Aberrant cell state plasticity mediated by developmental reprogramming precedes colorectal cancer initiation

Impaired differentiation is a conserved event in multiple mouse models of intestinal neoplasia

To study the molecular underpinnings of aberrant cell state plasticity, we used two genetically engineered and two carcinogen-induced murine models of intestinal neoplasia. Lgr5^{eGFP-IRES-CreERT2};Apc^{loxP-exon14-loxP}(Lgr5^Cre;Apc^f/f) genetically engineered mice develop hundreds of adenomas in the small intestines following tamoxifen-induced conditional deletion of Apc (33) in Lgr5⁺ ISCs (fig. S1, A and B) (10, 34). Compared to controls, adenomas from Lgr5^Cre;Apc^f/f mice displayed elevated mRNA expression of canonical WNT downstream targets (Axin2 and Myc) and stem cell genes (Ascl2, Lgr5, Prom1, and Lrig1), including Sox9 (Fig. 1B), consistent with known biology (35, 36). In contrast, markers of differentiated intestinal cells (e.g., Cdx2) of both absorptive (e.g., Krt20) and secretory lineages (e.g., Muc2) (37) were down-regulated in Lgr5;Apc^f/f intestinal adenomas (Fig. 1B). To validate these findings, we examined another previously published genetically engineered mouse model in which inducible Apc knockdown (KD) with or without mutant K-ras^G12D activation in Lgr5⁺ cells displayed robust formation of colonic adenomas (27). In agreement with the Lgr5^Cre;Apc^f/f model, genes associated with stem cell activity were up-regulated, whereas intestinal differentiation genes were suppressed in Apc KD colonic adenomas by bulk RNA sequencing (RNA-seq; fig. S1C). We also analyzed protein expression of Sox9 in early intestinal adenomas from Lgr5^Cre;Apc^f/f mice that were crossed with R26^LSL-tdTomato (R26^tdT or tdT^LSL) mice to label Apc^KO cells. Compared to the adjacent normal intestinal epithelium in which Sox9 expression is restricted to the crypt base, tdT⁺ Apc^KO lesions demonstrated elevated expression of Sox9 throughout the lesion, which coincided with the absence of differentiation markers Krt20 and Muc2 (Fig. 1C). These observations suggest that Apc inactivation is sufficient to impair intestinal differentiation and leads to inappropriate Sox9 expression in adenomas.

We next evaluated two mouse models of carcinogen-induced small intestinal and colonic neoplastic lesions to ask whether Sox9 activation and impaired differentiation are universal events in cancer initiation that can be achieved by random mutagenesisin vivo. The first, well-established model consists of exposing mice to the procarcinogen azoxymethane (AOM) and irritant dextran sodium sulfate (DSS) (Methods), which leads to colonic lesions ranging from low-grade dysplasia to intramucosal adenocarcinoma after several months. Consistent with our observations in the Lgr5^Cre;Apc^f/f;R26^tdT model, high-grade dysplastic colonic lesions demonstrated robust Sox9 overexpression and loss of Krt20⁺ enterocytes and Muc2⁺ goblet cells (Fig. 1D and fig. S1D). In the second model, mice were exposed to drinking water containing N-methyl-N-nitrosourea (MNU), a potent carcinogen with gastrointestinal tropism, and a subset developed poorly differentiated intestinal carcinomas after 1 year. Lgr5^Cre-eGFP mice treated with MNU showed marked up-regulation of Sox9 in premalignant and adenocarcinoma lesions (Fig. 1E and fig. S1E), consistent with an early role in cancer development. Differentiation markers Krt20 and Muc2 were absent in premalignant and adenocarcinoma lesions. Elevated expression of Lgr5 as indicated by enhanced green fluorescent protein (eGFP) immunohistochemistry (IHC) was found in intestinal adenocarcinomas but not their associated premalignant lesions, underscoring the central importance of stem cell activation in progression to malignancy (Fig. 1E). These results implicate impaired differentiation and Sox9 activation as a common pathway to CRC initiation.

To characterize the transcriptional program enabling impaired differentiation, we performed single-cell RNA-seq (scRNA-seq) on the Lgr5^Cre;Apc^f/f;R26^tdT genetic model. tdT⁺ epithelial cells were isolated from Lgr5^Cre;Apc^f/+;R26^tdT (controls that are phenotypically wild type) and Lgr5^Cre;Apc^f/f;R26^tdT mice by fluorescence-activated cell sorting (FACS) 28 days following tamoxifen induction (Fig. 1F). Unsupervised clustering followed by uniform manifold approximation and projection (UMAP) representation of single-cell gene expression profiles revealed that control Apc^WT cells displayed the expected distribution of cell lineages based on intestinal markers (fig. S2, A to C) (38). In contrast, Apc inactivation led to a fourfold enrichment in Paneth cells (5.9% versus 1.5%) and the emergence of cells that exhibited a new transcriptional state (63.1% versus 0%) at the expense of normal ISCs and differentiated cell types, including absorptive progenitors, mature enterocytes, goblet cells, and, to a lesser extent, enteroendocrine cells (Fig. 1G). tdT⁺ Apc^KO intestinal lesions demonstrated a greater number of Paneth cells by Lyz IHC (fig. S3A), which can also be found at earlier time points of Apc inactivation (35, 39). Apc deletion led to a greater percentage of Lgr5⁺ cells as indicated by eGFP FACS (fig. S2D), suggesting that ISC activity is up-regulated in Apc^KO cells, agreeing with a vast literature (40) and supported by the enrichment of published ISC signatures (table S1, Fig. 1H, and fig. S2A) (38, 41). A deeper evaluation revealed that select canonical ISC markers were not consistently found in Apc^KO cells, including Olfm4 (fig. S3, B and C). In contrast, Apc^KO cells expressed higher levels of Sox9 compared to normal ISC (Fig. 1H), which was supported by universal Sox9 expression in tdT⁺ intestinal lesions from Lgr5^Cre;Apc^f/f;R26^tdT mice (Fig. 1C). Furthermore, evaluating transcription factor perturbation gene sets that were enriched among up-regulated genes in Apc^KO cells (table S1) revealed that a Sox9 overexpression transcriptional program scored the strongest [Fig. 1I (adjusted P = 3.26 × 10⁻²²) and table S2]. These observations suggest that Apc deletion leads to a new transcriptional state with selective ISC transcriptional activity.

Reactivation of genes associated with fetal intestines upon Apc loss

Despite showing enrichment of most canonical ISC genes, the transcriptional profile of Apc^KO cells was found in one distinct cluster (Fig. 1G), expressing genes not found in normal intestinal cells; we, therefore, refer to these cells as aberrant stem cell–like (AbSC). To characterize the AbSC transcriptional program in greater depth, searching for features that explain its distinguished gene expression profile, we examined the differentially expressed genes in the AbSC cluster relative to all other intestinal clusters. Consistent with our previous results, genes associated with differentiated enterocytes (blue) were down-regulated, whereas ISC markers and WNT pathway targets (green) were up-regulated (Fig. 2A). Genes associated with interferon (IFN) signaling (fig. S4, A and B) were significantly up-regulated in AbSCs, which matched gene ontology analyses implicating IFN pathways and transcription factors (e.g., RELA and nuclear factor κB; fig. S4C and table S2). Genes associated with regeneration (pink) and fetal intestines (purple) were also uniquely and significantly up-regulated in the AbSC cluster (Fig. 2A). A regeneration gene expression signature derived from an experimental intestinal helminth infection model [Nusse et al. (42); table S1] was specifically up-regulated in AbSCs (Fig. 2B). Notably, this gene expression signature is also associated with IFN-γ signaling, indicating a potential link between regeneration and IFN-γ signaling (42).

Among regeneration genes within AbSCs, Ly6a and Ly6e consistently showed the strongest and most abundant expression (Fig. 2C and fig. S4D). While Apc inactivation led to minor overall chromatin accessibility changes at regeneration genes (fig. S4E), there was greater access at Ly6a and Ly6e genomic loci by assay for transposase accessible-chromatin (ATAC-seq; Fig. 2C and fig. S4D) (43). We decided to focus on Ly6a [also referred to as stem cell antigen-1 (Sca-1)] as a marker for validation studies because of its (i) selective high-level expression in a substantial fraction of AbSC cells (63.1%), (ii) absence from normal intestinal clusters, and (iii) association with stem cell properties in the hematopoietic system (44–46). Greater chromatin accessibility at the Ly6a genomic locus upon Apc deletion corresponded to elevated Ly6a mRNA expression by reverse transcription polymerase chain reaction (RT-PCR; fig. S4F). Lgr5^Cre;Apc^f/f;R26^tdT mice displayed selective Sca-1 expression in tdT⁺ intestinal lesions and, within these lesions, Sca-1 was expressed in 30 to 40% of tdT⁺ cells (Fig. 2D). Sca-1⁺ cells also expressed Sox9, which was found in approximately 85% of tdT⁺ cells (Fig. 2D) but did not consistently express Lgr5/eGFP or Olfm4 (fig. S4, G and H). To examine self-renewal properties of Sca-1⁺ AbSCs, we performed organoid-forming assays using tdT⁺ cells isolated by FACS. Consistent with in situ observations, ~28% of tdT⁺ cells from Lgr5^Cre;Apc^f/f;R26^tdT mice demonstrated Sca-1 protein expression by FACS, whereas tdT⁺ cells from Lgr5-tdT mice showed no expression (fig. S4I). Per 1000 cells plated, Sca-1⁺ cells showed a 7.4-fold greater organoid-forming ability than Sca-1⁻ cells isolated from Lgr5^Cre;Apc^f/f;R26^tdT mice (fig. S4, J and K). Dissociated Sca-1⁺ organoids passaged into secondary cultures maintained a 5.9-fold greater organoid forming ability than Sca-1⁻ cells (fig. S4L), suggesting population stability and preserved regenerative properties. These data suggest that sustained regenerative transcriptional activity is a feature of neoplastic initiation and contributes to self-renewal capacity.

Fetal intestinal genes were also up-regulated in AbSCs (Fig. 2A). Consistently, two relatively nonoverlapping fetal-like gene signatures [Vallone et al. (47) and Mustata et al. (48); table S1] mapped specifically to the AbSC cluster (Fig. 2E and fig. S5, A and B). To confirm these associations, we also interrogated bulk gene expression data directly extracted from intestines at different stages of embryonic development (1). AbSC down-regulated genes had stronger expression in late development [embryonic day 18 (E18)] and adult intestines compared to earlier stages of intestinal development (Fig. 2F, E12 to E16). By contrast, AbSC up-regulated genes are significantly enriched among genes preferentially expressed in embryonic relative to adult intestines (Fig. 2F), which include Sox9, Tacstd2 (encoding Trop2), and Ly6e. Among fetal intestinal genes, Tacstd2 showed the most specific expression in AbSCs, marking at least 11% of the cluster (Fig. 2G), which was confirmed by Trop2 IHC in intestinal lesions from the genetic mouse model (Fig. 2D). While greater chromatin accessibility was found at fetal genes upon Apc inactivation (fig. S5C), new accessibility was observed at the Tacstd2 genomic locus (Fig. 2G). These analyses suggest that expanded cell state plasticity by gaining access to and transcriptionally activating fetal intestinal genes is associated with intestinal neoplasia.

Regenerative and fetal-intestinal gene signatures appeared to occupy distinct compartments of the AbSC cluster, which was confirmed by a coexpression analysis of Tacstd2 and Ly6a (fig. S5D), higher resolution reclustering of AbSCs (fig. S5E), and relative nonoverlapping Sca-1 and Trop2 IHC in tdT⁺ intestinal lesions (Fig 2D). Pseudotime analysis of AbSCs indicated that gene expression associated with regeneration preceded that of fetal intestines (Fig. 2H). To examine this possibility, we performed scRNA-seq on tdT⁺ cells isolated from Lgr5^Cre;Apc^f/f;R26^tdT mouse intestines 10 days following tamoxifen-induction (fig. S5F). As expected, AbSC activity was significantly elevated 10 and 28 days following Apc inactivation (fig. S5G). While Ly6a expression was stably elevated at days 10 and 28 (fig. S5H), fetal intestinal gene expression, including Tacstd2, gradually increased over time (Fig. 2I and fig. S5I), consistent with regeneration preceding developmental reprogramming. These data also suggest that nongenetic factors are responsible for fetal gene activation following regenerative transcriptional activity.

Carcinogen-induced colon neoplasia demonstrates developmental reprogramming

To define features of AbSC in an independent model of colon neoplasia, we performed scRNA-seq of a neoplastic lesion and paired a normal sample from a mouse exposed to AOM/DSS. UMAP representation of single-cell gene expression profiles of epithelial cells revealed known cell types in the normal colon and a distinct transcriptional profile in the colonic lesion (Fig. 3A). AbSC gene signature derived from our genetic model demonstrated the strongest expression in a neoplastic cluster furthest from normal tissue profiles and weaker expression in the remaining neoplastic cells and normal ISCs (Fig. 3B). Neoplastic cells demonstrated greater Ly6a expression by scRNA-seq (fig. S6A) and Sca-1 staining by IHC, the latter of which showed specific expression in lesions without expression in adjacent or distal normal tissue (Fig. 3E and fig. S6B). Notably, expression of the fetal intestinal gene signature broadly (Fig. 3C) and Tacstd2 specifically (Fig. 3D and fig. S6C) was restricted to neoplastic cells. Consistent with these results, strong Trop2 expression was only found in premalignant and malignant lesions (Fig. 3E), corresponding to elevated Sox9 expression and attenuation of differentiation markers. These results demonstrate that a carcinogen-induced colon neoplasia model demonstrates developmental reprogramming in neoplastic cells, particularly marked by Trop2 positivity.

To validate and extend these findings, we analyzed a recently published bulk RNA-seq dataset derived from AOM/DSS-treated mice at four different time points of colon neoplastic evolution (49) (Fig. 3F). Compared to untreated controls, transcriptional profiles of colonic lesions showed enhanced stem cell activity and impaired differentiation over time (fig. S6, D and E). Colonic lesions at week 10, the latest time point evaluated, showed significant enrichment of regeneration and fetal intestinal gene signatures compared to untreated controls (Fig. 3, G and H). The regeneration program peaked at week 2 and then gradually declined over time (Fig. 3, G and I). In contrast, the fetal intestinal gene signature gradually increased over time, showing the strongest levels at week 10 (Fig. 3, H and I). These transcriptional program kinetics are consistent with histopathological evaluation of colonic lesions from our AOM/DSS model, which demonstrated robust Sca-1 expression in early compared to more advanced lesions, whereas Trop2⁺ expression followed the opposite pattern, with stronger staining in more advanced lesions compared to early ones (fig. S7). Furthermore, these results corroborate those found in our genetic model (Fig. 2, H and I), whereby developmental reprogramming lags behind regeneration activity in neoplastic progression (Fig. 3I).

Human familial adenomatous polyposis adenoma and derivative organoids display impaired differentiation and AbSC activity

To evaluate transcriptional abnormalities in human CRC initiation, we evaluated adenoma and paired normal from a 39-year-old patient with familial adenomatous polyposis (FAP), a hereditary condition in which a mutant copy of APC is inherited, hundreds to thousands of intestinal adenomas develop, and prophylactic colectomy is required to prevent progression to cancer (50). In agreement with our mouse models, colon adenomas displayed elevated expression of SOX9 and TROP2 compared to adjacent normal tissue (Fig. 4A). Markers of differentiated absorptive and secretory colonic cells were partially suppressed as shown by KRT20 and MUC2 IHC, respectively (Fig. 4A). This observation could reflect limited WNT pathway restriction due to hypomorphic mutant APC activity (51) or admixture of normal colonic tissue.

We next performed scRNA-seq on cryopreserved adenoma and paired normal tissue. UMAP representation of single-cell gene expression profiles revealed four epithelial clusters derived from adenoma and normal tissue (Fig. 4B and fig. S8A); however, unlike our mouse models and consistent with human histopathology, the separation between adenomatous and normal tissue was less distinct. To define these clusters, we first used established gene signatures to identify normal enterocyte and goblet cell clusters (Fig. 4, B and C, and fig. S8 B to D). We then defined the cluster found only in the adenoma sample as “Aberrant” (Fig. 4B). The remaining cluster was labeled “Intermediate” because (i) it was found in both normal and adenoma tissue and (ii) it did not engender a normal cell type gene expression profile (Fig. 4B) (38). Reassuringly, SOX9 expression was highest in the Aberrant cluster followed by the Intermediate cluster, whereas KRT20 expression was lowest in these two clusters (Fig. 4C and fig. S8C). In general, the expression of ISC signatures was greatest in the Aberrant cluster relative to Intermediate, enterocyte, and goblet clusters (figs. S9A and S8E), further building confidence in our classification. Notably, the AbSC gene expression signature was greatest in the Aberrant cluster and adenoma tissue (Fig. 4D and fig. S8F). A subset of regeneration and fetal-like intestinal signatures and genes, including TACSTD2, were elevated in the Aberrant cluster (figs. S9B and S8, G and H). We validated these findings using published scRNA-seq data from patients with CRC (n = 9) using normal colon (n = 3) as a comparator (fig. S10, A to E). Together, these data indicate that human adenomas demonstrate a partial block in differentiation and activation of an AbSC program involving developmental programs.

We next evaluated organoids derived from adenoma and normal colonic tissue from the same patient with FAP, asking whether features of AbSC activity are captured in this three-dimensional culture system that requires supplemented WNT3A, R-spondin, and Noggin (WRN) conditioned media rich in stem cell promoting factors (52). Normal colonic cultures contained organoids that appeared folded by phase contrast and generated crypts with high SOX9 expression, absent KRT20 expression, and absent MUC2 expression/Alcian blue (AB) staining (Fig. 4E, dotted line). These organoids also displayed differentiated regions with low or no SOX9 expression, high KRT20 expression, intermittent MUC2 expression, and strong AB staining (Fig. 4E). In contrast, most organoids in adenomas cultures (>95%) did not display differentiated regions as indicated by unfolded, spheroid morphology by phase contrast and weak KRT20 levels and absent MUC2 expression by IHC. Rather, adenoma organoids expressed uniform high SOX9 levels (Fig. 4E). These findings were confirmed in a broader survey of stem cell/WNT (LGR5, SOX9, ASCL2, and AXIN2) and regeneration/developmental genes (LY6E and TACSTD2), which demonstrated markedly higher mRNA expression in adenoma compared to normal organoid cultures (Fig. 4F). Ly6a was not profiled, as it does not have a direct human homolog. mRNA expression of colonic differentiation markers DPP4 and KRT20 were correspondingly reduced, albeit modestly because of enhanced stem cell cues conferred by WRN media (Fig. 4F).

We validated these results in Apc^WT and Apc^KO organoids derived from our genetic mouse model (fig. S9, C to E). After confirming Apc inactivation in tdT⁺ organoids conferred niche independence (the ability to grow without WRN media; fig. S9D), we performed bulk RNA-seq. In addition to the activation of stem cell genes and partial reduction in intestinal differentiation, Apc^KO organoids demonstrated robust activation of developmental intestinal genes, especially Tacstd2 (fig. S9E). Collectively, these data suggest that human adenoma and mouse Apc^KO organoids display features of developmental reprogramming as part of an AbSC transcriptional program.

Sox9 is required for intestinal neoplasia initiation

On the basis of its (i) enriched transcriptional activity in AbSCs, (ii) stronger expression in fetal intestines compared to adult intestines, and (iii) ability to functionally block differentiation in models of CRC (9), we hypothesized that Sox9 is required for CRC initiation. To evaluate this hypothesis, we bred Lgr5^Cre;Apc^f/f;R26^tdT mice with Sox9^f/f mice (53) to homozygosity, enabling conditional biallelic deletion of Sox9 in the Apc^KO setting upon tamoxifen induction. Experiments were initiated with three daily tamoxifen injections followed by weekly single-injection maintenance treatments, attempting to force robust genetic recombination. Successful Sox9 inactivation prevented tdT⁺ adenoma formation, reduced mRNA levels of stem cell genes (Ascl2 and Lgr5), and maintained intestinal homeostasis with crypt-restricted Lgr5^eGFP+ stem cells and Krt20⁺ differentiated villi (Fig. 5A). While Lgr5^Cre;Apc^f/f;Sox9^f/f;R26^tdT mice survived significantly longer than Lgr5^Cre;Apc^f/f;R26^tdT controls (Fig. 5B), they eventually succumbed to adenoma burden. These adenomas invariably expressed Sox9 despite tdT⁺ activation and Apc deletion, suggesting escape of Sox9 deletion, which we confirmed by PCR of flow-sorted tdT⁺ cells, demonstrating a significant fraction of unrecombined Sox9 (fig. S11A). To show that tdT⁺ adenomas from Lgr5^Cre;Apc^f/f;Sox9^f/f;R26^tdT mice were still dependent on Sox9, we generated organoids at the experimental end point and subjected them to ex vivo recombination using specific adenovirus (Ad) infections. Compared to organoids treated with mock Ad^GFP controls, Ad^GFP-Cre-treated organoids displayed reduced proliferation by CellTiter-Glo (CTG), which corresponded to robust Sox9 inactivation by recombination-specific PCR (Fig. 5C). We confirmed these results by performing a 10-day short-term experiment, isolating organoids from three mice from each genotype, and subjecting them to Ad^GFP-Cre infections followed by CTG (fig. S11B). Collectively, these experiments indicate that successful Sox9 inactivation prevents intestinal Apc^KO adenomas using genetically engineered mice and derivative organoids.

We next established colon organoids without endogenous Cre recombinase for controlled ex vivo Ad^GFP-Cre experiments, seeking further validation of Sox9 dependency in colonic adenomas. Colon organoids from R26^tdT, Apc^f/f;R26^tdT, and Apc^f/f;Sox9^f/f;R26^tdT mice were infected with Ad^GFP-Cre and subjected to proliferation assays by CTG. While Apc inactivation led to a 7.5-fold increase in organoid proliferation, concomitant Sox9 deletion reduced this proliferative advantage by threefold (Fig. 5D). To support these findings, we performed tdT⁺ quantification by FACS as a surrogate for cell viability following Ad^GFP-Cre treatment. tdT was activated in 54% of Apc^f/f;R26^tdT organoids 4 days after Ad^GFP-Cre infection, which was reduced by fourfold to 13% in Apc^f/f;Sox9^f/f;R26^tdT organoids (fig. S11C); similar results were achieved with lentiviral Cre activation (fig. S11D). Last, compared to Ad^GFP-treated controls, Ad^GFP-Cre treated Apc^f/f;Sox9^f/f;R26^tdT organoids demonstrated impaired proliferation by Ki67 IHC, which was not observed in R26^tdT or Apc^f/f;R26^tdT organoids (Fig. 5E and fig. S11E). Apc^f/f;Sox9^f/f;R26^tdT organoids with more efficient Ad^GFP-Cre infection marked by increased tdT activation demonstrated reduced Sox9 expression and lower proportion of Ki67⁺ cells relative to poorly infected ones (fig. S11E, red versus gray arrow). These experiments demonstrate a requirement for Sox9 in adenoma formation using genetically engineered mice and organoids.

scRNA-seq reveals that Sox9 is required for AbSC

To determine whether Sox9 is required for AbSC transcriptional activity, we performed scRNA-seq on FACS-isolated tdT⁺ cells from Lgr5^Cre;Apc^f/f;Sox9^f/+;R26^tdT mice 1 month after tamoxifen induction. We chose Lgr5^Cre;Apc^f/f;Sox9^f/+;R26^tdT mice to analyze tdT⁺ cells with consistent and reliable reduction in Sox9 expression given our difficulty in isolating viable tdT⁺ cells with efficient homozygous deletion of Sox9 (Fig. 5, B and C). Reassuringly, viable tdT⁺ cells from Lgr5^Cre;Apc^f/f;Sox9^f/+;R26^tdT mice demonstrated a considerable reduction in Sox9 levels compared to Lgr5^Cre;Apc^f/f;R26^tdT mice, which displayed higher Sox9 expression relative to Lgr5^Cre; R26^tdT controls (Fig. 6A and fig. S12, A and B). UMAP representation of single-cell gene expression profiles revealed clusters representing the normal epithelial cell types and AbSCs (Fig. 6A and fig. S12C). Notably, single copy deletion of Sox9 reduced the percentage of AbSCs from 60 to 10% in tdT⁺ Lgr5-Apc^KO cells (Fig. 6A and table S3) and significantly reduced AbSC gene signature activity (Fig. 6B), indicating a requirement for Sox9 in maintaining an AbSC transcriptional program. ISC gene signatures were also reduced in tdT⁺ cells from Lgr5^Cre;Apc^f/f;Sox9^f/+;R26^tdT mice (Fig. 6B and fig. S12, D and E). The sixfold reduction in AbSC cells was accompanied by the restoration of a normal distribution of ISC and differentiated intestinal cell types (Fig. 6A and table S3). We also observed reduced expression of IFN-γ, regeneration signatures, including Ly6a, and fetal-like intestinal gene signatures, including Tacstd2, in tdT⁺ cells from Lgr5^Cre;Apc^f/f;Sox9^f/+;R26^tdT mice (fig. S11, F to K). These results were confirmed in R26^tdT, Apc^f/f;R26^tdT, and Apc^f/f;Sox9^f/f;R26^tdT organoids using ex vivo Ad^GFP-Cre infections followed by RT-PCR (fig. S13). ATAC-seq profiling revealed that Sox9 inactivation led to reduced chromatin accessibility at regeneration and fetal genes in Apc^KO cells (Fig. 6C and fig. S14A), specifically at the Ly6a, Ly6e, and Tacstd2 genomic loci (Fig. 6D and fig. S14B), indicating rescue of expanded cell state plasticity. Collectively, these results demonstrate a requirement of Sox9 in AbSC activity, aberrant cell state plasticity, and fetal intestinal gene reactivation.

SOX9 KD in FAP adenoma organoid induces differentiation

To evaluate SOX9 activity and necessity in human colonic neoplasia, we used our FAP adenoma organoid model (Fig. 4E). We first defined genomic SOX9 binding patterns in paired normal colon and adenoma organoids by cleavage under targets and release under nuclease (CUT&RUN). A global increase in SOX9 binding was observed in adenoma organoids compared to normal colon organoids (fig. S15A). Greater SOX9 binding was observed at ISC and AbSC genes in adenoma compared to normal colon organoids (Fig. 7A and fig. S15B); while many sites are shared, there appears to be greater than fourfold unique SOX9 binding sites at AbSC genes in adenoma compared to normal (Fig. 7B), although the data are limited to one biological replicate. Consistent with selective ISC activity upon Apc inactivation, we noticed that SOX9 binding at ISC gene LRIG1 was unchanged in adenoma compared to normal colon organoids (fig. S15C), whereas greater SOX9 binding was found at a PROM1 intronic enhancer (fig. S15C), validating our previous work (9) and supporting SOX9-mediated AbSC activity in neoplasia. We also found greater SOX9 binding at both Mustata and Vallone fetal genes, as well as Nusse regenerative genes in adenoma compared to normal colon organoids (Fig. 7, B and C, and fig. S15, D to F). A total of 213 AbSC and 108 fetal genes were only bound by SOX9 in adenoma organoids (table S6). Consistently, stronger SOX9 binding was found at fetal genes, including TACSTD2 and CLU, in adenoma compared to normal colon organoids (Fig. 7D). Despite the enrichment in adenoma organoids, we think that there is considerably more SOX9 binding at fetal genes in normal colon organoids than expected, likely because of strong and sustained influences of WRN conditioned media.

We next asked whether SOX9 is required for AbSC activity in human FAP adenoma organoids. Human adenoma organoid cultures showed few differentiating organoids (~5%), which appear as folded structures by phase contrast, whereas organoid cultures derived from normal adjacent tissue displayed evidence of a differentiating population that amounted to about 50% by day 3 (Fig. 7E and fig. S16A). Using this system, we examined whether short hairpin RNA (shRNA)–mediated SOX9 KD can force differentiation in human adenoma organoids. Stable SOX9 KD shifted the adenoma culture to 40 to 70% differentiating organoids, representing up to a 14-fold increase compared to parental and nontargeting control (NTC) cultures by day 3 (Fig. 7E and fig. S16, B and C). Histopathology of fixed organoids demonstrated that SOX9 KD induced robust KRT20 expression by IHC and mucin production by AB staining and MUC2 IHC (Fig. 7F). Consistently, SOX9 KD reduced stem cell activity (LGR5, SOX9, and AXIN2) and regenerative/developmental reprogramming (LY6E and TROP2) while inducing differentiation (KRT20; Fig. 7G). In contrast, SOX9 KD led to a modest, statistically insignificant increase in differentiation of normal colon organoids, which corresponded to induction of KRT20 without impact on fetal genes (fig. S16, D and E). These data indicate that human FAP adenomas depend on SOX9 to maintain developmental reprogramming and prevent differentiation (Fig. 7H).

Study limitations

While we demonstrate that transcriptional activity associated with regeneration precedes fetal gene expression in a population of Apc^KO intestinal cells, we did not demonstrate whether this occurred in a single-cell lineage, which would require individual cell tracking (e.g., barcoding). We provide a potential reason underlying the partial impairment of differentiation in human adenomas from a patient with FAP; however, additional experiments are required to prove that this observation is due to hypomorphic mutant APC activity. While Cre expression from the endogenous Lgr5 locus leads to efficient inactivation of Apc, a positively selected alteration, it appears less proficient for inactivating both copies of Sox9, a negatively selected event. Future efforts should use a stronger Cre driver to more efficiently delete Sox9 in vivo. Our CUT&RUN data indicate that SOX9 binds to several regeneration and fetal genes in adenoma organoids; however, these results are limited by (i) a single biological replicate and (ii) the influence of WRN media, which is a potent inducer of a stem cell state. Direct tissue evaluation and/or evaluation of organoids grown in factor-reduced media across multiple biological replicates will provide stronger evidence that SOX9 directly regulates fetal reprogramming.

Animal studies

All procedures involving mice and experimental protocols were approved by the Institutional Animal Care and Use Committee (IACUC) of Dana-Farber Cancer Institute (11-009). Adenoma and paired normal-appearing tissue from a patient with FAP were collected after colectomy with informed consent under approval (protocol 13-189) by the Internal Review Board of the Dana-Farber Cancer Institute, Boston, MA, USA.

The generation of Lgr5^{eGFP-IRES-CreERT2} mouse was described earlier (69). Lgr5^{eGFP-IRES-CreERT2} mice were backcrossed in C57bL/6J mice and subsequently single-nucleotide polymorphism (SNP)–tested to ensure >97% pure background (Taconic). To inactivate Apc in intestinal tissue, we crossed Apc^f/f mice (a gift from C. Perret) to Lgr5^{eGFP-IRES-CreERT2}. These mice were further crossed to R26-LSL-tdTomato purchased from the Jackson Laboratory (JAX, stock #007905) (70) for conditional tdT labeling of Lgr5⁺ stem cells and their progeny. To delete Sox9, mice were crossed to Sox9^flox mice (JAX, stock #013106) (53).

To activate conditional alleles, experimental mice aged 6 to 8 weeks were injected intraperitoneally with a single-dose tamoxifen (50 μl of sunflower oil at 10 mg/ml) unless otherwise indicated. For MNU model, Lgr5^{eGFP-IRES-CreERT2} mice were treated with drinking water containing 240 parts per million of MNU (bioKEMIX) scheduled every other week for 10 weeks and followed for 1 year. For AOM/DSS model, AOM/DSS treatment was performed as previously described (71). Briefly, at 8 to10 weeks of age, mice were injected intraperitoneally with AOM (10 mg/kg; Sigma-Aldrich) and treated with 2% DSS (30 to 50 kDa; colitis grade; MP Biomedicals) in drinking water for 5 days. DSS treatment was repeated three times once per 4 weeks. Mice were euthanized 11 weeks after AOM injection.

Mice were euthanized at the end point of the experiment, and small intestine and colon were removed from mice, flushed with phosphate-buffered saline (PBS), and incubated in 10% formalin for 5 min. Subsequently, intestines were opened longitudinally, “Swiss-rolled,” incubated overnight in 10% formalin at room temperature, and processed for paraffin embedding. Fresh tissue was collected for organoid generation, RNA isolation, and protein collection and flash-frozen for long-term storage.

Human specimen

Adenoma and paired normal-appearing tissue from a patient with FAP were collected after colectomy under approval (protocol 13-189) by the Internal Review Board of the Dana-Farber Cancer Institute, Boston, MA, USA.

Histopathology

Paraffin-embedded intestines, organoids, or xenograft tumors were serially sectioned and mounted on a slide. Sections were subjected to hematoxylin and eosin (H&E), AB–periodic acid–Schiff (AB-PAS), as well as immunostaining, using standard procedures. For morphological analysis, sections were serially dehydrated in xylene and ethanol and stained with H&E for histological assessment or AB-PAS to identify goblet cells and mucus.

Immunofluorescence and IHC

For immunostaining, antigen retrieval was performed using a sodium citrate buffer (pH 6), Trilogy (Sigma-Aldrich Cell Marque), or tris-EDTA (pH 9). Slides were permeabilized using a 0.2% Triton X-100 for 30 min at room temperature and blocked with donkey serum for 1 hour. The primary antibodies used for IHC were rabbit anti-Sox9 [1:600; Cell Signaling Technology (CST), #82630; Trilogy antigen retrieval], rabbit anti-Mucin2 (1:200; Santa Cruz Biotechnology, sc-15334; tris-EDTA), anti-Krt20 (1:500; CST, #13063; tris-EDTA), anti–red fluorescent protein [RFP; 1:500; Rockland, 600-401-379; sodium citrate (pH 6)], anti-eGFP (1:1000; Abcam, ab290; tris-EDTA), anti-Olfm4 [1:1000; CST, #39141; sodium citrate (pH 6)], anti-Lyz, anti–Sca-1/Ly6a [1:150; Abcam, ab109211; sodium citrate (pH 6)], anti-Axin2 [1:800; Abcam, ab32197; sodium citrate (pH 6)], and rabbit anti-Ki67 [1:1000; CST, #12202; sodium citrate (pH 6)]. The binding of primary antibody was detected with 3,39-diamino-benzidine-tetrahydrochloride-dihydrate and counterstained with hematoxylin.

IHC quantification was performed using Fiji Is Just Image J (FIJI) application on 10× images from a Leica DM750 microscope. Ly6a- or Sox9-positive cells were quantified and expressed as a percentage of total tdT⁺ cells in intestinal lesions. For organoids, the percentage of Ki67-positive cells was quantified and expressed as a percentage of total cells in an organoid.

Flow analysis, sorting, and staining

Isolated intestines from euthanized mice were washed with ice-cold PBS; villi were scraped using glass slides and dissociated in 5 mM EDTA in PBS at 4°C for 20 min, shaking every 5 to 7 min for about 30 s. The epithelial fraction was collected by centrifugation and incubated at 37°C for 30 min in prewarmed 4× TrypLE. Single cells were filtered through a 70-μm filter after trypsinization, washed with complete media, and collected in FACS buffer.

For antibody staining, 100 μl of resuspended cells in FACS buffer were incubated with 1 μl of APC-labeled Sca-1 antibody (Thermo Fisher scientific, catalog no. 17-5981-82) for 20 min on ice, washed with PBS, and collected in FACS buffer. Cells were resuspended in 1 to 2 ml of FACS buffer with 4′,6-diamidino-2-phenylindole, passed through another 70-μm filter before transferring to 40-μm filtered FACS tubes and sorted. Sorted cells were collected in 5 ml of 50% fetal bovine serum (FBS) + 50% Dulbecco’s modified Eagle medium (DMEM) complete with 10 μM Y27632 Rho-kinase inhibitor.

Single-cell RNA-seq

Murine intestines and human adenoma samples were processed as described above and then subjected to scRNA-seq with or without cell hashing. If hashed, then the mouse cells were stained with TotalSeq-30301 Hashtag 1 (BioLegend, #155861) and TotalSeq-30302 Hashtag 2 (BioLegend, #155863) antibodies; no hashing was used for human samples. Viable cells were washed and resuspended in PBS with 0.04% bovine serum albumin at a cell concentration of 1000 cells/μl. About 17,000 viable mouse cells were loaded onto a 10x Genomics Chromium instrument (10x Genomics) according to the manufacturer’s recommendations. The scRNA-seq libraries were processed using Chromium single-cell 5′ library and gel bead kit (10x Genomics). Matched cell hashing libraries were prepared using a single-cell 5′ feature barcode library kit. Quality controls for amplified cDNA libraries, cell hashing libraries, and final sequencing libraries were performed using Bioanalyzer High Sensitivity DNA Kit (Agilent). The sequencing libraries for scRNA-seq were normalized to 4 nM concentration and pooled using a volume ratio of 4:1. The pooled sequencing libraries were sequenced on Illumina NovaSeq S4 300 cycle platform. The sequencing parameters were as follows: read 1 of 150 base pairs (bp), read 2 of 150 bp, and index 1 of 8 bp. The sequencing data were demultiplexed and aligned to mm10-3.0.0 using Cell Ranger version 3.1.0 pipeline (10x Genomics).

scRNA-seq analysis

Assay for transposase-accessible chromatin sequencing

ATAC libraries were prepared as described previously (72). Briefly, flow-sorted tdT⁺ cells (25,000 cells in duplicates) were lysed to prepare nuclear pellets, which then underwent transposition with TDE1 Enzyme (Illumina, 20034197). Tagmented DNA was purified using Zymo DNA Clean and Concentrator-5 Kit (catalog no. D4014), and the purified DNA was PCR-amplified with NEBNext 2X Master Mix and Illumina adapters (table S4). The libraries were purified after PCR using AMPure XP beads (Beckman Coulter). 150-bp paired-end reads were sequenced on a NovaSeq instrument (Illumina).

ATAC data processing

ATAC-seq data were analyzed as described previously (PMID: 28648363). Briefly, reads were mapped to the mm10/GRCm38 genome using Bowtie2 aligner version 2.3.5 (72). MACS2 was used to call peaks using the parameters “-t $bamfile -f BAMPE -n qc/macs/$base.macs2 -q 0.01 -g mm” (73). Bigwig files were generated using deepTools bamCoverage with the options “bamCoverage --binSize 10 --smoothLength 30 -p 4 --normalize using RPGC --effectiveGenomeSize 2730871774 --extendReads $fragLength -b $file -of bigwig” (PMID: 24799436). All bigwig and bed files were filtered using the ENCODE Blacklist. Only peaks with P < 0.00001 were considered for further analyses. Proximal peaks (−2 kb to 2 kb relative to transcription start site (TSS) called by MACS2 were linked/annotated to genes by ChIPseeker version 1.34.1 (74). Then, computeMatrix and plotHeatmap functions from deepTools (75) were used to visualize peaks.

CUT&RUN sequencing

CUT&RUN for SOX9 and immunoglobulin G (IgG) control in adenoma and normal organoids was done using CUTANA ChIC/CUT&RUN kit (Epicypher, #14-1048) following the manufacturer’s protocol. Briefly, 5 × 10⁵ cells were captured with concanavalin A, permeabilized using 0.01% digitonin, and incubated with 0.5 μg of antibody [anti-Sox9 (CST, #82630) and IgG (Epicypher, #13-0042)] in 50 μl of antibody buffer [20 mM Hepes at pH 7.5, 150 mM NaCl, 0.5 mM spermidine, 1× protease inhibitor cocktails (EDTA-free tablet; Roche), 2 mM EDTA, and 0.01% digitonin] overnight. After removing unbound antibody, pAG-MNase (20×) in 50 μl of cell permeabilization buffer was added to the cells and incubated for 10 min at room temperature. Then, CaCl₂ (2 mM) was added to activate MNase and incubated at 4°C for 2 hours. The reaction was stopped using 33 μl of 2× STOP buffer and Escherichia coli spike-in DNA (0.5 ng) was added as a control. The DNA from the released chromatin in the supernatant was purified and then quantified using Qubit dsDNA HS assay kit (Agilent Technologies). CUT&RUN libraries were constructed using a NEBNext Ultra II DNA library preparation kit as described previously (73) with a few modifications. Briefly, end repair and dA-tailing were conducted on 6 ng of CUT&RUN eluted DNA for 30 min at 20°C followed by 1 hour at 50°C. After adaptor ligation for 30 min at 20°C, the DNA fragments were purified by 1.75× volume of AMPure XP beads (Beckman Coulter) followed by 10 to 12 cycles of PCR amplification with Next Ultra II Q5 master mix. The PCR products were purified with 1× volume of AMPure XP beads. After quantitative and qualitative analysis, libraries with different indexes were pooled and sequenced on Illumina NovaSeq platform with paired-end 150-bp reads.

CUT&RUN data processing

CUT&RUN sequencing data were analyzed as described previously (74), following the standard pipeline (https://yezhengstat.github.io/CUTTag_tutorial/index.html). Briefly, paired-end 150-bp reads were aligned to GRCh38 human genome using Bowtie2 version 2.2.5 (75) with the following options: --local --very-sensitive --no-mixed --no-discordant --phred33 -I 10 -X 700. Technical replicates (n = 4 per condition) were merged before read alignment to increase the power of peak calling. macs2 (76) was used to call peaks from bam files. For SOX9 CUT&RUN peak calling, parameters —t input_file –p 1e-5 –f BAM –keep-dup all –n out_name was used to call narrow peaks. To check the SOX9 binding profile and enhancer activity of specific gene sets, proximal peaks (−2 kb to 2 kb relative to TSS) called by MACS2 were linked/annotated to genes by ChIPseeker version 1.34.1 (77). Then, computeMatrix and plotHeatmap functions from deepTools (78) were used to visualize peaks.

Intestinal organoid culture

Colonic glands were isolated, treated with EDTA, and then resuspended in 30 to 50 μl of Matrigel (BD Bioscience) and plated in 24-well plates. WRN containing DMEM/F12 with Hepes (Sigma-Aldrich) containing 20% FBS, 1% penicillin-streptomycin, and recombinant mouse epidermal growth factor (50 ng/ml; Life Technologies) was used for culturing Apc^KO colon organoids. For the first 2 to 3 days after seeding, the medium was also supplemented with 10 mM ROCK inhibitor Y-27632 (Sigma-Aldrich) and 10 mM SB431542 (Sigma-Aldrich), an inhibitor for the transforming growth factor–β (TGF-β) type I receptor to avoid anoikis. For passage, colon organoids were dispersed by trypsin-EDTA and transferred to fresh Matrigel. Passage was performed every 3 to 4 days with a 1:3 to 1:5 split ratio. For human colon organoid culture, the previous medium was supplemented with antibiotics Primocin (100 μg/ml; InvivoGen) and Normocin (100 μg/ml; InvivoGen), serum-free supplements 1× B27 [Thermo Fisher Scientific (Gibco)] and 1× N2 [Thermo Fisher Scientific (Gibco)], chemical supplements 10 mM nicotinamide (Sigma-Aldrich) and 500 mM N-acetyl cysteine (Sigma-Aldrich), hormone 50 mM [Leu15]-Gastrin (Sigma-Aldrich), growth factor FGF10 (100 μg/ml; recombinant human; Thermo Fisher Scientific), and 500 nM A-83-01 (Sigma-Aldrich), which is an inhibitor of the TGF-β receptors ALK4, ALK5, and ALK7.

Organoid fixation for FFPE

Confluent organoids in six-well plate were fixed in 10% formalin at 4°C on rocker overnight. The fixed organoids were washed with PBS, collected by centrifugation at 2000 rpm for 3 min. The supernatant was carefully aspirated; cell pellets were resuspended in 50 to 80 μl of 2% agar and then immediately kept on ice. The solidified agar with organoids was kept in a cassette and processed to make formalin fixed paraffin embedded (FFPE) blocks.

Cell proliferation assays

Cell viability was quantified by measuring cellular adenosine triphosphate content using the CTG Cell Viability assay (Promega) according to the manufacturer’s instructions. All experiments were performed in triplicate in 96-well plates.

SOX9 recombination PCR

Mouse tails were incubated at 95°C in 75 μl of the cell lysis buffer [5 ml of sterile ddH₂O, 7 μl of 50% NaOH, and 7 μl of 0.5 M EDTA (pH 8.0)], followed by 15 min at 4°C, and neutralized with 75 μl of neutralization buffer [40 mM tris-HCl (pH ~5)]. Supernatant (2 μl) was used for PCR amplification using primers designed for WT SOX9 and unfloxed SOX9 and floxed SOX9 (table S4) using Phire Tissue Direct PCR master mix (Thermo Fisher Scientific, #F170S). PCR cycling conditions were initial denaturation [94°C, 120 s; 10× touchdown cycles with the following three steps: 94°C, 10 s; 65°C (0.5°C increment each cycle), 10 s; and 72°C, 90 s; 25× cycles with the following three steps: 94°C, 10 s; 60°C, 10 s; and 72°C, 90 s; the last annealing: 72°C, 180 s]. The PCR products were analyzed on 1% agarose gel.

Organoid adenoviral transduction

To transduce colonic organoids, 0.5 M organoids were suspended in 500 μl of WRN medium with 10 μM Y27632 ROCK inhibitor in a 24-well ultralow attachment plate. Adenovirus (University of Iowa, Gene Transfer Vector Core, stock: Ad3786; high viral titer equal to 1010 plaque-forming units/ml) carrying GFP or GFP-Cre (1 μl) was added to the organoids, and the sealed plate is centrifuged at 600g for 1 hour at 32°C, followed by incubation at 37°C for 5 to 6 hours. After incubation, the transduced organoids were washed with complete medium and resuspended in about 200 μl of Matrigel and plated in Nunclon delta plates with WRN conditioned media with 10 μM ROCK inhibitor.

Generation of stable colon organoids

All genetically manipulated colon organoid lines were generated using the protocol described here (79). shRNAs against SOX9 were cloned into PLKO.1, TET-PLKO, and TET-Cellecta vectors. To generate lentiviruses, expression vectors were cotransfected into HEK293T cells with the lentiviral packaging constructs psPAX2 and pMD2.G (vesicular stomatitis virus glycoprotein) in a 1:1:1 ratio using X-tremeGENE 9 DNA Transfection Reagent (Roche) according to the manufacturer’s instructions. Cell culture medium was changed the following day, and lentiviral supernatant was harvested 48 and 72 hours later and filtered through a 0.45-μm filter (Millipore). Lentiviruses were aliquoted and stored at −80°C until use.

To transduce colonic organoids, spheroids in one well (24-well plate) were trypsinized, and a one-fourth to one-eighth volume of cell suspension was used for each infection. Cells were resuspended in 500 μl of lentiviral supernatant with polybrene (8 μg/ml) and 10 mM Y-27632, centrifuged at 600g at 32°C for 1 hour, and incubated for 6 hours in a cell culture incubator. The infected cells were suspended in 30 to 50 μl of Matrigel and cultured with Wnt/R-spondin–deprived medium containing 10 mM Y-27632 and 10 mM SB431542. Colon organoids were selected with puromycin (3 μg/ml) at 24 hours after infection.

RNA isolation and qPCR

Total RNA was isolated using the RNeasy Mini Kit (QIAGEN, Germantown, MD, USA), and cDNA was synthesized using the iScript Reverse Transcription Supermix for RT-qPCR (Bio-Rad) according to the manufacturer’s instructions. Gene-specific primers for SYBR Green real-time PCR were either obtained from previously published sequences or designed by Primer-BLAST (www.ncbi.nlm.nih.gov/tools/primer-blast/) and synthesized by Integrated DNA Technologies or ETON Bioscience. Real-time PCR was performed and analyzed using CFX96 Real-Time PCR Detection System (Bio-Rad Laboratories Inc., Hercules, CA) and using Power SYBR Green PCR Master Mix (Thermo Fisher Scientific) according to the manufacturer’s instructions. Relative mRNA expression was determined by normalizing to GAPDH expression, which served as an internal control. See table S4 for primers used for qPCR.

Immunoblot, antibodies, and inhibitors

Immunoblot analysis was performed as previously described (80). Briefly, cells were lysed in radioimmunoprecipitation assay buffer supplemented with a protease inhibitor cocktail (Roche). Whole-cell extracts were resolved by SDS–polyacrylamide gel electrophoresis, transferred to polyvinylidene difluoride membranes, and probed with indicated primary antibodies. Bound antibodies were detected with horseradish peroxidase (HRP)–conjugated secondary antibodies and chemiluminescent HRP substrate. The following primary antibodies were used for Western blotting (all from CST, Beverly, MA, USA, unless otherwise indicated): anti-SOX9 (#82630; 1:1000), anti-vinculin (#13901; 1:1000), and anti-RFP (Rockland, 600-401-379; 1:500).

Statistical analysis and reproducibility

Experiments were performed in triplicate. Data are represented as means ± SD unless indicated otherwise. For each experiment, either independent biological or technical replicates are as noted in the figure legends and were repeated with similar results. Statistical analysis was performed using Microsoft Office, Prism 7.0 (GraphPad), or RStudio statistical tools. Pairwise comparisons between groups (that is, experimental versus control) were performed using an unpaired two-tailed Student’s t test or Kruskal-Wallis test as appropriate unless otherwise indicated. For all experiments, the variance between comparison groups was found to be equivalent. Animals were excluded from analysis if they were euthanized because of health reasons unrelated to tumor volume end point.

This PDF file includes:

Other Supplementary Material for this : manuscript includes the following: