Colonic Organoids Derived from Human Pluripotent Stem Cells for Modeling Colorectal Cancer and Drug Testing

Introduction.

To model human disease of the large intestine using differentiated human embryonic stem cells or induced pluripotent stem cells (hESC/iPSCs), an effective protocol to derive colonic cells is required. Here, we report a robust strategy to efficiently generate colonic organoids (COs) from hESCs. Gene profiling confirmed that the COs represent colon rather than small intestine, containing stem cells, transit amplifying cells, and the expected spectrum of differentiated cells, including goblet cells, and endocrine cells. COs generated from iPSCs, derived from familial adenomatous polyposis patients with germline mutations in the gene encoding APC (FAP-iPSCs), show enhanced WNT activity and increased epithelial cell proliferation. Two potential drug candidates, XAV939 and rapamycin, decreased proliferation in FAP-COs, but also affected cell proliferation in wildtype COs, perhaps limiting therapeutic application for those compounds. In contrast, geneticin targeted abnormal WNT activity and thereby corrected the deregulated proliferation specific to FAP COs. These studies provide an efficient strategy to derive human COs, which can be used in disease modeling and drug discovery for colorectal disease.

The ability to differentiate hESCs or hiPSCs into colon stem cells could provide a platform to model patient-specific gastro-intestinal disease. Three dimensional self-organizing organoids were generated from hESCs and iPSCs representing intestinal¹, gastric², brain³, liver⁴, lung⁵, and pancreas⁶ “mini-organs”. While it is possible to generate colonic organoids (COs) from adult tissue, the derivation from hiPSC would significantly enhance the capacity to model human colorectal disease including cancer.

Applying pathways known to regulate normal mouse embryonic development, we developed a stepwise strategy for the progressive generation of definitive endoderm (DE), hindgut endoderm (HE), and subsequently COs (Fig.1a). For efficient generation of DE, culturing hESCs in 3 μM CHIR99021+100 ng/ml Activin A gave rise to more than 95% SOX17⁺ cells. (Supplementary Fig.S1a and Methods). To distinguish colon from small intestine, we reasoned that regionalization of HE would be required. In mouse embryos WNT signaling determines posterior fate in a dose-dependent manner by regulating expression of the posterior marker Cdx2⁷. Therefore, we screened WNT pathway modulators under chemically defined conditions to derive CDX2⁺ HE cells. To carry out our differentiation protocol in chemically defined conditions, we first tested replacing fetal bovine serum (FBS) with N2, B27 or knockout serum replacement (KOSR), which are commonly used chemically-defined supplements for optimizing cell growth in the absence of serum. B27 was most effective at increasing the percentage of CDX2⁺ cells, as measured by intracellular FACS (Supplementary Fig.S1b). Since GSK-3β lies downstream in the WNT canonical signaling pathway, a panel of 10 GSK-3β inhibitors was evaluated for their capacity to induce CDX2 expression. CHIR99021 treatment gave rise to more than 90% CDX2⁺ cells (Supplementary Fig.S1c). In addition, CHIR99021 showed significantly stronger capacity to upregulate the WNT pathway gene AXIN2, compared to human recombinant protein WNT3A (Supplementary Fig.S1d). Next, starting at day 8, HG cells were treated with colonic medium containing 3 μM CHIR99021, 300 nM LDN193189 and 500 ng/ml EGF for 12 additional days, to generate colonic epithelial cells. After dissociating and embedding in matrigel beads, single cells give rise to 2-3 cell clusters and become pseudostratified in embryonic gut-like spheroids (Supplementary Fig.S1e), and progressively cavitate into fully convoluted organoids (Fig.1b, c).

Figure 1. A defined strategy to derive colonic organoids from human embryonic stem cells (hESCs).

(a) Scheme of the stepwise differentiation strategy. The cell number at each differentiation stage is estimated based on a starting population of 1 million hPSCs. (b) Representative bright field images of developing hESC-derived COs. Scale bars = 100 μm. (c) Representative bright field image of a hESC-derived CO. Scale bar = 1 mm. (d) qRT-PCR analysis of colonic cell markers in hESC derived D38 small intestinal spheroids (SISs), D38 colonic spheroids (CSs), D48 small intestinal organoids (SIOs) and COs. (e) Immunohistochemistry analysis of hESC-derived SISs, CSs, SIOs and COs. E14.5 mouse embryonic SI and colon, and adult human duodenum tissue samples were used as the controls. Scale bars = 200 μm. (f) Immunocytochemistry analysis in sections of COs showing the presence of CDX2⁺, Ki67⁺, LGR5⁺, EPHB2⁺, KRT20⁺, VILLIN⁺, MUC2⁺, TPH1⁺, CHGA⁺, 5HT⁺, and GLP1⁺ cells. Scale bars = 100 μm. (g) The percentage of major cell types was quantified using MetaMorph Image Analysis Software (Molecular Devices). HE: hindgut endoderm, SISs, small intestinal spheroids, CSs, colonic spheroids, SIOs, small intestinal organoids, COs, colonic organoids. n=3 independent experiments for each condition. P values by unpaired two-tailed student t-test were *P<0.05, **P<0.01, ***P<0.001, ****P<0.0001. The center value is “median”. Error bar is S.D.

To test if these organoids have a colon phenotype, qRT-PCR was used to monitor the relative RNA expression of several colonic markers, including XPNPEP2⁷, HOXA7⁷, HOXD4⁸, HOXD10⁷, CAII⁹ and CAIV¹⁰ in the derived Day 38 (D38) spheroids (CSs) and D48 organoids (COs, Fig.1d). D38 small intestinal spheroids (SISs) and D48 organoids (SIOs) derived using a previous published protocol¹ were used as controls (Supplementary Table 1). Colonic markers were highly expressed in CSs and COs, relative to the SIS and SIO samples. In addition, we further performed immunohistochemistry analysis to confirm the expression of established colonic markers, CAII⁹, LY6A/E⁷ and CAIV¹⁰, in mouse embryonic colon tissue, human colon tissue, hESC-derived CSs and COs, but not mouse embryonic small intestine tissue, human duodenum tissue or hESC-derived SISs and SIOs (Fig.1e). The COs adopt typical crypt-like structures, in which proliferative cells at the bottom of the crypts express Ki67 (Fig.1f). More than 95% of cells in crypts express CDX2, and around 50% express the pan-differentiation marker Keratin 20 (KRT20) (Fig. 1f). Importantly, many of the cell types found in colonic/intestinal crypts are represented in the COs, including EPHB2⁺ transit amplifying (TA) cells¹¹, VILLIN⁺ epithelial cells¹², MUC2⁺ goblet cells¹³, and Chromogranin A (CHGA⁺) endocrine cells¹⁴ (Fig.1f and Supplementary Fig.S1f). Different subtypes of endocrine cells were detected, including the incretin glucagon-like peptide 1 (GLP1⁺)¹⁵ L-cells, and cells expressing tryptophan hydroxylase (TPH-1) and the neurotransmitter Serotonin (5HT)¹⁶ (Fig.1f). Of note, the expected proportion of the different cell types that make up the colon epithelium was observed (Fig.1g). qRT-PCR analysis confirmed that transcripts of stem cells markers LGR5 and BMI1¹⁷, quiescent stem cell marker HOPX¹⁸, TA cell marker EPHB2, and endocrine progenitor marker NEUROG3¹⁹, were significantly enhanced in D20 embryonic colon epithelium. Consistent with immunocytochemistry results, significantly increased levels of EPHB2, KRT20²⁰, VILLIN, MUC2, GLP-1 and TPH-1 were found in the later CO stage (Supplementary Fig.S1g). hESC-derived COs do not express anterior foregut endoderm marker SOX2²¹, posterior foregut endoderm markers PDX1²² and HNF6, or the proximal small intestine marker GATA4²³ (Supplementary Fig. S1h,i). The COs were predominantly epithelial, with a minor component of supportive mesodermal tissue, as no more than 5% vimentin⁺ mesenchymal cells were detected in COs, (Supplementary Fig.S1j). Compared to SIO samples, the transcript levels for Paneth cell markers, including LYZ, REGA, EPHB3, DEFA5, and small intestine markers, including GATA4 and OSR2, are significantly lower in COs (Supplementary Fig.S2a). Lysozyme was readily detected in human duodenum but not in hESC-derived COs (Supplementary Fig.S2b).

To further confirm the colonic identity of the derived COs, we compared the transcription profiles of COs and SIOs by RNA-seq. Colonic markers previously reported in a mouse study⁷ are highly expressed in hESC-derived COs, while small intestine markers are instead highly expressed in hESC-derived SIOs (Fig. 2a). Unbiased hierarchical clustering confirmed that hESC-derived COs closely resemble human colon, while hESC-derived SIOs closely resemble human duodenum and distal small intestine (Fig. 2b). Consistently, the genes expressed at relatively higher levels in human duodenum/distal small intestine and human colon followed a similar pattern of expression in hESC-derived SIOs and hESC-derived COs, respectively (Fig. 2c). Furthermore, gene set enrichment analysis (GSEA) showed that the gene set highly expressed in human colon is highly enriched in hESC-derived COs (Fig. 2d and Supplementary Fig.S2c). We also compared the expression profiles of our hESC-derived SIOs and COs with those previously reported for organoids derived from adult human duodenum²⁴, ileum²⁴, colon²⁵ and rectum²⁴. hESC-derived COs closely resemble the colon and rectum organoids. (Supplementary Fig.S2d). Finally, GSEA suggested that the gene set >50 fold upregulated in adult colon vs. fetal colon is highly enriched in hESC-derived COs (Fig.2e and Supplementary Fig.S2e).

Figure 2. Global gene expression analysis of hESC-derived COs or SIOs.

(a) Heatmaps derived from RNA-seq profiles of small intestine or colon marker genes in human duodenum, distal small intestine, colon (E-MTAB-1733) and hESC-derived COs and SIOs. (b) Heatmap showing sample-to-sample distances. Distance between two samples was calculated by comparing the normalized expression data with Euclidean distances. The darker color indicates no difference while the lighter color indicates larger distances. (c) Heatmap showing the expression data of marker genes of human duodenum, distal small intestine, and colon (E-MTAB-1733). Marker genes of each sample type were selected such that a gene has at least 3 times higher expression in that sample type compared to the other two sample types. (d) GSEA indicates that hESC-derived COs are significantly enriched for transcripts that are also highly expressed in human colon while hESC-derived SIOs are significantly enriched for transcripts that are also highly expressed in human duodenum. (e) GSEA indicates that an adult gene set (genes >50 fold upregulated in adult human colon/GSM1698576 vs. fetal human colon/GSM1698561) is significantly upregulated in hESC-derived COs and the fetal gene set (genes >50 fold upregulated in fetal human colon/GSM1698561 vs. adult human colon/GSM1698576) is significantly downregulated in hESC-derived COs. Enriched gene sets were selected based on statistical significance (FDR q value <0.25 and/or NOM p value <0.05).

Recently, COs derived from surgically resected human tissue were used to study the pathogenesis of colorectal cancer²⁶. Using our defined differentiation paradigm, we sought to use iPSC-derived COs to study the role of genetic factors in the progression of colorectal cancer. Familial adenomatous polyposis (FAP) is an inherited form of colorectal cancer, caused by mutation of the Adenomatous Polyposis Coli (APC) gene²⁷. Therefore, we created iPSCs from FAP patients and derived COs from FAP-iPSCs. Skin fibroblasts from two FAP patients were obtained, one (FAP8) with a 2 bp deletion of APC (4611_4612delAG) leading to a premature termination, and the other (FAP9) harboring a compound mutation: a C>T non-sense mutation (646C>T) resulting in early termination, and a C>T transition (2608C>T), which results in a proline 870 to serine substitution. Fibroblasts were obtained from an age and gender-matched control to generate iPSCs with wildtype APC alleles (wt-iPSCs). Fibroblasts were infected with lentivirus expressing OCT4, SOX2, KLF4 and cMYC and after two weeks iPSC colonies were picked and expanded²⁸. These showed typical iPSC colony morphology and expressed pluripotency markers (Supplementary Fig.S3a). Silencing of the viral transgenes in iPSCs was shown by qRT-PCR (Supplementary Fig.S3b), and the presence of the mutant alleles in FAP-iPSCs was confirmed using genomic DNA sequencing (Fig.3a). Finally, both in vitro differentiation to neurons (ectoderm), DE, and cardiomyocytes (mesoderm) (Supplementary Fig.S3c) and in vivo teratoma formation (Supplementary Fig.S3d) documented the pluripotency of the iPSCs.

Figure 3. FAP colonic organoids show enhanced WNT activity and increased cell proliferation.

(a) Mutations in the APC locus in FAP iPSC lines detected by genomic DNA sequencing. (b) Western blotting demonstrates reduced APC protein expression levels in FAP iPSCs. (c) Representative bright field images of wildtype (wt) and FAP COs. Scale bar= 1 mm. (d) Western blotting demonstrates the reduced APC protein expression levels in iPSC-FAP COs. (e) 3’3’-diaminobenzidine immunohistochemical staining and quantification of the percentage of nuclear β-catenin of wildtype and FAP iPSC-derived COs. β-catenin is enriched in nuclei only in the mutant lines (insets). Scale bar= 100 μm. (f, g) Gene ontology pathway analysis of genes that are at least two-fold higher expressed in FAP8 COs (f) or FAP9 COs (g) than wildtype COs. (h) qRT-PCR analysis of WNT pathway target genes in wildtype, FAP8 and FAP9 COs. (i) Maximal intensity projection images and (j) quantification of wildtype and FAP COs immunostained for CDX2 and CD1. (k) Maximal intensity projection images and (l) quantification of cells comparing wildtype and FAP COs immuno-stained for CDX2 and Ki67. Scale bar= 200 μm. n=3 independent experiments for each condition. P values by unpaired two-tailed student t-test were *P<0.05, **P<0.01, ***P<0.001, ****P<0.0001. The center value is “median”. Error bar is S.D.

Both FAP-iPSC lines harbor one allele with a nonsense mutation resulting in early termination of translation. Western blotting was used to measure the expression of APC proteins, which showed decreased expression levels in the FAP-iPSCs compared to wildtype control (Fig.3b). Loss of APC is known to enhance cell proliferation²⁹, so we evaluated expression of Ki67, in addition to CCND1 (which moves cells through the G1-S phase transition³⁰), to monitor the proliferation rate of iPSCs. In both FAP and wildtype iPSCs, essentially all cells are positive for these proliferation markers (Supplementary Fig.S4). No significant differences were detected comparing wildtype and FAP HE cells for either WNT target gene expression (Supplementary Fig.S5a) or cell proliferation (Supplementary Fig.S5b-e), consistent with relatively high WNT activity and proliferation rate even in wildtype cells at the HE stage.

Wildtype and FAP HE cells were further differentiated to COs (Fig. 3c and Supplementary Fig.S6a). Western blotting assays confirmed a decreased level of APC protein expression in FAP COs compared to wildtype (Fig. 3d). In stained sections of wildtype COs, β-catenin was clearly associated with the plasma membranes compared to nuclei. In contrast, there was significantly enhanced staining for β-catenin in the nucleus of FAP CO cells, consistent with activation of the WNT pathway (Fig. 3e and Supplementary Fig.S6b). RNA-seq and gene ontology pathway analyses highlighted the WNT pathway as strongly upregulated in FAP8 (Fig.3f) and FAP9 (Fig.3g) COs compared to wildtype COs, and genes associated with the WNT pathway were consistently upregulated in both FAP8 and FAP9 Cos, demonstrated by qRT-PCR (Fig.3h) and Ingenuity Pathway Analysis (Supplementary Fig.S6c and S6d). The nonsense mutation of FAP9 COs results in a shorter truncated APC protein compared to the frameshift mutation of FAP8 COs. Since the length of APC protein is correlated with WNT activity³¹, the expression levels of WNT target genes were predictably significantly higher in FAP9 COs compared to FAP8 COs (Fig.3h). Clustering analysis compared wildtype COs, FAP8 COs, FAP9 COs, normal mucosa samples, and adenomas from FAP patients. Transcript profiles from wildtype COs clustered with normal mucosa and those from FAP COs clustered with FAP adenomas (Supplementary Fig.S7a). Gene ontology pathway analysis found that similar pathways, including the WNT signaling pathway and colorectal cancer are upregulated in FAP COs and FAP adenomas (Supplementary Fig.S7b). GSEA indicates that FAP CO profiles are significantly enriched for transcripts that are also highly expressed in FAP colon adenoma (Supplementary Fig.S7c). Finally, wildtype and FAP COs were stained using antibodies to detect expression of CDX2 and CCND1 (Fig.3i), or CDX2 and Ki67 (Fig.3k) and z-stack maximal projections were captured by confocal microscopy and quantified to monitor proliferation of colonic epithelial cells. Consistent with activation of the WNT signaling pathway (Fig. 3f-h), FAP8 and FAP9 COs are comprised of a significantly higher percentage of CCND1⁺ or Ki67⁺ cells in the population of CDX2⁺ cells (Fig. 3j,l), indicating that colonic epithelial cells in FAP organoids have enhanced proliferation capacity, consistent with the early phenotype of FAP patients³².

A translational goal of disease modeling is to identify drugs that can reverse the pathogenic phenotype. We tested several compounds reported to ameliorate tumor-like phenotypes in colorectal cancer, including rapamycin³³, the Tankyrase inhibitor XAV939³⁴, and a combination of the ornithine decarboxylase inhibitor difluoromethylornithine (DFMO) and the selective COX-2 inhibitor Sulindac³⁵. Both XAV939 and rapamycin significantly decreased CDX2⁺ cell proliferation (Supplementary Fig. S8a, b). These effective compounds, were then used to treat wildtype and FAP COs. Consistent with the initial assay, the percentage of Ki67⁺ (Fig. 4a, b) or CD1⁺ (Fig.4c, d) cells in the CDX2⁺ cell population of both FAP8 and FAP9 COs was decreased when the cells were treated with 10 μM rapamycin or 5 μM XAV939. qRT-PCR experiments showed that XAV939 but not rapamycin, significantly down-regulated transcript levels for WNT target genes (Fig.4e). The results suggest that XAV939 functions through a WNT-dependent mechanism, while rapamycin functions through a WNT-independent mechanism. However, consistent with the known role of WNT to drive normal colonic epithelial stem cell proliferation, a similar impairment in cell proliferation was observed in wildtype organoids after exposure to these small molecules (Fig. 4a-d). This raises concern that rapamycin and XAV939 would not specifically target mutant tumor cells, and thereby could harm healthy colonic crypts.

Figure 4. A platform to evaluate drug candidates that can rescue APC mutation-induced hyper-proliferation in human colonic organoids.

(a) Maximal intensity projection images and (b) quantification of cells in wildtype and FAP COs treated with DMSO, 10 μM rapamycin or 5 μM XAV939, after being immuno-stained for CDX2 and Ki67. (c) Maximal intensity projection images and (d) quantification of cells in wildtype and FAP COs treated with DMSO, rapamycin or XAV939, after being immuno-stained for CDX2 and CD1. (e) qRT-PCR analysis of WNT pathway target genes comparing wildtype and FAP9 COs treated with DMSO, rapamycin or XAV939. (f-g) Representative western blot assay comparing wildtype and FAP iPSCs (f) and wildtype and FAP COs (g) treated with different concentrations of geneticin (Gen). (h) Maximal intensity projection images and (i) quantification of cells from wildtype and FAP COs treated with DMSO and geneticin immuno-stained for CDX2 and CD1. (j) Maximal intensity projection images and (k) quantification of cells from wildtype and FAP COs treated with DMSO and geneticin immunostained for CDX2 and Ki67. (l) qRT-PCR analysis of WNT pathway target genes in wildtype and FAP COs treated with DMSO or geneticin. wt: wildtype; rapa: rapamycin; XAV: XAV939; Gen: Geneticin. n=3 independent experiments for each condition. P values by unpaired two-tailed student t-test were *P<0.05, **P<0.01, ***P<0.001, ****P<0.0001. The center value is “median”. Error bar is S.D. Scale bar = 200 μm.

The FAP9 patient cells carry a nonsense mutation that results in an early stop codon. Certain ribosome-binding compounds, such as the aminoglycoside antibiotic G-418 (trade name geneticin), have the capacity to restore activity of some mutant genes by encouraging the ribosome to “read through” the premature stop codon, for example in the APC gene^36,37. Western blotting results suggested that 6 μg/ml geneticin fully restores the APC protein expression levels in FAP9 iPSCs (Fig.4f). A higher dose (25 μg/ml) was required to fully restore APC protein expression in FAP9 COs (Fig.4g). Geneticin does not affect APC protein expression levels in wildtype iPSCs or COs, suggesting that the compound functions through a mutation-specific mechanism. Geneticin-treated FAP9 COs show significantly decreased nuclear β-catenin levels (Supplementary Fig.S8c) as well as decreased colon epithelial cell proliferation (Fig.4h-k). Moreover, geneticin-treated FAP9 COs show significantly decreased WNT target gene expression (Fig.4l), but no change in the levels of the house-keeping gene, cyclophilin G (Supplementary Fig.S8d). The treatment does not affect cell apoptosis as evaluated by immunostaining for cleaved caspase-3 (Supplementary Fig.S8e, f). Since treatment with geneticin affects neither WNT activity nor cell proliferation in wildtype COs, this compound could potentially be used to specifically target cellular phenotypes including hyper-proliferation caused by certain nonsense mutations.

In summary, we established a robust strategy to derive human COs. Applying this 3D-differentiation strategy in FAP-iPSCs established a “disease-in-the-dish” platform to monitor the over-activation of WNT signaling associated with hyper-proliferation of colonic epithelial cells, highly relevant to colorectal cancer. A “read-through” drug, geneticin, could rescue APC protein expression levels, decrease WNT over-activation, and block colonic epithelial cell hyper-proliferation specifically in FAP COs, without compromising the normal phenotype of wildtype COs. In a mouse model, APC restoration promoted cellular differentiation and re-established crypt homeostasis in colorectal cancer³⁸, indicating the translational potential of our CO platform, and clinical application of such drug candidates for precision therapy of colorectal cancer patients.

Online Methods.

hESC and iPSC maintenance and colonic epithelial differentiation.

hESCs and iPSCs were grown on Matrigel (ThermoFisher)-coated 6-well plates in mTeSR1 medium (STEM CELL Technologies). Cells were maintained at 37C with 5% CO₂. To perform the definitive endoderm differentiation, hPSCs were cultured on feeders to 80%–90% confluence, then treated with 3 μM CHIR99021 (CHIR, Stem-RD) and 100 ng/ml ActivinA (R&D systems) in RPMI (Cellgro) supplemented with 2 mM GlutaMAX (Gibco) and 1X Pen-Strep (Gibco) for one day, and then 100 ng/ml activinA in RPMI supplemented with 0.2% BSA (Gibco), 2 mM GlutaMAX and 100 U/ml Pen-Strep. These cells were subjected to hindgut differentiation by treatment with 3 μM CHIR99021 and 500 ng/ml FGF4 (Peprotech) in RPMI supplemented with 1X B27 (Gibco), 2 mM GlutaMAX (Gibco) and 100 U/ml Pen-Strep (Gibco). The media was refreshed daily for 4 days, after which CDX2⁺ cells were identified as hindgut cells. From day 8, cells were cultured in colonic medium containing Advanced DMEM F12 (Invitrogen) supplemented with 1X B27 (Gibco), 2 mM GlutaMAX (Gibco), 100 U/ml Pen-Strep (Gibco), 3 μM CHIR99021, 300 nM LDN193189 (Axon) and 100 ng/mL EGF (R&D). The media was refreshed every two days for up to 3 months. At day 20, the cells were disaggregated to a single cell suspension with Accutase (Stem Cell Technologies) for 30 min at 37C. They were then resuspended in Matrigel (BD Biosciences) and after approximately 2 min. The Matrigel beads are made by using 70 μl Matrigel to form a solid dome-like structure and occupy only the center of the well of Nunclon Delta Surface plates (Thermo Scientific). Cells were subsequently fed colonic medium and 10 μM Y-27632 was added for the first two days. Throughout the ensuing 3 weeks, individual colonic stem cells produced colonic organoids (COs), which were passaged at a 1:4 density every ten days. COs were maintained and expanded in these conditions for up to 3 months. Three days before the assay, COs were switched to colonic medium without CHIR.

D38 small intestinal spheroids (SISs), D38 colonic spheroids (CSs), D48 small intestinal organoids (SIOs) and D48 colonic organoids (COs) were used in all assays.

Generation of iPSCs.

Both wildtype and FAP patient skin fibroblasts lines were purchased from Coriell Cell Repositories and cultured in DMEM supplemented with 10% FBS. Lentivirus was prepared as described. Three days after infection, the fibroblasts were passaged at 1 to 6 dilution onto a monolayer of feeders in 6 well plates in stem cell medium with 10 μM Y-27632. After two weeks, iPSC colonies were picked with pipette tips and transferred to individual wells and subsequently passaged and cultured in stem cell medium.

Neural differentiation.

iPSCs were plated in six-well plates on day 1 and ESC media was added daily for five days. Neuronal differentiation was started on day six in Medium A (DMEMF12 50/50: Neurobasal medium (1:1) + 1% penicillin streptomycin + 100 nM LDN193189 + 10 μM SB431542 supplemented with 20% knockout serum (KOSR) and 1X N2 supplement. Media was changed after two days. On day 10, they were treated with Medium A + 20% KOSR (no N2) and every alternate day KOSR was progressively decreased from 20% to 0% and N2 concentration increased from 0 to 1 X (0.25X increments). On day 18, the cells were disaggregated with Accutase and plated on Fibronectin/Laminin/Poly-ornithine coated 96 well plates at 60,000 cells per well in Medium B (Neurobasal + 1x N2 + 1x B27 supplement) + BDNF (1:1000) + dbCAMP. On day 20, cells were fixed and stained.

Cardiac differentiation.

For cardiac differentiation, hiPSCs cells were detached using 0.25%Trysin-EDTA for 1 min at room temperature and resuspended in hESC medium in ultralow cluster plates (Costar) to form embryoid bodies (EBs). The next day, the EBs were moved to the differentiation medium containing 80% knockout DMEM, 20% fetal bovine serum (Hyclone), 2 mM GlutaMax, 0.1 mM nonessential amino acids and 0.1 mM β-mercaptoethanol with 10 ng/ml BMP4 for 4 days. On day 4, EBs were plated on 0.1% gelatin-coated plates. The following day, differentiated cells were treated with WNT antagonist I (IWR-1, Stemgent) for 3 days. After day 10, the concentration of FBS was reduced from 20% to 2.5%. Differentiated cells were fixed with 4% PFA for 10 min, followed by staining with antibodies against Troponin T or MHC on day 15.

Pancreatic differentiation.

hESCs cells were dissociated with 0.5 mM EDTA and plated on Matrigel-coated 6-well plates at a ratio of 1:1-1:2 resulting in ~95% starting confluency (D0). The differentiation was started 24 hrs later. On day 1 on of the differentiation cells were exposed to basal medium RPMI 1640, 1x Glutamax (ThermoFisher Scientific), 50 ug/mL Normocin supplemented with 100 ng/ml Activin A (R&D), and 2 μM of CHIR99021 (GSK3β inhibitor 3, SelleckChem) for 24 hrs. The medium was changed on day 2 to basal RPMI 1640, 1x Glutamax (ThermoFisher Scientific), 50 ug/mL Normocin, 0.2% Fetal bovine serum (Corning) supplemented with 100 ng/ml Activin A (R&D) for 2 days. On day 4, the resulting definitive endoderm cells were cultured in basal RPMI 1640, 1x Glutamax (ThermoFisher Scientific), 50 ug/mL Normocin, 2% Fetal bovine serum (Corning) containing 50ng/mL FGF7 (Peprotech) for 2 days to acquire foregut fate. On day 6, the cells were induced to differentiate in pancreatic endoderm by being exposed to basal medium DMEM 4.5 g/L glucose (Corning), 1x Glutamax, 50 ug/mL Normocin and 2% B27 (GIBCO) containing 2 uM retinoic acid (RA; Sigma), 200 nM LDN193189 (LDN, Stemgent) and 0.25 μM SANT-1 for four days

Teratoma formation assay.

iPSCs were disaggregated with 0.25% Trypsin-EDTA (Corning, 25-053-Cl) at 37C for 5 min. Then, they were resuspended in PBS containing one third of matrigel. 50 μL of cell suspension was subcutaneously injected into nude mice. Animals were sacrificed after 2 months and teratomas extracted and fixed for 6 hr in 10% formalin. After overnight dehydration in 30% sucrose in 1X PBS, teratomas were embedded and frozen in Tissue-Tek^® O.C.T. Compound (Sakura Finetek) and sectioned using a Leica cryostat. To examine morphology, they were stained with Hematoxilin and Eosin. All animal protocols were approved by the Institutional Animal Care and Use Committee at Weill Cornell Medical College.

RNA extraction and quantitative real-time PCR (qPCR).

Total CO RNA was purified using the Absolutely RNA Nanoprep kit (Agilent Technologies, 400753), and reverse transcribed using a SuperScript^® III First-Strand Synthesis SuperMix cDNA reverse transcription kit (Thermofisher). Real-time PCR was performed with a LightCycler 480 (Roche) instrument with LightCycler DNA master SYBR Green I reagents (Roche). Primer sequences are listed in Supplementary Table 3.

RNA-seq.

The quality of RNA samples was validated by an Agilent Bioanalyzer. cDNA libraries were generated using TruSeq RNA Sample Preparation (Illumina). Each library was sequenced using single-reads in a HiSeq2000/1000 (Illumina).

Bioinformatics Analysis on hESC-derived SIOs and hESC-derived COs.

Distance between two samples was calculated by comparing the normalized expression data with Euclidean distances. A hierarchical clustering was performed based on the sample distances using hclust in R. The heatmap plot was generated using heatmap.2 in the R "gplots" package. The darker color indicates no difference while the lighter color indicates larger distances.

Gene expression data of the human duodenum, distal small intestine and colon samples were retrieved from online database (E-MTAB-1733). Marker genes of each sample type were selected such that a gene has at least 3 times higher expression in that sample type compared to the rest two sample types. The expression data were normalized per gene by subtracting their mean and then dividing by their standard deviation. In order to remove batch effect introduced by experiments, expression data were normalized separately for the duodenum, distal small intestine and colon samples, and for SIOs and COs samples prepared in the first batch and in the second batch. The heatmap plot was generated using heatmap.2 in the R "gplots" package, the red color indicates higher expression and the green color indicates lower expression. Pathway enrichment analysis on up/down-regulated genes was performed using the DAVID function annotation tool.

GSEA analysis was performed with the enrichment statistic equal to weighted and the metric for ranking genes equal to signal-to-noise. Global human fetal and adult colon gene expression data were obtained from GEO (GSM1698576 and GSM1698561). ^3939394111 False discovery rate (FDR) q value was < 0.25 or nominal (NOM) p values < 0.05 were considered significant.

Fragments per kilobase of transcript per million mapped reads (RPKM) expression values from two duodenum, two ileum, and two rectum-derived organoids were obtained from Table S2 of Forster et al (GSE56930)²⁴. RPKM values were log2-transformed and converted to z-scores. RMA-normalized expression values of three large intestine-derived organoids were obtained from Fujii et al²⁵ (GSM1936370, GSM1936372, GSM1936373, all in GSE74843) and then transformed into z-score as well. Finally, RPKM values from the first set (hESC_CO_1, hESC_CO_2, hESC_SIO_1, hESC_SIO_2) and the second set of hESC-derived organoids (hESC_CO_3", hESC_CO_4, hESC_SIO_3, hESC_SIO_4) were analyzed separately. Each of these sets of RPKM values were processed in the same way as the expression from the primary organoids. All the expression values from these datasets were merged into an expression table. Then, we performed an unsupervised hierarchical clustering using Euclidean distance. A sample-to-sample Euclidean distance matrix was plotted with pheatmap_1.0.8. All analyses were performed in R 3.3.1.

Bioinformatics Analysis on wildtype, FAP iPSCs-derived COs and adenomas from FAP patients.

Data derived from three colorectal adenomas from three patients diagnosed with FAP that were undergoing colonoscopy screening at the University of Texas MD Anderson Cancer Center was included in this study (GSE88945). Informed consent was obtained from all individuals and the study was approved by the Institutional Review Board. Pathological characterization, tissue preservation, and sample preparation for molecular analyses followed previously reported protocols⁴⁰. Expression data derived from normal left colonic tissue was obtained from a previously published study and was generated from patients undergoing screening colonoscopy with no polyps found on exam⁴¹ (GSE76987).

RNA-seq of the three FAP samples was performed on an Illumina HiSeq 2000 sequencer at the The University of Texas MD Anderson Sequencing Core Facility. Reads were mapped to human genome assembly hg19 using bowtie⁴²,⁴³. Gene expression was quantified by RSEM⁴⁴, and differentially expressed genes (DEGs) were detected by limma⁴⁵, based on expected counts from the RSEM output. Gene ontology and pathway enrichment analysis was performed using DAVID.

We started transforming RPKM counts of two FAP iPSC-derived colonic organoids (CO) and two wild-type (WT) iPSC-derived CO and RSEM expected counts of three adenomas of Familial Adenomatous Polyposis (FAP) patients and three normal colonic mucosa samples into z-score, respectively. Then, we performed an unsupervised hierarchical clustering of all samples using Euclidean distance and average agglomeration methods in hcluster function of package amap_0.8-14 in R version 3.3.1. The results were plotted with heatmap.2 in gplots_2.17.0.

We performed pathway analysis by obtaining differentially expressed transcripts between FAP adenomas and normal mucosa samples, and between FAP CO and wildtype CO with fold-change>8. The resulting differentially expressed transcripts were uploaded to David 6.7 for Pathway annotation clustering using default settings.

We obtained up and down regulated transcripts that were >2 fold-change and FDR <0.1 between FAP colonic adenoma and normal mucosa. Then, we performed a GSEA of the 2 gene sets with FAP and WT iPSC-derived CO expression data using GSEA version 2.2.0 and 5000 permutations. Significant results were selected based on FDR <0.25 and nom p-value <0.05.

Statistical Analysis.

n=3 independent experiments were carried out unless otherwise stated. n.s. indicates non-significant difference. p values calculated by unpaired two-tailed student t-test are *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001.

Immunofluorescence (IF) and immunohistochemistry (IHC).

COs were paraffin-embedded and sectioned at a thickness of 7 μm. Slides were rinsed in xylenes and with decreasing concentrations of ethanol to deparaffinize and rehydrate. Next, they were incubated for 10 min in a steamer in 10 mM sodium citrate, 0.05% Tween 20, pH 6.0 antigen retrieval buffer, cooled for 15 min and incubated in PBS with 5% horse serum and 0.5% Triton X-100 blocking buffer for 1 h. Then, they were incubated with the corresponding primary antibody in blocking buffer overnight at 4C. For tissue IF staining, this was followed by Alexa-488, Alexa-Fluor-555 and Alexa-Fluor-647-conjugated donkey secondary antibodies against mouse, goat or Rabbit (1:1000, Invitrogen). Nuclei were counterstained with DAPI.

For immuno-histochemical staining, slides were incubated after the primary antibody in 0.3% H₂O₂ for 30 min. Slides were then incubated with HRP-conjugated donkey anti-rabbit secondary antibody (Santa Cruz) for 30 min at room temperature (1:500 in PBS). Finally, slides were developed for 3 min with DAB Peroxidase (HRP) Substrate Kit, 3,3’-diaminobenzidine (Vector Laboratories SK-4100).

For IF in the dish, cells were fixed in the plates with 10% formalin at room temperature for 10 min and washed three times with 1× PBS. They were then blocked in PBS with 5% horse serum and 0.3% Triton X-100 for 1 h and incubated with primary antibodies overnight at 4C, followed by Alexa-488-, Alexa-Fluor-555-, and Alexa-Fluor-647-conjugated donkey secondary antibodies against mouse, goat or rabbit (1:1000, Invitrogen). Nuclei were counterstained with DAPI.

Primary antibodies and their dilution ratio are included in Supplementary Table 2. Alkaline Phosphatase (AP) activity was visualized using the VECTOR Red Alkaline Phosphatase Substrate Kit (Vector Laboratories SK-5100). IHC and IF images were captured under an Olympus BX50 microscope.

Whole-mount organoid staining, confocal microscopy imaging and quantification.

Whole-mount CO staining was performed by dislodging them out of the matrigel beads by up and down pipetting. Subsequently, they were washed in PBS and fixed in 10% formalin for 15 min. Next, they were washed twice with PBS and immersed in 5% horse serum, 0.5% triton X-100, PBS blocking solution for 1h. Finally, they were incubated with primary antibodies overnight at 4C in blocking solution. The next they, they were incubated in the corresponding secondary antibodies as mentioned above.

CO z-stack images were acquired with a 710 META Zeiss confocal microscope. Maximum intensity projection images were generated using the software Zen (Zeiss) and positive cells for the desired markers were quantified using Metamorph (Molecular Devices).

Flow Cytometry.

Cells were disaggregated with 0.25% Trypsin-EDTA (Corning, 25-053-Cl) at 37C for 5 min and stained using Cytofix/cytoperm (BD 554722) & Perm/Wash buffer (BD 554723). Samples were analyzed using a C6 flow cytometer (Accuri). Data were analyzed using cFlow Plus (Accuri Software).

Western blotting analysis.

Cells or COs were collected and lysed in ice-cold lysis buffer (20 mM Tris-HCl, pH 7.4, 150 mM NaCl, 1% NP-40, 1% Triton X-100, 0.05% SDS, 10% glycerol) containing 1 mM PMSF and proteinase inhibitors (Sigma-Aldrich). 20 μg of the lysates were run on SDS-PAGE gel and transferred to nitrocellulose membranes. The membranes were blocked with 5% skim milk in 1× PBST (0.1% Tween, 10 mM sodium phosphate, pH 7.2 and 150 mM NaCl in ddH2O), incubated overnight at 4C with a rabbit anti-APC antibody (Abcam, ab40778, 1:1000) and then for 45 min with an HRP-donkey anti-rabbit secondary antibody (Santa Cruz, 1:2000). The blot was developed by chemiluminescence using the ECL reagent (GE Amersham) for 5 min and membranes were exposed to X-ray film (Hyperfilm ECL, Amersham). All full-length blots were included in Supplementary Fig.S9.

Quantitative RT-PCR analysis.

Total RNA was isolated using Qiagen’s RNeasy mini kit according to instructions by the manufacturer. 0.5 μg of total RNA was used to generate first strand cDNA using the Superscript III First Strand Synthesis System (Thermofisher). First strand cDNA products were diluted 4-fold and used as qPCR templates. SYBR Green-based qPCR was carried out using the Roche 480 Lightcycler. Triplicate reactions were carried out for each sample. GAPDH was used as a control to normalize target gene expression.