P1 promoter-driven HNF4α isoforms are specifically repressed by β-catenin signaling in colorectal cancer cells

HNF4α is a nuclear receptor acting as a major regulator of gene expression in gut epithelial cells (Stegmann et al., 2006). Through its action on gene expression, HNF4α can regulate a number of key cellular functions including differentiation, proliferation, metabolism and inflammation (Babeu and Boudreau, 2014). Such HNF4α functions have been described in the mouse colon, where it was shown to act as a morphogen at an early age and to maintain integrity of the epithelial barrier during adulthood (Babeu et al., 2009; Garrison et al., 2006). In humans, dysregulation in HNF4α function has been associated with various diseases, including maturity-onset diabetes of the young (Yamagata et al., 1996) and inflammatory bowel diseases (Darsigny et al., 2009; UK IBD Genetics Consortium et al., 2009). HNF4α is hence recognized as a key transcription factor for colonic epithelial functions in both normal and pathological conditions.

Belonging to the nuclear receptor superfamily, HNF4α shares a protein structure commonly associated with this class of transcription factors. As is the case for other members of this superfamily, the structure of HNF4α can be modified through production of various gene transcripts by a combination of alternative splicing and alternative promoter usage (Babeu and Boudreau, 2014). Twelve isoforms, divided in P1 or P2 promoter-driven isoform classes, have been identified in humans. These two isoform classes mainly differ by their N-terminal domain, where P2 isoforms lack the AF-1 transactivation domain (Babeu and Boudreau, 2014). Because this specific region is involved in the binding of specific cofactors through which HNF4α modulates transcription, it is postulated that P1 and P2 isoforms could have different regulatory roles. Recent evidence lends further support to these differential roles, with P1 and P2 isoforms displaying different expression patterns in tissues (Tanaka et al., 2006), as well as the outcome of phenotypic alterations observed in mouse models genetically modified to express only one or the other classes of isoforms (Briançon and Weiss, 2006; Chellappa et al., 2016). Although some progress has been achieved to support these potential differing roles for P1 and P2 isoforms in mice, their specific roles in a human biological context remain a matter of speculation.

HNF4α has previously been associated with colorectal cancer and its expression is crucial for the survival of colorectal cancer cell lines in culture (Schwartz et al., 2009; Zhang et al., 2014) and promotes intestinal polyposis in mice (Darsigny et al., 2010). The HNF4A gene locus is amplified in colorectal tumors, while HNF4α overexpression is associated with specific subtypes of colorectal cancer (Zhang et al., 2014; Guinney et al., 2015). Although these findings initially pointed to an oncogenic function for HNF4α, additional investigations revealed a dichotomous role for HNF4α in colorectal cancer, where it can also act as a tumor suppressor gene. Indeed, although HNF4α global expression appears to be induced in colorectal tumors (Darsigny et al., 2010), P1 isoform expression was reported to be impaired at the protein level (Tanaka et al., 2006; Chellappa et al., 2012). P1 isoforms were also found to display tumor suppressor functions in a mouse colitis-associated cancer model (Chellappa et al., 2016), coincidently with the previously reported tumor suppressor role for P1 isoforms in mouse liver (Hatziapostolou et al., 2011). Although both share a common DNA binding domain, P1 and P2 isoforms can regulate distinct sets of genes in intestinal epithelial cells (Vuong et al., 2015; Chellappa et al., 2016).

Although they originate from a common gene, P1 and P2 promoter-driven HNF4α isoforms are distinctly regulated, given their different protein expression profile in colorectal cancer as well as along the mouse colonic crypt epithelium (Chellappa et al., 2016). P1 isoform protein stability can be specifically inhibited in colorectal cancer cells by SRC kinase, through phosphorylation of a tyrosine residue located in their specific N-terminal domain (Chellappa et al., 2012). Methylation of the P1 promoter in a human progenitor cell line negatively impacts P1 isoform expression (Ancey et al., 2017). Although these mechanisms might contribute to the specific inhibition of P1 isoforms in different cellular settings, additional investigation is needed to explain the high frequency of P1 inhibition in colorectal cancer (Tanaka et al., 2006), as well as their specific localization in the upper differentiated half of the colonic crypt (Chellappa et al., 2016).

The WNT/β-catenin pathway is an important signaling cascade that ultimately controls gene expression in normal intestinal epithelial and colorectal cancer cells. The activity of this pathway is tightly regulated and restricted to the lower half of colonic crypts, where it maintains stem cell activity and proliferation (Krausova and Korinek, 2014). As a result, mutations leading to the activation of the WNT/β-catenin pathway in colorectal tumors are frequent (∼90% of cases) and are considered as one of the early steps of tumorigenesis (Barker and Clevers, 2006). Upon activation, this pathway leads to the stabilization of the β-catenin (also known as CTNNB1) protein which, in turn, can translocate to the nucleus to modulate gene expression either by interacting with TCF/LEF transcription factors (Nusse, 2012), by regulating mRNA stability (Kim et al., 2012) or by interacting with transcriptional complexes (Nishiyama et al., 2012). Among the gene targets of the WNT/β-catenin pathway, several are functionally involved during colorectal cancer initiation and progression including the MYC-encoding transcription factor and CCND1 (Herbst et al., 2014).

The present study aimed to assess the consequences of specifically inhibiting P1 or P2 isoform expression on human colorectal cancer cell transcriptome. Results revealed that P1 promoter-driven isoform expression was reduced via β-catenin-dependent mechanisms. These findings identify a novel cascade by which the expression of P1 isoforms might be rapidly shut down in the early stages of colon tumorigenesis, thereby allowing a change in HNF4α-dependent gene signature in promoting colorectal cancer progression.

It has recently been observed that HNF4α P1 and P2 isoforms are localized in different epithelial compartments of the mouse colonic crypt (Chellappa et al., 2016). To investigate whether this pattern of expression was conserved in human colonic crypts, P1 and P2 isoform immunofluorescence was performed on serial sections of normal resection margins obtained from patients having undergone surgery for colon cancer. Immunofluorescence using an antibody recognizing all isoforms revealed a comparable distribution of HNF4α expression in the vast majority of epithelial cells of the colonic crypts (Fig. 1). However, isoform-specific antibodies revealed an opposite expression profile for P1 and P2 isoforms along the colonic crypts, in which P1 isoform detection was more pronounced in epithelial cells located in the upper differentiated half of the crypts, whereas P2 isoforms were more expressed in the lower half of the crypts (Fig. 1). A similar distribution for HNF4α isoforms was also observed in human fetal colonic crypts, thus indicating that this specific distribution is already established during early development and subsequently maintained until adulthood (Fig. S1).

Because HNF4α stimulates intestinal epithelial differentiation (Lussier et al., 2008), we next investigated the expression profiles of both P1 and P2 isoforms during differentiation of Caco-2/15 cells. This colorectal cancer cell line has been extensively used as a model of enterocyte differentiation in culture and has been found to mimic the transit-amplifying proliferative status when cultured at subconfluency, as well as to initiate spontaneous enterocyte differentiation upon reaching confluence (Beaulieu and Quaroni, 1991). Transcripts of the sucrase-isomaltase (SI) gene, a marker of intestinal epithelial differentiation, were induced 3 days after reaching confluence as determined by quantitative PCR (qPCR) (Fig. 2A). P1 isoform transcripts were increased upon reaching confluence (day 0, Fig. 2A), whereas P2 isoform transcripts remained stable during this transition. Similar results were obtained when the normal rat intestinal epithelial cell line IEC6/L1 was induced to differentiate in co-culture conditions (Lussier et al., 2008) (Fig. S2). These observations were also reflected at the protein level as determined by western blotting using an anti-HNF4α antibody recognizing all isoforms (Fig. 2B). Taken together, these results confirmed that HNF4α P1 isoforms are preferentially expressed in differentiated colonocytes, whereas P2 isoforms are enriched in proliferative cells.

Previous reports have observed that the protein expression of HNF4α isoforms was modified in colorectal tumors in which P1 isoforms were found to be inhibited comparatively to normal samples (Tanaka et al., 2006; Chellappa et al., 2012). However, these latter studies only investigated HNF4α isoforms at their protein level, leaving unknown whether their modulation in colorectal cancer is linked to a change at the transcript level. Nevertheless, we previously reported that total expression of HNF4α was modulated at both the protein and transcript level in matched tumor and resection margin samples from a local colorectal cancer patient cohort (Darsigny et al., 2010). We therefore took advantage of the availability of these samples (see Materials and Methods) to determine whether HNF4α isoforms were also modulated at the transcript level in colorectal cancer. A first series of experiments was undertaken to validate the previous findings at the protein level by screening 36 patient samples by immunofluorescence. There was a significant reduction in P1 isoform expression in 78% of tumor samples when compared with the paired resection margin (Fig. 3A,B). In contrast, expression of P2 isoforms was maintained or increased in 92% of these samples (Fig. 3A,B). This analysis was thereafter further extended at the gene transcript level. In a local cohort of 67 patients, 90% (>2-fold) displayed a reduction in P1 isoform transcripts in their tumors when compared with their paired resection margin (Fig. 3C). On the other hand, P2 isoform transcripts were found to be either stable (37%) or increased more than 1.5-fold (45%) in the tumors when compared with their paired resection margin. The profile of HNF4α isoform expression was subsequently assessed in eight different colorectal cancer cell lines. Western blot analysis clustered these cell lines into two categories based on their global HNF4α expression level. Caco-2/15, T84, Colo205 and LoVo cell lines displayed high HNF4α expression levels, whereas HT-29, SW480 and DLD-1 displayed low HNF4α expression (Fig. 4A), and HCT116 cells were negative for HNF4α expression (Fig. 4A). Using high-resolution gels, P2 isoforms were found to be the major HNF4α isoform detected among all tested colorectal cell lines (Fig. 4A). P1 isoform expression was mainly restricted to Caco-2/15 and T84 cell lines (Fig. 4A), both of which retained a potential for cell polarization in culture (Beaulieu and Quaroni, 1991; Dharmsathaphorn et al., 1984). This pattern of expression was reminiscent of the gene transcript profiles for both P1 and P2 isoforms (Fig. 4B). Collectively, these observations support that variations observed in HNF4α isoform protein levels in colorectal cancer are driven by specific changes in isoform transcript production.

Specific modulations of HNF4α isoform expression in colorectal tumors suggest that P1 and P2 classes could function differently in this context. In support of the latter, mice genetically modified to express only P1 or P2 isoforms have been shown to exhibit different susceptibilities in a model of colitis-associated colon cancer (Chellappa et al., 2016). In addition, α2- (P1 class) or α8- (P2 class) specific isoforms are able to regulate different gene networks when artificially overexpressed in the colorectal cancer cell line HCT116 (Vuong et al., 2015). It remains, however, unclear whether knockdown of each class of P1 or P2 isoform could differently impact the transcriptome in human colorectal cancer cells. Caco-2/15 cells, the only cell line to express equivalent amounts of P1 and P2 isoforms (Fig. 4A), were used to inhibit expression of populations of either P1 or P2 isoforms. To this end, an RNA interference (RNAi) strategy was developed to specifically target each class of isoform in which short hairpin RNA (shRNA) molecules were designed to target the unique 5′ transcript region of each P1 or P2 isoform population (Fig. S3A). A western blot confirmed the specificity of targeting of P1 or P2 isoforms in Caco-2/15 cells following lentiviral infections (Fig. 5A). However, the short nature of the specific 5′ transcript sequence of the P2 isoform prevented the optimization of more than one shRNA sufficiently efficient to reduce expression of P2 isoforms (data not shown). RNA sequencing (RNA-seq) transcriptomic analysis, after application of a cutoff of P≤0.01, identified 2495 transcripts modulated in shP1 populations and 3667 transcripts in shP2 populations. All of these transcripts were modulated by at least 1.5-fold with 80% displaying a 2-fold or higher change. A Venn diagram distribution of transcriptomic changes among the different shRNA cell populations showed only a 24% overlap of transcript modulation in shP1 or shP2 condition (Fig. 5B). Gene signatures established from shP1 or shP2 Caco-2/15 cell populations were first explored as to whether they were generally recognizable as HNF4α-dependent gene targets. A gene set enrichment analysis (GSEA) of the data with reported gene sets of HNF4α-dependent modulated genes revealed a strong enrichment of downregulated targets in shP1 or shP2 conditions (Fig. 5C). This indicated that these signatures included a significant number of targets previously identified as being dependent on HNF4α expression. Transcriptomic data were then compared to available transcriptomic signatures obtained from intestinal differentiated (George et al., 2008; Kosinski et al., 2007) or proliferative cells (Merlos-Suárez et al., 2011). When compared with both human ileum and colon transcriptomic signatures obtained from differentiated cells, strong enrichments of downregulated gene transcripts were associated with Caco2/15 cells depleted in P1 isoforms in contrast to P2-depleted cells (Fig. 5C). For example, Caco2/15 cells depleted in P1 isoforms showed significant reduction in AFP, MTTP, LCT, APOA4, MUC13, APOB, FABP1, APOA1, TFF1, TTR and APOC3 differentiation-associated gene transcripts, as compared with nontarget control Caco2/15 cells (Fig. 5D). This indicated that loss of P1, but not P2, isoforms resulted in a reduction of gene transcripts associated with intestinal differentiation. On the other hand, downregulated transcripts in both P1- or P2-depleted conditions were found enriched when compared with transcriptomic signatures associated with proliferative intestinal crypt cells (Fig. 5C). Overall, these associations suggest that P1 isoforms can regulate both proliferation and differentiation-associated genes in Caco2/15 cells, whereas P2 isoform genes are predominantly associated with proliferation. In support of the latter, stable depletion of either P1 or P2 isoforms led to reduced cellular growth of these cells in culture (Fig. S4).

To gain further insight into the manner by which HNF4α P1 or P2 isoforms can influence the biological functions of colorectal cancer cells via their specific gene targets, a bioinformatics approach was used to restrain off-target genes from identified shP1 and shP2 transcripts, taking advantage of the published overall genomic binding site of HNF4α in Caco-2 cells (Verzi et al., 2010). By combining the RNA-seq data herein with the HNF4α chromatin immunopreciptation with DNA sequencing (ChIP-seq) data in Caco-2 using the BETA software (Wang et al., 2013) available on the Cistrome platform (Liu et al., 2011), a list of predicted direct target genes was obtained based on their modulation by the specific shRNAs used in the present experiments and the proximity of a HNF4α-binding site on these genes (Table S1). Applying a stringent filter to retain only predicted targets genes having a BETA rank product of 0.02 or less, 527 potential direct target genes were identified for P1 isoforms and 629 for P2 isoforms. Approximately 46% of these predicted direct target genes were common between P1 and P2 isoforms (Fig. 5E). An Ingenuity Pathway Analysis (IPA) from these predicted target genes was then performed by comparing P1- and P2-depleted conditions. Focusing on enrichment of targets in canonical pathways, P1 and P2 isoforms were able to regulate common as well as specific pathways of their own (Fig. 5F). Among the common pathways influenced by both isoform classes, certain expected roles for HNF4α were confirmed, such as its implication in the farnesoid X receptor (FXR) metabolic pathway (Thomas et al., 2013) and acute phase response signaling (Wang and Burke, 2007), as well as new functions such as ephrin B signaling and sphingosine metabolism. P1 isoform depletion led to specific changes in cell metabolism-related processes including lipid metabolism [liver X receptor (LXR) activation], amino acid metabolism (arginine biosynthesis) and pyrimidine metabolism (uridine-5′-phosphate biosynthesis). In addition, the tricarboxylic acid cycle pathway was found specifically associated with gene transcripts modulated in P1-depleted cells (Fig. 5F). On the other hand, P2-depleted cells showed strong associations with signaling and molecular mechanisms associated with cancer (Fig. 5F).

Taken together, these results suggest that inhibition of P1 isoform expression in cancer might alter cell metabolism in order to sustain the needs of cancer cells. On the other hand, the maintenance of P2 isoform expression could globally support aberrant oncogenic signalization in order to promote cancer cell survival and progression.

The mechanisms responsible for the specific inhibition of P1 isoform gene transcript expression during colon cancer were further explored. Because P1 isoform expression was observed to be weak in lower-compartment colonic crypts as well as during colon cancer, it was hypothesized that the WNT/β-catenin pathway could potentially be functionally involved in repressing P1 isoform expression. Colo205 and DLD-1 cell lines were stably infected with an inducible shRNA system targeting β-catenin (shβcat) or not (shNT). Supplementation of doxycycline during 48 h led to a reduction in active β-catenin protein levels in both cell lines as compared with controls (Fig. 6A). Inhibition kinetics of β-catenin over 48 h was next performed at different time points to measure the impact of reducing β-catenin on the expression of HNF4α P1 and P2 isoform transcript levels. After 12 h of induction, β-catenin transcripts levels dropped by 80% to reach a maximum inhibition of 90% after 24 h of doxycycline treatment (Fig. 6B). In a concomitant manner, P1 isoform transcript levels began to increase at 24 h to reach a 4.0- and 3.3-fold induction at 48 h in Colo205 and DLD-1 cells, respectively (Fig. 6B). No substantial modulation in P2 isoform transcript levels was observed under these conditions (Fig. 6B). In a similar manner, P1 isoform protein expression was induced in Colo205 cells 48 h after the inhibition of β-catenin by shRNA (Fig. 6C). The APC^Min mouse model harboring a mutation in the APC gene was next used to further determine whether the specific inhibition of P1 isoforms transcripts by β-catenin could also occur in colonic epithelium in vivo. When colonic polyps were compared with normal adjacent tissue, global HNF4α transcripts, as well as specific P1 isoform transcripts, were reduced by 50% and 65%, respectively (Fig. 6D), whereas P2 isoform transcript expression did not significantly change under these conditions (Fig. 6D). By immunofluorescence, P1 isoforms were also observed to be downregulated at the protein level in epithelial cells of colonic polyps from APC^Min mice compared with epithelial cells of the normal margins (Fig. 6E). Again, P2 isoform expression was found to be stable between polyps and normal crypt cells.

To further explore whether downregulation of β-catenin could be associated with changes in chromatin marks at both HNF4A promoters, chromatin immunoprecipitation (ChIP) was performed in Colo205 and DLD-1 cells with or without induction of β-catenin inhibition. The P2 promoter was more enriched in H3K27Ac and H3K4me3 active chromatin marks than the P1 promoter in both cell lines, while both H3K27me3 and H3K36me3 repressive chromatin marks were only weakly enriched for both promoters (Fig. 7A). Inhibition of β-catenin expression did not change the pattern of these chromatin marks for both P1 and P2 promoters (Fig. 7A). Chromatin data for both H3K27Ac and H3K4me3 active chromatin marks were extracted from available public ChIP-seq analyses performed on colorectal cancer cell lines. HNF4A P2 promoter was enriched for both chromatin marks in most colorectal cancer cell lines (Fig. 7B). HNF4A P1 promoter was enriched with these marks in Caco-2, Colo205 and Lovo cells (Fig. 7B), a pattern consistent with P1 isoform gene transcript expression (Fig. 4B). To explore whether these modifications occur in intestinal epithelial cells positive or not for β-catenin activity, ChIP-seq public data previously obtained on isolated mouse Lgr5⁺ stem cells and villi enterocytes were analyzed at the Hnf4a gene locus. The H3K27Ac-positive chromatin mark was enriched at the P2 promoter in Lgr5⁺ stem cells when compared with differentiated enterocytes (Fig. 7C). Interestingly, the H3K4me3-positive chromatin mark was enriched at both P1 promoter and enhancer regions in differentiated enterocytes when compared with Lgr5⁺ stem cells (Fig. 7C). This observation prompted us to investigate whether changes in chromatin marks were occurring in colorectal cancer cells at the HNF4A P1 enhancer following inhibition of β-catenin expression. A weak but significant increase in H3K27Ac active chromatin mark was observed at the P1 enhancer when β-catenin was depleted, whereas both H3K4me1 and H3K4me3 chromatin marks remained unchanged under these conditions (Fig. 7C). Taken together, our results suggest that activation of the β-catenin pathway can lead to the specific inhibition of P1 isoforms via mechanism(s) that do not imply chromatin marks change at P1 promoter and enhancer regions.

HNF4α is a nuclear receptor influencing a vast network of genes regulating cell differentiation, proliferation, inflammation and metabolism. Its activity can be modulated by small molecules (Kiselyuk et al., 2012; Schwartz et al., 2009), and its influence on carcinogenesis (Schwartz et al., 2009; Chang et al., 2014) – as well as its restricted expression in the human body (Tanaka et al., 2006) – render HNF4α a potentially valuable therapeutic target. Although many studies have linked HNF4α with colorectal cancer, controversy regarding its role as a tumor suppressor gene or as an oncogene have dampened its potential application for cancer treatment. Recent evidence suggests, however, that such controversy in the role of HNF4α in colorectal cancer might be the result of the different functions between its two isoform classes. This difference between P1 and P2 isoforms function has been studied in mice genetically modified to express only HNF4α P1 or P2 isoforms. Mice expressing only HNF4α P1 isoforms showed a moderate impaired glucose tolerance, while mice expressing only P2 isoforms presented dyslipidemia (Briançon and Weiss, 2006). These mice furthermore demonstrated a different susceptibility to induced colitis and to a colitis-associated cancer model in which P1 isoform expression specifically appeared to be protective in this context (Chellappa et al., 2016). In addition, P1 and P2 isoforms displayed a potential to regulate different gene networks (Vuong et al., 2015). Based on these previous findings, it hence appeared that the nature of HNF4α isoform expression could be of functional importance for intestinal epithelium maintenance, leading us to further investigate this possibility in the setting of human colorectal cancer.

Our findings strongly support that HNF4α P1 and P2 isoform classes are differentially expressed in human normal colon crypts similarly to that previously reported in mice (Chellappa et al., 2016). This expression pattern appears to already be established during fetal human colonic development. Of particular interest, the WNT/β-catenin pathway is well described to be strictly active in the bottom of the crypts, a region found herein to be negative for P1 isoform expression. This exclusive expression profile is clearly in keeping with our findings supporting that β-catenin acts as a specific inhibitor of P1 isoforms without influencing P2 isoform expression. Based on the transcriptomic signatures obtained from Caco2/15 cells, specific inhibition of P1 isoform class is predictive of cell and energy metabolism alterations, thereby leaving P2 isoforms to support oncogenic signaling and proliferation. These observations are furthermore well correlated with P1 and P2 isoform expression profiles observed in colorectal cancer patient samples.

HNF4α P1 isoforms have been reported to be downregulated in colorectal cancer (Tanaka et al., 2006; Chellappa et al., 2012). The sole mechanism proposed to explain these observations was that HNF4α P1 isoforms become unstable owing to oncogenic SRC tyrosine kinase-dependent phosphorylation during colorectal cancer (Chellappa et al., 2012). To our knowledge, the current study is the first to monitor HNF4α P1 and P2 isoform gene transcript expression in paired colorectal cancer samples. The present observations identified that P1 inhibition occurs at the gene transcript level both in patient colorectal tumors as well as in all colorectal cancer cell lines screened herein. This downregulation of P1 isoform gene transcripts was highly frequent (90% of the 67 patients screened) and likely to be the result of an early event during polyposis based on the APC^Min mouse polyps data and in the few human adenomas included in our patient cohort (data not shown). Therefore, it seems unlikely that SRC kinase action represents the major mechanism through which P1 isoforms are downregulated in colorectal cancer, given the upstream nature of gene transcript modulation observed herein. Moreover, HNF4α has been reported to negatively autoregulate its own promoters (Schwartz et al., 2009), making the downregulation of its mRNA an improbable consequence of HNF4α P1 protein degradation by SRC phosphorylation. For all the above reasons, we thus propose a model in which P1 isoform expression is mainly downregulated in colorectal cancer through the specific action of the WNT/β-catenin pathway, which is still considered the first critical event during the development of colorectal cancer.

The exact mechanism through which β-catenin specifically inhibits HNF4α P1 isoform expression remains unclear. Although the canonical WNT/β-catenin pathway is mainly known to ultimately act as an activator of its target genes, there are also examples of genes repressed by β-catenin. The various mechanisms described for β-catenin repression of its target genes have included the recruitment of a co-repressor complex or compaction of chromatin through interaction with histone H1 (Nishiyama et al., 2012). However, ChIP at the P1 promoter and enhancer upon β-catenin inhibition did not identify dramatic changes for histone modification marks in colorectal cancer cells. Therefore, it would appear that the mechanism involved in the present setting is not through the restraining of P1 promoter accessibility by chromatin compaction. It has also been demonstrated that β-catenin can regulate target expression through modulation of mRNA stability (Kim et al., 2012). Again, P1 isoform mRNA stability, as determined by actinomycin D treatment, was not altered when β-catenin was inhibited in Colo205 cells (data not shown). Because the P1 pre-mRNA sequence is included in the P2 pre-mRNA sequence, we were unable to discriminate the putative transcriptional effect of β-catenin by qPCR. Both P1 and P2 promoters harbor predicted binding sites for TCF/LEF as determined by the MatInspector software (data not shown), indicating that the mechanism responsible for the specific inhibition of HNF4α P1 gene transcripts by β-catenin is invariably complex and thus requires further in-depth investigations.

An important question that will also need to be addressed is the reason why β-catenin solely inhibits HNF4α P1 isoforms in colorectal cancer. Previous studies in which HNF4α was inhibited in colorectal cancer cell lines revealed such inhibition as a necessity for cancer cell survival and polyp formation (Schwartz et al., 2009; Darsigny et al., 2010; Zhang et al., 2014). Given the new knowledge gained from the present results, whereby P2 isoforms are the ones expressed in most colorectal cancer cell lines, it could be suggested that β-catenin must necessarily inhibit the proper isoforms otherwise cancer progression would be compromised. It has been reported that HNF4α can interact with the WNT/β-catenin pathway by physical interaction with TCF4 (Gougelet et al., 2014), LEF1 (Colletti et al., 2009) and β-catenin (Gougelet et al., 2014), as well as by cooperation or competition for binding sites (Vuong et al., 2015). These interactions appear in many cases to inhibit WNT/β-catenin target genes (Vuong et al., 2015; Yang et al., 2013), which would be consistent with the proliferation increase observed in the intestinal epithelium of Hnf4a knockout mice (Babeu et al., 2009). It is therefore possible that in the colon, HNF4α P1 isoforms could interfere more strongly with WNT/β-catenin activated genes than would P2 isoforms. This would be consistent with the expression pattern of HNF4α isoforms, whereby P2 isoforms are located in cells in which WNT/β-catenin activity is the highest.

In conclusion, we demonstrate herein that the activation of the WNT/β-catenin pathway is responsible for the specific inhibition of HNF4α P1 isoforms in colorectal cancer cells. This inhibition occurs at the gene transcript level, although future studies are required to elucidate the exact mechanisms involved in this process. Given the additional finding that the P1 isoform gene signature is associated with an intestinal epithelial cell differentiation program, whereas the P2 isoform gene signature is rather associated with pro-oncogenic functions, our results further identify the WNT/β-catenin pathway as a novel molecular switch of functional importance in the regulation of HNF4α isoforms during colorectal cancer.

Colorectal cancer cell lines DLD-1, Colo205, HT-29, HCT116, SW480, T84 and Lovo were obtained from the American Type Culture Collection (ATCC) and cultured according to respective recommendations. They were authenticated and tested for contamination by 17 short tandem repeat (STR) profiling by the cell authentication service at ATCC. DLD-1 and Colo205 cell lines were cultured in RPMI medium, HT-29 and HCT116 in McCoy's medium, SW480 in 1× L1 medium, T84 in F12 medium/Dulbecco's modified Eagle medium (DMEM) and Lovo in F12K medium (Wisent). The colorectal cancer cell line Caco2/15 was kindly provided by Dr Jean-François Beaulieu (Université de Sherbrooke, Sherbrooke, Canada) and cultured in DMEM. All media were supplemented with 10% fetal bovine serum (Wisent), 10 mM HEPES and 1× GlutaMAX (Gibco). The IEC6/L1 cell line was co-cultured as previously described (Lussier et al., 2008). The Caco2/15 differentiation kinetics were performed by seeding cells at 20% confluence and processing the latter the next day for the undifferentiated condition or after reaching 90% confluence (determined as day 0) or beyond for the differentiated condition. For the induction of shRNA against β-catenin (shβcat) or its nontarget control (shNT), cells were cultured in the presence of 50 ng/ml doxycycline (Sigma-Aldrich) for 48 h. Mock-treated cells were cultured in the presence of an equivalent volume of 1× PBS.

shRNA targeting the HNF4α P1 isoform class (shP1) or the nontarget control (shNT) were all purchased from Sigma-Aldrich (clone ID NM_000457.3-1489s1c1 for shP1 and SHC016 for shNT). The shRNA targeting the HNF4α P2 isoform class (shP2) was generated by cloning the shRNA sequence 5′-CAGTGGAGAGTTCTTACGACACTCGAGTGTCGTAAGAACTCTCCACTGTTTTT-3′ between AgeI and EcoRI in the pLKO.1 vector [Addgene plasmid #8453, deposited by B. Weinburg (Stewart et al., 2003)] yielding an equivalent backbone to the shP1 vector. The inducible shRNA vector targeting β-catenin and its corresponding control were kindly provided by Dr Michael Schlabach (Novartis Institute for Biomedical Research, Cambridge, MA, USA) and described elsewhere (Scholer-Dahirel et al., 2011). Lentiviruses were produced in 293T cells according to the manufacturer's recommendations (ViraPower Lentiviral Expression System, Invitrogen). Infections were conducted in 35 mm dishes with 350 µl of viruses (5% of total virus production) in the presence of 4 µg/ml polybrene (Sigma-Aldrich), and stable cell lines were obtained after selection with 0.5–2.0 µg/ml puromycin. For transient infections of Caco2/15 cells with shP1, shP2 or control shNT, six-well plates at 60% confluence were infected for 48 h as previously described and subsequently incubated for another 24 h in fresh medium without lentivirus before being processed for RNA and protein extraction.

Samples of colon cancer tumors with their paired margins were obtained from 71 patients who had undergone surgery at the Hôtel-Dieu hospital of the Centre intégré universitaire de santé et de services sociaux de l'Estrie – Centre hospitalier universitaire de Sherbrooke (CIUSSS de l'Estrie – CHUS, Sherbrooke, Canada) and who did not receive neoadjuvant therapy. Tissues were collected based on obtaining informed written consent according to the protocol approved by the Institutional Human Subject Review Board of the Centre Hospitalier de l'Université de Sherbrooke. Tissues were snap frozen in liquid nitrogen within 15 min after resection and half of the samples were embedded in a paraffin block for histological assessment of the sample. RNA was extracted from the frozen tissues using a Qiagen RNeasy kit. Clinical and pathological characteristics of the samples used are described in Table S2. A human fetal colon sample was obtained from a 20-week-gestation fetus following normal elective pregnancy termination. The sample was collected in accordance with the approved protocol from the Institutional Human Subject Review Board. All clinical investigation was conducted according to the principles expressed in the Declaration of Helsinki.

Formalin-fixed paraffin-embedded (FFPE) sections of normal human colon tissues for the localization of HNF4α isoforms in crypts by immunofluorescence were obtained from the Biobanque des maladies digestives du Centre de recherche du CHUS (CIUSSS de l'Estrie – CHUS). Samples from three different individuals were analyzed. All patients gave informed consent for use of their tissue samples and the ethics review board at the CIUSSS de l'Estrie – CHUS approved the use of these samples for this study. Briefly, following resection for colon cancer, the colon was washed thoroughly and fixed within 40 min of resection in 10% buffered formalin and embedded in paraffin. Consecutive 4-μm-thick FFPE sections were used for immunofluorescence, and normal colon crypts were located by the hospital pathologist using a consecutive slide stained with Hematoxylin and Eosin.

Polyps from 4-month-old Apc^Min mice on a C57BL/6J background were isolated from the colon as previously described (Boudreau et al., 2007) and used to collect RNA or to prepare paraffin blocks for immunofluorescence analysis. Mice were treated following a protocol approved by the Institutional Animal Research Review Committee of the Université de Sherbrooke (approval ID number 102-14) in conformity with the Canadian Council on Animal Care.

Immunofluorescence on tissue samples was performed as previously described (Babeu et al., 2009) with the following modifications. After rehydration, antigen retrieval was achieved by boiling samples in Tris-EDTA pH 9.0 solution antibodies against HNF4α (1:200, sc-6556, Santa Cruz Biotechnology), P1 isoforms (1:2000 for tumor and paired margins, 1:4000 for normal colon samples, K9218, R&D Systems) and P2 isoforms (1:100 for normal colon samples and 1:150 for tumor and paired margins, H6939, R&D Systems). Primary antibodies were incubated overnight at 4°C and subsequently incubated for 45 min at room temperature with anti-mouse Alexa Fluor 488 (1:400, A11017, Invitrogen), anti-mouse Alexa Fluor 594 (1:500, 8890, Cell Signaling Technology), anti-goat Alexa Fluor 555 (1:400, A21432, Invitrogen) or anti-rabbit Rhodamine (1:200, AQ301R, Chemicon). Immunofluorescence images of tumor samples and their paired resection margins were taken with a Leica DM LB2 microscope and processed with Photoshop CS4 software. Immunofluorescence images of HNF4α in normal colon crypts were taken with a Zeiss LSM 800 confocal microscope and reconstructed using ZEN blue 2.3 software.

For the scoring of HNF4α isoform expression in colorectal cancer, paired tumor and margin samples were obtained from 36 patients and analyzed by immunofluorescence as described above. Immunofluorescence for P1 and P2 isoforms was performed on consecutive slides obtained by serial sectioning of paraffin blocks and achieved through co-immunofluorescence with anti-HNF4α antibody and 4′,6-diamidino-2-phenylindole (DAPI) staining to validate their expression. Three to five images were acquired per sample and expression score assessed on a scale of 0 to 2. A score of 1 was calibrated to correspond to the staining generally observed in margins, which consisted of clear staining in the nucleus of some cells with an intensity ranging from low to moderate. The score 0 corresponded to samples with a faint or a complete absence of staining, and the score 2 was attributed to samples having clear staining in some cell nuclei but with a high intensity, or to samples with low staining intensity but present in all epithelial or tumor cells.

Total protein was extracted from cells as previously described (Boudreau et al., 2007). Ten to 40 µg of proteins were boiled in Laemmli buffer (161-0737, Bio-Rad) and separated on 10% acrylamide gels. Proteins were transferred onto a polyvinylidene fluoride membrane and immunoblotting performed as described (Boudreau et al., 2007) using the following antibodies: anti-HNF4α (1:3000, sc-6556, Santa Cruz Biotechnology), anti-nonphospho β-catenin (1:1500, 8814S, Cell Signaling Technology) and anti-actin (1:10,000, MAB1501R, Millipore).

Total RNA was extracted using the RNeasy kit (Qiagen) and complementary DNA (cDNA) synthesis was performed with Avian Myeloblastosis Virus Reverse Transcriptase (AMV-RT) (101 091 180 01, Roche) according to the manufacturer's recommendations. qPCR was performed with 10 ng cDNA by the RNomics Platform at the Université de Sherbrooke (https://rnomics.med.usherbrooke.ca). Quantification was performed in triplicate using the qBASE algorithm (Hellemans et al., 2007) and the expression of the following reference genes: MRPL19, SDHA and RPL13 for the colorectal cancer cell lines, MRPL19 and SDHA for Caco2/15 cell differentiation kinetic, MRPL19 for human samples, and Hmbs, Pum1 and Sdha for mouse colonic polyps.

Total RNA was extracted from cells using the RNeasy kit (Qiagen) and analyzed by a Bioanalyzer (Agilent) to ensure mRNA integrity. RNA-seq was performed by McGill University and the Génome Québec Innovation Center. Briefly, cDNA libraries were prepared with the TrueSeq mRNA kit (Illumina) on 250 ng total RNA and sequenced using the Illumina Hiseq 2000/2500 sequencer (Illumina). Base calls were made using the Illumina CASAVA pipeline and readings aligned to the 1000 genome b37 reference using a combination of Tophat and Bowtie software (Trapnell et al., 2009). Transcript abundance was determined by the Cufflinks program (Roberts et al., 2011) and differential gene expression analysis performed with the DESeq and edgeR programs (Anders and Huber, 2010; Robinson and Oshlack, 2010), comparing experimental samples with the shNT control. For subsequent analysis, only transcripts with an adjusted DESeq P-value of 0.01 or less were considered.

GSEA was performed with the GSEA software v3.0 (Subramanian et al., 2005) using a dataset the RNA-seq genes modulated at a P-value ≤0.01 by shP1 or shP2 in Caco2/15 cells when compared with the nontarget control. Gene set signatures for human ileum villi, human colon differentiated cells and proliferative cells in intestine crypts were derived from published data (Kosinski et al., 2007; Merlos-Suárez et al., 2011; Tremblay et al., 2006; George et al., 2008). Gene sets for HNF4α targets were obtained from the Molecular Signature Database using the ‘LUCAS_HNF4A_TARGETS_DN’ and ‘LUCAS_HNF4A_TARGETS_UP’ lists established from HEK293 overexpressing HNF4α and the ‘SUMI_HNF4A_TARGETS’ list established from HepG2 (Subramanian et al., 2005). Analyses were performed using default settings, except that gene set permutation was used with a minimum of at least 1000 permutations. A heatmap distribution of the results was achieved using Morpheus software (Broad Institute).

Identification of potential direct target genes was achieved using BETA software v.1.0.0 (Wang et al., 2013) on the Cistrome platform (Liu et al., 2011). The genomic binding sites of HNF4α were obtained from previously published ChIP-seq data performed in the differentiated Caco-2 cell line (Verzi et al., 2010) and available through the Gene Expression Omnibus (GEO) GSM575229 dataset. The genes for which expression was modulated following HNF4α P1 or P2 isoform inhibition were obtained from our RNA-seq analysis of Caco2/15 cells as described above. For prediction of direct targets, only ChIP-seq peaks with a false discovery rate (FDR) <0.02 were considered as a binding site for HNF4α, and only genes modulated with an adjusted DESeq P-value ≤0.01 were considered as potential targets. The distance from gene TSSs within which binding sites for HNF4α were considered was set to 20 kb, and CTCF boundaries were used to filter peaks around genes. Only targets having a rank product score of 0.02 or less were considered as potential direct targets of HNF4α isoforms.

Analysis of canonical pathway enrichment was performed using IPA software (Qiagen) using only molecular relationships established from mammals and experimentally observed as database reference. Only pathways with a P-value <0.05 were considered significant and redundant pathways were manually filtered based on the amount of overlapping targets between them.

Colo205 and DLD-1 stable cell lines containing the inducible shRNA vector targeting β-catenin or the inducible nontarget shRNA control vector were amplified and seeded in two 150 mm dishes (Colo205) or five 100 mm dishes (DLD-1). Upon reaching 50% confluence, puromycine selection was removed and cells were incubated for 48 h in the presence of 50 ng/ml doxycycline (Bio Basic). Medium was removed and cells were fixed with 1% methanol-free formaldehyde (Thermo Fisher Scientific) in 1× PBS for 15 min at room temperature. Cells were washed twice for 10 min in PBS-glycine 125 mM, scraped in cold 1× PBS and collected by centrifugation prior to being snap frozen in liquid nitrogen. Validation of HNF4α P1 isoform induction following β-catenin inhibition was confirmed for each sample by qPCR before performing ChIP assays. Each sample was lysed and sonicated to solubilize and shear crosslinked DNA on an average length between 100 and 500 bp. Then, 25 µg chromatin per condition was pre-cleared with 40 µl Protein G magnetic beads (16-662, Millipore) for 2 h at 4°C and incubated overnight with 40 µl Protein G magnetic beads and either 4 µg anti-histone H3K4me3 ChIP grade (AB8580, Abcam), 4 µg anti-histone H3K36me3 ChIP grade (AB9050, Abcam), 4 µg anti-histone H3K27Ac ChIP grade (AB4729, Abcam), 4 µg anti-trimethyl-histone H3 (Lys27) (07-449, Millipore) or 4 µg rabbit IgG ChIP grade (AB37415, Abcam) as negative controls. Protein/DNA complexes were eluted, reverse crosslinked and treated with proteinase K and RNaseA. Purified DNA was used as a template for qPCR with a LightCycler 96 System (Roche Applied Science). The primer sequences used to detect histone modifications were 5′-TTGAAAGGAAGGCAGAGAGG-3′ and 5′-GACGAGGGTTTTGGAGAGTC-3′ for HNF4A P1; 5′-CCTCCCCGTGTGTTTCTTAC-3′ and 5′-TTAGGGAAGCGGTCACATTG-3′ for HNF4A P2; 5′-CATCATCGGGTGAACAAACA-3′ and 5′-TGGCCACAGACTGAACCATA-3′ for the P1 enhancer region. Primer sequences used for the negative region control were 5′-GAGGTCAGGGTGCTGTGATT-3′ and 5′-AGCTGCATTCCTCTGGAGAC-3′. These sequences target the CPNE4 gene not expressed in the gut and are undetectable in Caco2/15 cells based on RNA-seq data. Calculation of enrichment was performed by normalizing ChIP DNA to input DNA and then subtracting the percentage of input for each ChIP by the one of IgG controls.

Public ChIP-seq data of histone modifications for colorectal cancer cell lines or mouse intestinal cells were obtained from the Cistrome Data Browser (Liu et al., 2011) available at http://cistrome.org/db. ChIP-seq data were investigated with the Human Genome Browser (Kent et al., 2002) using GEO samples GSM2058025 for Colo205 (Cohen et al., 2017), GSM2037784 and GSM2283764 for DLD-1 (Abraham et al., 2017; Rokavec et al., 2017), GSM2042874 for HT-29 (Savic et al., 2016), GSM2058026 and GSM945304 for HCT-116 (Cohen et al., 2017; Thurman et al., 2012), GSM1890746 for SW480 (McCleland et al., 2016), GSM2224582 and GSM1208811 for Lovo (Liu et al., 2017; Yan et al., 2013), and GSM945162 for Caco-2 (Thurman et al., 2012). GEO samples GSM2386645, GSM2386633, GSM2386657 and GSM2386642 were used for mouse enterocyte and intestinal stem cell investigations (Kazakevych et al., 2017). The P1 enhancer region was determined based on previous studies (Zhong et al., 1994; Bailly et al., 2001).

All statistical analyses were performed with PASW 18.0 or GraphPad Prism 7. All qPCR histograms are represented as mean±s.d., and statistical analyses were performed using a paired Student's t-test for mouse polyps analysis or a one-way ANOVA test for time-course analysis. A one-sample Student's t-test was used in the analysis of HNF4α isoform expression ratio in patient samples or mouse colon polyps. The difference in HNF4α isoform expression in tumors obtained from immunofluorescence scores was determined using a Wilcoxon signed-rank test.

We thank Dr Michael Schlabach (Novartis Institute for Biomedical Research) for providing the inducible shRNA vector targeting β-catenin and its corresponding control, Pierre Pothier for editing the manuscript, the McGill University and Génome Québec Innovation Centre for RNA-seq services, the RNomics Platform at the Université de Sherbrooke for qPCR services, and the Electron Microscopy and Histology Research Core of the Faculté de médecine et des sciences de la santé (FPSS) at the Université de Sherbrooke for histology and phenotyping services. We acknowledge support provided by the Biobanque des maladies digestives du Centre de recherche du CHUS, CIUSSS de l'Estrie – CHUS, certified by the Canadian Tissue Repository Network and affiliated with the Réseau de recherche sur le cancer. F.B., S.G. and J.C.C. are members of the Fonds de recherche Santé Québec-funded Centre de Recherche du CHUS.