SMAD4 Suppresses WNT-Driven Dedifferentiation and Oncogenesis in the Differentiated Gut Epithelium

The intestinal epithelium is strictly compartmentalized—with stem and proliferative cells nestled into the intestinal wall in the crypts of Lieberkühn and differentiated cells lining the lumen on villi in the small intestine or mucosal surface in the colon. Cells continuously transit from the crypt to the mucosal surface, where they are shed into the lumen. Stem cells and their niche-supporting Paneth cells are the only cell types that remain anchored to the crypt base and escape this cellular transit. This pattern of movement prevents accumulation of oncogenic mutations in differentiated cells, as differentiated cells will be lost into the lumen. Hence, stem cells are thought to be the cell of origin of colon cancer—a model elegantly supported by in vivo mouse studies (1). However, histologic evidence from human patients with colon cancer supports the notion that dysplastic cells arise from differentiated cells (2). Such a scenario requires the differentiated cells to acquire stem-like properties for oncogenesis (3, 4). The tumor cell of origin often dictates the epigenetic status of the tumor and can influence treatment strategies (5, 6).

Recent studies have shown that combinations of genetic alterations can trigger stem cell activity from differentiated intestinal epithelial cells in mice. These include simultaneous mutations in APC and KRAS as well as simultaneous activation of the WNT and NF-κB pathways (7, 8). The potential for plasticity in normal differentiated intestinal epithelial cells is also supported by experiments showing that upon loss of endogenous crypt base columnar (Lgr5⁺) stem cells, the progenitor cell populations of distinct intestinal lineages can dedifferentiate and acquire the fate of Lgr5⁺ cells, which are the actively cycling stem cells of the intestine (9). These studies have shown that cells previously referred to as “reserve” or “+4” cell populations include enteroendocrine progenitor cells that can dedifferentiate into active stem cells (10–13). In these dedifferentiation models, Lgr5⁺ stem cells are typically replaced by neighboring progenitor cells in the crypt, rather than cells luminally positioned on the more differentiated villus.

SMAD4 is a transcriptional effector of the BMP- and TGFβ-signaling pathways. The Cancer Genome Atlas (TCGA) research network has identified mutations in SMAD4 to be among the most frequently mutated genes in colon cancers (14), and loss of Smad4 heterozygosity in the Apc^min/+ mouse colon tumor background promotes invasive progression of colon adenomas (15). BMP signaling is required to maintain the differentiated state in the intestinal epithelium (16–18), as expression of BMP inhibitors beginning at embryonic stages leads to emergence of crypt-like structures in the villi of the adult tissue. However, the mechanisms and subsequent events leading to the loss of differentiation in the intestinal epithelium are incompletely understood, particularly whether adult-onset disruption of BMP signaling would lead to tumorigenesis from differentiated tissues.

The following study expands our understanding of differentiated-cell-derived tumorigenesis by demonstrating that simultaneous loss of SMAD4 and activation of the WNT pathway triggers stem cell properties and adenoma formation in the differentiated epithelium of the adult intestine.

The Villin-Cre^ERT2 transgene (19), Smad4^flox/flox (20), and Catnb^lox(ex3)/⁺ (21) alleles were integrated to generate the conditional compound mutants and controls. The Lgr5-EGFP-ires-CreERT2 allele (22) was used for mosaic Cre induction. Tamoxifen was administered at 0.05 g/kg mice, intraperitoneally, for 4 consecutive days to induce Cre recombination. All the experiments conducted had the approval of the Rutgers Institutional Animal Care and Use Committee.

Intestines were fixed with 4% paraformaldehyde overnight at 4°C and washed in PBS prior to dehydration and paraffin embedding. Mice were injected with 100 μg of EdU (Thermo Fisher Scientific, cat #C10337) 1 hour prior to sacrificing when the tissues were prepared for EdU labeling of proliferating cells. Five-micron sections were prepared from paraffin for histologic analysis. For immunostaining, antigen retrieval was performed with 10 mmol/L sodium citrate buffer (pH6) in a pressure cooker. Tissues were permeabilized with 0.5% triton. Antibodies used in the study were Ki-67 (Abcam, ab16667, 1:300), Smad4 (Santa Cruz Biotechnology, SC-7966, 1:500), OLFM4 (Cell Signaling Technology, 39141, 1:2,000), Sox9 (Millipore, AB5535, 1:2,000), BrdUrd (AbD Serotec, MCA2060GA, 1:500), lysozyme (DAKO, A 0099, 1:2,000), Keratin 20 (Cell Signaling Technology, D13063, 1:400), CHGA (SC-376827, 1:1,000, and DCAMKL (Abcam, ab37994, 1:1,000 (IHC-P), 1:200 (IHC-F), β-catenin (Santa Cruz Biotechnology, SC-7199, 1:250; Cell Signaling Technology, 9587, 1:1,000). Goat serum for blocking and secondary antibodies were used according to the manufacturer's protocol (ABC kit, Vector Labs, PK4001), and development of the color reaction after addition of 3, 3′-diaminobenzidine (Amresco, 0430) was monitored under a light microscope. Colabeling for Goblet cells and proliferation markers was performed by periodic-acid-schiff (PAS) staining followed by Ki-67 immunohistochemistry. Alkaline phosphatase staining kit II (Stegment, cat# 00-0055) was used to mark enterocytes, followed by Ki-67 immunohistochemistry. Click-it reagent kit (Thermo Fisher Scientific, Cat# C10337) was used to visualize the EdU-positive proliferating cells after fluorescent IHC for the various differentiation markers, KRT20 (1:200), DCAMKL (1:200) ChGA (1:200), and OLFM4 (1:200).

Nikon Eclipse E800 microscope and Retiga 1300CCD (Q-Imaging) cameras were used for bright field microscopy. For fluorescent microscopy, Zeiss Axiovert 200M microscope and Retiga-SRV CCD (Q-Imaging) QC imaging were used for image acquisition. Keynote, the apple software, was used for adjustments of sharpness and contrast and applied equally for comparative images. ImageJ software was used for merging fluorescent images.

Villi from the proximal half of the duodenum were isolated in the following manner with extreme care to avoid contamination from crypts. Mice of the indicated genotypes were treated with tamoxifen for 4 consecutive days. The proximal half of the duodenum was harvested on day 10 after the first tamoxifen injection, flushed with cold PBS, opened longitudinally and laid on clean 100 mm petri plate with villi facing up. Two histologic glass slides were used to isolate the villi by scraping: one to hold down the intestine, and the other to scrape. Only two villi-yielding scrapes were carried out (to avoid crypts). The villi collected were washed with PBS and isolated from smaller debris using a 70-μm filter with cold PBS to wash the villi and filter out any crypts if present. Fifty villi in 25 μL of Matrigel (Corning, 356230) were plated per well in a 48-well plate. The villi were cultured in ENR media (23) with 2 μg/mL primocin. Crypts were isolated from proximal duodenum and cultured in BME-R1 Matrigel (Trevigen, Cat#3433-005-R1) according to previously described methods (23).

RNA was extracted from jejunal epithelia of the indicated genotypes in triplicates following 4 consecutive days of tamoxifen injection. Uninjected mice served as control. After flushing the freshly harvested jejunum with cold PBS, the epithelia were dissociated from underlying mesenchyme by incubating with 3 mmol/L EDTA/PBS at 4°C as described previously (24) and filtered through 70-μm filter to isolate the crypts. The crypts were washed with PBS and pelleted to remove excess PBS prior to addition of TRIzol for RNA extraction. Sequencing was performed using Illumina's TruSeq RNA Library Prep kit v2 and data processed with version 2.2.1 of Cufflinks suite of tools (25). CuffDiff was used to call significant differentially expressed genes between the control and the Smad4^KO, as well as the control and double mutant. We used quartile normalization and per-condition dispersion. CuffNorm was used to calculate fragments per kilobase of transcript per million (FPKM) values using quartile normalization. These results were used for further downstream analysis. With the R package cummeRbund (version 2.18.0), the CuffDiff results were analyzed for significant differentially expressed genes, defined as having an alpha value of 0.05 or lower. GO term analysis was performed with DAVID (version 6.8). Using FPKM values calculated by CuffNorm, gene set enrichment analysis (GSEA) was used to examine differentially expressed genes using the signal-to-noise metric, as described (26). All data have been deposited in GEO (GSE102171).

Colorectal adenocarcinoma (COAD) samples from the TCGA project were used to investigate the relationship between SMAD4 and β-catenin signaling in human tumors. TCGA COAD Illumina HiSeq RSEM normalized gene expression (n = 434) and reverse phase protein array (RPPA; n = 493) data were acquired from the Broad Institute GDAC Firehose website (https://gdac.broadinstitute.org/) on February 23, 2018; the data set was filtered to include only primary COAD samples (n = 379). For protein-based analyses, each protein of interest was compared with SMAD4 by a Spearman rank correlation. To assess the relationship between SMAD4 mRNA levels and WNT signaling, each individual gene as well as the combined mean expression of the 67 genes identified by analyses of the Smad4^KO;β-catenin^GOF mouse model was analyzed by a Spearman rank correlation. The mean expression of the 67-gene WNT signature (Supplementary Tables S1 and S2) as well as specific significant genes are shown in the resulting heat maps. All resulting P values and correlations are reported in Supplementary Tables S1 and S2. A list of all samples is presented in Supplementary Tables S3 and S4.

At least 30 villi per biological replicate were used to evaluate the appearance EdU-positive (proliferative) or OLFM4-positive (stem cells) over time in the villi of Smad4^KO; β-catenin^GOF following tamoxifen induction. Between 296 and 334 cells expressing the differentiation markers were counted to determine the number of differentiated cells coexpressing differentiation and proliferation markers in the villi of Smad4^KO;β-catenin^GOF mice at 7 days after tamoxifen injection. The proximal two thirds of the duodenum was used in all the above cases. Ectopic crypts were quantified by taking into account the fraction of villi that display ectopic crypts in the differentiated compartment along the various regions of intestine. In the Lgr5-Cre model, each SMAD4-negative adenoma was scored for whether the underlying crypts were SMAD4 positive or SMAD4 negative. Adenomas were counted on one swiss-roll preparation of the proximal two thirds of the duodenum per biological replicate.

Smad4 is a critical transcriptional effector of TGFβ/BMP signaling and a tumor suppressor (27). Loss-of-function mutations in SMAD4 are found in juvenile polyposis syndrome in humans (28). Functional studies in mice implicating Smad4 function in the intestine were often conducted using heterozygous germ line mutations (15, 29, 30), mosaic mutations (31) or T-cell–specific Smad4 deletion (32). To study SMAD4 function specifically in the adult intestinal epithelium, we used the Villin-Cre^ERT2 transgenic model (19), which allows inactivation of floxed alleles of Smad4 (20) in the intestinal epithelium upon tamoxifen injection. Four days after the onset of tamoxifen treatment of Smad4^f/f;Vil-Cre^ERT2 mice, SMAD4 protein expression in the intestinal epithelium was observed by immunohistochemistry and immunoblotting and found to be below the level of detection; qRT-PCR and RNA-seq corroborated severe reduction of Smad4 transcripts after tamoxifen treatment (Fig. 1A and B; Supplementary Fig. S1). Histologic analysis (Fig. 1A) revealed an overtly normal intestinal epithelium without notable alterations in the various differentiated lineages: enterocytes (alkaline phosphatase activity), tuft cells (DCAMKL), goblet cells (PAS), enteroendocrine cells (ChgA), and Paneth cells (lysozyme). Immunostaining for OLFM4, a marker of crypt base columnar stem cells, suggested the stem cell population was unaltered. Consistent with these observations, the Smad4 knockout mice remained viable and appeared healthy during the period of observation for over 6 months. Thus, mice are viable upon losing SMAD4 in the mature intestinal epithelium and exhibit overtly normal intestinal homeostasis, indicating that SMAD4 is surprisingly dispensable in the adult intestinal epithelium, at least over a span of several months.

Although normal cell lineage distributions and tissue morphology were observed in the Smad4^f/f; Villin-Cre^ERT2 intestinal epithelium at 4 days after tamoxifen induction, notable differences were observed in the transcriptome of these tissues. RNA-seq was performed in 3 biological replicates from jejunal epithelium of control and Smad4^KO mice. Six hundred and forty-one genes were defined as differentially regulated (q < 0.05), with 343 transcripts increasing and 298 transcripts decreasing in the mutant including Smad4 transcripts (Fig. 1B). Canonical readouts of the SMAD4 signaling pathway also exhibited markedly reduced transcript levels, including negative feedback regulators Smad6 and Smad7, and members of the ID family of transcription factors (Fig. 1C). These findings confirmed disruption of SMAD4 function in our model. Gene ontology analysis revealed that transcripts reduced in the Smad4-mutant tissues were associated with properties of differentiated intestinal epithelial cells such as lipid metabolism, oxidative phosphorylation, and localization to the brush border (Fig. 1D). Other transcriptomic signatures that were disrupted in the Smad4-mutant epithelium included decreased levels of transcripts associated with differentiation and increased levels of transcripts associated with the WNT signaling pathway (Fig. 1E–G).

Thus, while Smad4 loss has no immediate consequence on intestinal epithelial morphology or cellular distributions, there is a shift at the transcriptomic level, with reduced expression of BMP/TGFβ signaling target genes, and elevated expression of signature genes of the WNT pathway (33).

Intestinal neoplasia is observed when Smad4 is deleted in T lymphocytes (32), and loss of heterozygosity in mice with germline mutations in Smad4 results in duodenal polyps within 9 months of age (29). However, the epithelial-specific function of Smad4 in suppressing tumors upon adult-onset loss is unclear. To determine if the observed molecular alterations in Smad4^KO intestinal epithelium can lead to tumorigenesis, we sacrificed tamoxifen-injected Smad4^f/f;Vil-Cre^ERT2 mice within a period 6 to 9 months after SMAD4 loss (Fig. 2A). Four of 6 mice with the Smad4^KO intestinal epithelium developed tumors, which were primarily observed in the small intestine (Fig. 2B) and histologically resembled conventional adenomas (Fig. 2C and D). Tumors featured highly glandular structures, nuclear β-catenin, increased expression of OLFM4, a marker for crypt-base columnar stem cells, and SOX9, a marker enriched in crypts (Fig. 2E)—all properties of conventional adenomas. Thus, while acute SMAD4 loss is well tolerated (Fig. 1), prolonged deficiency of SMAD4 in the intestinal epithelium leads to adenomas (Fig. 2).

The high expression of nuclear β-catenin in the tumors of the Smad4^KO epithelium (Fig. 2E) and the elevated WNT gene expression signature upon acute SMAD4 loss (Fig. 1) suggested that further elevation of the WNT pathway might be required for tumor development. We explored the relevance of this possibility by examining RNA expression data (n = 379) from the TCGA project using COAD samples. The WNT signature that we find elevated upon loss of SMAD4 in our mouse model (Fig. 1G) was inversely correlated (P = 3.1E−05; r = −0.21) with SMAD4 mRNA expression (Fig. 2F); 56/67 genes in the WNT signature were negatively correlated, with 34/67 being statistically significant (Supplementary Table S1). These data collectively demonstrate a significant inverse relationship between SMAD4 and β-catenin/WNT signaling in human colon tumors. To further test the possibility that WNT pathway activation would facilitate tumor formation in the Smad4 mutant, we challenged Smad4^KO mice with elevation of WNT signaling using a single allele of an inducible β-catenin–activating mutation, Catnb^lox(ex3), in the intestinal epithelium. The heterozygous Catnb^lox(ex3)/+ gain-of-function (GOF) allele permits deletion of the third exon of Ctnnb1, which encodes phosphorylation sites that target β-catenin for degradation (21). We chose this model, instead of other available β-catenin transgenic models, because it allows expression of the stabilized β-catenin from its endogenous locus and efficiently promotes proliferation in the intestinal epithelium (Supplementary Fig. S2). Thus, tamoxifen treatment of Smad4^f/f;Catnb^lox(ex3)/+;Vil-Cre^ERT2 mice results in concomitant SMAD4 loss and WNT activation (referred to throughout as Smad4^KO;β-catenin^GOF). Remarkably, 14 days after tamoxifen treatment, ectopic, crypt-like structures are observed in the villi of these double-mutant mice (Fig. 3A). The ectopic crypt-like structures appearing in Smad4^KO;β-catenin^GOF-mutant mice were similar to structures observed previously upon fetal introduction of BMP-signaling antagonists (16, 18), but with drastically reduced latency. We first saw the proliferative marker Ki-67 in cells with the morphologic appearance of enterocytes in the villi at 4 days after initiation of tamoxifen treatment (Fig. 3B). By 7 days, clusters of proliferative cells were invaginating toward the villus mesenchyme, and by 10 days, these structures nestled more deeply into the villus body and resembled crypts. These cup-shaped structures continued to elongate and/or branch for another week, at which time the mice became moribund and were humanely euthanized. Immunohistochemical analyses revealed markers of stem cells (OLFM4), Paneth cells (lysozyme), and SOX9-expressing cells in the ectopic crypts (Fig. 3C). Ectopic crypts were most frequently found in the proximal intestine but were also present throughout the small intestine and colon (Fig. 3D)—the ectopic crypts in the small intestine being more perceptible owing to the presence of villi (Fig. 3E). Importantly, ectopic crypts are not observed in the villi of control, Smad4^KO, or β-catenin^GOF single-mutant mice alone (Supplementary Fig. S3).

As the ectopic crypts advanced to 14 days post-tamoxifen treatment, cells of all the differentiated lineages could be observed, suggesting that stem cells residing within ectopic crypts are multipotent (Supplementary Fig. S4). We further assayed for crypt-like properties of these villus-resident structures by culturing the villi ex vivo (Fig. 4A), in conditions mimicking the endogenous crypt environment (Sato, 2009). Villi isolated from the Smad4^KO;β-catenin^GOF mice gave rise to organoids (Fig. 4B). Organoid precursors within the villi were visible within 4 days of culture, developed into organoids within 7 days of culture (Fig. 4C and D), and resembled wild-type organoids grown in WNT-conditioned media in their propensity to form spheres (34). Notably, villi from control, Smad4^KO, or β-catenin^GOF single-mutant mice did not form organoids (0 of 300 villi) in Matrigel, consistent with lack of ectopic crypts in the villi of these mutants (Fig. 4C). These organoids could be subcultured repeatedly (Fig. 4D). Thus, combined inactivation of SMAD4 and activation of β-catenin lead to the appearance of functional, ectopic crypts that express characteristics and functions of normal intestinal crypts, including the ability to give rise to organoid cultures.

To better appreciate the potential mechanisms through which concomitant loss of SMAD4 and activation of β-catenin triggers ectopic crypt formation, we investigated transcriptomic changes in the double-mutant mice compared with controls at 4 days after the onset of tamoxifen treatment (at the onset of ectopic proliferation in the villus). A total of 1,678 genes were downregulated, while 1,005 transcripts were upregulated in the double mutant (q < 0.05). Gene ontology analysis revealed that genes ascribed functions typical of differentiated cells were decreased, while genes with elevated transcripts were associated with secretory processes (Fig. 5A). We also noted an increased expression of intestinal stem cell signature genes in the double mutants, consistent with reports indicating that loss of SMAD4 (35) or activation of the WNT pathway promotes expansion of stem cells (Fig. 5B). We next queried the expression data for gene signatures that have been observed in previous models of villus-resident proliferation. Upon loss of the cell-cycle suppressor Rb, villus cells have been observed to proliferate (36, 37). Changes in the transcriptome in Rb-mutant mice were correlated to those observed in our Smad4^KO;β-catenin^GOF model (Fig. 5C). It has also been shown that activation of the NF-κB pathway, in conjunction with elevated WNT signaling, can trigger ectopic crypt formation in villi (7). Interestingly, transcriptomic data from mice in which the NF-κB signaling pathway is perturbed (38) also correlated with changes observed in our Smad4^KO;β-catenin^GOF model (Fig. 5D), though the correlation was less robust. The relevance of these observed relationships could be corroborated in human patient samples. RPPA data from the TCGA project were used to assess the relationship between SMAD4 and β-catenin protein expression in 493 tumor samples and determined, consistent with our mouse model, that expression of these proteins are inversely correlated in human COAD (Spearman correlation; P = 4.96E−25, r = −4.42E−01; Fig. 5E). We further determined that both the phosphorylated (S536) p65 subunit of NF-κB (P = 8.02E−15, r = −3.40E−01) and phosphorylated (S807, S811) Rb (P = 3.23E−13, r = −3.20E−01) were negatively correlated with SMAD4 expression (Fig. 5E; Supplementary Table S2).

Together, these findings suggest that activation of β-catenin in the absence of SMAD4 may share common mechanisms used by Rb-loss or NF-kB activation to drive proliferation in the differentiated epithelium, and thus contributes to a growing appreciation of the proliferative capacity in differentiated tissues.

Induction of the Smad4^KO;β-catenin^GOF mutations in the intestine in a pan-epithelial manner with Villin-Cre^ERT2 renders the mice moribund within weeks, as virtually all epithelial cells become mutant. To monitor the progress of ectopic crypts over time, we induced these mutations in a mosaic manner using the Lgr5-Cre^ERT2-IRES-GFP expression cassette (22), which induces recombinase activity in ∼30% of the Lgr5⁺ stem cells, as we have observed in our previous studies (24). Thus, tamoxifen injection of Smad4^f/f;Catnb^lox(ex3)/+;Lgr5-Cre^ERT2-IRES-GFP mice results in Smad4^KO;β-catenin^GOF mutations in only about one third of the intestinal crypts, while the majority of crypts retain the wild-type protein expression for Smad4 and β-catenin. The reduced numbers of cells harboring mutations in this model should allow the mouse to survive longer than the Villin-Cre^ERT2 model. The mosaic mutation was tolerated in the mice for up to 4 weeks. Histology revealed adenoma development in both small and large intestine (Fig. 6A). SMAD4 immunohistochemistry indicated that while the villus-resident adenomas lacked SMAD4 expression, the normal-looking villi, and the majority of the crypts retained Smad4 expression (Fig. 6A and B). This observation suggests that despite induction of the mutation in the crypt stem cells, SMAD4-mutant cells within the stem cell compartment were replaced by adjacent SMAD4-positive stem cells over time. Notably, the villus-resident adenomas appeared to be discontinuous from the normal crypt base, as an average 46 SMAD4-negative adenomas were observed in villi overlaying morphologically normal, SMAD4-positive crypts (per duodenal sample, SD = 5.65, n = 2), whereas SMAD4-negative crypts were in contact with adenomas in only 2.5 adenomas per mouse (SD = 0.7). Analogous structures were also observed in the luminal surface of the differentiated colon epithelium (Fig. 6A), suggesting that Smad4 suppresses WNT-driven adenomas in the differentiated cells. Developing adenomas resident in the differentiated compartments of the small and large intestine expressed characteristics of adenomatous polyps (Fig. 6C); however, elevated levels of WNT signaling did not appear to rise to the levels observed in aged SMAD4^KO mice, based upon β-catenin immunoreactivity (Fig. 2E; Supplementary Fig. S5). Notably, the location of adenomas within the villi of Smad4^f/f;Catnb^lox(ex3)/+;Lgr5-Cre^ERT2 mutants contrasts with most intestinal tumor models in which the lesions expand from the crypt base (1). This suggests that SMAD4 plays an active role in suppressing WNT-driven adenoma formation within the differentiated intestinal epithelium.

Villi in the normal intestinal epithelium harbor only postmitotic, differentiated cells. Because stem cells can give rise to all of the cell types in the intestinal epithelium, we sought to determine whether the dedifferentiating cells in Smad4^KO;β-catenin^GOF-mutant villi first reenter cell cycle or first express markers of stem cell fate. We investigated the kinetics with which villus cells either incorporate a marker of DNA replication (EdU) or express the stem cell marker OLFM4. EdU labeling was done with a brief, 1 hour pulse chase to assess cells recently in S-phase. EdU-labeled cells first appeared in the villi at 4 days after the onset of tamoxifen treatment of the Smad4^KO;β-catenin^GOF mice, whereas OLFM4-positive cells did not appear until 24 hours later, and at a much lower frequency (Fig. 7A). This trend continued at 6 days after initiation of tamoxifen treatment, at which time OLFM4-positive cells were observed in most duodenal villi but still were outnumbered by EdU⁺ cells (Fig. 7B). These results suggest that villus cells reenter the cell cycle before acquiring stem cell properties.

The initial appearance of proliferating cells in the Smad4^KO;β-catenin^GOF villi occurred without any apparent changes in differentiation of intestinal lineages. To pinpoint the identity of villus cells first expressing proliferative markers, dual staining for proliferation and differentiation markers for the various intestinal lineages was conducted. Of each lineage interrogated, only cells expressing enterocyte markers (alkaline phosphatase or KRT20), were found coexpressing Ki-67 or taking up the uridine analog, EdU (Fig. 7C and D). The coexpression of proliferation markers with enterocyte lineage markers indicates that cells committed or partially committed to the enterocyte lineage were particularly capable of contributing to villus-resident mitosis. Conversely, markers of tuft (DCAMKL), enteroendocrine (CHGA), Paneth (lysozyme), or goblet cells (PAS) did not colocalize with proliferation markers (Fig. 7C). Thus, loss of SMAD4 allows villus-resident enterocyte-like cells to undergo proliferation upon elevation of WNT signaling. Formally, lineage-specific Cre drivers would be required to demonstrate that these nonenterocyte cell lineages are not the source of ectopic crypts, whereas enterocytes are the definitive source. Still, these costaining results suggest that SMAD4 protects against WNT-driven dedifferentiation of enterocytes.

Elegant studies in mice have demonstrated the potential of active WNT signaling to drive oncogenesis from stem cells, but not in early progenitor or differentiated intestinal epithelial cells (1). Histologic evidence from human patients with colon cancer, however, suggests that dysplastic cells arise from differentiated cells (2). Our studies find that active WNT signaling can drive dedifferentiation and adenoma formation in differentiated intestinal epithelial cells upon concomitant SMAD4 loss in the adult.

SMAD4 can serve as the transcriptional effector of both BMP and TGFβ signaling. Although our findings do not distinguish which ligand SMAD4 functions through to suppress dedifferentiation, recent work would support BMP ligands as the suppressors of dedifferentiation. Loss of Tgfbr1 was recently shown to accentuate dedifferentiation driven by APC-loss and KRAS activation; however, Tgfbr1 loss in conjunction with activation of WNT signaling does not lead to dedifferentiation (39). Notably, it is the expression of BMP antagonists, GREM1 and NOGGIN, that have been demonstrated to trigger ectopic crypt formation (16, 18). Suppression of hedgehog signaling also triggers ectopic crypt formation in the developing mouse, in a mechanism proposed to operate via BMP signaling. Therefore, taken together with our work, these findings point to BMP signaling rather than TGFβ signaling, in a SMAD4-dependent mechanism, as the primary suppressor of dedifferentiation in the intestinal epithelium. Our finding also refines the mechanism to a cell-autonomous role for SMAD4 in the epithelium. Disentangling how the cells are interpreting the BMP versus TGFβ ligands to suppress dedifferentiation will be an important next step in understanding how the epithelium protects against oncogenesis.

In this study, we observe that enterocytes, but not other differentiated lineages, reacquire stem cell marker expression and the ability to reenter the cell cycle upon loss of SMAD4 and gain of WNT signaling. Certainly, other intestinal lineages have been shown to be capable of acquiring stem cell activity upon targeted loss of endogenous stem cells or upon conditions of tissue regeneration, including tuft cells, Paneth cells, and secretory precursor cells (11–13, 40, 41). We hypothesize that the SMAD4 phenotype is specific to the enterocyte lineage in that it plays specific roles in promoting enterocyte differentiation, whereas other differentiation-promoting factors, unique to these other lineages, may require suppression in order for dedifferentiation of these lineages. Enterocytes have been shown to have the capacity to dedifferentiate upon ablation of the Lgr5⁺ stem cell population, but the same study demonstrated that enterocytes do not transform upon simultaneous activation of WNT and KRAS (9). In light of these findings, our study results suggest that SMAD4 may mediate a specific role in blocking WNT-KRAS-mediated oncogenesis from enterocytes.

How does SMAD4 block adenoma formation in differentiated tissues? SMAD4 has recently been shown to suppress stem cell renewal in the crypt base (35, 42); its loss in differentiated tissues, upon exposure to stem cell–promoting WNT signaling, could thus create conditions favorable for the stem cell fate, consistent with our observations in the mutant transcriptome (Fig. 5B). We also observe increased expression of genes functioning in extracellular exosome formation (Fig. 5A), which could play a role in carving a stem cell niche in the villus. Perturbations of NF-κB signaling (7) or loss of the cell-cycle inhibitor Rb (36) can also lead to proliferation in the villus, triggering transcriptomic changes that correlate with the transcriptomic changes we observe in our mutants (Fig. 5C and D), and which are also correlated with SMAD4-low human tumors (Fig. 5E). Future studies can investigate whether these pathways might work upstream, downstream, or in parallel with BMP/SMAD4 signaling to regulate dedifferentiation of enterocytes and will broaden our understanding of the potential cells of origin of colorectal cancer.

No potential conflicts of interest were disclosed.

Conception and design: A.O. Perekatt, N. Patel, M.L. Gatza, M.P. Verzi

Development of methodology: A.O. Perekatt, A. Wu, N. Patel, J. Xing

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A.O. Perekatt, P.P. Shah, V. Gandhi, Q. Feng, N. Patel, L. Chen, S. Joshi, N. Gao

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A.O. Perekatt, P.P. Shah, S. Cheung, N. Jariwala, N. Kumar, Q. Feng, L. Chen, S. Joshi, A. Zhou, J. Xing, M.L. Gatza, M.P. Verzi

Writing, review, and/or revision of the manuscript: A.O. Perekatt, P.P. Shah, S. Cheung, N. Jariwala, J. Xing, M.L. Gatza, M.P. Verzi

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): P.P. Shah, S. Cheung, Q. Feng, M.M. Taketo

Study supervision: A.O. Perekatt, J. Xing, N. Gao, M.L. Gatza, M.P. Verzi

Other (participate in conducting the experiment): A. Wu

Other (provided scientific input): E. White

This research has been funded by grants from the NIH, NCI (R01CA190558), and the New Jersey Health Foundation (PC46-16) to M.P. Verzi, the New Jersey Commission on Cancer Research (DFHS13PPC034) to A.O. Perekatt, the Aresty Undergraduate Research Program (N. Patel, S. Cheung, and P.P. Shah), the Department of Life Sciences Summer Undergraduate Research Fellowship (P.P. Shah), MacMillan Cancer Research grants to (P.P. Shah and S. Cheung), the National Cancer Institute of the National Institute of Health (CA166228), and the V Foundation for Cancer Research (V2016-013) to M.L. Gatza. M.P. Verzi, E. White, and M.L. Gatza are members of the Cancer Institute of New Jersey (P30CA072720), which provided core resources for sequencing. The work was also supported by an Initiative for Multidisciplinary Research Teams award from Rutgers University, Newark, NJ (N. Gao and M.P. Verzi).