Enhancement of TGF-β Signaling Responses by the E3 Ubiquitin Ligase Arkadia Provides Tumor Suppression in Colorectal Cancer

Abstract

Transforming Growth Factor (TGF)-β signaling provides tumor protection against colorectal cancer (CRC). Mechanisms that support its tumor suppressive properties remain unclear. The ubiquitin ligase Arkadia/RNF111 enhances TGF-β signaling responses by targeting repressors of the pathway for degradation. The co-repressors SnoN/Ski, critical substrates of Arkadia, complex with the activated TGF-β signaling effectors Smad2/3 (pSmad2/3) on the promoters of target genes and block their transcription. Arkadia degrades this complex including pSmad2/3 and unblocks the promoter. Here we report that Arkadia is expressed highly in the mouse colonic epithelium. Heterozygous Akd^+/− mice are normal but express less Arkadia. This leads to reduced expression of several TGF-β target genes suggesting that normal levels of Arkadia are required for efficient signaling responses. Critically, Akd^+/− mice exhibit increased susceptibility to AOM/DSS carcinogen-induced CRC as they develop four-fold more tumors than wild type mice. Akd^+/− tumors also exhibit a more aggressive pathology, higher proliferation index and reduced cytostasis. Therefore Arkadia functions as a tumor-suppressor whose peak expression is required to suppress CRC development and progression. The accumulation of nuclear SnoN and pSmad2, along with the down-regulation of TGF-β target genes observed in Akd^+/− colon and tumors, suggest that Arkadia’s tumor suppressing properties are mediated by its ability to de-repress TGF-β signaling. Consistent with this likelihood, we identified mutations in primary colorectal tumors from human patients that reduce Arkadia function and are associated with the accumulation of nuclear SNON. Collectively, our findings reveal that Arkadia enhances TGF-β signaling responses and supports its tumor suppressing properties in CRC.

Introduction

Colorectal cancer (CRC) is one of the most common malignancies worldwide and accounts for over 600,000 deaths each year (1). Mutations that inactivate transforming growth factor-β (TGF-β) signaling components are associated with CRC, consistent with a role of this pathway in tumor suppression (2, 3). Specifically, mutation or deletion of the receptor genes TGFßR1 and TGFßR2, and the effector genes SMAD2, SMAD3 and SMAD4 of the pathway, have been reported in CRC (2, 4). Germ-line mutations of SMAD4 are also associated with juvenile polyposis in the colon (5). In addition, mouse models in which these core signal transduction components are inactivated result in an increased susceptibility to develop adenocarcinomas under oncogenic stress (3). However, analyses of tumor samples have revealed that the TGF-β pathway also exhibits oncogenic properties in CRC, as it promotes survival, invasion and metastasis (6). Mechanisms that impair TGF-β-mediated tumor suppression but do not abolish signaling may allow TGF-β-mediated metastasis to occur more frequently later in tumorigenesis. Regulators of such mechanisms could be used as early markers for susceptibility and prognosis and help in both the choice of drug treatment and in the development of novel anti-cancer agents.

The TGF-β ligands, including TGF-β-1/2/3, Nodal and Activin, bind to specific cell-surface receptors and activate, via phosphorylation, the Smad2/3 effectors (pSmad2/3). PSmads complex with Smad4 and translocate to the nucleus where they function as transcription factors (7, 8). Within the constantly regenerating colonic epithelium, the normal function of the TGF-β pathway is to maintain homeostasis by counteracting proliferation and promoting differentiation and apoptosis (9). Interestingly, TGF-β factors specify cell fate and differentiation in a dose-dependent fashion during development (10). However, reduction of TGF-β signaling in the adult colonic epithelium and its involvement in CRC is not clear.

TGF-β signaling is tightly regulated by both extracellular and intracellular mechanisms. Key intracellular regulators include the inhibitors Smad6/7, which mediate the degradation of receptors and interfere with the phosphorylation of effector Smads; and the nuclear co-repressors SnoN and Ski, which directly interact with the effector Smad proteins and recruit a co-repressor complex containing HDAC and N-CoR to targeted gene promoters to block expression (11). Gene amplification and overexpression of SMAD7 and SNON/SKI proteins have also been linked to CRC, attributed to loss of TGF-β signaling (12, 13). We previously showed that Arkadia, a nuclear RING-domain E3 ubiquitin ligase, is a positive regulator of the TGF-β/Nodal signaling branch. It mediates the ubiquitin-proteasome degradation of all the above-mentioned negative regulators of the pathway and therefore constitutes a potent ‘de-repressor’ that enhances TGF-β target gene transcription (14-17). Degradation of SnoN/Ski by Arkadia depends upon their specific interaction with pSmad2/3 (15, 18, 19). As Arkadia also interacts and degrades pSmad2/3, its function results in clearing the promoters from used/blocked effectors, thereby allowing fresh effectors to bind and activate transcription (20). Consistent with this, absence of Arkadia results in increased levels of stable SnoN/Ski and pSmad2/3, but as these are together in a complex, the promoters of target genes are occupied by stable, repressed pSmad2/3 leading to repression (20). Arkadia is broadly expressed in the mouse embryo and its absence leads to loss of a subset of Nodal signaling responses necessary for the development of anterior/head structures, which are also lost with the genetic reduction of Nodal (20-22). However, whether Arkadia enhances TGF-β signaling responses in the adult colonic epithelium, and how this affects CRC development, remained unknown.

Using a deep sequencing screen of human Arkadia (AKD) mRNA from tumors of CRC patients, we identified somatic mutations that reduce AKD function. We demonstrated that reduction in Arkadia levels increased susceptibility to develop CRC in a mouse model and showed that this mechanism involves increased stability of SnoN and pSmad2, and a reduction of TGF-β-mediated target gene transcription. Collectively, our data reveals that Arkadia is required for peak efficiency of a subset of TGF-β transcriptional responses in the colonic epithelium and in colorectal tumors and thereby supports the tumor suppressive arm of this pathway.

Materials and Methods

Deep sequencing

Total RNA was extracted from FFPE tumor and adjacent normal tissue sections as previously described (23) and reversed transcribed. Five to six overlapping PCR amplifications spanning the 300bp C-terminus of AKD were performed per patient/tissue type; each representing an ‘amplicon library’ (primer sequences in supplemental section). Preparation of DNA-carrying beads was performed as previously described (24). Beads were purified and loaded onto a 16-gasket picotiter plate for high throughput pyrosequencing using the GS20 454-sequencer (Roche). Each amplicon library yielded an average of 4,000 sequencing reads. Data was analyzed as previously described (24). The Ensembl entry ENST00000380504 was used as the reference sequence for AKD (corresponding to protein Q6ZNA4-1).

Plasmid construction and cell culture assays

Site-directed mutagenesis to generate the different point mutations was performed from a full-length human AKD cDNA clone (Invitrogen, ID: 30336202; see supplemental section for primer details). Fragments were subsequently PCR amplified and cloned in-frame into the expression plasmid pTriex2-GFP as XhoI-digested products (20). Transfections and luciferase assays were performed in HEK293T cells (Cancer Research UK) as previously described (15). For immunoblotting, cells were harvested 24 hours post-transfection and lysates analyzed for GFP proteins. The HEK293T cell line was authenticated in July 2011 by the Health Protection Agency using STR multi-loci genotyping.

Immunohistochemistry (IHC)

IHC was performed as previously described (25) using the following rabbit polyclonal antibodies: anti-Arkadia (1:100, LifeSpan Biosciences, Inc); anti-SnoN H-317 (1:80, Santa Cruz) and anti-pSmad2 (1:50, Millipore). The ABC–DAB detection system was used (ABC detection kit, Pierce; DAB substrate, Thermo Scientific). Blocking solution was used instead of primary antibodies for negative controls. Immunoreactivity was graded on a scale of 0-3 (score 0: negative; score 1: weak; score 2: moderate; score 3: strong) according to intensity of staining and percentage of immunopositive cells as previously described (26). All sections were counter-stained with hematoxylin.

Immunoblotting

Experiments were performed as described previously (20), using the following antibodies (dilutions): rabbit anti-GFP (1:1000, Invitrogen); rabbit anti-pSmad2 (1:500, Cell Signaling); rabbit anti-SnoN, H-317 (1:5000, Santa Cruz); rabbit anti-p21, C-19 (1:200, Santa Cruz); rabbit anti-Histone H3, ab-1791 (1:10000, abcam); rabbit monoclonal anti-Smad2 (1:1000, Epitomics) and mouse monoclonals anti-PCNA, PC-10 (1:3000, Santa Cruz) and anti-active ß-catenin (anti-ß-catⁿ), 8E7 (1:500, Millipore). Densitometric quantification of bands was performed using ImageJ software.

Colorectal tumor Induction protocol and histological analysis

20-week old wild type (wt) and Akd^+/− mice in a 129SVcc inbred genetic background were injected with a single intraperitoneal injection of the carcinogen Azoxymethane (AOM, 7.4mg/kg; Sigma), one week after which mice were subjected to 2% Dextran Sodium Sulfate (DSS, MW: 36,000-50,000; MP Biomedical) in their drinking water for a period of five days. Two more cycles of the five-day DSS treatment were given, each separated by a 16-day period on normal drinking water. Twelve weeks after the final DSS treatment, mice were sacrificed and their colons analyzed for tumors. This protocol was repeated using mice in a 129SVcc/CD1 hybrid background. Tumor counts were made under a dissecting microscope. Colons were fixed in 4% paraformaldehyde and paraffin embedded. X-gal staining was performed as previously described (21). Sections of 4μm thickness were either H&E stained or used in IHC. Histological analysis was performed for all wt tumors and at least 3 representative tumors per Akd^+/− mouse.

Statistical analysis

Statistical analysis was performed using the SPSS for Windows, release 12.0 (SPSS inc, Chicago, IL, USA). Chi-square analysis was used to test the significance of differences in immunoreactivity scores or histopathological parameters between wt and Akd^+/− tumors. All other data was analyzed using Student’s t test and presented as mean ± SEM. P<0.05 was considered as significant and denoted with a single asterix. P≤0.01 and P≤0.001 were denoted with two and three asterisks, respectively.

Results

Arkadia (AKD) mutations in selected human colorectal tumors with high SNON

We recently described a cohort of primary human CRCs, of which 83/87 (95.4%) overexpressed SNON (26). Nuclear SNON expression was identified in 42/87 (48.3%) of these tumors and, importantly, this correlated with advancing tumor grade. Interestingly, we showed that SNON accumulation in these tumors was not a result of elevated SNON mRNA, suggesting increased protein stability. Since SNON is a substrate of AKD, itself a nuclear protein, it is possible that inactivating mutations in AKD may account in part for this stable nuclear SNON phenotype. To test this possibility, we performed a deep sequencing screen for mutations in AKD mRNA extracted from human CRC paraffin-embedded tumors. We selected five CRC patients with tumors displaying the highest levels of nuclear SNON protein but a relatively low level of SNON mRNA expression (Fig. 1A) (26). Normal expression levels of AKD mRNA were present in these samples (data not shown).

Figure 1. Deep sequencing of AKD C-terminus from CRC patients.

(A) Scatter plot correlating the relative level of SNON expression (qPCR) and nuclear SNON staining (IHC) from human CRC samples. Patients with high SNON stability highlighted in black were selected for deep sequencing. (B) Protein alignment of the C-terminus of AKD from human (Q6ZNA4), mouse (Q99ML9), chick (Q90ZT8), Xenopus (Q0V9R0), Drosophila (Q9VGI6) and C.elegans (Q9XUM8). The grey lines indicate the NRG (NRGASQG), TIER (TIERCTY) and NLS (PHKYKKV) domains and the black line highlights the RING domain. Numbers indicate four key mutations identified. (C) Luciferase CAGA₁₂ reporter assay values from three biological repeat experiments, each in quadruplicate. The Q899STOP mutation exhibits a dominant-negative effect, similar to the W972R control. (D) Immunoblot shows relative stability of GFP-AKD proteins (top bands) compared to GFP control (bottom bands). GFP-Q899STOP is stable at approximately 140kDa. H3 was used as a loading control. 40μg of protein extract was loaded in each lane.

We hypothesized that mutations inactivating the ubiquitin ligase activity of AKD, but do not disrupt substrate binding, would protect the substrates from ubiquitination and subsequent degradation and therefore prevent wild type AKD function. The domains that are required for the activity of AKD are located at the highly conserved C-terminal 100 amino acids (aa). This region comprises the NRG, which is at the end of the domain required for substrate (pSMAD2 and SNON/SKI) recognition, followed by a conserved TIER domain (of unknown function), a nuclear localization signal (NLS) and a RING domain, required for the ubiquitin ligase activity (20). Inactivation of the RING domain or deletion of the TIER domain in Arkadia converts its function to a dominant-negative, as shown in luciferase reporter experiments performed in HEK293T cells using the SMAD-dependent reporter CAGA₁₂-Luc (Supplemental Fig. S1). Interestingly, the COSMIC database has catalogued two different point mutations at the C-terminal of AKD from Ovarian cancer (27). This supported our hypothesis that mutations in the C-terminus are more likely to affect AKD function and led us to focus the mRNA sequencing screen to the last 300bp of AKD.

The paraffin-embedded tumor sections that we used for our sequencing analysis comprised a heterogeneous population of cells including both tumorigenic epithelial cells and non-tumorigenic stromal cells, such as fibroblasts and inflammatory cells. In addition, not all cells exhibited an accumulation of nuclear SNON, and therefore a potential mutation in AKD. Furthermore, since AKD mutations may arise from only one allele in these cells, they are likely to be represented at a low frequency in the sample. To determine the true frequency of such mutations in the mRNA (cDNA), we applied the GS20 system to achieve thousands of bi-directional sequencing reads for the C-terminal of AKD from each sample (24). Mutations were identified in two of the five patients (patients 6 and 32, Supplemental Table S1). Flowgram analysis was performed to validate these mutations (Supplemental Fig. S2). Table 1 summarizes four of these mutations identified in the screen, each corresponding to a conserved aa residue (Fig. 1B).

Table 1. Summary of the Q899STOP, P908S, D937N and T943I mutations of AKD identified in the deep sequencing screen.

Mutation	Codon Change	Amino Acid Change	Total Freq % (Reads)	Patient
1	CAG>TAG	Q899STOP	2.76 (4,167)	6
2	CCA>TCA	P908S	9.94 (4,167)	6
3	GAC>AAC	D937N	4.30 (3,833)	32
4	ACT>ATT	T943I	1.59 (3,833) 1.76 (5,282)	32 6

We tested the functionality of each of these mutations in HEK293T cells using the CAGA₁₂-Luc reporter assay (Fig. 1C). Overexpression of GFP-tagged AKD carrying the P908S, D937N or T943I mutation enhanced reporter expression to a similar level as wt GFP-AKD (which is more than two-fold above endogenous signaling, as assessed by overexpression of empty vector controls). In contrast, the Q899STOP mutant reduced signaling three-fold compared to controls (Fig. 1C), suggesting that it exhibits a dominant-negative function similar to the control RING-W972R mutant AKD. Consistent with our hypothesis, this mutation introduces a STOP codon at the end of the substrate recognition domain, resulting in a truncated protein lacking the enzymatic RING domain. The P908S mutation was observed at the highest frequency (9.94%). As this is positioned within the NLS of AKD, we examined the localization of its GFP-tagged form in HEK293T transfections but found it to be nuclear like wt AKD (data not shown).

We have previously shown that deletion of the RING or NRG stabilizes the normally unstable wt Arkadia (20). By co-expressing a GFP control plasmid with each GFP-tagged AKD mutant in HEK293T cells, we observed an equal abundance of short GFP in all lysates following immunoblotting. However, only the GFP-Q899STOP AKD mutant gave a larger stable and visible band (Fig. 1D). The stabilization of Q899STOP AKD most likely contributes to its ability to strongly suppress wt AKD function when this mutation occurs in one allele. Considering that only a proportion of tumor cells displayed stabilized nuclear SNON (26), combined with the basis of screening for single-allele mutations in a heterogeneous tumor sample, the observed frequency of 2.76% for the Q899STOP mutation has a good association to be causal to the accumulation of SNON.

Expression from both alleles of Arkadia in the colonic epithelium maintains peak levels of TGF-β target-gene expression

We decided to directly test the role of Arkadia in a mouse model for CRC, and for this we first examined its expression in the colon. We used Akd^+/− gene trap mice expressing β-geo under the endogenous promoter (21) to show that Arkadia is highly expressed in the crypt compartments of the colon (Fig. 2A and 2B). Immunohistochemistry (IHC) with an anti-Arkadia antibody showed a defined nuclear distribution of Arkadia protein in epithelial cells along the crypt (Fig. 2C). Both methods showed weaker expression in the surface epithelium. This expression data suggests a role for Arkadia in the colonic epithelium, where TGF-β is involved in maintaining tissue homeostasis.

Figure 2. Expression of Arkadia in the mouse colon.

(A) wt and (B) Akd^+/− colon sections stained with X-gal. (C) anti-Arkadia IHC in wt colon. (D) qPCR expression of Akd and (E) six TGF-β target genes in wt and Akd^+/− colon tissue. (F) Immunoblot with pSmad2 antibody and H3 loading control. Expression of SnoN in (G) wt and (H) Akd^+/− colon tissue. Elevated SnoN staining is observed in Akd^+/− colons, particularly in the differentiated epithelial cells at the surface of the crypts. All images are shown at x40 magnification; bar 25μm.

The gene trap insertion terminates transcription of the Akd allele before the first exon (21) and quantitative PCR (qPCR) confirmed that Akd^+/− colons express half the level of Arkadia compared to wt colons (Fig. 2D). Analysis of TGF-β target gene expression revealed significant down-regulation of genes such as p15^INK4b, PAI-1 (Serpine1), Smad7 and TMEPAI (Pmepa1) in the Akd^+/− colon, whereas targets such as SnoN and p21^WAF showed no difference (Fig. 2E). Immunoblotting revealed higher levels of pSmad2 in Akd^+/− compared to wt colon (Fig. 2F). We also performed IHC for SnoN and observed higher levels in Akd^+/− compared to wt colon (Fig. 2G and 2H). These observations are consistent with the known molecular function of Arkadia to degrade SnoN-Smad complexes and enhance TGF-β target gene expression (14, 15, 20), as it does in embryos, ES and other cell lines (14, 20).

Interestingly, the above data suggests that even a reduction of Arkadia function by the loss of one wt allele is associated with the reduction of TGF-β signaling responses, which was not known before. Multipotent stem cells reside at the base of the crypts and give rise to proliferating daughter epithelial cells that migrate upwards before finally acquiring a differentiated epithelial phenotype at the surface with TGF-β signaling counteracting proliferation and maintaining homeostasis (28). Notably, the increase of SnoN in the Akd^+/− colon occurs predominantly in the more differentiated cells of the surface epithelium (compare Fig. 2G and 2H). This suggests that the down-regulation of TGF-β signaling responses occurs within the epithelium and most likely impairs the maintenance of balance between proliferation and differentiation.

Akd^+/− mice exhibit increased susceptibility to CRC

To determine whether loss of one allele of Arkadia increases susceptibility to develop tumors, we induced CRC using the AOM/DSS method (29, 30) in age-matched wt and Akd^+/− mice in two different genetic backgrounds (129SVcc inbred and 129SVcc/CD1 hybrid) under a specific pathogen free environment (Fig. 3A). AOM (Azoxymethane) is a potent carcinogen that reproduces the activating mutations in β-catenin and KRAS (31, 32), and the inactivating mutations in APC that are observed in human CRC (33, 34). Crucially, 5/12 (42%) wt mice did not develop any tumors throughout the duration of the protocol whereas all Akd^+/− mice developed tumors (Fig. 3B). Akd^+/− mice developed almost four-fold the number of colorectal tumors as wt mice in a 129SVcc background (mean number of tumors 7.6 ± 1.0 SEM; P=0.0001, Fig. 3C, left panel). We found very similar results in the 129SVcc/CD1 background, with a near three-fold increase in tumor multiplicity in Akd^+/− mice (P=0.0053, Fig. 3C, right panel). These results show that Akd^+/− mice are more susceptible to carcinogenesis and that wild type levels of Arkadia provide a protection against CRC. This supports the hypothesis that mutations that simply reduce AKD function in patients may contribute to the development of CRC.

Figure 3. AOM/DSS induction of CRC in wt and Akd^+/− mice.

(A) Macroscopic view of representative X-gal stained mouse colons (129SVcc background) at end of protocol (distal end at top; bar 1cm). (B) Percentage of mice that developed tumors. (C) Mean number of tumors observed from wt (n=12) and Akd^+/− (n=7) mice in a 129SVcc background and wt (n=6) and Akd^+/− (n=6) mice in a 129SVcc/CD1 background. (D-I) H&E staining of colons and tumors from wt and Akd^+/− treated mice. wt: (D) normal colon, (E) adenoma with severe dysplasia and (F) intramucosal carcinoma. Akd^+/−: (G) intramucosal carcinoma with mucus production, (H) invasive adenocarcinoma and (I) vessel invasion. Black arrowheads mark sites of invasion. (D-H) shown at x2.5 magnification; bar 250μm and (I) at x20 magnification; bar 50μm.

Enhanced tumor progression in Akd^+/− mice

To characterize the tumor pathology from wt and Akd^+/− mice, all 23 tumors from wt mice and a total of 24 tumors from Akd^+/− mice from a 129SVcc background were subjected to histopathological analysis. The lesions identified in the colon of treated mice included tubular adenomas with low or high-grade dysplasia, intramucosal carcinomas and invasive carcinomas (Table 2). Although the majority of tumors from wt mice developed to intramucosal carcinomas, a significant proportion of tumors (13%) did not develop beyond adenomas with high-grade dysplasia (Fig. 3D-F). In contrast, all 24/24 Akd^+/− tumors analyzed developed to intramucosal carcinomas (Fig. 3G), with one Akd^+/− mouse even harboring an invasive carcinoma (Fig. 3H). We also observed a vessel invasion within a carcinoma from another Akd^+/− mouse (Fig. 3I).

Table 2. Histopathology of wt and Akd^+/− colorectal tumors.

Tumors from all wt (n=12) and all Akd^+/− (n=7) mice (129SVcc background) were subjected to histopathological analysis. Colons were examined along the entire distal-proximal axis. VI = vessel invasion found in one tumor.

Mouse	Tumors	Tumors Analyzed	Adenomas	Intramucosal Carcinomas	Invasive Carcinomas
wt_1	5	5	-	5	-
wt_2	1	1	-	1	-
wt_3	2	2	-	2	-
wt_4	3	3	1	2	-
wt_5	3	3	-	3	-
wt_6	-	-	n/a	n/a	n/a
wt_7	-	-	n/a	n/a	n/a
wt_8	-	-	n/a	n/a	n/a
wt_9	-	-	n/a	n/a	n/a
wt_10	-	-	n/a	n/a	n/a
wt_11	6	6	2	4	-
wt_12	3	3	-	3	-
Total	23	23	3 (13%)	20 (87%)	-
Akd+/−_1	10	3	-	3	-
Akd+/−_2	9	3	-	3VI	-
Akd+/−_3	9	3	-	2	1
Akd+/−_4	3	3	-	3	-
Akd+/−_5	5	3	-	3	-
Akd+/−_6	7	4	-	4	-
Akd+/−_7	10	5	-	5	-
Total	53	24	-	23 (96%)	1 (4%)

The above analysis shows that lesions progress more rapidly through the adenoma-carcinoma sequence in Akd^+/− mice, which may explain why a high proportion of wt mice did not develop any gross tumors. Statistical analysis confirmed that loss of one allele of Akd significantly correlated with the progression of lesions from adenoma to intramucosal and invasive carcinomas (P=0.03) (Table 2).

Tumors from Akd^+/− mice exhibit reduced TGF-β-mediated cytostasis and increased proliferation

We next examined the tumors at a molecular level to identify the mechanisms involved in their enhanced progression. We analyzed the expression of various genes by performing Immunoblot and qPCR analysis on proteins and RNA extracted from six wt (T¹-T⁶) and six Akd^+/− (T^A-T^F) tumors (Fig. 4A-C). Using the proliferation marker PCNA, we found that tumors from Akd^+/− mice (hereafter termed Akd^+/− tumors) exhibited significantly higher levels of proliferation compared to tumors from wt mice (hereafter termed wt tumors; Fig. 4A).

Figure 4. Immunoblot analysis of wt and Akd^+/− tumors.

(A) Immunoblotting with SnoN, pSmad2, active beta-catenin (β-catⁿ) and PCNA. Lanes N, treated adjacent normal colon tissue; lanes T, representative wt (T¹-T³) and Akd^+/− (T^A-T^C) tumor samples. H3 was used as a loading control. Protein levels normalized to H3 are shown in the three charts. (B) Analysis of p21^WAF, SnoN and pSmad2 protein levels in six wt (T¹-T⁶) and six Akd^+/− (T^A-T^F) tumors. 10μg of protein extract was loaded in each lane for all gels. Due to the enhanced stability of pSmad2 in Akd^+/− tumors, we also observed more stable levels of total Smad2. (C) qPCR expression levels of various genes (as indicated) in the above analyzed tumors. The difference is PAI-1 expression was calculated as P=0.0532. Average fold changes are shown in the charts on the right.

As activation of Wnt/β-catenin signaling is critically involved in the proliferation of the regenerating colonic epithelium and in CRC, we examined whether differences in its activation underpinned this hyperproliferation. An antibody that specifically detects the active nuclear form of β-catenin (β-catⁿ) (35) showed no difference between Akd^+/− and wt tumors (Fig. 4A). Furthermore, the expression of the Wnt/β-catenin target oncogene c-Myc was also not significantly different between Akd^+/− and wt tumors (Fig. 4C). This data suggests that an alternative mechanism underlies the more aggressive pathology observed in Akd^+/− tumors.

As reduction in cytostasis can also lead to hyperproliferation, and as this critically depends on TGF-β signaling in the colonic epithelium, we examined the levels of the p21^WAF cyclin-dependent kinase inhibitor (CDKI) in tumors (36). Indeed, this factor was significantly lower in all six Akd^+/− tumors and in 2/6 wt tumors (Fig. 4B). However, we did not observe a corresponding decrease in mRNA expression of p21^WAF in Akd^+/− tumors (Fig. 4C), suggesting a possible reduction in p21^WAF protein stability in these tumors. In contrast, transcription of the cytostatic target p15^INK4b was significantly reduced in Akd^+/− tumors (Fig. 4C). This reduction in p15^INK4b expression pre-exists in the colon of untreated Akd^+/− mice (Fig. 2E). Together, the data suggest that reduced cytostasis is responsible for the hyperproliferation and enhanced progression of Akd^+/− tumors.

Furthermore, examination of the levels of Arkadia’s substrate SnoN showed that it is elevated in tumors and adjacent normal colon of Akd^+/− mice compared to wt mice (Fig. 4A and 4B). SnoN mRNA expression, however, was not significantly different between Akd^+/− and wt tumors (Fig. 4C), suggesting an increased stability of the SnoN protein in Akd^+/− tumors. Consistent with the molecular function of Arkadia, pSmad2, which complexes with SnoN and is also degraded by Arkadia, was also increased in Akd^+/− tumors (Fig. 4A and 4B). As the expression of the TGF-β target gene PAI-1, like p15^INK4b, was also reduced in Akd^+/− tumors (Fig. 4C), collectively, the above data strongly suggest that repression of the pathway is increased due to the reduction or loss of Arkadia from cells of Akd^+/− tumors.

Enhanced tumor progression in Akd^+/− mice is associated with reduction and not loss of Arkadia function

To address the distribution and level of TGF-β pathway repression in tumors we performed IHC to visualize SnoN and pSmad2 proteins in six Akd^+/− tumors and four wt tumors. We found that pSmad2 staining was stronger in Akd^+/− tumors (score-2 or 3 in 4/6 tumors) compared to wt tumors (score-1 in 4/4 tumors) (Fig. 5A and 5B). This difference, however, did not quite reach statistical significance, most likely because pSmad2 levels also depend on the amount of active ligand that each tumor expresses, which may vary between tumors. There was no difference in the cellular localization of pSmad2 in wt and Akd^+/− tumors, as both were nuclear. In contrast, SnoN levels in Akd^+/− tumors were classified as either moderate (score-2 in 2/6 tumors) or strong (score-3 in 4/6 tumors), while expression in wt tumors was frequently weak (score-1 in 3/4 tumors). This difference was statistically significant (chi-square, P=0.036). Notably, SnoN cellular localization was cytoplasmic in wt tumors (3/4 tumors cytoplasmic) and largely nuclear in Akd^+/− tumors (Fig. 5C and 5D). Both SnoN and pSmad2 were found to be nuclear in the cells that constitute the lining of the tubular structures of the adenocarcinoma from Akd^+/− mice suggesting increased repression of TGF-β signaling in these cells (Fig. 5B and 5D).

Figure 5. IHC for SnoN, pSmad2 and Arkadia in wt and Akd^+/− tumors.

Representative sections of adenocarcinomas stained with anti-pSmad2: (A) wt and (B) Akd^+/−; anti-SnoN: (C) wt and (D) Akd^+/−; and anti-Arkadia: (E) wt and (F) Akd^+/−. All images are shown at x40 magnification; bar 25μm.

To examine whether complete loss of Arkadia expression is associated with the enhanced progression of Akd^+/− tumors, we performed qPCR on wt and Akd^+/− tumors and found that Akd^+/− tumors consistently express Akd but only at half the level of wt tumors (Fig. 4C). To examine whether this reduction is an average expression of all tumor cells rather than loss in a subset of cells, we performed IHC for Arkadia in tumors (this antibody is raised against the C-terminus of AKD and is expected to recognize functional Arkadia proteins). We observed the presence of nuclear Arkadia protein in almost every cell lining the tubular structures of tumors from both genotypes (Fig. 5E and F), even in the most aggressive Akd^+/− tumors (Fig. 5F shows Arkadia staining on the same invasive adenocarcinoma pictured in Fig. 3H). Therefore, even in the most advanced tumors, Arkadia is not lost and this therefore cannot account for the enhanced growth of Akd^+/− tumors. Collectively, the above data suggest that reduction of Arkadia rather than loss is responsible for the enhanced tumor progression.

Our findings also mirror reports that Smad4 and Tgfβr1 haploinsufficiency promotes CRC development in a genetic background carrying Apc mutations (37-40), which alone causes adenomas (37, 41). Together, these studies emphasize that gene dosage of important components of the TGF-β pathway acts as key determinants in CRC susceptibility and supports the hypothesis that the tumor suppressing properties of this pathway depend on its peak levels. Furthermore, TGF-β signaling responses and cytostasis are reduced in Akd^+/− mice suggesting that Arkadia’s tumor suppressing properties in both the normal colonic epithelium and colorectal tumors are mediated by the enhancement of this pathway.

Discussion

TGF-β signaling exerts an anti-tumorigenic function in the colonic epithelium, whereas in advanced tumors it can promote metastasis and progression. What regulates these different downstream TGF-β signaling effects remains largely unknown. In this study, we provide evidence that the E3 ubiquitin ligase Arkadia, whose major function is to enhance TGF-β signaling by degrading negative regulators of the pathway, exhibits tumor suppressing properties in CRC (Fig. 3 and Table 2). We found that two wild type alleles of Arkadia are required to maintain sufficiently low levels of these negative regulators (and therefore a high activity of the TGF-β pathway) and lower susceptibility for CRC development and progression under oncogenic stress in mice (Fig. 4 and Fig. 5). We did not find evidence of complete loss of Arkadia in invasive tumors from Akd^+/− mice suggesting that the tumor suppressing properties of TGF-β signaling depend on enhanced downstream responses. Furthermore, we screened for mutations that lower AKD function in patients with CRC exhibiting stabilization of SNON (26), a key repressor and substrate for AKD ubiquitination and degradation. We found several point mutations in the C-terminal functional domains of AKD that may account for the nuclear accumulation of SNON (Fig.1 and Table 1). As this is associated with a more advanced tumor grade, AKD emerges as a likely tumor-suppressor in CRC.

In the regenerating colonic epithelium, homeostasis is thought to be maintained by the WNT/β-catenin pathway that promotes proliferation of epithelial cells (and therefore exhibits oncogenic potential) and the TGF-β pathway, which promotes differentiation, cytostasis and apoptosis (tumor suppressive properties) (9). Mutations in WNT/β-catenin signaling, such as the constitutive activation of β-catenin or the inactivation of APC that regulates β-catenin activity, lead to hyperproliferation and adenoma formation. However, for progression to CRC, it is clear that there is a requirement for cooperation with other pathways (3, 28, 42). Mutations that inactivate TGF-β signaling have been shown to contribute to CRC progression (37-39, 41). However, whereas activation of WNT/β-catenin can be generated in a single hit (i.e. one allele of either APC or β-catenin), inactivation of a component of the TGF-β pathway usually has to occur in both alleles in the same cell. This is a very rare event and most likely occurs late after the accumulation of mutations that increase genomic instability. Gene amplification of negative regulators of the TGF-β pathway, such as SMAD6/7 and SNON/SKI, has been found in CRC (12, 13). However, amplification rarely occurs early enough to correspond to a tumor-promoting event, nor have single mutations leading to the overexpression of these factors been described.

In contrast, as Arkadia mediates the degradation of all the above negative regulators, and not only that of SNON, the simple reduction of its function by point mutations may lead to repression of TGF-β signaling responses and constitute an early tumor-promoting event. This is supported in this study by the identification of several missense mutations in AKD from CRC patients with stabilized nuclear SNON. One mutation in particular Q899STOP, introduces a stop codon at the C-terminus that eliminates the final 100aa harboring the ubiquitin ligase activity but preserves the substrate recognition domains (Fig. 1B). This truncated AKD is more stable (Fig. 1D), binds to the substrates and protects them from ubiquitination by wt AKD (Fig. 1C). In fact, a mutation that introduces a stop codon anywhere in the C-terminus between residue 889 and 978 (essential for enzymatic activity; Fig. 1B) would be expected to disrupt AKD function and exhibit a dominant-negative effect. Other missense mutations within this highly conserved region may also inactivate the enzymatic activity of Arkadia and result in a dominant-negative effect. The COSMIC database of somatic mutations in cancer reports two missense mutations in AKD (R904S and K915I), both in heterozygosity from primary Ovarian tumors (27). The R904S mutation is located in the center of the TIER domain (Fig. 1B). Deletion of the TIER domain results in a dominant-negative form of Arkadia (Supplemental Fig. S1), suggesting that R904S AKD may also act in this fashion. Therefore, each AKD allele exhibits a large region sensitive to single-hit mutations, which could act as an early event in the progression of adenomas to CRC and perhaps in other tumors that depend on TGF-β tumor suppression.

In the carcinogenesis model, both wt and Akd^+/− tumors displayed a similar level of nuclear β-catenin. However, only Akd^+/− tumors showed nuclear accumulation of SnoN and pSmad2 (Fig. 5), suggesting that repression of TGF-β signaling by SnoN, and most likely Ski (not examined in this study), is associated with susceptibility to CRC development and increased progression. In human CRC, we also reported a correlation between nuclear SNON/SKI accumulation and nuclear β-catenin (26). This suggests that tumors with WNT/β-catenin activation, combined with mutations that increase SNON, lead to a more aggressive CRC pathology. Therefore, as mutations inactivating just one allele of AKD can lead to stable nuclear SNON, combination with nuclear β-catenin may provide an early prognostic marker.

In this study, we have shown a strong association between susceptibility to CRC and repression of TGF-β signaling responses caused by loss of one allele of Arkadia in mice and a dominant-negative functioning mutant of AKD from a human CRC patient. Interestingly, in Akd^+/− colons and tumors, expression of the TGF-β target genes p21^WAF and SnoN are not obviously affected whereas p15^INK4b PAI-1, TMEPAI and Smad7 are reduced (Fig. 2E and 4C). Similarly, in Akd-null embryos and ES cells, not all TGF-β/Nodal target genes show equally reduced expression levels (20). Most likely SnoN/Ski do not repress all target gene promoters, suggesting that their targets are involved in the tumor suppressing function of TGF-β signaling. As mutations in AKD do not completely inactivate all TGF-β signaling responses, this will increase the risk of TGF-β-mediated metastasis later, which most likely involves a different subset of target genes. Collectively, our data suggest that somatic mutations in a single AKD allele that reduce its function could act as an early tumor-promoting event in human CRC development and progression.