CYP26A1 Links WNT and Retinoic Acid Signaling: A Target to Differentiate ALDH+ Stem Cells in APC-Mutant CRC

1. Introduction

Our goal has been to understand how mutation in the APC gene causes overpopulation of aldehyde dehydrogenase-positive (ALDH+) cancer SCs (CSCs) and decreased differentiation of SCs into neuroendocrine cells (NECs). The APC tumor suppressor gene is a negative regulator of WNT-signaling, which controls SC population size [1,2]. ALDH is a key enzyme in retinoic acid (RA) signaling, which is a pathway that promotes SC differentiation [3,4,5,6,7]. Thus, we chose to investigate whether there is a functional connection between WNT and RA signaling, and if APC mutation generates an imbalance in a WNT:RA-linked mechanism that contributes to ALDH+ SC overpopulation by impeding retinoid-induced differentiation [8].

We previously reported that inducing wild type (wt)-APC in HT29 CRC cells, which contain homozygous mutant APC, leads to decreased cell proliferation and increased apoptosis [9,10,11]. In mice, inactivation of APC leads to intestinal adenoma formation which provided evidence for crypt SCs as the cells-of-origin of intestinal cancer [12,13]. Restoring Apc in murine tumors led to increased enterocyte differentiation, tumor regression, and re-established crypt-villus homeostasis in CRC [13].

Previously we found that RA receptors are selectively expressed in ALDH+ SCs, which indicates RA signaling mainly occurs in ALDH+ SCs [14]. Since ALDH is a key enzyme in the RA pathway and RA signaling occurs via ALDH+ SCs, it provides a mechanism to selectively target CSCs based on a therapeutic strategy involving retinoid-induced differentiation [15]. Indeed, retinoids have been previously studied as chemo-preventive drugs and systemic therapies in treatment of CRC and other cancers [16,17]. Furthermore, we found that RA ligands such as ATRA induce differentiation of ALDH+ SCs into NECs [8,18]. On the other hand, APC mutation leads to incomplete differentiation in CRC cells [8,19]. Altogether, these findings suggest that APC mutation attenuates RA signaling in ALDH+ SCs, which contributes to their overpopulation that drives CRC development [8].

To identify mechanisms that functionally link WNT and RA signaling, we investigated how inducing wt-APC expression to decrease WNT signaling affects ATRA’s ability to induce NEC differentiation and reduce ALDH+ CSCs in APC mutant CRC.

2. Materials and Methods

2.1. Cell Culture and Inducing wt-APC Expression

HT29 colon carcinoma cells containing a ZnCl2-inducible wt-APC or β-galactosidase (control) expression vector were provided by Dr. Bert Vogelstein’s lab (Johns Hopkins University) [9] We refer to the modified HT29 cells as HT29-APC or HT29-GAL cells. HT29 cells were chosen for most of the study because they contain a homozygous APC mutation and do not have any known retinoic acid receptor gene mutations. Modified cells were cultivated in McCoy’s 5A modified media (Gibco/Thermofisher Scientific) supplemented with 10% fetal bovine serum (FBS; Atlanta Biologicals/R&D Systems), 1% penicillin:streptomycin (pen-strep; VWR), and 600 μg/mL hygromycin B (Thermofisher Scientific). All variations of HT29 cells grew equally well in terms of cell density. Respective expression plasmids were induced with 120 μM ZnCl2 (Sigma-Aldrich) for the times indicated. HT29-APC cells treated with ZnCl₂ are referred to wt-APC and those not treated with ZnCl₂ are referred to mt-APC. All cells were maintained in a 37°C incubator with 5% CO2.

2.2. Cell Proliferation

HT29 cells were plated at 12,000 cells per well into 96-well plates. Cells were allowed to adhere for 24 hrs and serum-starved for an additional 24 hrs. Cells were then treated for 48 hrs with ZnCl₂ and/or ATRA (Stem Cell Technologies) as indicated. Controls involved no ZnCl2 or DMSO (0.1%; Fisher Scientific) vehicle. After indicated time, cells were washed in PBS and stained in 0.5% crystal violet (Sigma-Aldrich) solution for 20 minutes. Excess crystal violet was then washed off with deionized water and left to dry at room temperature for at least 24 hrs. Dried crystal violet stain was then re-suspended in 10% acetic acid solution and optical density measured at 570 nm using a Tecan Multi-Plate Reader, data was quantified using Tecan i-Control software, and analyzed using Microsoft Excel. Values reflect cell proliferation as our standard curve shows that absorbance values from crystal violet staining are directly proportional to cell number. The proliferation index is calculated by normalizing cell proliferation index of treated to untreated cells. Biological replicates were performed in triplicate with 6 technical replicates per condition, and the average plotted with standard error of the mean (SEM) indicated.

2.3. WNT/β-Catenin Activity

WNT/β-catenin activity was measured using the TCF/LEF Reporter (BPS Biosciences) and Two-Step or Dual Glo (Firefly and Renilla) Luciferase Assays (BPS Biosciences or Promega) according to manufacturer instructions. In brief, cells were plated onto clear bottom, white 96-well plates and adhered for 24 hrs. Next, cells were transfected with the TCF/LEF reporter plasmids using Lipofectamine 3000 (Thermofisher Scientific). Cells were then treated as indicated for 24 hrs. The Firefly and Renilla luciferase assays were performed, and luminescence recorded using a Tecan Multi-Plate Reader, i-Control software, and data analyzed using Microsoft Excel. Luciferase activity was measured by calculating the ratio of Firefly to Renilla luminescence signals. Biological replicates were performed in triplicate and with 6 technical replicates per condition. The average luciferase activity was plotted with SEM indicated.

2.4. NanoString Profiling

To measure the effects of manipulating WNT and RA signaling, we utilized the nCounter PanCancer Pathways Panel from NanoString to provide high-throughput data including differential gene expression and subsequent pathway analysis. The PanCancer Pathways Panel includes 770 genes involved in several cancer pathways including WNT, hedgehog, apoptosis, cell cycle, RAS, PI3K, STAT, MAPK, Notch, TGF-β, chromatin modification, transcriptional regulation, and DNA damage control [20] and we included an additional 55 genes of interest to evaluate SC marker, differentiation marker, ALDH isoforms, RA receptors, and HOX genes.

HT29-APC cells were plated at 700,000 cells per well onto 6-well plates and allowed to adhere for 24 hrs. Cells were then serum-starved for 24 hrs. Cells were then treated with 15 µM ATRA or 0.1% DMSO and 120 µM ZnCl₂ or no ZnCl₂ for 24 hrs. Total RNA was isolated from cell lysates using the Trizol method. Samples were frozen at -80°C and sent to Wistar Institute’s Genomic Core Facility for analyzing quality of RNA and subsequent NanoString profiling using the Pan-cancer Panel. Results were obtained in RCC file format and data analyzed using NanoString nSolver software. A heat map was then generated to show differential expression of 785 mRNAs amongst the four treatment conditions (+ZnCl₂ & +ATRA, + ZnCl₂ & -ATRA, - ZnCl₂ & +ATRA, ZnCl₂ & -ATRA). Ratios were determined using nSolver, significance assessed, and plotted in Microsoft Excel. Gene lists of interest were analyzed using the STRING database for protein-protein interaction networks and functional enrichment [21].

2.5. Western Blotting & Densitometry

Standard western blotting procedures were performed. Cells were plated, treated as indicated with 120 µM ZnCl2 (Millipore Sigma), ATRA, or DMSO and lysed with RIPA (Millipore Sigma) containing Halt Protease and Phosphatase Inhibitor (Thermofisher) and cysteine protease inhibitor E-64 (Millipore Sigma). Cells lysates were scraped from dishes and centrifuged at 12,000 rpm for 15 minutes and stored at -20°C. Protein concentrations were determined using the Pierce BCA Protein Assay Kit (Thermofisher) and 100 µg total protein was prepared in Laemmli buffer (Bio-Rad) and 2.5% or 5% β-mercaptoethanol (Bio-Rad). Samples were loaded onto 4-20% mini-precast gradient gels (Bio-Rad) for SDS-PAGE at 200V for 45 minutes. Proteins were then transferred to PVDF membrane (Thermofisher) at 90V for 2 hrs. Blots were then blocked in 5% BLOTTO (nonfat dried milk in tris buffered saline (TBS) with 0.1% Tween 20 [TBST]) for 1 hr. Next, blots were incubated in primary antibody prepared in BLOTTO overnight on a rocking platform at 4°C. Blots were then washed in TBST 3x for 10 minutes. Secondary antibody was prepared in BLOTTO and blots incubated for 1-2 hrs at room temperature. Blots were then washed in TBST 3x for 10 min and incubated in Super Signal West Dura Extended Duration Substrate (Thermofisher) for 5 minutes to detect horseradish peroxidase conjugation. Blots were then imaged using the LiCor Odyssey Fc developer and Image Studio software. Densitometry was done using Image Studio and graphs plotted using Microsoft Excel. Experiments were performed 3 times and standard error of the mean (SEM) was calculated and plotted as shown.

2.6. Flow Cytometry and Fluorescence Activated Cell Sorting

NEC and ALDH+ (ALDEFLUOR+) cells were quantified and/or isolated using GLP2R antibody conjugated to Alexafluor 594 (Novus Biologicals) with IgG control (R&D) or ALDEFLUOR Assay (Stem Cell Technologies) using manufacturer’s instructions. Flow cytometry was performed on a BD FACS Aria II and FACS performed on a BD LSR Fortessa. Data was processed using FACS Diva software and graphs plotted in Microsoft Excel with error bars indicating SEM. Experiments were done 3 times.

2.7. Determining synergistic, Additive, or Antagonistic Anti-Proliferative Effects of CYP26A1 and WNT Signaling Inhibitors

We investigated the anti-proliferative effects of the CYP26A1 inhibitors Liarozole and Talarozole in combination with agents (Sulindac or Piroxicam) that have anti-tumor activity against APC-mutant tissues [reviewed in 16]. HT29 and HCT116 cells were cultivated in McCoy’s 5A media supplemented with 10% FBS and 1% pen-strep. SW480 cells were cultivated in Leibovitz L15 media supplemented with 10% FBS and 1% pen-strep. Cells were plated, allowed to adhere for 24 hrs, and serum-starved for 24 hrs. The cells were then treated for 48 hrs with the various agents as single drugs or in combination. For CYP26A1 inhibitors, IC50 doses of Liarozole and Talarozole are as follows: HT29 cells- 80 and 30 µM, HCT116 cells- 110 and 22 µM, and SW480 cells- 80 and 25 µM, respectively. For WNT inhibitors, IC50 doses of Sulindac and Piroxicam are as follows: HT29 cells- 680 µM and 550 µM, HCT116 cells- 450 µM and 600 µM, SW480 cells- 220 µM and 270 µM, respectively. In dose response experiments, Sulindac and Piroxicam doses were varied based on the IC50 dose (75% of IC50, IC50, 125% of IC50). For clarity, all concentrations are listed in Table S2. The proliferation index was determined using crystal violet assay (see above). Synergistic, additive, or antagonistic effects were calculated based on the Bliss independence dose-response model [22]. For example, these effects were calculated as the ratio of observed (observed values for the combination of drug a & b) divided by predicted (values for individual drug a multiplied by individual values for drug b), based on the Bliss Independence dose-response model. Further analyses were performed as described in main and supplementary figure legends.

2.8. Patient Survival Studies

RNA-seq, genotype, and clinical data on CRC patients were obtained from Genomic Data Commons (GDC) data portal. For individual patients, records of their APC genotypes, survival data, as well as CYP26A1 expression in both human CRC and normal human colon samples were paired together. These patients were then grouped based on whether they carry APC mutation and the change (either increase or decrease) of CYP26A1 expression between CRC and normal colon tissue samples.

2.9. Statistical Analyses

Statistics were calculated either in Excel using the t-test function or GraphPad Prism using one-way ANOVA, two-way ANOVA, or multiple unpaired t-tests as indicated in the figure legends. For patient survival studies, Kaplan-Meier survival curves of different groups were compared using log-rank tests.

3. Results

3.1. Inducing wt-APC Decreases WNT Signaling and Reduces Expression of WNT Target Genes

We first wanted to show that the HT29-APC cell line established by Morin et al [9] was functioning correctly in the ability of ZnCl₂-treatment to induce wt-APC expression. Using two independent experimental approaches, we confirmed that expression of wt-APC occurs in response to ZnCl₂ treatment (Figure S1). We initially measured WNT activity utilizing the TopFlash luciferase reporter assay which measures TCF/LEF transcriptional activity. There was a 90% reduction in WNT activity in ZnCl₂-treated cells compared to untreated cells (Figure S1A). We then confirmed that reduced expression occurs in several proteins encoded by WNT/ β-catenin target genes [23] including MET [24], c-JUN [25], CD44 [26], and c-MYC [27] (Figure S1B).

3.2. ATRA Promotes WNT/β-Catenin Activity, wt-APC Attenuates ATRA’s Effect

Although ATRA reduces the number of ALDH+ SCs, reduces sphere-forming ability, and promotes differentiation of SCs in CRC [14], here we found that total cell proliferation in cells is reduced less than 20% at 48 hrs when treated with 20 µM ATRA (Figure 1A). However, when wt-APC expression was induced in HT29 cells, ATRA led to further reduction of total cell proliferation by 42% compared to untreated controls (Figure 1A). Dose response analysis also shows that cells with wt-APC are 2.4 times more responsive to the anti-proliferative effects of ATRA than cells with mt-APC (based on line slopes; Figure 1B). Additionally, we observed that WNT/β-catenin activity is increased by ATRA in HT29 cells by up to 10 times as compared to control (Figure 1C); but, this increase in WNT activity is attenuated with the expression of wt-APC (Figure 1C).

3.3. Inducing wt-APC Decreases ALDH+ Stem Cells and Increases NEC Differentiation

Here, we show an increase in the percentage of GLP2R+ NECs when wt-APC expression is induced (Figure 2A). The relative number of GLP2R+ cells quadruples when wt-APC is induced and increases 6-fold with combined wt-APC expression and ATRA as compared to mt-APC and no ATRA (Figure 2B). In parallel, the percentage of ALDH+ (ALDEFLUOR+) cells decreases by approximately 20% with ATRA alone and almost 60% with the combination of ATRA treatment and wt-APC induction (Figure 2C). In a similar manner, mRNA expression of ALDH1A1 decreased by 5-fold with the ATRA and wt-APC combination as compared to control (Figure 2D). These results further support the existence of an interplay between the WNT and RA signaling pathways.

We also investigated whether inducing wt-APC in HT29-APC cells changes the expression of NEC differentiation cell markers. Indeed, several NEC markers increase in expression with the expression of wt-APC in HT29 cells. Increased protein expression of CHGA, GLP2R, NSE (ENO2), and SSTR1 occurred as shown in (Figure 3). Overall, we found that inducing wt-APC in cells decreases ALDH+ CSCs and increases NEC differentiation.

3.4. Expression Profiling Identified CYP26A1 as a Link between WNT and RA Signaling Pathways

NanoString profiling revealed differential expression of 785 mRNAs across four treatment groups from results on ZnCl₂ to induce wt-APC expression and ATRA to increase RA signaling (+ZnCl₂ & +ATRA, +ZnCl₂ & -ATRA, -ZnCl₂ & +ATRA, -ZnCl₂ & -ATRA). Hierarchical, agglomerative clustering was achieved and shown as a heat map that was generated using nSolver software (Figure 4A).

To examine the effect of inducing wt-APC expression and increasing RA signaling at the same time, we calculated the ratio of treated cells (+ZnCl₂ & +ATRA) to corresponding control group (-ZnCl₂ & -ATRA) using the nSolver software. A set of 248 genes was identified as being differentially expressed (Figure S2). To identify predicted signaling pathway interactions, we analyzed the set of 248 genes using the STRING database [21]. The large number of predicted interactions found in this network analysis (Figure S2) suggests that a substantial level of crosstalk occurs when the WNT and RA signaling pathways are modulated in HT29 cells.

By analyzing the list of 248 genes using the DAVID functional tool [29] and ranking the list of over 1500 hits, we identified those genes that are involved in RA signaling using the GOTERM_BP_DIRECT list. We then re-analyzed this list of 14 genes using the STRING database and found several predicted interactions. Of the 14 genes listed as involved in RA signaling, 12 were identified to have protein interactions: CYP26A1, RARA, HDAC2, KLF4, TNF, LEP, PDGFA, MYB, DKK1, WNT7B, WNT11, and FZD7 (Figure 4B). Notably, five WNT target genes were identified: CYP26A1 [30], WNT7B, WNT11, FZD7, and DKK1 (listed on NanoString’s Gene to Pathway Summary). However, CDKN2D and HOXA2 did not show any predicted connections with the other proteins identified in our current STRING network.

The highest change in expression occurred in CYP26A1 with a 15.52-fold increase when both wt-APC expression is induced and cells are treated with ATRA (Figure 4C). Moreover, HT29 cells treated with ATRA alone experience a 30.8-fold increase in CYP26A1 expression and inducing wt-APC attenuates this increase by 50% (Figure 5A). Western blot experiments validated the 50% decrease in CYP26A1 expression when cells are induced to express wt-APC and treated with 20 µM ATRA (Figure 5B,C).

3.5. CYP26A1 Inhibitor Agents Sensitize CRC Cells to the Anti-Proliferative Effect of Drugs That Downregulate WNT Signaling.

The anti-proliferative effects of CYP26A1 inhibitors, Liarozole and Talarozole, were evaluated alone and in combination with drugs that downregulate WNT signaling, Piroxicam and Sulindac (Figure 6 and Figure S3). Three CRC cell lines with mutations that activate WNT signaling were used, two with mutant APC (HT29 & SW480) and one with mutant β-catenin/CTNNB1 (HCT116). Each of the three cell lines have different RA receptor genotypes. HT29 cells have wild-type retinoid receptors, SW480 cells have mutations in RARA and RXRG, and HCT116 cells have mutant RARA [16]. Sulindac and Piroxicam have anti-tumor activity against APC mutant tissues, and Sulindac has been shown to target β-catenin and downregulate WNT signaling [reviewed in 16].

CRC cells were treated with the various agents as single drugs or in combination (Figure 6 and Figure S3). The Sulindac and Piroxicam doses were varied based on the IC50 (half-maximal inhibitory concentration) dose for treated cell lines. The data reveal that anti-tumor agents Sulindac and Piroxicam decrease the proliferation of CRC cells. The data also shows that the CYP26A1 inhibitors Liarozole and Talarozole alone reduce CRC proliferation. Moreover, when Sulindac or Piroxicam was combined with Liarozole or Talarozole, the drug combination produced additive or synergistic effects (Figure 6C,F). Further dose response analysis on the anti-proliferative effects of varying Piroxicam and Sulindac concentrations showed that Liarozole & Piroxicam, Liarozole & Sulindac, and Talarozole & Sulindac combinations produced the highest dose responsiveness in HT29, SW480, and HCT116 cells, respectively (Figure S4A,B).

3.6. Analysis of Human CRC Cases Indicates CYP26A1 Predicts Survival of Patients with Wild-type APC Tumors

To evaluate the effects of CYP26A1 on CRC patient survival, we conducted multi-omics bioinformatics analysis using RNA-seq, mutation, and clinical data obtained from Genomic Data Commons (GDC). Patients were grouped based on their APC genotypes and changes (increase or decrease) of CYP26A1 expression level between CRC and normal human colon samples. We found that the majority of CRC patients express increased levels of CYP26A1 in their CRC samples compared to normal colon tissue from the same patient, and this difference is statistically significant (Figure 7A). By itself, this change in CYP26A1 expression level is not a statistically significant predictor of patient survival (Figure 7B). However, when genotype of APC was used to further sub-group the patients, we found that CYP26A1 expression is a significant predictor of patient survival only when patient tumors are wild-type APC (Figure 7C). But estimated survival is not statistically significant between increased and decreased CYP26A1 expression groups among those patient tumors carrying APC mutations (Figure 7D). Similarly, when patients are sub-grouped based on CYP26A1 level, APC mutation is not a significant predictor of patient survival (Supplementary Figure 7A,B).

4. Discussion

We previously established that overpopulation of ALDH+ CSCs correlates with zygosity state of APC mutation during the stepwise tumor development in FAP patients [1]. We also found that APC mutation causes failure of ALDH+ SCs to mature into NECs [8,18]. Together, these findings provide a clue to a functional interaction between the WNT and RA signaling pathways that maintain stemness and promote differentiation, respectively [8,14,18]. Accordingly, we conjectured that a link exists between WNT and RA pathways, and, when APC is mutant, an imbalance in a WNT:RA-linked mechanism promotes CRC development [8]. Our main finding herein identified CYP26A1 as a link between WNT and RA signaling that can be targeted to decrease ALDH+ SCs and increase retinoid-induced differentiation of APC-mutant CRC cells. The functional relevance of this finding is that 1) CYP26A1 expression is controlled by WNT signaling via APC [30], and 2) CYP26A1 enzyme controls intracellular RA metabolism and regulates RA signaling [16,17]. Our previous finding that RA signaling mainly occurs via ALDH+ SCs [14] raises two key questions addressed below:

• Q1 How is RA signaling regulated in ALDH+ SCs?
• Q2 How does dysregulation of RA signaling due to APC mutation contribute to overpopulation of ALDH+ SCs that drives development of CRC?

Studying the effect of wt-APC on ATRA response indicates that WNT signaling, via its target gene CYP26A1, regulates RA signaling in the differentiation of ALDH+ SCs.

We found that inducing wt-APC expression, which decreases WNT-signaling, reduced expression of several WNT target genes and increased sensitivity of HT29 cells to the anti-proliferative effects of ATRA (Figure 1 and Figure S1). We also found that ATRA treatment of HT29 cells significantly increased CYP26A1 expression (30.8-fold) and inducing wt-APC attenuated this increase by 50% (Figure 5A). This increase of CYP26A1 by ATRA treatment of APC mutant HT29 cells may be explained by two factors 1) CYP26A1 is a RA target gene [31,32,33] which should be induced by ATRA. 2) CYP26A1 is also a WNT target gene [30] and ATRA increases WNT/β-catenin activity in HT29 cells even though they already have constitutively activated WNT due to mutated APC (Figure 1C). Notably, increased CYP26A1 expression in ATRA treated HT29 cells decreases when they are induced to express wt-APC (Figure 5). This finding indicates that the ability of wt-APC to reduce WNT/β-catenin activity is the main mechanism that counterbalances the ability of ATRA to increase CYP26A1 expression. Consequently, since wt-APC inhibits WNT signaling, it should maintain the balance between WNT and RA signaling. Indeed, that is precisely what we observed because induction of wt-APC reversed ATRA’s ability to increase: i) WNT/β-catenin activity (Figure 1C) and ii) CYP26A1 expression (Figure 5). These findings show that modulation of CYP26A1 levels by WNT signaling regulates RA signaling in APC mutant cells. Thus, our findings indicate that the anti-proliferative effect of ATRA is greatly enhanced by decreased expression of CYP26A1 due to the ability of wt-APC to reduce WNT signaling.

Since the RA pathway components are mainly expressed in ALDH+ SCs [14], WNT signaling, via its target gene CYP26A1, regulates RA signaling in ALDH+ SCs. Indeed, inducing wt-APC expression reduced expression of CYP26A1 and enhanced the ability of ATRA to decrease CRC cell proliferation (Figure 1 and Figure 5). Moreover, inducing wt-APC: i) suppressed ALDH1A1 expression, ii) decreased ALDH+ SCs, and iii) increased neuroendocrine cell differentiation. These findings show that APC’s role in the control of WNT signaling regulates, via its target gene CYP26A1, RA signaling in the differentiation of ALDH+ CSCs.

To further explore the mechanism by which WNT signaling might regulate RA signaling, we examined differential mRNA expression in HT29 cells that were induced to express wt-APC and that were treated with ATRA compared to their respective controls. NanoString profiling revealed that among the 248 genes showing a significant change in expression, CYP26A1 showed the greatest increase in expression (Figure 4C and Figure S2). Additionally, our NanoString and bioinformatics analyses identified a novel CYP26A1-based network of protein interactions which reveals how WNT and RA pathways are inter-connected. Specifically, we found that a unique signaling cascade that links components of the RA signaling pathway (CYP26A1, RARA) with components of the WNT pathway (DKK1, WNT7B, WNT11, FZD7) via HDAC2, KLF4, and TNF (Figure 4B). Thus, a CYP26A1-based network of both RA and WNT signaling components is predicted to play a role in the functional link between these two pathways.

Studying the effect of APC mutation on NEC differentiation indicates that decreased RA signaling contributes to overpopulation of ALDH+ SCs that drives development of CRC.

We found that inducing wt-APC expression, which decreases WNT-signaling and lowers CYP26A1 expression, leads to a decrease in cell proliferation (Figure 1A,B) and an increase in NEC differentiation (Figure 2A,B and Figure 3). Inducing wt-APC also decreased ALDH+ SC numbers and reduced expression of the ALDH1A1 SC marker (Figure 2C,D). Altogether, these findings show that elevated CYP26A1 levels prevent differentiation of CSCs in APC mutant cells by increasing RA clearance which reduces RA signaling (Figure 8). This mechanism provides an explanation for how decreased RA signaling contributes to ALDH+ SCs overpopulation that drives CRC development.

Our results concur with the finding in Shelton et al [30] in which CYP26A1 was upregulated in the intestine of APC mutant zebrafish embryos (and Apc^Min/+ mouse adenomas, human FAP adenomas, and human sporadic CRCs) but not in zebrafish embryos with wt-APC. Defects in gut differentiation was also reversed with pharmacologic inhibition or knockdown of CYP26A1 in APC mutant zebrafish embryos [30]. Further studies on zebrafish showed that APC has a dual role in regulating Wnt and RA signaling [34,35]. Therefore, expression of wt-APC or pharmacologic inhibition of CYP26A1 may provide a novel therapeutic strategy to sensitize CRC cells to ATRA [36]. For example, a recent study by Penny et al. involved treating Apc^Min/+ mice with the CYP26A1 inhibitor Liarozole [37]. The administration of Liarozole to Apc^Min/+ mice increased endogenous RA signaling (by blocking ATRA metabolism), and dramatically reduced intestinal adenoma numbers in these Apc-mutant mice. We also found that the treatment of human CRC cells with Liarozole decreased proliferation, sphere formation, and the size of the ALDH+ SC population [14]. This suggests that decreasing the intracellular metabolism of ATRA using agents that inhibit CYP26A1 activity, might be a way to increase ATRA levels and therapeutically augment RA signaling in order to decrease CSC numbers in APC-mutant colon cancer tissues [16].

Clinical significance of our results that show CYP26A1 inhibitors, which block RA metabolism, sensitize CRC cells to anti-proliferative effect of drugs that downregulate WNT signaling.

Our findings likely have vast clinical importance. For instance, given that RA receptor mutant CRC cells show similar response to the CYP26A1 inhibitors (Liarozole and Talarozole) as do non-mutant cells (Figure 6), this indicates that the inhibition of CYP26A1 leads to a high enough intracellular RA level to induce growth inhibition regardless of RA receptor genotype. We then investigated the effects of these CYP26A1 inhibitors in combination with agents (Sulindac or Piroxicam) that have anti-tumor activity against APC-mutant tissues and Sulindac is known to target β-catenin and downregulate WNT signaling [reviewed in 16]. Notably, when Sulindac or Piroxicam was combined with CYP26A1 inhibitors Liarozole or Talarozole, the drug combination produced additive or synergistic effects (Figure 6C,F).

That the drug combinations reduce CRC proliferation to a greater extent than the CYP26A1 inhibitors or anti-WNT agents alone indicates that inhibiting CYP26A1 to increase RA signaling combined with the effect of inhibiting WNT signaling holds great promise as a therapeutic approach in oncology.

For human CRC cases, CYP26A1 predicts patient survival according to APC genotype.

Bioinformatics analysis of human CRC cases provides further evidence that CYP26A1 has clinical significance (Figure 7). We found that the majority of CRC patients express an increased level of CYP26A1 in their tumor tissues compared to normal colon from the same patient. Considering that CYP26A1 is a key enzyme responsible for RA degradation, patients who express lower levels of CYP26A1 should have higher levels of RA in their tumor cells, which should up-regulate RA-signaling activity and endow CRC patients with better survival. This prediction is supported by our results showing that CYP26A1 is only a significant determinant of patient survival when patient tumors carry wild-type, but not mutant, APC. This finding also supports the existence of crosstalk between WNT and RA-signaling that is dependent upon CYP26A1 expression level (hence cellular RA level), and APC genotype of the patients.

5. Conclusions

This work has vast clinical significance because it shows how CRC SCs might be sensitized to retinoid-induced cellular differentiation. We present a mechanism and therapeutic strategy to explain how differentiation-inducing agents such as ATRA may be used to promote CSC differentiation. Indeed, RA is an effective therapeutic agent in the treatment of acute promyelocytic leukemia (APL) [38,39,40,41,42]. RA therapy has also been shown to improve the survival of neuroblastoma patients [43]. While using retinoids to treat other cancers has shown limited success [16,17], therapeutically inhibiting CYP26A1, as we have shown herein, to reduce RA metabolism may prove useful in designing approaches to promote differentiation of CSCs with retinoids. Therefore, our findings are important because CRC development is driven by overpopulation of CSCs which are resistant to conventional chemotherapies and radiation [44,45,46]. Thus, finding ways to promote retinoid-induced differentiation of CSCs in vivo may lead to new and more effective therapeutic strategies for CRC patients.