2.1	QSOX2
rnh-1.0	RNASEH1
sec-24.2	SEC24B
ant-1.1	SLC25A5
ZK1240.3	TRIM17
ZK1240.3	TRIM22
ZK1240.3	TRIM25
ZK1240.3	TRIM38
ttyh-1	TTYH3
ftt-2	YWHAB
ftt-2	YWHAG

Ablation of the 33 C. elegans genes in let-60(gf); daf-2(lf); cfim-1(lf) triple mutants identified 4 genes that suppressed and 2 genes that enhanced the Muv phenotype. In addition to pak-1/PAK1 (observed in our initial screen; Fig. 2E), other genes that suppressed the Muv phenotype were mrp-5/ABCC, hst-1/NDST1, and mml-1/MLXIP (Fig. 4F), which have established prosurvival or chemoresistance roles in cancers (44–46). In line with a previous study, we also observed that daf-18/PTEN is subject to APA regulation in both systems, and its ablation enhanced the Muv phenotype of let-60(gf); daf-2(lf); cfim-1(lf) triple mutants (Fig. 4F) (47). In summary, these results indicate that cfim-1/CFIm25 plays important and conserved roles in regulating 3′UTRs of genes that modulate Ras signaling.

ABCC1 contributes to CFIm25-dependent cell migration and EMT

One of the top candidates from our screen of the 33 cfim-1–regulated genes was mrp-5, which is orthologous to the ABCC family of transporters in humans. Multiple family members (ABCC1, ABCC2, and ABCC5) were observed to undergo 3′UTR shortening after CFIm25 depletion in human cancer cells (Fig. 4G and fig. S7C). Since we observed increased expression of proximal isoforms for these transporters in both C. elegans and human cells, we next asked whether this is functionally relevant in the context of oncogenic Ras.

To corroborate our observations in Fig. 4F, we depleted mrp-5 in C. elegans let-60(gf); daf-2(lf); cfim-1(lf) triple mutants and found that it significantly suppressed VPC induction scores from 3.47 to 3.07 (Fig. 5A). Similarly, cosuppression of ABCC1 with CFIm25 showed the strongest suppression of migration of HCT116 G13D cells, compared with depletion of ABCC2 and ABCC5. KD of ABCC1 alone exhibited only subtle effects on cell migration (Fig. 5B and fig. S7D). Codepletion of CFIm25 and ABCC1 also increased E-cadherin levels significantly and partially decreased phospho-ERK levels, indicating that ABCC1 and CFIm25 cooperate in regulating the Ras pathway and EMT (Fig. 5C). We also found that the codepletion of CFIm25 and ABCC1 reversed the cell cycle arrest phenotype, accumulation of γ-H2AX, and partially restored cell growth rate (Fig. 5, D to F). Depletion of ABCC1 alone had only modest effects on cell cycle progression and cell growth. Thus, ABCC1 is a potent effector of cellular behavior when the APA landscape is shifted in response to CFIm25 depletion. We also assessed the functional relevance of ABCC1 and CFIm25 in DLD-1 G13D cells and found that their codepletion also reversed the increased cell migration observed with CFIm25 depletion alone (fig. S7E), indicating that CFIm25 and ABCC1 have broader roles in the context of Ras biology.

Fig. 5. Conserved transcripts regulate oncogenic Ras activity and tumorigenesis.

(A) VPC induction scores of let-60(gf); daf-2(lf); cfim-1(lf) mutants on RNAi against mrp-5. Percent (%) Muv indicates the percentage of animals with more than three VPCs induced. N indicates the number of animals measured. Y-axis origin is set to 3 to represent the normal number of VPCs that are induced to divide. ****P < 0.0001, unpaired Student’s t test. (B) Quantification of migration of KRAS G13D cells with either single depletion or codepletion of CFIm25 and ABBC1/2/5. Each data point represents quantification of a field of view acquired across two independent experiments. Representative images are shown for each condition. Validation of KD is shown in fig. S7D. ****P < 0.0001, **P < 0.01, and *P < 0.05, one-way ANOVA with Tukey’s post hoc test. (C) Representative Western blot of E-cadherin and phospho-ERK expression in KRAS G13D cell line transfected with the indicated siRNAs and quantification of E-cadherin protein levels normalized to tubulin across three independent experiments. (D) Cell cycle analysis of KRAS G13D transfected with the indicated siRNAs. Table depicts the proportion of cells in each phase ± SD across two independent experiments. (E) Quantification of γ-H2AX intensities in KRAS G13D cells transfected with indicated siRNAs. ****P < 0.0001, not significant, one-way ANOVA with Tukey’s post hoc test. (F) Alamar Blue assay of KRAS G13D cells transfected with the indicated siRNAs. ****P < 0.0001 and **P < 0.01, one-way ANOVA with Tukey’s post hoc test. (G) Representative IC₅₀ calculation for KRAS G13D cells transfected with indicated siRNAs and treated with etoposide at increasing concentrations. Cell viability was measured using the Alamar Blue assay. (H) Average IC₅₀ values for KRAS G13D cells calculated across four independent experiments. **P < 0.01, ns-nonsignificant, one-way ANOVA with Tukey’s post hoc test. (I) Percent confluency of KRAS G13D transfected and treated with increasing concentrations of Etoposide- 0, 3, 8 and 23μM (averages from Fig. 5H). ****P < 0.0001, ***P < 0.001, *P < 0.05, ns-nonsignificant, Two-way ANOVA with Tukey’s post hoc test.

Since ABCC1 has well-known roles in drug efflux, we asked whether CFIm25-dependent regulation of ABCC1 has any effect on resistance to chemotherapeutics. Since depletion of CFIm25 causes accumulation of cells in S phase, we chose to assess the response to topoisomerase poisons that act in this phase of the cell cycle. Depletion of CFIm25 in HCT116 cells caused increased resistance to the topoisomerase inhibitor etoposide compared with controls. Codepletion of CFIm25 and ABCC1 restored sensitivity to etoposide back to control levels (Fig. 5, G to I). In contrast, we did not observe significant effects with another topoisomerase inhibitor, Camptothecin (fig. S7F). This could indicate specific dependencies displayed by CFIm25-depleted cells in response to different chemotherapeutics.

Collectively, these results show that increased expression of the short mrp-5/ABCC1 isoform in response to CFIM-1/CFIm25 depletion is functionally important and can regulate vulva development in C. elegans and cancer cell behavior in the context of oncogenic Ras. Although ABCC1 also underwent 3′UTR shortening in CFIm25-depleted KRAS WT cells, these cells did not exhibit increased migratory properties, indicating that CFIm25-dependent APA of transcripts such as ABCC1 cooperates specifically with cellular alterations induced by oncogenic KRAS.

ABCC1 cooperates with TGFB2 signaling to regulate cell migration

Given that ABCC1 can regulate cell migration, we sought to uncover the pathways that cooperate with ABCC1 function. We surveyed our RNA-seq dataset for gene expression changes in EMT-related pathways and found that several members of the TGFB (Transforming growth factor beta) pathway were altered after CFIm25 KD. These included the TGF ligands TGFB2, GDF5, and GDF15, as well as the downstream effector of the TGFB pathway, SMAD3 (Fig. 6A). We therefore asked whether CFIm25-depleted cells were dependent on TGFB signaling for cell migration. Treating CFIM25-depleted cells with A83-01, an inhibitor of the TGFB receptor, significantly reduced cell migration (Fig. 6B). We then asked whether CFIm25 and ABCC1 can cooperate with TGFB signaling and assessed the expression of the TGFB ligands TGFB1 and TGFB2 under CFIm25 KD and CFIm25/ABCC1 double-KD conditions. Quantitative reverse transcription polymerase chain reaction (qRT-PCR) analysis revealed that TGFB2, but not TGFB1, was up-regulated in both CFIm25 and CFIm25/ABCC1 codepleted cells despite the dual inhibition of CFIm25 and ABCC1 causing reduced cell migration (Fig. 6C). TGFB ligands function in an autocrine or paracrine manner to influence alterations in cellular behavior through the SMAD transcription factors (48). Phosphorylated SMADS localize to the nucleus where they can activate the transcription of target genes involved in a wide array of cellular processes, including cell migration. We therefore assessed the expression of the downstream effectors of the pathway, SMAD2 and SMAD3. Cells were stained with a SMAD2/3 antibody, and consistent with our gene expression data, CFIm25-depleted cells had higher SMAD2/3 levels, with increased accumulation in the nucleus. CFIm25/ABCC1 codepleted cells, however, showed reduced SMAD2/3 expression and nuclear localization (Fig. 6D). These observations indicate that ABCC1 functions to enhance TGFB activity specifically in the context of CFIm25 loss. To further validate the requirement for TGFB signaling, we performed migration assays with CFIm25-depleted and CFIm25/ABCC1 codepleted cells and found that supplementing media with TGFB2 significantly rescues the migration of CFIM25/ABCC1 codepleted cells (Fig. 6E).

Fig. 6. ABCC1 cooperates with TGFB signaling to regulate cell migration.

(A) Analysis of overall gene expression for TGFB2, GDF5, GDF15, and SMAD3 in KRAS G13D RNA-seq data. n = 3 ± SD. (B) Quantification of migration of KRAS G13D cells transfected with indicated siRNAs and treated with A83-01 (0, 0.25, and 2.5 μM). Each data point represents a field of view captured across three independent experiments. ****P < 0.0001, not significant, two-way ANOVA with Tukey’s post hoc test. (C) qRT-PCR analysis of CFIm25, ABCC1, TGFB1, and TGFB2 in cells transfected with the indicated siRNAS, n = 2 ± SD. (D) Representative confocal images and quantification of SMAD2/3 (green) in KRAS G13D cells transfected with the indicated siRNAS. Phalloidin (red) was used to stain actin, and DAPI (blue) was used to stain nuclei. Scale bars, 10 μm. Each data point represents a single nuclear quantification from two independent experiments. ****P < 0.0001, one-way ANOVA with Tukey’s post hoc test. (E) Quantification of migration of KRAS G13D cells transfected with indicated siRNAs in the presence or absence of TGFB2 (10 ng/ml). Each data point represents a field of view captured across three independent experiments. ****P < 0.0001, two-way ANOVA with Tukey’s post hoc test

3′UTR shortening mediates posttranscriptional regulation of ABCC1

Having genetically and functionally established a role of ABCC1 for CFIm25-dependent cellular responses, we went on to assess the relevance of the short and long isoforms of ABCC1. Since quantification of total ABCC1 mRNA showed no changes in overall levels (Fig. 7A), we assessed the localization and expression of ABCC1 protein by antibody staining in HCT116 KRAS G13D cells. As expected, we observed that ABCC1 localized to the plasma membrane by costaining the membrane protein GLUT1. In control cells, we also observed higher junctional localization of ABCC1, especially where they contacted each other, and a reduced expression pattern at cell “edges” that were not in contact with neighboring cells. In CFIm25-depleted cells, which display more spindle-like morphologies and lack cell-cell junctions, ABCC1 was evenly distributed along the plasma membrane (Fig. 7B and fig. S8A). ABCC1 KD caused a robust reduction in protein levels. Similar expression pattern changes were observed in DLD-1 KRAS G13D cells (fig. S8B), indicating that this redistribution of ABCC1 expression in CFIm25-depleted cells may hold important implications for cells undergoing EMT.

Fig. 7. 3′UTR sequence elements mediate posttranscriptional regulation of ABCC1.

(A) Analysis of overall gene expression for ABCC1 in KRAS G13D RNA-seq data. n = 3 ± SD. (B) Representative confocal images (maximum projection) of KRAS G13D cells transfected with the indicated siRNAs and stained for ABCC1 (green), GLUT1 (red), and DAPI (blue). Scale bars, 10 μm. Quantification of ABCC1 (intensity) was calculated by measuring the intensity profile across an ROI on cell membranes. In siRLuc samples, this included across cell junctions and cell edges. In siCFIm25 samples, random regions on cell membranes were selected. Additional images and quantifications for each condition are in fig. S8A. (C) Schematic representation of ABCC1 3′UTR luciferase reporter plasmids. Numbers denote the beginning and end of ABCC1 proximal or distal 3′UTR sequences. Quantification of Firefly luciferase activity for proximal and distal constructs normalized to Renilla luciferase (internal control). (D) qRT-PCR analysis for Firefly Luciferase expression performed for cells transfected with either proximal or distal ABCC1 3′UTR constructs normalized to Renilla Luciferase expression. *P < 0.05, unpaired Student’s t test. (E) Quantification of Firefly luciferase activity for proximal, distal, and distal with miR site deletion. **P < 0.01, ****P < 0.0001, unpaired t-test; not significant.

Given that there was no change in total ABCC1 mRNA levels, we hypothesized that the specific 3′UTR isoforms might contribute to its posttranscriptional regulation. As a proxy for measuring protein output of the specific 3′UTR isoforms, the short and long 3′UTRs were cloned into a luciferase reporter construct (Fig. 7C). Luciferase activity was reduced by ~2-fold for the ABCC1 distal construct compared to the proximal construct, indicating that the long 3′UTR contains sequence elements that limit protein output, either through RNA degradation or reduced translation of the long isoform. qRT-PCR analysis of the luciferase transcript revealed that the long isoform showed decreased expression, confirming that this ABCC1 isoform is less stable than the short one (Fig. 7D). Analysis of the distal 3′UTR of ABCC1 using TargetScan revealed overlapping consensus binding sites for microRNAs miR-145 (position 1728 to 1734) and miR-199a (position 1726 to 1733), which has been associated with regulating ABCC1 expression in breast cancers and sensitivity to chemotherapy (49). We deleted these microRNA binding sites in the distal 3′UTR construct but did not observe any changes in luciferase activity levels. These observations suggest that alternative mechanisms, either through other microRNAs or RBPs, serve to regulate the stability of the distal ABCC1 isoform in this cell type. Given that CFIm25 depletion promotes an increase in proximal isoform expression accompanied by a comparable decrease in distal isoform, as observed in the RNA-seq data (Fig. 4G), we predict that this results in insignificant changes to overall transcript levels. However, the fraction of proximal isoform affords a more stable transcript as seen by luciferase activity and mRNA levels, which, in turn, has strong effects on EMT and cell migration (Fig. 8). Collectively, these results indicate that disrupting the balance of short and long ABCC1 isoforms has profound consequences on the behavior of Ras-driven cells from nematode to human.

Fig. 8. Schematic model of CFIM-1/CFIm25 regulating oncogenic Ras signaling.

CFIM-1/CFIm25 functions to regulate alternative polyadenylation of mRNA transcripts. Loss of CFIM-1/CFIm25 promotes 3′UTR shortening of genes such as mrp-5 in C. elegans and its ortholog ABCC1 in human cancer cells. ABCC1 cooperates with TGFB2 signaling to regulate oncogenic Ras signaling, thereby influencing EMT, cell migration, and drug response.

DISCUSSION

Understanding the cooperation between posttranscriptional regulatory mechanisms and activated oncogenes is an important challenge for identifying cancer dependencies that extend beyond genetic alterations. Here, we show that alternative polyadenylation functions as a highly conserved process in the modulation of oncogenic Ras output. Specifically, the APA factor CFIM-1/CFIM25 functions to dampen hyperactive Ras-dependent output by biasing 3′UTR usage of multiple transcripts.

Loss of cfim-1 results in 3′UTR shortening of numerous genes in C. elegans that increase the Muv phenotype with oncogenic gf mutations in the Ras ortholog let-60. Loss of cfim-1 alone was insufficient to induce the Muv phenotype. Instead, cfim-1 ablation enhanced Muv when the strong let-60(n1046) allele was suppressed by a reduction-of-function allele in the IGFR daf-2 or in the weaker let-60(ga89) single mutant. Similarly, in isogenic colon cancer cells, CFIm25 depletion promoted the 3′UTR shortening of key oncogenes and induced EMT-like phenotypes both in Ras mutant and, to a lesser extent, in WT cells. However, increased migration and invasion were only observed in cells expressing mutant KRAS, suggesting that these alterations in the APA landscape cooperate with oncogenic Ras to modulate signaling output in both C. elegans and human cells. Studies have shown that thresholds of mutant KRAS activity dictate the progression of cancer and metastasis (50–52). We postulate that APA functions to lower the threshold for activated oncogene signaling in both organ development and cancer cell migration. However, how is oncogenic Ras output enhanced by biasing the APA landscape to shorter mRNA transcripts?

Although CFIm25 depletion has been associated with increased prevalence of cancer phenotypes in different tissues (33, 53–55), our study provides mechanistic insights into how conserved genes such as the mrp-5/ABCC family of transporters that are subject to APA by CFIm25 cooperate with oncogenic Ras. Specifically, ablation of mrp-5/ABCC1 suppressed the Muv phenotype in worms and migration of human cancer cells caused by cfim-1/CFIm25 depletion, indicating that these are highly conserved determinants of oncogenic Ras output. The role of ABCC1 in cancers has been traditionally associated with multidrug resistance through the efflux of chemotherapeutic compounds such as etoposide and vincristine (46, 56). High levels of ABCC1 transcript or protein have also been shown to correlate with drug resistance and increased metastasis of breast and other types of cancers (57), although the precise mechanisms of altered ABCC1 expression are unknown. Our study reveals that ABCC1 can regulate resistance to the topoisomerase inhibitor etoposide and cancer cell migration. Cancer cell migration, specifically, is dependent on TGFB signaling. Studies have shown that TGFB has a variety of roles in inducing EMT and drug resistance (48). Therefore, the identification of TGFB dependency in CFIm25-depleted cancer cells highlights a novel vulnerability that could be exploited for therapeutic purposes. Further studies are warranted to uncover the precise biochemical interaction between the ABCC1 membrane transporter and TGFB signaling.

The alterations in drug response and cell migration occur without overall changes in ABCC1 transcript. While we were unable to identify the precise microRNAs or RBPs that regulate transcript stability, it is apparent that the longer 3′UTR sequence contains elements that posttranscriptionally contribute to reduced expression of this isoform. Thus, the increase in proximal isoform expression and more uniform plasma membrane distribution of ABCC1 in CFIm25-depleted cells appear to hold important functional consequences for cells during EMT progression.

CFIm25 depletion affected the APA of hundreds of transcripts, suggesting that this posttranscriptional process serves to modulate expression thresholds of genes that function in multiple processes that orchestrate Ras-driven cellular behaviors. An example that we show in our study is the dependency on TGFB signaling. Therefore, identifying proteins/pathways affected by APA provides a unique opportunity to discover druggable targets to treat a variety of tumors. Given that these alterations would not be detectable using standard genome (DNA) sequencing technology, it highlights the importance of sequencing mRNA transcripts in patient samples.

Although ABCC1 is often amplified at the transcript or protein level in cancers, our study shows that ABCC1 is subject to a novel mode of regulation through alternative 3′UTR usage that does not alter transcript levels. We predict that this probably holds true for many novel genes that are not detected in standard gene expression studies. In support of this, a recent study using 3′-end sequencing of patient-derived chronic lymphocytic leukemia samples identified novel tumor suppressor genes altered through APA. These genes are not frequently mutated in cancer but have strong tumor suppressor functions, again highlighting the need to extend cancer-related studies beyond genomic DNA sequencing and standard gene expression analyses (58). This could be particularly important for pediatric cancers, where the mutational load is low compared with adult tumors (59). Model organisms such as C. elegans offer complementary and inexpensive in vivo systems to efficiently study the cooperation between activated oncogenes and APA factors. In addition, the array of genetic tools available in C. elegans offers many advantages for uncovering conserved genes that enforce oncogenic output relevant to human cancer. Our discovery of mrp-5/ABCC1 serves as one such example and highlights the advantage of combining C. elegans with human cancer cells to accelerate the discovery of therapeutic targets to exploit cancer vulnerabilities.

MATERIALS AND METHODS

Crosses to generate double and triple mutants

All crosses were performed by plating 3 to 5 hermaphrodites with 10 to 15 males on petri dishes filled with nematode growth media (NGM) containing a bacterial lawn. Approximately five F1 hermaphrodites at the L4 stage of development were then plated onto NGM plates. F2 progeny were collected at the L4 stage of development and singled out onto separate NGM plates. F2 worms were incubated until they laid eggs to propagate the strain, and then the F2 parents were collected and lysed. The worm lysate was used for PCR reactions to confirm the presence of the desired genetic lesion.

Quantification of the multivulva phenotype

The multivulva phenotype was scored on the basis of the degree of vulva induction. Vulva induction refers to the number of VPCs induced to form part of the vulva at the L2 stage of development. This is measured by visualizing the VPC daughter cells at the L4 stage of development. Visualization was performed using a Leica DMRA2 compound light microscope with differential interference contrast (DIC) optics at ×63 magnification. Worms were mounted on glass slides with 4% agarose pads and anesthetized using 20 mM tetramisole. Primary VPCs undergo three rounds of cell division and comprise eight daughter cells, secondary daughter cells undergo eight rounds of division and comprise seven daughter cells, and tertiary VPCs undergo one round of division and comprise two daughter cells. These daughter cells and the pattern they form are easily visualized by DIC microscopy at ×63 magnification. A normal vulva develops from one primary and two secondary vulva cells, which are assigned an induction score of 3.0 since three VPCs have been induced. Worms with a Muv phenotype often have tertiary VPCs induced to partial primary and secondary fates, which have only three to four vulva cell daughters developing into pseudovulvae. These are given an induction score of 0.5 as “half” of a vulval cell has been induced in these cases. In addition to induction scores, the percentage of Muv worms of various genotypes was calculated. In cases where induction scores were calculated, % Muv was calculated by dividing the number of worms with induction scores more than 3.0 by the total number of worms measured. In cases where no induction scores were calculated, % Muv was calculated by quantifying all worms with at least one pseudovulva and dividing this number by the total number of adult worms. All worms scored for vulva induction were raised at 20°C, with the exception of the let-60(gf); daf-2(lf); cfim-1(lf) strain, which was raised at 16°C.

RNAi feeding protocol

RNAi cultures were obtained from the Source Bioscience C. elegans RNAi library. Bacteria were struck onto LB plates containing ampicillin and tetracycline (working concentrations of 50 and 5 mg/ml, respectively). Plates were incubated overnight at 37°C. Single colonies were picked and inoculated in liquid LB^amp+tet overnight at 37°C with shaking. Following the overnight inoculation, isopropyl β-d-1-thiogalactopyranoside (IPTG) was added to each culture (0.5 μM) and incubated at 37°C with shaking for 4 hours. The bacterial culture was then concentrated 2× or 10× by centrifugation and mixing. The culture was then plated on solid NGM media with added carbenicillin and IPTG (working concentrations of 25 μg/ml and 25 mM, respectively). Worms were added to these RNAi plates at the L4 stage, and measurements were taken of the F1 progeny.

Candidate RNAi screening protocol

For the screening of DAF-16 target genes, single colonies of each RNAi from the Source Bioscience C. elegans RNAi library were inoculated into liquid LB media and induced with IPTG as described in the “RNAi feeding protocol” section. Bacterial cultures were plated onto 12-well plastic plates filled with solid NGM RNAi media so that each well contained a different RNAi bacteria strain. Each plate also included one well with negative control bacteria (HT115) and a positive control bacterial strain expressing daf-16 RNAi. Bacterial lawns were grown overnight at 37°C, and two L4 stage daf-2(e1370); let-60(n1046) double mutant worms were picked into each well. Worms were grown until pseudovulvae could be seen with a dissecting microscope in the daf-16 RNAi–positive control well. Plates were then scanned through the rest of the wells, noting RNAi treatments that resulted in worms being noticeably more Muv than the negative controls and appearing similar to worms treated with daf-16 RNAi. Each plate was replicated, and a positive hit was only scored if a Muv phenotype could be seen in both replicates. Positive hits were subjected to a more stringent analysis by calculating vulval induction scores described in the “Quantification of the multivulva phenotype” section. The candidate RNAi strains were grown and concentrated 2× and then plated onto 10-cm solid NGM RNAi media plates. We plated three to four daf-2(e1370); let-60(n1046) L4 worms onto these plates and quantitated vulva induction scored for 15 to 20 F1 progeny under ×63 magnification.

For the screen of genes with shortened 3′UTRs, RNAi cultures were grown overnight and induced with IPTG for 4 hours. Cultures were then concentrated 10× and plated onto NGM plates. Four to five let-60(n1046); daf-2(e1370); cfim-1(on9) L4 worms were picked onto each RNAi plate, and % Muv was calculated in the F1 generation. Only those RNAi treatments that were able to produce a significant increase in Muv induction compared to the negative controls HT115 were considered positive hits.

Generation of CRISPR strains

The cfim-1(lf) CRIPSR strain was created with CRISPR technology using the self-excising drug selection cassette as previously described (60). The “aaacctccg” sequence at position 79 from the cfim-1 start codon was mutated by deleting the 5–base pair (bp) “aaacc” and changing “tccg” to “tagt” to create frameshift and a premature stop codon (TAG).

RNA extraction

For C. elegans samples, RNA was extracted from L4 stage worms using TRIzol (Invitrogen). RNA from human cancer cells was extracted 48 hours after transfection using the RNeasy Kit from QIAGEN according to the manufacturer’s instructions.

RNA-seq library preparation and sequencing

To construct RNA-seq libraries from C. elegans samples, we used an automated QuantSeq 3′mRNA-seq (Lexogen GmbH, Vienna) using the Agilent NGS Workstation (Agilent Technologies, Santa Clara) as per the manufacturer’s protocol. Briefly, 250 ng of total RNA was used to generate complementary DNA (cDNA). cDNA was amplified with 17 PCR cycles as determined by qPCR analysis using the PCR Add-on Kit (Lexogen). External RNA Controls Consortium (ERCC) RNA spike-ins were added following the manufacturer’s instructions. The resulting libraries were quantified with Qubit DNA HS (Thermo Fisher Scientific, Waltham), and fragment sizes were analyzed on the Agilent Bioanalyzer using the High-Sensitivity DNA assay before sequencing. Sequencing was performed at The Centre for Applied Genomics (Hospital for Sick Children, Toronto) across two lanes of an Illumina HiSeq2500 rapid run flowcell (Illumina, San Diego) with SR100 bp.

For sequencing of human cancer cell samples, total RNA was treated with the DNA-free DNA Removal Kit (catalog no. AM1906, Thermo Fisher Scientific Inc., Waltham, USA) to remove contaminant DNA. Deoxyribonuclease-treated total RNA was then quantified using Qubit RNA BR (catalog no. Q10211, Thermo Fisher Scientific Inc., Waltham, USA) fluorescent chemistry, and 1 ng was used to obtain RNA integrity number (RIN) using the Bioanalyzer RNA 6000 Pico Kit (catalog no. 5067-1513, Agilent Technologies Inc., Santa Clara, USA). The lowest RIN was 9.1; median RIN score was 9.8. A total of 1000 ng per sample was then processed using the TruSeq Stranded mRNA Library Prep (catalog no. 20020595, Illumina Inc., San Diego, USA; protocol v. 1000000040498 v00) including poly(A) selection, with 8 min of fragmentation at 94°C and 15 cycles of amplification. Top stock (1 μl) of each purified final library was run on an Agilent Bioanalyzer Double-stranded DNA (dsDNA) High-Sensitivity chip (catalog no. 5067-4626, Agilent Technologies Inc., Santa Clara, USA). The libraries were quantified using the NEBNext Library Quant Kit for Illumina (catalog no. E7630L, New England Biolabs, Ipswich, USA) and were pooled at equimolar ratios after size adjustment. The final pool was run on an Agilent Bioanalyzer dsDNA High-Sensitivity chip and quantified using the NEBNext Library Quant Kit. The quantified pool was hybridized at a final concentration of 2.2 pM and sequenced on the Illumina NextSeq 500 platform using two High-Output v2 flowcells for an average of 85 million 1 × 85 bp reads.

Analysis of 3′UTRs

C. elegans

Sequencing read processing

Single-end sequencing reads of 144 bp were obtained. Reads were trimmed using Cutadapt as follows: (i) First, 12 bp of reads were trimmed, (ii) base pairs with a sequencing quality lower than 30 were trimmed from the ends of reads, and (iii) poly(A) stretches at the end of reads were trimmed using 10 A’s as an “adaptor” sequence. Only reads with at least 5 A’s trimmed at the ends are used for subsequent analysis. ERCC transcripts were only used to ensure library quality and were not included in the subsequent analyses.

PA site identification

The 3′ most nucleotide of each read was used to create a track of reads pile-up. R package “derfinder” (version 1.12.0) (61) was used to identify expressed regions separately for reads that mapped to the positive or the negative strand. Essentially, any consecutive positions with more than 1 RPM (reads per million mapped reads) were identified as an expressed region. Regions were extended 5 bp upstream. Regions that are within 5 bp from each other were then merged.

Poly(A) clusters were then mapped to C. elegans gene annotation (WBcel235, obtained from Ensembl 85). Meanwhile, sequence composition downstream of each ER was examined to identify expressed regions that potentially originated from internal poly(A) priming events. Specifically, two criteria are used: (i) within 30 bp downstream and 5 bp upstream from each expressed region end, if the pattern of 18 As (allowing five mismatches) can be found; (ii) within 10 bp downstream and 5 bp upstream from each expressed region end, if the frequency of As is larger or equals to 60%. In addition, 35 bp upstream from each expressed region end, we searched for the two most common poly(A) signals, AATAAA and ATTAAA.

PA site annotation

Expressed regions that map to exons are annotated to the associated genes. Expressed regions that do not map to exons are annotated with the following steps:

1) Expressed regions that are within 3 kb from the nearest transcript end site (TES) upstream are annotated to the corresponding gene.

2) Expressed regions that map to introns are annotated to the associated genes.

3) In the rare case that an expressed region maps both to an intron and downstream of a gene, if the expressed regions maps within 200 bp of a gene’s TES, then it is annotated to that gene; otherwise, it is annotated to the intron-containing gene.

Last, all annotations are merged, and the number of reads mapped to each expressed region is then calculated.

Identify alternative PA sites per gene

We systematically processed PA sites that mapped to each gene and used stringent criteria to analyze only confident PA sites. We consider a few different situations:

1) PA sites that are likely derived from internal poly(A)s are discarded even if the PA site is mapped to exons.

2) An exception to (i): If an internal poly(A)–derived site (determined in the “PA site annotation” section) is the longest potential PA site or the second longest PA site followed by a real PA site and also contains a significant proportion (more than 50%) of all reads mapped to this gene, then it is considered a “valid PA site.” In these two cases, although we cannot precisely pinpoint PA sites [because of internal poly(A) priming within the 3′UTR], we have high confidence that this poly(A) signal is derived from the longest 3′UTR variant (because no signal is detected downstream) and can be used to assay alternative PAs (APAs).

3) If more than 75% of the reads mapped to this gene map to internal poly(A)–derived sites, then this gene is not considered for APA analysis.

4) If potential PA sites are within 15 bp of each other, then merge them.

5) After merging, only keep the PA sites containing more than 5% of all the mapped reads of this gene and have more than 10 reads in at least three samples for further analysis. The purpose is to filter out extremely weak PA signals.

6) Last, genes with more than one PA site after the filtering are subjected to DEXSeq analysis.

We order and name the PA sites based on their relative positions along the gene body, and “E001” refers to the most proximal PA site. Proximal polyadenylation site usage (PPAU) is calculated as the number of reads mapping to the most proximal PA site/total number of reads mapping to all valid PA sites of the gene.

APA analysis

To characterize genes that undergo APA, we used the R package “DEXSeq” (version 1.24.2) (62) and examined every pair of comparisons. We applied a cutoff of adjusted P value of <0.05. To summarize the results, we only considered the comparisons between N2 (the WT) and cfim-1 mutants. For each gene in each comparison, we annotate on the basis of its APA condition. The top four primary situations are as follows: (i) The most proximal PA site is expressed higher in WT; (ii) at least one of the more distal PA sites is expressed higher in mutants; (iii) the most proximal PA site is expressed higher in WT, and at least one more distal PA site is expressed higher in mutants; and (iv) the most proximal PA site is expressed higher in mutants, and at least one more distal PA site is expressed higher in the WT. Other more complicated APA cases were not specifically classified and are represented as “others.”

Gene expression analysis

First, the number of reads mapped to each gene was counted by adding the reads in expressed regions that mapped to (i) any exons and (ii) being annotated to this gene and passed the internal poly(A) filtering criteria (described in the “PA site identification” section). Next, normalized read count was obtained using the fpm() function from R package “DESeq2” (version 1.24.0) (63). Data for PPAU and classification of APA events identified can be found in files S9 and S10.

Homo sapiens

Quantification of 3′UTR isoforms

To estimate 3′UTR isoform abundances, we used the alternative polyadenylation analysis tool QAPA version 1.2.0 (41). In short, QAPA constructs a refined set of 3′UTR isoforms from the GENCODE basic gene annotations using external databases of polyadenylation sites (64). Next, QAPA selects 3′UTRs from the most downstream exon of the gene to resolve ambiguities in quantification of 3′UTR isoforms with reads generated by RNA-seq. Using this refined 3′UTR annotation library, we quantified 3′UTR isoforms using raw RNA-seq reads with the alignment-free isoform quantification tool Sailfish version 0.10.0 (65). On the basis of mRNA quantification from biological replicates per each cell type, we calculated the average, SD, and coefficient of variation for each 3′UTR isoform.

APA analysis

Abundance of 3′UTR isoforms was quantified in transcripts per million (TPM column in Sailfish output). To classify genes into APA patterns, we considered log₂ fold change (FC) and Z scores between CFIm25 (NUDT21) KD and control conditions for both WT and mutant KRAS samples using the following equation

$$\begin{array}{c}Z=\sqrt{n}\mid {\text{TPM}}_{\text{mean}}^{\text{KD}}-{\text{TPM}}_{\text{mean}}^{\text{CTRL}}\mid /\sqrt{{\text{TPM}}_{\text{var}}^{\text{KD}}+{\text{TPM}}_{\text{var}}^{\text{CTRL}}}\\ {\text{log}}_2{\text{FC}}_{}={\text{log}}_2\frac{{\text{TPM}}_{\text{mean}}^{\text{KD}}}{{\text{TPM}}_{\text{mean}}^{\text{CTRL}}}\end{array}$$ (1)

where n is number of samples (3 in this experiment) and TPM_mean and TPM_var are the TPM average and SD based on replicates. Shift patterns were categorized into three groups based on Z and log₂FC_TPM: switches, proximals, and distals.

Switches are defined as genes with a reciprocal shift in any pair of the 3′UTR isoforms. One isoform changes by log₂FC_TPM > log₂(1.5) and Z > 2, and another isoform changes by log₂FC_TPM < −log₂(1.5) and Z > 2.

Proximals are defined as genes with a shift only in the shortest 3′UTR, corresponding to proximal pA site. Only the shortest 3′UTR changes by abs(log₂FC_TPM) > log₂(1.5) and Z > 2.

Distals are defined as genes with a shift in only one 3′UTR isoform, corresponding to any distal pA site. Only one distal 3′UTR changes by abs(log₂FC_TPM) > log₂(1.5) and Z > 2.

Last, we select genes with an increase in proximal 3′UTR among the genes captured in the way described above. Proximal 3′UTR changes by log₂FC_TPM > log₂(1.5) and Z > 2. A shift of log₂(1.5) means that abundance is higher in CFIm25 KD than in control. These genes were used for comparisons with C. elegans 3′UTR results. QAPA data, classification of APA events detected, PPAU, and gene expression data can be found in files S11 to S14.

Gene ontology analysis

C. elegans genes were classified into biological categories by selecting the gene ontology biological function on Genemania. For overlapping with human dataset, C. elegans genes were converted into their human orthologs in Ortholist 2.0 (43). The resulting orthologs were used for subsequent IPA analysis or overlaid with human dataset to identify commonly regulated genes. IPA was performed on both human and worm-converted genes to classify genes into signaling pathways. Ortholist analysis can be found in file S15.

Cell culture

HCT116 KRAS G13D, KRAS WT cells, and DLD-1 cells were maintained in McCoy’s 5A medium supplemented with 10% fetal bovine serum (FBS). Pancreatic BXPC3, CFPAC-I, and HPAF-II cell lines were grown in Dulbecco’s modified Eagle’s medium supplemented with 10% FBS. Cell lines were validated by STR profiling.

siRNA transfection

For siRNA KD experiments, 300,000 cells per well were plated on six-well plates and transfected with pooled siRNA at a final concentration of 25 nM using Lipofectamine RNAiMAX according to the manufacturer’s instructions. All siRNAs were obtained from Dharmacon.

Growth assay

For growth assays, cells were harvested 24 hours after transfection and counted, and 3000 cells per well were plated on 96-well plates. Growth was measured using Presto Blue (Invitrogen), and fluorescence was measured using a plate reader according to the manufacturer’s instructions. Readings were taken at 0, 24, 48, and 72 hours after plating for time course experiments.

Cell cycle analysis

Cell cycle analysis was performed using the Click-iT Plus EdU Alexa Fluor 647 Flow Cytometry Assay Kit (Thermo Fisher Scientific). To obtain the cell cycle profile, HCT116 cells with and without CFIm25 KD (72 hours) were pulse labeled with 10 μM EdU for 2 hours. The cells were then harvested and fixed by adding ice-cold ethanol dropwise to the cells. The cells were then permeabilized and washed, and Click-iT EdU was detected using the manufacturer’s protocol. Cells were lastly suspended in 0.5 ml of FxCycle propidium iodide/ribonuclease staining solution (Invitrogen). Samples were analyzed on a Fortessa flow cytometer.

Cell migration and invasion assays

Cells were harvested 24 hours after transfections and counted, and 80,000 cells were plated on top of a transwell 8 μM insert in low (0.1% FBS) media. Complete media (10% FBS) was placed in the well below the insert. Each condition was performed in triplicate. Cells were allowed to migrate for 20 hours before fixing and staining with 1% Crystal Violet in 20% methanol for 20 min at room temperature. For migration assays related to TGFB signaling, transfected cells were pretreated with A83-01 overnight at indicated concentrations before being plated for migration assays in low FBS media supplemented with the indicated concentrations of A83-01 (Selleckchem). Double-transfected cells were plated for migration in low FBS media with or without TGFB2 (10 ng/ml; Cell Signaling Technologies). The top layer of the transwell membrane was scraped to remove excess stain and washed with phosphate-buffered saline (PBS). Inserts were allowed to dry before imaging on a Nikon epifluorescence microscope using a 10× objective. At least five fields of view were acquired per insert, and the area of migrated cells was calculated using ImageJ and represented as the mean area of migration. For invasion assays, Matrigel (150 μg/cm²) was added to the top of each insert and allowed to solidify before plating cells. Fixation and imaging were performed as described above.

Drug assays

Cells were transfected on 6-well plates, and a total of 5000 cells per well were plated onto 96-well plates after 24 hours. Cells were allowed to adhere overnight and then treated with increasing doses of etoposide (Tocris Bioscience) and Camptothecin (Selleckchem) for a further 48 hours. Cell viability was measured using Alamar Blue assay. Median inhibitory concentration (IC₅₀) curves were generated, and calculations were performed using GraphPad Prism 6.0 software.

To calculate the confluency of cells after drug treatment, cells were transfected on six-well plates and treated with etoposide at increasing doses for 48 hours. Cells were then stained with 1% Crystal Violet in 20% methanol and air-dried. Five random fields of view were acquired for each condition, and the area covered by cells was quantified using ImageJ.

Immunofluorescence

Cells were transfected for 24 hours and then harvested, and 100,000 cells were plated on poly-l-lysine–coated glass coverslips for a further 48 hours. Cells were fixed in 4% paraformaldehyde, permeabilized with 0.1% Triton X/PBS for 5 min, and blocked for 1 hour with 10% goat serum/1% bovine serum albumin solution. Coverslips were then incubated overnight at 4°C with the primary antibody anti-Vimentin (1:1000; BioLegend). Coverslips were washed and stained with the secondary antibody goat anti-chicken Alexa Fluor 488 for 1 hour at room temperature. Cells were stained with 4′,6-diamidino-2-phenylindole (DAPI; 1:1000) for 5 min. Coverslips were washed and mounted using ProLong Gold Antifade Mountant. Cells were imaged on a Quorum spinning disk confocal microscope using the 40× objective.

For γ-H2AX staining, cells were fixed and stained with anti–γ-H2AX antibody (1:500; Sigma-Aldrich), followed by staining with a goat anti-mouse Alexa Fluor 568 and DAPI, and imaged on a Quorum spinning disk confocal microscope using the 40× objective. Quantification of mean γ-H2AX signal was performed using the Volocity analysis software. Total nuclear area was identified using DAPI staining, and mean red intensities were calculated. Regions of interests (ROIs) were randomly selected to quantify background signal, which was subtracted from the total intensity values.

For SMAD2/3 staining, cells were fixed and stained with anti-SMAD2/3 antibody (1:200; Cell Signaling Technologies), followed by staining with a goat anti-rabbit Alexa Fluor 488, phalloidin-568 (1:1000; Invitrogen), and DAPI. Cells were imaged on a Quorum spinning disk confocal microscope using the 40× objective. Quantification of mean SMAD2/3 in nuclei was performed using the Volocity analysis software. Total nuclear area was identified using DAPI staining, and mean green intensities were calculated. ROIs were randomly selected to quantify background signal, which was subtracted from the total intensity values.

For assessing ABCC1 expression, cells were fixed in a 1:1 ratio of methanol and acetone. Cells were then stained with anti-ABCC1 antibody (1:50; Monosan) for 1 hour at room temperature followed by staining with a donkey anti-rat Alexa Fluor 488 antibody. Cells were also costained with anti-Glut1 antibody (1:200; Abcam) followed by a goat anti-rabbit Alexa Fluor 568. Cells were imaged on a Quorum spinning disk confocal microscope using the 40× objective. For quantification, ROIs were randomly selected along cell membranes. The “plot profile” function on ImageJ was used to calculate the intensity of signal along ROIs.

Western blotting

Cells were lysed in 1% NP-40 lysis buffer containing 5 M NaCl, 10% NP-40, and 1 M tris (pH 8.0) supplemented with protease and phosphatase inhibitors. Lysates were cleared by centrifuging at maximum speed. Protein concentration was determined by using the BCA assay (Thermo Fisher Scientific). Protein (30 μg) was loaded and run on 10% Mini-PROTEAN TGX Precast Protein Gels (Bio-Rad). Separated proteins were transferred onto nitrocellulose membranes (Bio-Rad) and blocked with 5% bovine serum albumin for 1 hour, followed by incubation with primary antibodies overnight at 4°C. Following three washes with 0.1% Tween 20 in tris-buffered saline, membranes were incubated with horseradish peroxidase–conjugated secondary antibodies for 1 hour at room temperature. Blots were washed and developed using enhanced chemiluminescence substrate (Thermo Fisher Scientific). The following antibodies were used: anti-NUDT21 (1:1000; Proteintech), anti-tubulin (1:5000; Sigma-Aldrich), anti–E-cadherin (1:1000; Cell Signaling Technology), and anti–phospho-ERK (1:1000; Cell Signaling Technology).

Quantitative reverse transcription polymerase chain reaction

cDNA was reverse-transcribed using the SuperScript III Reverse Transcriptase Kit (Thermo Fisher Scientific), and qPCR was performed using the ssoAdvanced Universal SYBR Green Supermix (Bio-Rad). Primer sequences are included in Table 4.

Table 4. List of qRT-PCR primers.

	Forward (5′-3′)	Reverse (5′-3′)
18S	GATTAAGTCCCTGCCCTTTGTACA	GATCCGAGGGCCTCACTAAAC
CFIm25	TGCAGCTGGGAACAACTT	TGACGACCCAGTATCTCTGT
ABCC1	TGGACCTGTTTCGTGACATC	AGGGATTAGGGTCGTGGAT
ABCC2	CAGCTGGTATGACAGCATCA	GCTTTCTGCAGCTCTCTCTT
ABCC5	CGAAGGGTTGTGTGGATCTT	TTTCACCATGAAGGCTGGTC
Firefly Luciferase	GCAGAAGAAGCTGCCTATCA	GGCAGATGAGATGTCACGAA
Renilla Luciferase	AAAGCGAAGAGGGCGA GAA	TGCGGACAATCTGGAC GAC
TGFB1	GATGTCACCGGAGTTGTGCG	AGTGAACCCGTTGATGTCCACT
TGFB2	GCCTGAACAACGGATTGAGC	ATCGAAGGAGAGCCATTCGC

Luciferase assays

The long and short 3′UTRs cloned into the dual-luciferase vector miTarget vector were obtained from GeneCopoeia. HCT116 cells were plated on six-well plates and transfected with 50 ng of each plasmid. Luciferase expression was measured after 48 hours using the Dual Luciferase Assay System (Promega). Transfected cells were also collected for RNA extraction and qRT-PCR analysis as described above.

Site-directed mutagenesis

Site-directed mutagenesis of the long ABCC1 3′UTR construct was performed using the QuikChange II XL Site-Directed Mutagenesis Kit from Agilent according to the manufacturer’s protocol.

Statistical analysis

All statistical analysis was performed as indicated on GraphPad Prism.

Creation of schematic representations

All schematic representations were created with BioRender.com.

Supplementary Materials

This PDF file includes:

Figs. S1 to S8

Legends for Supplementary files S9 to S15

Other Supplementary Material for this manuscript includes the following:

Supplementary files S9 to S15

View/request a protocol for this paper from Bio-protocol.