Lineage Reversion Drives WNT Independence in Intestinal Cancer

The WNT signaling pathway is a key developmental regulator and is essential for homeostasis of numerous adult cell types, including hematopoietic (1, 2) and intestinal stem cells (3, 4). Genetic alterations that hyperactivate the WNT pathway, including disruption of the APC tumor suppressor, mutational activation of CTNNB1 (β-catenin), or chromosomal translocations involving RSPO genes, occur in more than 90% of colorectal cancers (5, 6) and likely facilitate the initial growth of these tumors (7–9). We and others have shown that, in many contexts, WNT hyperactivation is essential for tumor maintenance. For instance, restoration of endogenous APC expression in genetically engineered or tumor-engrafted mice leads to rapid and sustained tumor regression (10, 11), whereas blocking WNT ligand secretion with porcupine (PORCN) inhibitors or blocking RSPO3 directly in human or murine tumors with RSPO3 fusions drives cell-cycle arrest, differentiation, and tumor clearance (9, 12, 13). Thus, most normal and transformed cells of the intestine are thought to be WNT-dependent.

Targeting the WNT pathway in cancer has clear therapeutic potential, and this has led to the recent initiation of multiple phase I clinical trials (NCT02521844, NCT02649530, NCT03447470) for treatment of RSPO fusion cancers with PORCN inhibitors. However, due to the current lack of clinically approved WNT antagonists and scarcity of RSPO fusion model systems, it has not been possible to explore potential mechanisms of therapy failure and resistance to WNT-targeted therapy. To further understand WNT dependence and the therapeutic potential of WNT inhibition, we engineered an array of organoid-based models capturing the most frequently observed genetic alterations in this molecular subtype. We found that although no individual oncogenic change influences sensitivity to WNT suppression, the combined alteration of multiple colorectal cancer–associated lesions enables rapid acquired resistance to WNT blockade. This occurs via a TGFβ-induced lineage conversion that reverts adult intestinal epithelium to a fetal intestinal state. The lineage switch is driven and maintained by YAP/TAZ signaling and is not reversible. Consequently, WNT-independent cells become exquisitely and selectively sensitive to suppression of YAP/TAZ.

PTPRK–RSPO3 fusions drive the development of WNT-dependent murine intestinal adenomas that are extremely sensitive to treatment with the PORCN inhibitor WNT974 (9). Human RSPO fusion tumors carry an array of cooperating oncogenic lesions, including frequent mutational activation of KRAS or BRAF, as well as disruption of TP53 and/or SMAD4 (Fig. 1A). In many cases, three or four of these oncogenic events occur within the same tumor, but how such co-occurring alterations contribute to WNT dependence is not known. Unlike APC-mutant colorectal cancer, there are very few human RSPO3 colorectal cancer cell lines or organoid models, and until recently (14) no engineered human cell systems with which to interrogate RSPO disease biology. To recapitulate some of the genetic complexity of human RSPO3 fusion tumors, we used CRISPR to develop a series of murine organoid models harboring common oncogenic genotypes (Fig. 1B). We first generated multiple independent organoid lines, each carrying an endogenous Ptprk–Rspo3 (R) fusion (9) and an endogenous Kras^G12D (K) mutation (from the Kras^LSL-G12D allele; ref. 15). We then used Cas9 and single guide RNAs (sgRNA) targeting Trp53 (P) and Smad4 (S) to create triple (KRP and KRS) and quadruple (KRPS) mutants (Fig. 1B; see Methods for details). As described previously (16, 17), organoids containing loss-of-function alterations in p53 and SMAD4 were selected by treating for 7 days with Nutlin-3 (5 μmol/L) or TGFβ (5 ng/mL), respectively. We confirmed the presence of each mutation by sequencing or assessed protein disruption by Western blot analysis (Fig. 1C–E). R, KR, and KRP cultures resembled wild-type intestinal organoids, with proliferative crypt-like budding protrusions and a KRT20-positive differentiated core (Fig. 1F). In contrast, KRS and KRPS were proliferative spheroid structures with very few KRT20-positive differentiated cells.

Similar to organoids carrying only Rspo3 fusions (9), KRP and KRS organoids showed rapid cell-cycle arrest (loss of EdU incorporation) and differentiation (induction of KRT20) 4 days following PORCN inhibition with 500 nmol/L WNT974 (Fig. 2A and B; Supplementary Fig. S1A). Quadruple KRPS mutants showed a marked decrease in EdU incorporation and increased KRT20 expression; however, unlike KRP and KRS, a subpopulation of KRPS cells (∼4%–5%) continued to proliferate and expand in the presence of drug (Fig. 2A and B). Drug resistance was not specific to WNT974, as we observed an identical response with an independent PORCN inhibitor (ETC-159; Fig. 2B). Furthermore, resistance to both drugs was long-lived, as rechallenge with WNT974 or ETC159 2 to 4 weeks after drug withdrawal caused only a moderate reduction in proliferation (Fig. 2B), and no change in cell morphology over continual culture with WNT974 (Supplementary Fig. S1B). Like KRPS cells, quadruple mutant organoids expressing an endogenous murine Braf^V619E allele (18), equivalent to human BRAF^V600E (“BRPS”), also showed rapid resistant outgrowth in the presence of WNT974 (Supplementary Fig. S1C).

To determine whether the (KRPS/BRPS) genotype-dependent response was specific to PORCN inhibition or reflected a more general bypass of WNT dependence, we generated organoids carrying Kras^G12D and Trp53 mutations and an inducible Apc-targeted shRNA that enables doxycycline (dox)-regulated control of APC expression (shAKP). In this context, withdrawal of dox drives APC-mediated suppression of hyperactive WNT signaling and, in the absence of exogenous RSPO1/WNT ligand, promotes cell-cycle arrest and differentiation (10). Consistent with previous work (10), APC restoration in shAKP cells induced cell-cycle arrest and differentiation, preventing long-term RSPO-independent growth (Fig. 2A, bottom). However, similar to KRPS cells, a subpopulation of SMAD4 mutant “shAKPS” cells were able to expand in RSPO-free media following APC reexpression (Fig. 2A). Furthermore, these RSPO/WNT-independent shAKPS cells showed no acute response to treatment with WNT974 (Supplementary Fig. S1D), confirming they did not escape APC restoration by upregulating WNT ligand expression. Together, these data suggest that the combination of Kras/Braf, Trp53, and Smad4 mutations can enable the generation of WNT-independent intestinal organoids, following pharmacologic or genetic suppression of WNT signaling.

In total, we generated 20 WNT-independent organoid lines using both shRNA and CRISPR-based approaches (Supplementary Table S1), which showed two distinct patterns of resistance. A subset of the lines (4/20) showed robust reactivation of the WNT pathway (“WNT^hi”; Fig. 2C; Supplementary Table S1). In one instance, this was associated with a 25-fold genomic amplification of Ctnnb1 (Supplementary Fig. S2A), and in another, a hotspot S33C mutation in Ctnnb1 (Supplementary Fig. S2B). Using base editing, we confirmed that, as expected, organoids with mutational activation of Ctnnb1 (S33F) were dramatically enriched following treatment with WNT974 (Supplementary Fig. S2C). The majority of resistant cultures (16/20) maintained very low levels of WNT target gene expression while cultured in WNT974 (“WNT^lo”; Fig. 2C; Supplementary Fig. S3), suggesting they did not escape treatment by downstream activation of WNT signaling. We further saw no evidence of genomic WNT pathway alterations in those WNT^lo resistant cases examined by whole-exome sequencing (Supplementary Table S1). Together, these data reveal the unexpected appearance of intestinal cells that remain responsive to WNT-suppressive stimuli, yet do not require WNT pathway activation for growth and proliferation. We defined organoids with the “WNT^lo resistant” phenotype as KRPS^R, BRPS^R, and shAKPS^R.

Transcriptome analysis of WNT974-naïve and WNT974- resistant isogenic organoid pairs by RNA sequencing revealed a number of differentially regulated genes and pathways in both KRPS^R and shAKPS^R genotypes, including a series of inflammation-associated gene signatures (Fig. 3A). We noted that during the generation of KRPS, BRPS, and shAKPS cultures, each sample was treated transiently with TGFβ to select for SMAD4-mutant populations (Fig. 1B) and reasoned that TGFβ treatment might be altering transcriptional networks and conditioning or “priming” cells to become WNT-independent. To determine whether TGFβ priming was important for WNT independence, we generated KRPshS organoids in which SMAD4 expression was silenced by an shRNA, enabling the enrichment of cells with SMAD4 depletion, without functional (TGFβ) selection (Fig. 3B). In contrast to spheroid KRPS cells, KRPshS cultures retained a budding organoid morphology similar to R, KR, or KRP cells (Fig. 3C). However, short-term treatment (3 days) with TGFβ (5 ng/mL) induced a rapid morphologic shift toward spheroid structures (Fig. 3C and D) that was maintained 2 weeks following withdrawal of TGFβ (Fig. 3D), up to at least 6 months in culture (not shown). The spheroid change was driven by signaling through the canonical TGFβ receptor complex as cotreatment with a TGFβR1 inhibitor (LY2157299) or CRISPR-mediated genetic disruption of Tgfbr2 completely abolished the morphologic response (Fig. 3C and D).

In many contexts, SMAD4 association with SMAD2/3 is required for canonical TGFβ signaling (19). Indeed, the magnitude of TGFβ target gene expression following acute TGFβ stimulation was reduced in SMAD4-depleted cells (KRPshS) compared with those with wild-type SMAD4 (KRP; Fig. 3E). However, most downstream targets were still strongly induced compared with isogenic untreated cells, highlighting a widespread SMAD4-independent transcriptional response to TGFβ in these cells (Fig. 3E; Supplementary Fig. S4A), as has been reported in other settings (20, 21). In support of the notion that TGFβ drives SMAD4-independent canonical signaling, we saw a marked increase in SMAD2/3 phosphorylation but no change in the activation of noncanonical TGFβ pathways (MAPK, AKT, p38, JNK; Supplementary Fig. S4B). Furthermore, TGFβ-mediated transcriptional changes in SMAD4-depleted cells were SMAD2/3-dependent, as simultaneous silencing of these SMADs suppressed target gene induction and spheroid transformation (Fig. 3F and G). Most importantly, and consistent with results from CRISPR-derived KRPS cultures, TGFβ-treated KRPshS organoids developed resistance to WNT974, whereas TGFβ-naïve KRPshS organoids remained WNT-dependent and could not survive continuous treatment with the drug (Fig. 3H). Similarly, disruption of Tgfbr2 prevented the emergence of WNT974-resistant cells (Fig. 3H). Thus, TGFβ priming via canonical TGFβ receptors can initiate the induction of WNT independence.

As noted above, withdrawal of TGFβ from the culture media had no impact on TGFβ-induced spheroid morphology (Fig. 3D). Likewise, treatment of WNT-independent organoids with a TGFβR1 inhibitor (LY2157299) had no effect on morphology or growth of organoids (Supplementary Fig. S4C and S4D). This observation raised the possibility that SMAD4 loss was important to enable TGFβ priming, but was dispensable beyond this event. To directly test the requirement for SMAD4 loss in regulating WNT independence, we generated KRPshS organoids in which the expression of the SMAD4 shRNA was controlled by a dox-regulated element (TRE) and a reverse tet-transactivator. We generated WNT974-resistant organoids in the presence of dox, exactly as described above, and restored SMAD4 expression by withdrawal of dox from the media (Supplementary Fig. S5A and S5B). As expected, reexpression of SMAD4 sensitized organoids to acute exposure to TGFβ (Supplementary Fig. S5C and S5D); however, it had no impact on organoid morphology or the maintenance of WNT independence in the absence of exogenous TGFβ ligand (Supplementary Fig. S5E). Proliferation was slightly increased following dox withdrawal, perhaps owing to subtle effects of dox on organoid growth (Supplementary Fig. S5F).

We next used a similar approach to generate WNT-resistant organoids in which p53 expression was controlled by a dox-regulated shRNA (KRSshP*; Supplementary Fig. S5A). In contrast to the effect of SMAD4 reexpression, restoration of endogenous p53 significantly reduced cell proliferation in both drug-naïve and WNT974-resistant cells, though WNT-independent cells were 2 to 3 times more sensitive to p53 restoration (Fig. 3I and J; Supplementary Fig. S6A). Notably, cell-cycle arrest occurred in both the presence and absence of WNT974, suggesting that p53 was not simply restoring sensitivity to WNT974, but that WNT-independent cells were hypersensitive to p53-mediated tumor suppression. Cell-cycle arrest downstream of p53 is often mediated by p21 (22). Indeed, we noted that resistant cells showed increased chromatin accessibility at the Cdkn1a locus (Supplementary Fig. S6B), and both p21 transcript and protein were more dramatically induced in resistant cells compared with their drug-naïve counterparts (Fig. 3K; Supplementary Fig. S6C). Silencing of p21 partially blocked cell-cycle arrest following p53 restoration (Fig. 3L), suggesting that bypassing p21-induced arrest contributes to but is not the only factor underlying the importance of p53 disruption in initiating and maintaining WNT independence.

Together, these data support the notion that oncogenic alterations play a critical role in the development of WNT independence, but that the requirements for initiation and maintenance of this program are different. Transient TGFβ priming in KRPS/shAKPS intestinal cells drives SMAD2/3-dependent TGFβ pathway activation that avoids SMAD4-dependent cell death (21), but continual TGFβ signaling is not essential to maintain the WNT-independent state. Conversely, p53 disruption is required for both initiation and maintenance of WNT pathway independence.

TGFβ is an abundant cytokine in the tumor microenvironment, and modulation of TGFβ receptor has been shown to affect colorectal cancer progression in animal models by altering tumor–stroma signaling (23, 24). We sought to determine whether exposure to tumor microenvironmental signals was sufficient to prime TGFβ-naïve organoids and induce WNT independence in vivo.

Long-term treatment with WNT974 can cause intestinal and bone damage in mice (25, 26). To enable potent and sustained WNT suppression in tumor cells without effects on surrounding tissues, we used the regulatable shApc model, whereby withdrawal of dox from chow drives rapid APC restoration and potent cell-intrinsic WNT pathway suppression (10, 11). We generated TGFβ-naïve shAKPS cells by simultaneous delivery of sgRNAs targeting both Trp53 and Smad4, and selected for p53 disruption by Nutlin-3 treatment. Surrogate selection for Trp53 mutation produced a polyclonal population of cells that were TGFβ-naïve, but contained greater than 95% Smad4 disruption (Supplementary Fig. S7A and S7B). As expected, TGFβ-naïve shAKP and shAKPS organoids and TGFβ-treated shAKPT organoids were sensitive to WNT suppression by APC restoration and never became WNT-independent in vitro (Supplementary Fig. S7C).

Each of the organoid lines was injected in mice pretreated with dox (200 mg/kg in chow). When tumors reached approximately 100 mm³ (10–14 days postinjection), animals were either maintained on dox or returned to normal chow to restore APC. As described previously, shAKP organoids showed cell-cycle arrest (loss of Ki-67; Fig. 4A and B), marked reduction of WNT target expression (SOX17; Fig. 4A), and increased differentiation (KRT20; Supplementary Fig. S8) in response to APC restoration, and showed little to no tumor growth following the dox switch (Fig. 4C). As expected, tumors derived from TGFβ-treated shAKPS organoids continued to grow following APC restoration (Fig. 4A). At harvest, these tumors showed evidence of WNT pathway suppression (SOX17; Fig. 4A), but in contrast to shAKP tumors, were highly proliferative (Fig. 4A and B) and showed little evidence of differentiation (KRT20; Supplementary Fig. S8), similar to dox-treated tumors. These tumors also showed frequent phosphorylated SMAD3, highlighting widespread TGFβ-mediated signaling throughout the tumor mass (pSMAD3; Supplementary Fig. S8). Tumors derived from TGFβ-naïve shAKPS organoids, that were entirely WNT-dependent in vitro (Supplementary Fig. S7C), showed a mixed response in vivo, with clusters of arrested differentiated cells interspersed with poorly differentiated and proliferative tumor cells (Fig. 4A). This is consistent with the outgrowth of a subset of WNT-independent tumor cells, as observed in TGFβ-primed ex vivo cultures (Fig. 2A and B). Similar to TGFβ-primed shAKPS tumors, TGFβ-naïve tumors showed widespread pSMAD3 (Supplementary Fig. S8), indicating activated TGFβ signaling. As expected, genetic disruption of Tgfbr2 prevented SMAD3 phosphorylation and WNT-independent tumor growth (Fig. 4A–C; Supplementary Fig. S8). Similar effects were seen with KRP-, KRPS-, and KRPT-derived tumors, following 2 weeks of WNT974 treatment (Fig. 4D; Supplementary Fig. S9A and S9B). In all, these data show that the tumor microenvironment is sufficient to prime cells for WNT independence, and that this is likely driven by canonical TGFβ signaling.

As WNT independence was initiated but not maintained by TGFβ, we sought to identify downstream signaling pathways that were altered following TGFβ exposure and maintained following TGFβ withdrawal. To do this, we again compared pathway enrichment using the hallmark gene set enrichment analysis (GSEA) dataset (27), as well as a range of unique gene signatures that have recently been identified from single-cell and bulk transcriptome analyses of the mouse intestine (28–31). As expected, TGFβ signaling was the most enriched pathway following acute (3 days) TGFβ treatment, but was lost after long-term (>30 days) TGFβ-free culture (Fig. 5A). Strikingly, the only molecular signature that was increased and robustly maintained in post-TGFβ organoids was a collection of genes upregulated in embryonic or fetal intestinal organoids, compared with adult-derived cultures (Fig. 5A and B; ref. 31). Accordingly, genes downregulated in fetal intestinal cultures were also strongly suppressed following TGFβ priming (Fig. 5A). Many of the genes downregulated in TGFβ-primed organoids specified differentiated intestinal lineages, including enterocytes, enteroendocrine cells, and Paneth cells (Fig. 5A and B). These changes identified by transcriptome analysis were readily apparent in tumors in vivo, with KRPS and shAKPS tumors, but not KRP, shAKP, KRPT, or shAKPT tumors, showing loss of adult lineages (Paneth cells marked by LYZ1) and induction of fetal markers (SPP1 and ANXA1; Fig. 5C; Supplementary Fig. S10).

To more closely interrogate the dynamics of cell fate change following TGFβ priming and the development of WNT independence, we performed single-cell RNA sequencing (scRNA-seq) in KRP, and KRPshS TGFβ-naïve, TGFβ-primed, and WNT974-resistant organoids. Mapping of marker gene expression across 12 identified cell clusters revealed multiple distinct populations (Supplementary Fig. S11A–S11C), including LGR5-positive adult stem cells, enterocytes, and goblet, Paneth, and enteroendocrine cells (Fig. 5D; Supplementary Fig. S12). Depletion of SMAD4 (in KRPshS organoids) induced expansion of a WNT-high and LGR5-high stem cell compartment compared with KRP (Fig. 5E; Supplementary Fig. S12), as reported in other intestinal organoid and in vivo models (32, 33). Despite the expansion of the stem cell pool, SMAD4 loss alone did not alter lineage differentiation (Fig. 5E, top; Supplementary Fig. S12). In contrast, TGFβ priming of KRPshS cells caused a dramatic shift in transcriptional identity, depleting multiple differentiated intestinal lineages (Fig. 5E and F). In addition, we noted the expansion of a cell population with marker gene expression reminiscent of recently identified “revival stem cells” (expressing CLU, ANXA1, CXADR, and BASP1; hereafter “RevSCs”) that are activated in vivo following intestinal injury (ref. 34; Fig. 5D; Supplementary Fig. S12; Supplementary Fig. S13A). WNT974-resistant KRPshS cells were transcriptionally distinct from TGFβ-primed cultures, with some overlap in the RevSC-like cluster. They showed further depletion of differentiated intestinal lineages (Supplementary Fig. S13B) as well as low WNT and adult stem cell signatures (Fig. 5E, middle; Supplementary Fig. S12; Supplementary Fig. S13A and S13B). Most notably, WNT974-resistant organoids showed marked elevation of the fetal intestinal signature (Fig. 5E, bottom), consistent with a role for this transition in the development of WNT independence. Importantly, all scRNA-seq profiling was performed more than 6 weeks after the withdrawal of TGFβ and 30 days after withdrawal of WNT974. Thus, transcriptional reprogramming and the associated lineage switch is maintained over long-term culture and does not require sustained TGFβ signaling or active WNT suppression.

YAP1 and TAZ (WWTR1) are closely related transcriptional coactivators in the Hippo signaling pathway that have been linked to the acquisition of stem and progenitor cell properties in breast, pancreas, and neuronal tissue (35), and for tissue regeneration in the colonic epithelium (36). YAP1 and TAZ are direct transcriptional targets of SMAD2/3 (refs. 37, 38; Supplementary Fig. S14A) and were induced following TGFβ treatment (Fig. 6A). YAP transcriptional signatures were strongly induced by TGFβ treatment and remained high following TGFβ withdrawal (Fig. 6A and B). Remarkably, the expression of a small set of recently reported canonical YAP target genes (39) was sufficient to accurately cluster the organoid genotypes into pre-TGFβ, post-TGFβ, and WNT-independent subtypes (Supplementary Fig. S14B). Moreover, this comparison, along with single-cell analysis, showed that the relative amplitude of the YAP signature was increased in WNT-independent organoids relative to TGFβ-primed cells (Fig. 6C; Supplementary Fig. S14B). Finally, analysis of chromatin accessibility TGFβ-primed and WNT974-resistant organoids by assay for transposase-accessible chromatin using sequencing (ATAC-seq) revealed the consensus TEAD binding motif as the most significantly enriched of all known transcription factor binding motifs in WNT-independent cells (Fig. 6D); the TEAD family are transcriptional co-factors for YAP and TAZ. Together, these molecular data suggest that transcriptional reprogramming associated with WNT independence is likely driven by direct YAP/TAZ/TEAD DNA binding, and that robust YAP/TAZ activity may be critical for the establishment of WNT independence. YAP/TAZ induction was also clearly apparent in vivo, with KRPS tumors showing markedly elevated nuclear YAP/TAZ staining compared with KRP and KRPT genotypes (Supplementary Fig. S14C). We further observed prominent nuclear (nonphosphorylated) active YAP in 7 of 8 human RSPO2/RSPO3 fusion colorectal cancers examined, all of which contained concomitant mutations in KRAS or BRAF (Supplementary Fig. S15). A subset of these tumors also showed weak-to-moderate expression of two markers associated with fetal intestinal identity, ANXA1 and MSLN (Supplementary Fig. S15A and S15B). Interestingly, 2 of 3 tumors that showed expression of all three markers carried mutations in Kras (or Braf), Trp53, and Smad4 (Supplementary Fig. S15B).

To directly assess whether YAP/TAZ signaling was the key driver of WNT independence downstream of TGFβ, we silenced YAP and TAZ in TGFβ-naïve KRPshS organoids using tandem shRNAs. YAP/TAZ silencing in these cells had no obvious impact on morphology or organoid growth under basal conditions, but, like Smad2/3 knockdown, partially inhibited the transition to spheroid morphology (Supplementary Fig. S16A) and blocked key molecular and transcriptional changes associated with lineage reversion (Fig. 6E–G). YAP/TAZ activity was not only required for spheroid formation and transcriptional changes downstream of TGFβ, but was sufficient to induce WNT independence. Expression of either stabilized YAP (YAP^S5A) or TAZ (TAZ^S4A) in WNT-dependent KRP organoids was sufficient to downregulate adult intestinal lineage markers, upregulate fetal intestinal signature genes, and drive spheroid transformation in the absence of TGFβ priming (Fig. 6H; Supplementary Fig. S16B). Moreover, KRP-YAP^S5A and KRP-TAZ^S4A organoids were refractory to WNT inhibition with WNT974, showing only a moderate decrease in EdU incorporation (Fig. 6I), comparable to KRPS^R organoids (Fig. 2B).

To determine whether YAP (or TAZ) activation could induce WNT independence in human colorectal cancer, we identified a BRAF/TP53/SMAD4–mutant patient-derived organoid line (“MSK121Li”) with a homozygous disruptive mutation in RNF43 (40). RNF43 is an E3 ligase that modulates WNT signaling by controlling the abundance of the Frizzled/LRP WNT receptors at the cell surface. Like RSPO fusions, RNF43-mutant cells are predicted to be sensitive to PORCN inhibitors. Indeed, unlike APC-mutant cells (“MSK132P”) that were completely insensitive to WNT974 (Fig. 6J and K), MSK121Li cultures showed morphologic changes reminiscent of mucinous-like differentiation (Fig. 6J), rapid cell-cycle arrest (Fig. 6K), and could not be maintained for more than 2 passages (∼10 days) in the presence of WNT974. Similar to murine organoids, expression of YAP^S5A or TAZ^S4A blunted the acute WNT974-induced cell-cycle arrest (Fig. 6L, left) and enabled the outgrowth of WNT974-resistant organoids when maintained in drug over 30 days (Fig. 6M). As expected, YAP/TAZ-expressing cells maintained very low levels of WNT target gene expression in the presence of WNT974 (Fig. 6N), suggesting they had bypassed WNT dependence. Similarly, expression of YAP^S5A or TAZ^S4A in the RSPO3 fusion–positive human colorectal cancer cell line SNU1411 led to a 2.5-fold increase in proliferation following treatment with WNT974 (Fig. 6L, right).

To determine whether YAP/TAZ-mediated signaling was required for maintenance of WNT independence, we silenced YAP, TAZ, or both YAP and TAZ in KRP, KRPS^N, and KRPS^R cells and measured proliferation by fluorescence-based competition assay. TAZ knockdown alone had very little impact on the proliferation of organoids relative to untransduced cells, whereas YAP silencing had a more dramatic effect on KRPS^R cells (Fig. 6O). Combined YAP/TAZ suppression induced a rapid and near-complete depletion of shRNA-expressing cells by 8 days (Fig. 6O). In contrast, YAP/TAZ knockdown caused only a minor (∼20%) reduction in EdU incorporation in KRP cells (Supplementary Fig. S16C), highlighting the switch from WNT to YAP dependency following TGFβ-primed transcriptional rewiring and WNT independence. Interestingly, TGFβ-primed, WNT974-naïve (KRPS^N) organoids were acutely sensitive to inhibition of both WNT (Fig. 2B) and YAP (Fig. 6O), suggesting they represent a dual-dependent lineage intermediate.

Together, these data suggest that YAP/TAZ transcriptional activity is the central requirement for TGFβ-primed lineage reversion, and that, following the transition to WNT independence, YAP/TAZ signaling becomes essential for survival and proliferation.

WNT pathway hyperactivation is the most frequent initiating factor in human colorectal cancer and thought to be one of the most important molecular drivers in this disease. Here, we show that intestinal tumor cells carrying common cancer-associated genetic alterations can become primed to evade targeted WNT inhibition and rapidly evolve to become completely WNT-independent. WNT independence is initiated by canonical TGFβ signaling, which drives YAP/TAZ-dependent transcriptional reprogramming and lineage reversion. This lineage switch is irreversible, and consequently WNT-independent cells become reliant on YAP/TAZ signaling and can be selectively depleted by suppressing YAP/TAZ.

The role of YAP in intestinal biology is complex. YAP is required for early phases of tumor initiation in APC-mutant cells (41, 42), but in normal intestinal epithelium, YAP induction leads to crypt loss and intestinal dysfunction (43). Likewise, activating genetic alterations in YAP, TAZ, and TEAD are rare in colorectal cancer, although YAP signaling can be important for proliferation and metastasis in WNT-driven colorectal cancer xenografts (44). Our data show that YAP/TAZ signaling can be a potent driver of survival and proliferation in intestinal cancer cells, independent of WNT activation. This mirrors a recently described role for YAP/TAZ in normal tissue regeneration following colitis (36) and irradiation-induced mucosal damage (34). In both cases, YAP-mediated repair is associated with upregulation of fetal markers, although it is unclear whether the regenerating epithelium, like fetal organoids (31), is WNT-independent. Taken together, it seems likely that TGFβ priming in cancer cells can promote WNT independence by hijacking a normal physiologic wound-repair process to drive lineage reversion.

Despite the similarities, resistance to WNT blockade in cancer cells has two critical and fundamental differences to the wound-repair program. First, TGFβ priming, unlike injury-induced YAP activation, is dependent on the presence of multiple oncogenic mutations (e.g., KRAS/BRAF, TP53, and SMAD4), indicating that this mode of reprogramming is likely cancer-specific. Second, YAP-mediated wound repair is a tightly controlled and transient process (34, 36), whereas lineage reversion of KRPS/AKPS cells is irreversible. Even following months of culture in the absence of WNT suppression, WNT-independent organoid cultures remain resistant to WNT974 and reliant on YAP/TAZ for survival. Understanding the factors that dictate the differences between reversible YAP activation in injury repair and permanent YAP reprogramming in WNT inhibitor resistance may identify key cellular switches that could be harnessed to prevent or reverse WNT independence in therapy-refractory tumors. For instance, while blocking the TGFβ receptor or restoring SMAD4 has no impact on the maintenance of WNT independence, it is possible that cytokine-dependent YAP induction guides reprogramming on a different course to transient wound repair.

Unlike with SMAD4, we show that disruption of p53, which affects cellular reprogramming and lineage plasticity in multiple settings (45–48), is critical for the initiation and maintenance of lineage reversion. Consistent with this, mutations in TP53 have been associated with the development of WNT/RSPO niche independence in human gastric and pancreatic organoids (49, 50), although the precise mechanisms that drive pathway independence in each tissue type may be distinct. Interestingly, TP53 disruption is extremely common in patients with colitis-associated colorectal cancer, whose tumors show relatively infrequent WNT-activating mutations (51). It is tempting to speculate that induction of an inflammation-induced YAP regenerative program in p53 mutant cells would enable WNT-independent cancer growth. It is notable that the RSPO3 fusion colorectal cancers we identified with elevated active YAP1, ANXA1, and MSLN expression each contained TP53 (and KRAS/BRAF) mutations, but did not all contain SMAD4 alterations. Although the number of samples is limited, it is likely that SMAD4 loss is not an absolute requirement for YAP/TAZ activation in colorectal cancer, which can be induced by factors other than TGFβ (44). Hence, although we describe one mechanism of lineage reversion and WNT independence, we suspect that there will be other genetic combinations, contexts, and/or extracellular stimuli that induce a similar WNT independent protumorigenic outcome. Similar to our work, previous studies have identified elevated expression of a variety of YAP and fetal-like markers in human colorectal cancer, including ANXA1 (52), SPP1 (53), and MSLN (54, 55). Whether these markers alone can distinguish WNT-independent or “primed” human tumors is unknown.

Our functional studies indicate that both YAP and TAZ, individually, are capable of inducing lineage reversion and WNT independence, whereas knockdown experiments suggest YAP may be a more dominant driver of this process. Both YAP and TAZ are controlled post-translationally by phosphorylation and degradation (56), but it remains unclear exactly how these processes support ongoing pathway activation to maintain WNT independence. One plausible mechanism is through engagement with the extracellular matrix and activation of YAP/TAZ via FAK and SRC, as has been described in wound repair (34, 36). Interestingly, while YAP and TAZ were sufficient to drive WNT independence in our patient-derived organoid model, TGFβ was not sufficient to induce this response in the same way as murine organoids, at least in the one patient-derived organoid sample we assessed. We do not yet know the reason for this, although it is possible that culture in mouse-derived extracellular matrix (Matrigel) does not provide the necessary extracellular cues to trigger appropriate intracellular signaling to YAP, as was observed in the context of wound repair (36). Such a mechanism would also provide a reasonable explanation for the lack of PORCN inhibitor resistance reported in previous xenograft studies (12, 13). It is also likely that there are other unidentified factors that influence the activation of YAP/TAZ and WNT inhibitor resistance. Identifying those genes and pathways required to maintain WNT independence will help reveal these signaling networks and may highlight new therapeutic opportunities. Regardless of the mechanism of activation, the dependence on YAP/TAZ for survival of WNT-independent cells suggests that therapies directly targeting these effectors would be an attractive approach to prevent or reverse WNT inhibitor resistance. As potent and selective YAP/TAZ-targeted agents become available, it will be important to determine whether they can be used safely in combination, or sequentially, with WNT inhibitors.

The work described here was guided by the clinical genetics of colorectal cancer, and in particular RSPO fusion colorectal cancer. Because of the paucity of pre- and post-treatment clinical samples, we exploited engineered murine organoids to interrogate the mechanisms of WNT dependence, but this work has obvious implications for clinical application of WNT therapies. There is evidence that lineage changes during colonic wound repair are similar in mice and humans (36), but, owing to lack of clinical treatment data, we do not yet know the exact conditions or frequency with which lineage reversion occurs in human colorectal cancer or how it will affect clinical treatment response. To date, WNT-targeted drugs have shown a range of on-target dose-limiting toxicities, including bone fractures (57), although recent work has demonstrated that a combination of RANKL and PORCN inhibitors provides potent WNT pathway suppression while avoiding bone-related adverse events (26). On the basis of these data, new combination trials have been launched, specifically targeting RSPO3-fusion colorectal cancer. It is too early to know the outcome of these new trials, but given the prevalence of p53, SMAD4, and RAS/RAF alterations in RSPO fusion cancers, it will be critical to pay close attention to how genotype influences clinical tumor behavior and to assess features of lineage reversion in emergent resistant disease. Ultimately, such genetic biomarkers may serve to further stratify patients to improve clinical outcomes.

Sequences encoding guide RNAs (Supplementary Table S2) were cloned into the BsmBI site of a Cas9-P2A-Cre (LCC) lentiviral vector. For tandem guide RNA cloning, each U6-sgRNA cassette was amplified from px458 vector and ligated into the EcoRI site of LCC. For shRNA cloning, 97mer oligonucleotides (Supplementary Table S2) were PCR amplified as previously described (58, 59) using specific primers (Supplementary Table S2) and cloned into the XhoI/EcoRI sites of SGEN (ref. 59; Addgene #111171). For tandem shRNA cloning, shRNAs were cloned sequentially into the EcoRI site downstream of the existing shRNA in SGEN, as described previously (60). For cDNA cloning, cDNAs were PCR amplified using specific primers from gBlocks (Supplementary Table S2) synthesized at Integrated DNA Technologies, and ligated into the BglII/MluI sites of pLMN (61) using Gibson assembly.

All studies involving animals were approved by the Institutional Animal Care and Use Committee (IACUC) at Weill Cornell Medicine (New York, NY), under protocol number 2014-0038. For APC restoration studies, animals were fed with dox chow (200 mg/kg; Harlan Teklad) 1 day before transplantation. Organoids (∼100,000 cells for each flank) were implanted subcutaneously into athymic nude mice. Animals were distributed randomly into “On Dox” and “Off Dox” groups when tumors reached a volume of approximately 100 mm³. For WNT974 treatment studies, organoids (∼100,000 cells for each flank) were implanted subcutaneously into athymic nude mice. After 2 weeks, animals were randomized for WNT974 treatment. WNT974 (5 mg/kg, Selleckchem #S7143) was mixed with 0.5% methylcellulose and 0.1% Tween 80 and then administered by daily oral gavage for 2 weeks. All animals were humanely euthanized after 5 weeks. Macrodissected tumors were weighed, fixed in fresh 4% paraformaldehyde (PFA), and processed for histology.

Tissue, fixed in freshly prepared 4% PFA for 24 hours, was embedded in paraffin and sectioned by IDEXX RADIL. Sections were rehydrated and unmasked (antigen retrieval) by either: (i) heat treatment for 5 minutes in a pressure cooker in 10 mmol/L Tris/1 mmol/L EDTA buffer (pH 9) containing 0.05% Tween 20 or (ii) for lysozyme staining only, Proteinase K treatment (200 μg/mL) for 10 minutes at 37°C. For IHC, sections were treated with 3% H₂O₂ for 10 minutes and blocked in TBS/0.1% Triton X-100 containing 1% BSA. MSLN staining was performed in a Leica Bond automated platform, with pretreatment (ER2) for 30 minutes. For immunofluorescence, sections were not treated with peroxidase. Primary antibodies, incubated at 4°C overnight in blocking buffer, were: rabbit anti-Ki-67 (1:100, Sp6 clone, Abcam #ab16667), rabbit anti-KRT20 (1:200, Cell Signaling Technology, #13063), rabbit anti-Lysozyme (1:1,000, Dako, #EC 3.2.1.17), rat anti-BrdU (1:200, Abcam #ab6326), rabbit anti-YAP/TAZ (1:200, Cell Signaling Technology, #8418), mouse anti-SPP1 (1:200, Santa Cruz Biotechnology, #sc-21742), anti-pSMAD3 (1:200, Abcam #ab52903), anti-ANXA1 (1:200, Invitrogen, #PA5-27315, anti-MSLN (1:100 Vector Labs, clone 5B2, # VP-M649), anti-active YAP (1:2,000, Abcam #205270). For IHC, sections were incubated with anti-rabbit or anti-rat ImmPRESS HRP-conjugated secondary antibodies (Vector Laboratories, #MP7401) and chromagen development performed using ImmPact DAB (Vector Laboratories, #SK4105). Stained slides were counterstained with Harris hematoxylin. For immunofluorescent stains, secondary antibodies were applied in TBS for 1 hour at room temperature in the dark, washed twice with TBS, counterstained for 5 minutes with DAPI, and mounted in ProLong Gold (Life Technologies, #P36930). Secondary antibodies used were: anti-rabbit 568 (1:500, Molecular Probes, #a11036). Images of fluorescent and IHC-stained sections were acquired on a Zeiss Axioscope Imager (chromogenic stains), Nikon Eclipse T1 microscope (IF stains). Raw .tif files were processed using FIJI (ImageJ) and/or Photoshop (Adobe Systems Inc.) to create stacks, adjust levels, and/or apply false coloring.

Isolation, maintenance, and staining of mouse intestinal organoids has been described previously (62, 63). Briefly, for isolation, 15 cm of the proximal small intestine was removed and flushed with cold PBS. The intestine was then cut into 5-mm pieces, vigorously resuspended in 5 mmol/L EDTA-PBS using a 10 mL pipette, and placed at 4°C on a benchtop roller for 10 minutes. This was then repeated for a second time for 30 minutes. After repeated mechanical disruption by pipette, released crypts were mixed with 10 mL DMEM Basal Media [Advanced DMEM F/12 containing penicillin/streptomycin, glutamine, 1 mmol/L N-Acetylcysteine (Sigma Aldrich A9165-SG)] containing 10 U/mL DNAse I (Roche, 04716728001), and filtered sequentially through 100-μm and 70-μm filters. One milliliter FBS (final 5%) was added to the filtrate and spun at 125 × g for 4 minutes. The purified crypts were resuspended in basal media and mixed 1:10 with growth factor–reduced Matrigel (BD Biosciences, 354230). 40 μL of the resuspension was plated per well in a 48-well plate and placed in a 37°C incubator to polymerize for 10 minutes. Two hundred microliters of small intestinal organoid growth media [basal media containing 50 ng/mL EGF (Invitrogen PMG8043), 100 ng/mL Noggin (PeproTech 250-38), and 500 ng/mL R-spondin (R&D Systems, 3474-RS-050, or from conditioned media] was then laid on top of the Matrigel. Where indicated, dox was added to experiments at 500 ng/mL.

For subculture and maintenance, media was changed on organoids every two days and they were passaged 1:4 every 3 to 5 days. To passage, the growth media was removed and the Matrigel was resuspended in cold PBS and transferred to a 15 mL falcon tube. The organoids were mechanically disassociated using a p1000 or a p200 pipette and pipetted 50 to 100 times. Seven milliliters of cold PBS was added to the tube and pipetted 20 times to fully wash the cells. The cells were then centrifuged at 125 × g for 5 minutes and the supernatant was aspirated. They were then resuspended in GFR Matrigel and replated as above. For freezing, after spinning the cells were resuspended in Basal Media containing 10% FBS and 10% DMSO and stored in liquid nitrogen indefinitely.

For fixed staining, organoids were grown in 40 μL of Matrigel plated into an 8-well chamber slide (Lab-Tek II, 154534). Where indicated, 10 μmol/L EdU was added to the growth media for 6 hours before fixing. The growth media was removed and the cells were fixed in 4% PFA-PME (50 mmol/L PIPES, 2.5 mmol/L MgCl₂, 5 mmol/L EDTA) for 20 minutes. They were then permeabilized in 0.5% Triton for 20 minutes and blocked in IF Buffer (PBS, 0.2% Triton, 0.05% Tween, 1% BSA) for 1 hour or immediately processed for EdU staining performed according to directions provided with the Click-iT Edu Alexa Fluor 647 Imaging Kit (Invitrogen C10340). For alkaline phosphatase staining, fixed cells were washed twice with TBS and then incubated with the BCIP/NBT Substrate Kit (Vector Laboratories, SK-5400) for 15 minutes in the dark. The chambers were then washed twice with TBS and then imaged using brightfield microscopy. For immunofluorescent staining, cells were incubated in primary antibodies overnight in IF buffer: rabbit anti-KRT20 (1:200, Cell Signaling Technology, #13063), rabbit anti-Lysozyme (1:200, Dako, #EC 3.2.1.17). They were then washed three times with TBS 0.1% Tween. Secondary antibodies (1:1,000, same reagents as above) were incubated with or without Alexa-647 Phalloidin (Molecular Probes, #A22287) for 1 hour. The solution was removed and DAPI in PBS was added for 5 minutes and washed twice with TBS 0.1% Tween. The chambers were then removed and coverslips were mounted using Prolong Gold antifade medium (Invitrogen P36930). Images were acquired using Zeiss LSM 880 laser scanning confocal microscope, and Zeiss image acquiring and processing software. Images were processed using FIJI (ImageJ) and Photoshop CS software (Adobe Systems Inc.). For determining the proportion of spheroids to organoids, total number of both live organoids and spheroids was counted in representative microscopy fields using 4× objective lens. At least three different fields for each sample were counted. The percentage of spheroids is calculated by dividing the number of spheroids to total number of organoids.

Murine small intestinal organoids were cultured in transfection medium containing CHIR99021 (5 μmol/L) and Y-27632 (10 μmol/L) for 2 days prior to transfection. Single-cell suspensions were produced by dissociating organoids with TrypLE express (Invitrogen #12604) for 5 minutes at 37°C. After trypsinization, cell clusters in 300 μL transfection medium were combined with 100 μL DMEM/F12-Lipofectamine 2000 (Invitrogen #11668)-DNA mixture (97 μL-2 μL-1 μg), and transferred into a 48-well culture plate. The plate was centrifuged at 600 × g at 32°C for 60 minutes, followed by another 6-hour incubation at 37°C. The cell clusters were spun down and plated in Matrigel. For transduction, organoids were pretreated and dissociated as for transfection, then mixed with viral supernatant and transferred to a 48-well plate for spinoculation. The plate was centrifuged at 600 × g at 32°C for 60 minutes, followed by another 4- to 6-hour incubation at 37°C. The cell clusters were spun down and plated in Matrigel.

To select for organoids with Ptprk–Rspo3 rearrangements, exogenous RSPO1 was withdrawn from the culture media 1 week after transfection. For selection of organoids with Kras or Braf mutations, organoids were treated with 1 μmol/L gefitinib for 2 weeks. For selection of organoids with loss-of-function Trp53 mutations, organoids were cultured in medium containing Nutlin-3 (5 μmol/L). For selection of organoids with loss-of-function Smad4 or Tgfbr2 alterations, cells were cultured in medium containing TGFβ1 (5 ng/mL) for 1 week. To generate TGFβ-naïve Smad4/Tgfbr2 organoids, cells transduced with both Trp53 and Smad4/Tgfbr2 sgRNAs were selected using Nutlin-3 only.

Human cancer samples were collected and analyzed under appropriate Institutional Review Board protocols and waivers (#06-107, 12-245, 16-1261). DNA sequencing was performed with the MSK-IMPACT assay, a targeted exome capture assay with deep sequencing coverage (64). RSPO2/3 fusions were detected with a custom Archer-targeted RNA seq–based next-generation sequencing assay (65). Human cancer samples were collected and analyzed under appropriate Institutional Review Board protocols and waivers (#06-107, 12-245, 16-1261). DNA sequencing was performed with the MSK-IMPACT assay, a targeted exome capture assay with deep sequencing coverage (61). RSPO2/3 fusions were detected with a custom Archer-targeted RNA seq–based next-generation sequencing assay (62). Cases with both APC/CTNNB1/RNF43 and RAS hotspot/BRAF p.V600E mutations were not analyzed by Archer for RNA-level fusions. Therefore, some selection bias is present. All studies were conducted in accordance with the Declaration of Helsinki. Classification of tumor immunohistochemistry was performed while blinded to sample genotype.

Patient-derived organoids were derived under Memorial Sloan Kettering Institutional Review Board biospecimen research protocols 14-244. All studies were conducted in accordance with the Declaration of Helsinki, and all patients provided preprocedure written informed consent. Tissue was processed within 1 hour of surgical resection and organoids derived as described previously (40). Conservation of truncal driver mutations was verified by MSK-IMPACT of formalin-fixed, paraffin-embedded tissue block and cognate organoids.

Small intestine organoids were grown in 300 μL of Matrigel in 1 well of 6-well plate for 4 days post passage. Organoids were then recovered from the Matrigel using cell recovery solution (Corning #354253). Organoid pellets were lysed with RIPA buffer. Antibodies used for Western blot analysis were: mouse anti-p53 (1:1,000, Cell Signaling Technology, #2524), rabbit anti-SMAD4 (1:1,000, Cell Signaling Technology, #46535), rabbit anti-SMAD2/3 (1:1,000, Cell Signaling Technology, #8685), rabbit anti-YAP (1:1,000, Cell Signaling Technology, #14074), rabbit anti-YAP/TAZ (1:1,000, Cell Signaling Technology, #8418), rabbit anti-RUNX2 (1:1,000, Cell Signaling Technology, #12556), mouse anti–E-Cadherin (1:1,000, Cell Signaling Technology, #14472), rabbit anti-phospho-AKT (Ser473; 1:1,000, Cell Signaling Technology, #4060), rabbit anti-N-MYC (1:1,000, Cell Signaling Technology, #84406), rabbit anti-c-JUN (1:1,000, Cell Signaling Technology, #9165), rabbit anti-JUNB (1:1,000, Cell Signaling Technology, #3753), rabbit anti-ETS-1 (1:1,000, Cell Signaling Technology, #14069), rabbit anti-GAPDH (1:1,000, Cell Signaling Technology, #5174), rabbit anti-Cyclophilin B (1:1,000, Cell Signaling Technology8, #43603), rabbit anti-pan-TEAD (1:1,000, Cell Signaling Technology, #13295), rabbit anti-phospho-SMAD2 (Ser465/467; 1:1,000, Cell Signaling Technology, #3108), mouse anti-phospho-SAPK/JNK (Thr183/Tyr185; 1:1,000, Cell Signaling Technology, #9255), rabbit anti-phospho-p38 MAPK (Thr180/Tyr182; 1:1,000, Cell Signaling Technology, #4511), mouse anti-TEAD4 (1:1,000, Abcam, #ab58310), rabbit anti-phospho-SMAD3 (Ser423/425; 1:1,000, Abcam, #ab52903).

Organoids were pulsed with 10 μmol/L EdU for 3 hours in regular tissue culture incubator. Cells were pelleted and trypsinized using 0.25% Trypsin-EDTA (Thermo Fisher Scientific #25200056) at 37°C for 5 minutes, resuspended using FACS buffer (PBS + 2% FBS) then filtered through cell strainer (Corning #352235). EdU was stained using Click-iT EdU Alexa Fluor 647 Flow Cytometry Assay Kit (C104424) following the manufacturer's instructions. Flow cytometry was conducted on Invitrogen Attune NxT following the manufacturer's instructions. For fluorescence-based competition assay, cells were transduced with lentiviral vectors containing GFP-linked shRNA/s or cDNAs. Two days after viral transduction, the frequency of GFP was measured by flow cytometry two days post transduction (D2) and remaining cells were replated in Matrigel culture. Every three days, cells were split for flow cytometry analysis and replating. The percentage of GFP-positive cells at each passage is normalized to D2.

RNA was extracted using TRIzol according to the manufacturer's instructions, and contaminating DNA was removed by DNase treatment for 10 minutes and column purification (Qiagen RNAeasy). cDNA was prepared from 1 μg total RNA using qScript reverse transcription kit (Quantabio, #95047). Quantitative PCR detection was performed using SYBR green reagents (Quantabio #101414-288) and specific primers listed in Supplementary Table S2.

Each gDNA sample based on Qubit quantification was mechanically fragmented on an Covaris E220 focused ultrasonicator (Covaris). Two hundred nanograms of sheared gDNA was used to perform end repair, A-tailing, and adapter ligation with Agilent SureSelect XT (Agilent Technologies) library preparation kit following the manufacturer's instructions. Then, the libraries were captured using Agilent SureSelectXT Mouse All Exon probes, and amplified. The quality and quantities of the final libraries were checked by Agilent 2100 Bioanalyzer and Invitrogen Qubit 4.0 Fluorometer (Thermo Fisher Scientific), and libraries were pooled at 8 samples per lane and sequenced on an Illumina HiSeq 4000 sequencer (Illumina Inc.) at PE 2 × 100 cycles. Copy-number alterations were identified and plotted using cnvkit (v0.9.6) and single nucleotide variant called using MuTect2.

Total RNA was isolated using TRIzol, DNAse treated, and purified using the RNeasy Mini Kit (Qiagen). Following RNA isolation, total RNA integrity was checked using a 2100 Bioanalyzer (Agilent Technologies). RNA concentrations were measured using the NanoDrop system (Thermo Fisher Scientific, Inc.). Preparation of RNA sample library and RNA-seq were performed by the Genomics Core Laboratory at Weill Cornell Medicine. Messenger RNA was prepared using TruSeq Stranded mRNA Sample Library Preparation Kit (Illumina), according to the manufacturer's instructions. The normalized cDNA libraries were pooled and sequenced on Illumina NextSeq500 sequencer with single-end 75 cycles.

The quality of raw FASTQ files were mapped to mouse reference GRCm38 using STAR two-pass alignment (v2.4.1d; default parameters; ref. 66), and transcript abundance estimates were performed using Kallisto (67), aligned to the same (GRCm38) reference genome. Kallisto transcript count data for each sample were concatenated, and transcript per million (TPM) data were reported for each gene after mapping gene symbols to ensemble IDs using R packages “tximport”, “tximportData”, “ensembldb”, and “EnsDb.Mmusculus.v79”. Differential gene expression was estimated using DESeq2 (68). For data visualization and gene ranking, log fold changes were adjusted using lfcShrink in DESeq2, to minimize the effect size of poorly expressed genes. GSEA (v3.0) was performed on preranked gene sets from differential expression between control and treated groups. We used a custom curated gene set including the Broad “HALLMARK” collection and published bulk and single-cell RNA-seq from murine intestine (https://github.com/lukedow/Genesets.git). We used R (v3.6.1) and R Studio (v1.2.1335) to create all visualizations, perform hierarchical clustering, and perform principal component analysis. Volcano plots, heat maps, and other visualizations were produced using the software packages Enhanced Volcano (https://bioconductor.org/packages/devel/bioc/html/EnhancedVolcano.html); pheatmap (https://cran.r-project.org/web/packages/pheatmap/index.html); ggplot2 (https://cran.r-project.org/web/packages/ggplot2/index.html).

ATAC-seq library preparation, sequencing, and post-processing of the raw data was performed at the Epigenomics Core at Weill Cornell Medicine by using the OMNI-ATAC-seq method described previously (69). Briefly, organoids were dissociated from Matrigel using cell recovery solution (Corning #354253), and 50,000 live cells were spun down and incubated for 3 minutes at 4°C in 25 μL of a detergent buffer containing 0.2% Igepal CA-630 (Sigma-Aldrich), 0.2% Tween 20 (Sigma-Aldrich), and 0.02% Digitonin (Promega Corporation). Nuclei were centrifuged at 500 × g for 10 minutes and immediately resuspended in 25 μL of buffer containing 2.5 μL of Tn5 transposase (Illumina, Inc., catalog no. 15027865) for a 30-minute incubation at 37°C. Fragments generated by the Tn5 transposase were purified using the DNA Clean and Concentrate Kit from Zymo Research (Zymo Research, catalog no. D4014). Uniquely indexed libraries were obtained by amplification of the purified fragments with indexed primers using 9 cycles of PCR (5 minutes × 72°C, 5 cycles each 10 seconds × 98°C, 30 seconds × 63°C, 1 minute × 72°C). Resulting libraries were subjected to a two-sided size clean-up using SPRI beads (Beckman Coulter) to obtain sizes between 200 and 1,000 bp, and pooled for sequencing. The pool was clustered at 9 pmol/L on a paired-end read flow cell and sequenced for 50 cycles on an Illumina HiSeq 2500 to obtain approximately 40 M reads per sample. Primary processing of sequencing images was done using Illumina's Real Time Analysis software (RTA) as suggested by Illumina. CASAVA 2.17 software was used to perform image capture, base calling, and demultiplexing of the raw reads. FASTQ files generated were aligned to the mouse mm10 genome build using the BWA aligner.

Raw ATAC-seq reads were processed using ENCODE ATAC-seq pipeline. This includes Bowtie2 (70) alignment against mm10 and MACS2 (71) as peak caller. Duplicated reads and reads from mitochondria were filtered out. Peaks for all samples were merged together into a union of peaks, and the ones that existed in at least two samples were kept for further analysis. Peaks were annotated using the ChIPseeker (72) “annotatePeak” function with “TxDb.Mmusculus.UCSC.mm10.knownGene”. Numbers of reads mapped to each peak were counted with featureCounts (73), and raw read counts were normalized using rlog in DESeq2 (68). Differentially accessible peaks were defined as peaks with P_adj smaller than 0.05 and absolute value of log₂(FC) >1. Known and de novo motifs were identified using findMotifsGenome.pl from Homer (74) and “-size given-mask” was applied on the differentially accessible peaks, with nondifferentially accessible peaks as background.

Organoids were dissociated from Matrigel using cell recovery solution (Corning #354253), then trypsinized using 10× trypsin (Thermo Fisher Scientific #15090046) for 15 minutes at 37°C. Single cells were resuspended using organoid culture medium and stained with DAPI. DAPI-negative single cells were sorted using BD FACSAria II machine with 130-μm nozzle. Sorted single-cell suspensions were transferred to the Genomics Core Facility at Weill Cornell Medicine to proceed with the Chromium Single Cell 3′ Reagent Kit v3 (10x Genomics, product code # 1000075) using 10× Genomics' Chromium Controller. A total of 10,000 cells were loaded into each channel of the Single-Cell A Chip to target 5,000 cells in the end. Briefly, according to manufacturer's instructions, the sorted cells were washed with 1× PBS + 0.04% BSA, counted by Bio-Rad TC20 Cell Counter, and cell viability was assessed and visualized. A total of 10,000 cells and Master Mixes were loaded into each channel of the cartridge to generate the droplets on Chromium Controller. Beads-in-Emulsion (GEM) were transferred and GEMs-RT was undertaken in droplets by PCR incubation. GEMs were then broken and pooled fractions were recovered. After purification of the first-strand cDNA from the post GEM-RT reaction mixture, barcoded, full-length cDNA was amplified via PCR to generate sufficient mass for library construction. Enzymatic fragmentation and size selection were used to optimize the cDNA amplicon size. TruSeq Read 1 (read 1 primer sequence) was added to the molecules during GEM incubation. P5, P7, a sample index, and TruSeq Read 2 (read 2 primer sequence) were added via End Repair, A-tailing, Adaptor Ligation, and PCR. The final libraries were assessed by Agilent Technology 2100 Bioanalyzer and sequenced on Illumina NovaSeq sequencer with pair-end 100 cycle kit.

Reads were pseudoaligned and transcript abundances estimated using kallisto (v0.46.0), and gene count matrices were produced using bustools (v0.39.3; ref. 75) and BUSpaRse. Count matrices were processed and analyzed using the Seurat package (v3.1.1) in R (v3.6.1) and R Studio (v1.2.1335). Low-quality cells (less than 1,000 identified genes and greater than 10% mitochondrial reads) were removed from the dataset after normalization. Individual scRNA-seq datasets were merged function and clusters were resolved on the combined data. Expression of cell-type markers was plotted by calculating the mean expression of the genes indicated for each individual cell, using a custom function (“FeaturePlot.1”; https://github.com/nyuhuyang/SeuratExtra").

Raw exome-seq, RNA-seq, ATAC-seq, and scRNA-seq data have been deposited in the sequence read archive (SRA) under accession PRJNA578488.

O. Elemento reports personal fees and other support from Volastra Therapeutics (equity) and other support from OneThree Biotech (equity) outside the submitted work. J.F. Hechtman reports grants from Boehringer Ingelheim outside the submitted work. R. Yaeger reports grants from Boehringer Ingelheim during the conduct of the study; and grants and personal fees from Array BioPharma outside the submitted work. L.E. Dow reports personal fees from Mirimus Inc. (advisory board member) outside the submitted work. No potential conflicts of interest were disclosed by the other authors.

T. Han: Conceptualization, formal analysis, investigation, methodology, writing-original draft, writing-review and editing. S. Goswami: Formal analysis, methodology. Y. Hu: Software, formal analysis, methodology. F. Tang: Formal analysis, methodology. M.P. Zafra: Methodology. C. Murphy: Formal analysis, methodology. Z. Cao: Resources, methodology. J.T. Poirier: Formal analysis, methodology. E. Khurana: Formal analysis, supervision. O. Elemento: Supervision. J.F. Hechtman: Resources, formal analysis. K. Ganesh: Resources. R. Yaeger: Resources. L.E. Dow: Conceptualization, data curation, formal analysis, supervision, funding acquisition, writing-original draft, project administration, writing-review and editing.

We thank Wouter Karthaus, Charles Sawyers, and Darjus Tschaharganeh for technical and experimental advice. We thank Marie Parsons, Francisco Barriga, Michal Nagiec, Ashley Laughney, and J. Joshua Smith for advice and comments on preparation of the manuscript. This work was supported by a Research Scholar Award from the American Cancer Society (RSG-17-202-01-TBG), a project grant from the NIH/NCI (1R01CA222517-01A1), project grants from the Starr Cancer Consortium (#I10-0095 and #I11-0040) and a Stand Up To Cancer Colorectal Cancer Dream Team Translational Research Grant (SU2C-AACR-DT22-17). Stand Up To Cancer (SU2C) is a division of the Entertainment Industry Foundation. Research grants are administered by the American Association for Cancer Research, a scientific partner of SU2C. We thank the Weill Cornell Genomics Resource Core Facility who performed library preparation and sequencing for WES, RNA-seq, and scRNA-seq, and the Weill Cornell Epigenomics Core Facility who performed library preparation and ATAC-seq. M.P. Zafra is supported in part by NCI grant NIH T32 CA203702. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.