Studies in embryonic development have guided successful efforts to direct the differentiation of human embryonic and induced pluripotent stem cells (PSCs) into specific organ cell types in vitro1,2. For example, human PSCs have been differentiated into monolayer cultures of liver hepatocytes and pancreatic endocrine cells3,4,5,6 that have therapeutic efficacy in animal models of liver disease7,8 and diabetes9, respectively. However, the generation of complex three-dimensional organ tissues in vitro remains a major challenge for translational studies. Here we establish a robust and efficient process to direct the differentiation of human PSCs into intestinal tissue in vitro using a temporal series of growth factor manipulations to mimic embryonic intestinal development10. This involved activin-induced definitive endoderm formation11, FGF/Wnt-induced posterior endoderm pattering, hindgut specification and morphogenesis12,13,14, and a pro-intestinal culture system15,16 to promote intestinal growth, morphogenesis and cytodifferentiation. The resulting three-dimensional intestinal ‘organoids’ consisted of a polarized, columnar epithelium that was patterned into villus-like structures and crypt-like proliferative zones that expressed intestinal stem cell markers17. The epithelium contained functional enterocytes, as well as goblet, Paneth and enteroendocrine cells. Using this culture system as a model to study human intestinal development, we identified that the combined activity of WNT3A and FGF4 is required for hindgut specification whereas FGF4 alone is sufficient to promote hindgut morphogenesis. Our data indicate that human intestinal stem cells form de novo during development. We also determined that NEUROG3, a pro-endocrine transcription factor that is mutated in enteric anendocrinosis18, is both necessary and sufficient for human enteroendocrine cell development in vitro. PSC-derived human intestinal tissue should allow for unprecedented studies of human intestinal development and disease.
The epithelium of the intestine is derived from a simple sheet of cells called the definitive endoderm17. As a first step to generating intestinal tissue from PSCs (summarized in Supplementary Fig. 1), we used activin A, a nodal-related TGF-β molecule, to promote differentiation into definitive endoderm as previously described11, resulting in up to 90% of the cells co-expressing the definitive endoderm markers SOX17 and FOXA2 and fewer than 2% expressing the mesoderm marker brachyury (Supplementary Fig. 2a). Using microarray analysis we observed a robust activation of definitive endoderm markers, many of which were expressed in mouse definitive endoderm from embryonic day (e)7.5 embryos (Supplementary Fig. 3 and Supplementary Table 1a, b). We investigated the intrinsic ability of definitive endoderm to form foregut and hindgut lineages by culturing for 7 days under permissive conditions and observed that cultures treated with activin A for only 3 days were competent to develop into both foregut (albumin (ALB)+ and PDX1+) and hindgut (CDX2) lineages (Fig. 1b, control). In contrast, treatment with activin A for 4–5 days resulted in definitive endoderm cultures that were intrinsically anterior in character and less competent in forming posterior lineages (Supplementary Fig. 2b).
Having identified the window of time when definitive endoderm fate was plastic (day 3 of activin A treatment), we used WNT3A and FGF4 to promote hindgut and intestinal specification. Studies in mouse, chick and frog embryos have demonstrated that Wnt and FGF signalling pathways are required for repressing anterior development and promoting posterior endoderm formation into the midgut and hindgut12,13,14. Consistent with this, conditioned media containing WNT3A was recently shown to promote Cdx2 expression in mouse embryonic stem (ES)-cell-derived embryoid bodies19. In human definitive endoderm cultures, neither factor alone was sufficient to robustly promote a posterior fate (Supplementary Fig. 2c); but high concentrations of both FGF4 and WNT3A (FGF4+WNT3A) induced expression of the hindgut marker CDX2 in the definitive endoderm after 48 h (Supplementary Fig. 4). However, 48 h of FGF4+WNT3A treatment did not stably induce a CDX2+ hindgut fate and expression of anterior markers PDX1 and albumin reappeared after cells were cultured in permissive media for 7 days (Fig. 1a, c). In contrast, 96 h of exposure to FGF4+WNT3A resulted in stable CDX2 expression and absence of anterior markers (Fig. 1a, d). These findings indicate a previously unidentified requirement for the synergistic activities of both the FGF and Wnt pathways in specifying the CDX2+ mid/hindgut lineage.
Remarkably, FGF4+WNT3A-treated cultures underwent morphogenesis that was similar to embryonic hindgut formation. Between 2 and 5 days of FGF4+WNT3A treatment, flat cell sheets condensed into CDX2+ epithelial tubes, many of which budded off to form floating hindgut spheroids (Fig. 2a–c, Supplementary Fig. 5a–f and Supplementary Table 2a). Spheroids were similar to e8.5 mouse hindgut and consisted of uniformly CDX2+ polarized epithelium surrounded by CDX2+ mesenchyme (Fig. 2d–g). Spheroids were completely devoid of albumin and PDX1-expressing foregut cells (Supplementary Fig. 5h, i). In vitro gut-tube morphogenesis was never observed in control or WNT3A-only treated cultures. FGF4-treated cultures had a twofold expansion of mesoderm and generated 4–10-fold fewer spheroids (Supplementary Fig. 2c and Supplementary Table 2a), which were weakly CDX2+ and did not undergo further expansion (data not shown). Together our data support a mechanism for hindgut development where FGF4 promotes mesoderm expansion and morphogenesis, whereas FGF4 and WNT3A synergy is required for the specification of the hindgut lineage.
Importantly, this method for directed differentiation is broadly applicable to other PSC lines, as we were able to generate hindgut spheroids from both H1 and H9 human ES cell lines and from four induced PSC (iPSC) lines that we have generated and characterized (Supplementary Figs 3, 5 and 6). The kinetics of differentiation and the formation of spheroids were comparable between these lines (Supplementary Table 2). Two other iPSC lines tested were poor at hindgut spheroid formation and line iPSC3.6 also had a divergent transcriptional profile during definitive endoderm formation (Supplementary Fig. 3 and Supplementary Table 2c).
Whereas in vivo engraftment of PSC-derived cell types, such as pancreatic endocrine cells, has been used to promote maturation9, efficient development and maturation of organ tissues in vitro has proven more difficult. We investigated whether hindgut spheroids could develop and mature into intestinal tissue in vitro using recently described three-dimensional culture conditions that support growth and renewal of the adult intestinal epithelium15,16. When placed into this culture system, hindgut spheroids developed into intestinal organoids in a staged manner that was notably similar to fetal gut development (Fig. 3, Supplementary Fig. 5g and Supplementary Fig. 7). In the first 14 days the simple cuboidal epithelium of the spheroid expanded and formed a highly convoluted pseudostratified epithelium surrounded by mesenchymal cells (Fig. 3a–c), similar to an e12.5 fetal mouse gut (Fig. 3f). After 28 days, the epithelium matured into a columnar epithelium with villus-like involutions that protrude into the lumen of the organoid (Fig. 3d, e). Comparable transitions were observed during mouse fetal intestinal development (Fig. 3f, g and Supplementary Fig. 7). The spheroids expanded up to 40 fold in mass as they formed organoids (data not shown) and were split and passaged over 9 additional times and cultured for over 140 days with no signs of growth failure. The cellular gain during that time was up to 1,800 fold (data not shown), resulting in a total cellular expansion of 72,000 fold per hindgut spheroid. This directed differentiation was up to 50 fold more efficient than spontaneous embryoid body differentiation methods20 (Supplementary Fig. 8) and resulted in organoids that were almost entirely intestinal (Supplementary Fig. 2e–g) as compared to embryoid bodies that contained a mix of neural, vascular and epidermal tissues (Supplementary Fig. 8).
Marker analysis showed that after 14 days in culture, virtually all of the epithelium expressed the intestinal transcription factors CDX2, KLF5 and SOX9 broadly and was highly proliferative (Fig. 3b, c). By 28 days, CDX2 and KLF5 remained broadly expressed in over 90% of the epithelium (Supplementary Fig. 2), whereas SOX9 became localized to pockets of proliferating cells at the base of the villus-like protrusions (Fig. 3d, e) similar to the intervillus epithelium of fetal mouse intestines at e16.5 (Fig. 3g and Supplementary Fig. 9). 5-bromodeoxyuridine (BrdU) pulse chase and analysis of organoids using a Z-stack series of confocal microscopic images showed that epithelial BrdU incorporation was largely restricted to SOX9-expressing cells in crypt-like structures that penetrated into the underlying mesenchyme (Supplementary Fig. 9). At 28 days, LGR5 is not expressed and ASCL2 (ref. 21) is broadly expressed and not restricted to the SOX9+ proliferative zone. However, organoids cultured until 56 days expressed both ASCL2 and LGR5 in restricted epithelial domains that appear to overlap with the SOX9+ zone (Fig. 3h–j and Supplementary Fig. 10). This domain is similar to developing intestinal progenitor domains in vivo, which ultimately give rise to the stem cell niche in the crypt of Lieberkühn15. iPSCs were equally capable of forming intestinal progenitor domains (Supplementary Fig. 9e). Thus, PSC-derived intestinal epithelium continued to mature in vitro and develop proliferative domains with nascent intestinal stem cells.
Between 18 and 28 days in culture, we observed cytodifferentiation of the stratified epithelium into a columnar epithelium containing brush borders and all of the major cell lineages of the gut as determined by immunofluorescence and quantitative polymerase chain reaction with reverse transcription (RT–qPCR) (Fig. 4a–d and Supplementary Fig. 11). By 28 days of culture, villin (Fig. 4a) and DPPIV (not shown) were localized to the apical surface of the polarized columnar epithelium and transmission electron microscopy revealed a brush border of apical microvilli indistinguishable from those found in mature intestine (Fig. 4d and Supplementary Fig. 1). Enterocytes had a functional peptide transport system and were able to absorb a fluorescently labelled dipeptide (Fig. 4e)22. Cell counting revealed that the epithelium contained approximately 15% MUC2+ goblet cells, which secrete mucins into the lumen of the organoid, 18% lysozyme-positive cells, which are indicative of Paneth cells, and ∼1% chromogranin-A-expressing enteroendocrine cells (Fig. 4 and Supplementary Fig. 11g). MUC2 and lysozyme staining indicated that the goblet and Paneth cells in 28-day organoids are immature (Fig. 4a, b). However, in organoids that were passaged over 100 days, all cells had acquired a more mature phenotype and Paneth cells were often localized in crypt-like structures (Supplementary Fig. 12b, c). RT–qPCR confirmed the presence of additional markers of differentiated enterocytes (FABP2; also known as IFABP) and Paneth cells (MMP7) (Supplementary Fig. 11). Individual organoids seemed to be a mix of proximal intestine (GATA4+/GATA6+) and distal intestine (GATA4−/GATA6+; HOXA13-expressing) (Supplementary Figs 11 and 13)23. Thus, directed differentiation of PSCs into intestinal tissue in vitro is highly efficient in generating three-dimensional intestinal tissue containing crypt-like progenitor niches, villus-like domains and all of the differentiated cell types of the intestinal epithelium.
Intestinal organoids contained a mesenchymal layer that developed along with the epithelium in a staged manner similar to embryonic development10,24 (Supplementary Fig. 14). Mesenchyme probably came from the 2% of mesoderm cells that were present after activin differentiation, which expanded up to 10% in FGF4-treated hindgut cultures (Supplementary Fig. 2). At 14 days, organoids broadly expressed mesenchymal markers including FOXF1 and vimentin (Supplementary Fig. 14), similar to an e12.5 embryonic intestine (Supplementary Fig. 7). We also observed vimentin/smooth muscle actin (SMA; also known as ACTA2) double-positive cells indicative of intestinal subepithelial myofibroblasts25. By 28 days, we observed a layer of SMA+/desmin+ double-positive cells, indicating smooth muscle, and desmin+/vimentin+ fibroblasts26. The fact that intestinal mesenchyme differentiation coincided with differentiation of the overlying epithelium indicates that epithelial–mesenchymal crosstalk may be important in the development of PSC-derived intestinal organoids.
The molecular basis of congenital malformations in humans is often inferred from functional studies in model organisms. For example, neurogenin 3 (NEUROG3) was investigated as a candidate gene responsible for congenital loss of intestinal enteroendocrine cells in humans18 because of its known role in enteroendocrine cell development in mouse27,28,29,30. However, it has been impossible to directly investigate the role of NEUROG3 during human intestinal development. We therefore performed gain- and loss-of-function analyses to investigate the role of NEUROG3 during human enteroendocrine cell development (Fig. 4 and Supplementary Fig. 15). NEUROG3 was overexpressed in 28-day human organoids using adenoviral (Ad)-mediated transduction31. After 7 days, approximately 5% of cells were GFP+ and Ad-NEUROG3–GFP-infected organoids contained fivefold more chromogranin A+ endocrine cells than control organoids (Ad-enhanced GFP (eGFP)) (Fig. 4f–h and Supplementary Fig. 15), demonstrating that NEUROG3 expression is sufficient to promote an enteroendocrine cell fate. To knockdown endogenous NEUROG3, we generated human ES cell lines by transducing cells with NEUROG3 short hairpin (sh)RNA-expressing lentiviral vectors. NEUROG3 mRNA levels were knocked down by 63% and this resulted in a 90% reduction in the number of enteroendocrine cells (Fig. 4i–k and Supplementary Fig. 15d–f), demonstrating that intestinal enteroendocrine cell development is highly dependent on NEUROG3 expression. This indicates that partial loss-of-function mutations in human NEUROG3 would be sufficient to cause a marked reduction in enteroendocrine cell numbers.
This is the first report, to our knowledge, demonstrating that human PSCs can be efficiently directed to differentiate in vitro into human tissue with a three-dimensional architecture and cellular composition remarkably similar to the fetal intestine. Moreover, PSC-derived human intestinal tissue undergoes maturation in vitro, developing intestinal stem cells and acquiring both absorptive and secretory functionality. This system allows for functional studies to investigate the molecular basis of human congenital gut defects in vitro and to generate intestinal tissue for eventual transplantation-based therapy for diseases such as necrotizing enterocolitis, inflammatory bowel diseases and short-gut syndromes. The ability to generate human intestinal tissues should also greatly facilitate future studies of intestinal stem cells and drug design to enhance absorption and bioavailability.
Generation of human intestinal organoids
Human ES cells and iPSCs were maintained on Matrigel (BD Biosciences) in mTesR1 medium without feeders. Differentiation into definitive endoderm was carried out as previously described11. Briefly, a 3-day activin A (R&D systems) differentiation protocol was used. Cells were treated with activin A (100 ng ml−1) for three consecutive days in RPMI 1640 medium (Invitrogen) with increasing concentrations of 0%, 0.2% and 2% HyClone defined fetal bovine serum (dFBS; Thermo Scientific). For hindgut differentiation, definitive endoderm cells were incubated in 2% dFBS-DMEM/F12 with 500 ng ml−1 FGF4 and 500 ng ml−1 WNT3A (R&D Systems) for up to 4 days. Between 2 and 4 days of treatment with growth factors, three-dimensional floating spheroids formed and were then transferred into three-dimensional cultures previously shown to promote intestinal growth and differentiation15,16. Briefly, spheroids were embedded in Matrigel (BD Bioscience) containing 500 ng ml−1 R-Spondin1 (R&D Systems), 100 ng ml−1 Noggin (R&D Systems) and 50 ng ml−1 EGF (R&D Systems). After the Matrigel solidified, medium (advanced DMEM/F12; Invitrogen) supplemented with l-glutamine, 10 μM HEPES, N2 supplement (R&D Systems), B27 supplement (Invitrogen), and penicillin/streptomycin-containing growth factors was overlaid and replaced every 4 days.
Maintenance of PSCs
Human ES cells and induced pluripotent stem cells were maintained on Matrigel (BD Biosciences) in mTesR1 medium32,33. Cells were passaged approximately every 4 days, depending on colony density. To passage PSCs, they were washed with DMEM/F12 medium (no serum) (Invitrogen) and incubated in DMEM/F12 with 1 mg ml−1 dispase (Invitrogen) until colony edges started to detach from the dish. The dish was then washed 3 times with DMEM/F12 medium. After the final wash, DMEM/F12 was replaced with mTesR1. Colonies were scraped off of the dish with a cell scraper and gently triturated into small clumps and passaged onto fresh Matrigel-coated plates.
Differentiation of PSCs into definitive endoderm
Differentiation into definitive endoderm was carried out as previously described11. Briefly, a 3-day activin A (R&D systems) differentiation protocol was used. Cells were treated with activin A (100 ng ml−1) for three consecutive days in RPMI 1640 media (Invitrogen) with increasing concentrations of 0%, 0.2% and 2% HyClone defined fetal bovine serum (dFBS; Thermo Scientific).
Differentiation of definitive endoderm in permissive media
After differentiation into definitive endoderm, cells were incubated in DMEM/F12 plus 2% dFBS with either 0, 50 or 500 ng ml−1 FGF4 and/or 0, 50 or 500 ng ml−1 WNT3A (R&D Systems) for 6, 48 or 96 h. Cultures were then grown in permissive medium consisting of DMEM plus 10% FBS for an additional 7 days.
Directed differentiation into hindgut and intestinal organoids
After differentiation into definitive endoderm, cells were incubated in 2% dFBS-DMEM/F12 with either 50 or 500 ng ml−1 FGF4 and/or 50 or 500 ng ml−1 WNT3A (R&D Systems) for 2–4 days. After 2 days with treatment of growth factors, three-dimensional floating spheroids were present in the culture. Three-dimensional spheroids were transferred into an in vitro system to support intestinal growth and differentiation previously described15,16. Briefly, spheroids were embedded in Matrigel (BD Bioscience; no. 356237) containing 500 ng ml−1 R-Spondin1 (R&D Systems), 100 ng ml−1 Noggin (R&D Systems) and 50 ng ml−1 EGF (R&D Systems). After the Matrigel solidified, medium (advanced DMEM/F12; Invitrogen) supplemented with l-glutamine, 10 μM HEPES, N2 supplement (R&D Systems), B27 supplement (Invitrogen), and penicillin/streptomycin-containing growth factors was overlaid and replaced every 4 days.
Generation and characterization of iPSC lines
Normal human skin keratinocytes (HSKs) were obtained from donors with informed consent (Cincinnati Children’s Hospital Medical Center (CCHMC) Institutional Review Board protocol CR1_2008-0899). Normal HSKs were isolated from punch biopsies following trypsinization and subsequent culture on irradiated NIH3T3 feeder cells in F medium34. For iPSC generation, normal HSKs were transduced on two consecutive days with a 1:1:1:1 mix of recombinant RD114-pseudotyped retroviruses expressing OCT4, SOX2, KLF4 and MYC35,36 in the presence of 8 µg ml−1 polybrene. Twenty-four hours after the second transduction the virus mix was replaced with fresh F medium and cells were incubated for an additional three days. Cells were then trypsinized and seeded into 6-well dishes containing 1.875 × 105 irradiated mouse fibroblasts per well and Epilife medium. On the following day, medium was replaced with DMEM/F12 50:50 medium supplemented with 20% knockout serum replacement, 1 mM l-glutamine, 0.1 mM β-mercaptoethanol, 1× non-essential amino acids, 4 ng ml−1 basic fibroblast growth factor, and 0.5 mM valproic acid. Morphologically identifiable iPSC colonies arose after 2–3 weeks and were picked manually, expanded and analysed for expression of human PSC markers NANOG, DNMT3B, and using the antigen antibodies Tra1-60 and Tra1-8137,38. Early passage iPSC lines were adapted to feeder-free culture conditions consisting of maintenance in mTeSR1 (Stem Cell Technologies) in culture dishes coated with Matrigel (BD Biosciences) and lines were karyotyped.
Microarray analysis of human ES cells, iPSCs and definitive endoderm cultures
For microarray analysis, RNA was isolated from undifferentiated and 3-day activin-treated human ES cell and iPSC cultures and used to create target DNA for hybridization to Affymetrix Human 1.0 Gene ST Arrays using standard procedures (Affymetrix). Independent biological triplicates were performed for each cell line and condition. Affymetrix microarray Cel files were subjected to RMA normalization in GeneSpring 10.1. Probe sets were first filtered for those that are overexpressed or underexpressed and then subjected to statistical analysis for differential expression by 2 fold or more between undifferentiated and differentiated cultures with P < 0.05 using the Students t-test. Log2 gene expression ratios were then subjected to hierarchical clustering using the standard correlation distance metric as implemented in GeneSpring.
Adenoviral-mediated expression of NEUROG3
Adenoviral plasmids were obtained from Addgene and particles were generated as previously described31. Transduction was done on 28-day organoids that were removed from Matrigel, manually bisected then incubated in Ad-GFP or Ad-NEUROG3 viral supernatant and medium at a 1:1 ratio for approximately 4 h. Organoids were then re-embedded in Matrigel and incubated overnight with viral supernatant and medium at a 1:1 ratio. The next day, fresh organoid medium was placed on the cultures and was changed as described until the end of the experiment.
shRNA knockdown human ES cell lines
GipZ shRNA lentiviral vectors were obtained from Open Biosystems (GipZ-NEUROG3 Open Biosystems clone no. v2lhs_309089; v2lhs_309091; v2lhs_309093; v2lhs_309092 and GipZ-Control; Openbiosystems clone no. RHS4346). The CCHMC Viral Vector Core produced high-titre lentiviral particles for each plasmid. Low-passage H9 human ES cells were dissociated into a single-cell suspension using Accutase, were spun down and resuspended in mTesR1containing 10 μM Y-27632. Cells were plated at low density and incubated with lentivirus for 24 h. For the NEUROG3 shRNA knockdown line, particles from all four vectors were used. mTesR1 was replaced daily, and after 72 h selection for puromycin- (2–4 μg ml−1) resistant human ES cells was carried out. Puromycin-resistant colonies were routinely maintained and passaged in mTesR1+puromycin (4 μg ml−1).
β-Ala-Lys-AMCA was purchased from BioTrend Chemicals and was resuspended in water. Intestinal organoids were cut in half using a scalpel and were incubated for four hours in advanced DMEM/F12 plus 24 μM β-Ala-Lys-AMCA. Following incubation, organoids were washed several times in PBS, embedded in OCT freezing medium and were frozen at −70 °C. Ten-micrometre cryosections were cut and processed for standard immunohistochemistry.
Tissue processing, immunohistochemistry and microscopy
Tissues were fixed for 1 h to overnight in 4% paraformaldehyde or 3% glutaraldehyde for transmission electron microscopy (TEM). Cultured PSCs and definitive endoderm cells were stained directly. Hindgut and intestinal organoids were embedded in paraffin, epoxy resin LX-112 (Ladd Research), or frozen in OCT. Sections were cut at 6–10 µm for standard microscopy and 0.1 µm for TEM. TEM sections were stained with uranyl acetate. Paraffin sections were deparaffinized, subjected to antigen retrieval, blocked in the appropriate serum (5% serum in 1× PBS plus 0.5% Triton-X) for 30 min, and incubated with primary antibody overnight at 4 °C. Slides were washed and incubated in secondary antibody in blocking buffer for 2 h at room temperature (23 °C). For a list of antibodies used and dilutions, see Supplementary Table 3. Slides were washed and mounted using Fluormount-G. Confocal images were captured on a Zeiss LSM510 and Z-stacks were analysed and assembled using AxioVision software. An Hitachi H7600 transmission electron microscope was used to capture images.
RNA isolation, RT–qPCR
RNA was isolated using the Nucleospin II RNA isolation kit (Clonetech). Reverse transcription was carried out using the SuperScriptIII Supermix (Invitrogen) according to manufacturer’s protocol. Finally, qPCR was carried out using Quantitect SybrGreen MasterMix (Qiagen) on a Chromo4 Real-Time PCR (BioRad). PCR primers sequences were typically obtained from qPrimerDepot (http://primerdepot.nci.nih.gov/). Primer sequences are available upon request.
Gene Expression Omnibus
Data have been deposited at NCBI under accession number GSE25557.
We thank members of the laboratory, D. Wiginton and C. Wylie for input. We also thank M. Kofron, T. Stefader and R. Lang for assistance with imaging. Vectors and antibodies were from D. Melton (Addgene no. 19410, 19413), S. Yamanaka (17217–17220), C. Baum (OCT4, KLF4, SOX4, MYC lenti), and I. Manabe (KLF5 antibody). This work was supported by the Juvenile Diabetes Research Foundation JDRF-2-2003-530 (J.M.W.) and NIH, R01GM072915 (J.M.W.); R01DK080823A1 and S1 (A.M.Z. and J.M.W.); R03 DK084167 and R01 CA142826 (N.F.S.), F32 DK83202-01 and T32 HD07463 (J.R.S.). We also acknowledge core support for viral vectors, microarrays (supported by P30 DK078392), karyotyping and the Pluripotent Stem Cell Facility (supported by U54 RR025216).
This file contains Supplementary Figures 1-15 with legends, Supplementary Tables 1-3 and additional references.
About this article
Nature Reviews Genetics (2019)