Recent Advances and Prospects in the Differentiation of Pancreatic Cells From Human Embryonic Stem Cells

The regulation of pancreas development from the gut endoderm is a complex process that involves a carefully balanced interplay between several common signaling pathways. Studies in several model organisms indicate conservation of the main mechanisms of pancreas development from lower to higher vertebrates (Table 1). Gastrulation in the developing embryo (E7.5 in mice) results in the formation of three germ layers, namely the ectoderm, the endoderm, and the mesoderm. It appears that these germ layers do not further develop in isolation from each other, but signals emanate from one to pattern the other and vice versa (1). Several structures temporally participate in these early tissue interactions including the notochord (2), the cardiac mesoderm, the septum transversum mesenchyme and the lateral plate mesoderm (3), the aortic endothelial cells and the vitelline veins (4), and the developing pancreatic mesenchyme (5).

As far as the pancreas is concerned, specific growth and differentiation factors released by adjacent tissues control a set of transcription factors expression, resulting in the patterning of the ventral and dorsal prepancreatic endoderm. The initial interactions usually confer competence to respond to additional inductive signals that establish organ determination and specification at particular time points referred to as competence windows. Although our present knowledge of extrinsic factors that control patterning, proliferation, and differentiation cues throughout the developing pancreas is not yet complete, there is strong evidence emanating from all vertebrate developmental models that complex spatio-temporal combinations of common signaling pathways are crucial for appropriate specification of pancreatic cell fates from the endoderm (6,,–9). These involve, for instance, the hedgehog, Wnt, retinoid, and Notch pathways, as well as Activin/bone morphogenetic protein (BMP), fibroblast growth factor (FGF), vascular endothelial growth factor (VEGF), epidermal growth factor (EGF), and hepatocyte growth factor (HGF) signal transductions.

An overview of the roles played in pancreas development by the most relevant growth and differentiation factors is summarized in Table 2. A detailed sequential presentation is difficult to address because many signaling events are not clearly dissociated from each other, while others function redundantly or intermittently at several developmental stages. It is also worth noting that the developmental effects elicited by these common signals can vary significantly according to the cell type being stimulated or according to the stage of development when the signal occurs. This level of complexity is further enhanced by the interplay between transcription factors that lie downstream of these extracellular stimuli and are responsible for observed developmental features.

During the past two decennia, considerable progress has been achieved in understanding pancreas embryogenesis thanks to the numerous genetic studies involved in transcription factors. Most of them belong to the homeobox domain, the paired-box, or the basic helix-loop-helix families. The sequential expression of different transcription factors that control pancreas development has been extensively reviewed elsewhere (6,–8,10). Only a brief summary of their expression pattern is given in Fig. 1 with special focus on 1) the early pancreatic epithelial cells, 2) the pancreatic endocrine progenitors, and 3) the endocrine subtypes selection.

It is presently well recognized that attempts to generate pancreatic cells from pluripotent cells should first establish a definitive endoderm (DE) population as occurs in vivo. Wnt and Nodal signals were identified as sufficient to induce DE from mouse and human embryonic stem (hES) cells, but also from induced pluripotent stem (iPS) cells. In this model, high concentrations of activin A (ActA) are used to mimic the function of the endogenous endoderm inducer Nodal (11). Beside the activation of this pathway, initial treatment with Wnt3a or BMP4 was found to improve the efficiency of DE induction by setting up a transient mesendoderm progenitor population, especially when cultures are performed in feeder cell-free conditions (12,13). Because active PI3K signaling is inhibitory for DE differentiation from embryonic stem (ES) cells, antagonists of this pathway (LY294002, Wortmanin) were also included in some protocols, resulting in an increased efficiency (14,15). In addition, supplementation of sodium butyrate was found to support endoderm differentiation when combined with ActA (16), though it remains unclear how this chemical participates in the signaling network required for the definitive endoderm.

The standard and modified protocols for DE induction from ES cells allow for the derivation of at least 60–80% cells that express the characteristic markers FOXA2, SOX17, GSC, and CXCR4 (Table 3) but not the visceral endoderm marker SOX7 (13,14,16,,,–20). Whether the quest for a universal β-cell differentiation protocol is a desired myth or an achievable reality, it appears clear from the literature that combined Wnt3a and ActA treatments efficiently activate DE phenotype in a vast majority of hES cell lines. In other words, the molecular events that lead DE differentiation in vivo have been largely—if not completely—unraveled and implemented in vitro. Despite the recently described differences in the response of hES cell lines to differentiation cues (21), the variability and the low success rate in the outcome of initial pancreas differentiation protocols might first of all reflect the lack of such a universal developmentally based strategy that thoroughly reproduces in vivo events in the culture dish after the DE stage.

When maintained in culture without added growth factors or when transplanted under the kidney capsule of immunocompromised mice, hES-derived definitive endoderm cells could spontaneously generate hepatocyte-like cells characterized by the expression of AFP, albumin (ALB) (Fig. 2 A), or hepatocyte-specific antigen (12,17,20). This tendency for hepatic lineage commitment was also reported in several studies wherein hES-derived DE cells were exposed to BMP and/or FGF, two growth factors implicated in embryonic liver induction and specification (13,20,22,23). The contribution of these pathways was further demonstrated by the treatment with BMP antagonist Noggin and FGF receptor antagonist SU5402, leading to the complete abrogation of hepatic gene expression from DE progenies that otherwise was mainly composed of hepatocytes. As suggested from animal studies, these experiments also indicated that FGF did not function in the induction but essentially contributed to the amplification of liver progenitors initially induced by BMP signaling (20). Therefore, the molecular signals elicited by BMP and FGF are the main factors driving hepatocyte differentiation from hES-derived DE cells.

In vivo, BMP activity generated by the adjacent mesoderm–derived structures and operating in the ventral endoderm is associated with hepatic induction during early embryonic development. On the contrary, ventral pancreas progenitors are located away from the midline endoderm with active BMP signaling, whereas BMP inhibition in the dorsal endoderm plays a crucial function in the acquisition of a pancreatic fate (3,24). The requirement for BMP antagonism is reverted after pancreas induction since this organ requires BMP signaling later on for the maintenance of Pdx1 expression and further differentiation. This underscores the existence of a tight competence window and the versatile functions of growth factors during development, which complicates their in vitro implementation for directed differentiation of ES cells (24).

Retinoic acid (RA) plays a crucial role shortly after gastrulation in endoderm patterning and in the induction of PDX1 expression by pancreas progenitors in several organisms from Xenopus to humans, which indicates the conservation of its activity during evolution (25,–27). This role for RA has already been evaluated during mouse and human ES cell differentiation in vitro and has proven to be relevant for the induction of pancreatic gene expression, notably detectable levels of PDX1 (17,23,28,,,–32).

In line with the in vivo recognized functions of these two signaling pathways on pancreas induction, recent studies have combined BMP inhibition (Noggin supplementation) with retinoid signaling immediately after the DE induction stage and succeeded in producing large amounts (up to 80%) of PDX1⁺ pancreas progenitor cells (Table 3), (13,14,20). Concomitantly, the proportion of hepatocytes that could be detected in these conditions was significantly reduced (Fig. 2,A and B), indicating an efficient blockade of their induction upon BMP antagonism (20,33). This combination was efficient not only in different hES cell lines, but also in iPS cells (14), suggesting that the basic molecular events that should be monitored in order to design a universal pancreatic differentiation protocol are being progressively clarified. These minimally include the signals for the blockade of hepatic induction on the one hand and those required for pancreatic induction (namely PDX1 expression) on the other (Fig. 3). The necessity for such a combination was recently highlighted by treatment of hES-derived DE cells with Noggin in the absence of RA; this blocked the expression of liver markers, but did not generate PDX1⁺ cells (20,23). This requirement for combining RA treatment with BMP antagonism could also be seen from another angle. Indeed, retinoid signaling was shown to activate BMP expression in several in vivo and in vitro systems including embryonal carcinoma cells and differentiating mouse embryoid bodies (34,–36). It is not yet clear whether this inductive effect on BMP also occurs during hES cell differentiation. Nevertheless, considering the in vivo function of BMP signaling in the midgut endoderm, it appears reasonable that both retinoid signaling and BMP antagonism were required in these studies for strong induction of a pancreatic fate from the DE cells while preventing hepatic differentiation. However, it is likely that the effect of RA is context-dependent because when applied on hES-derived DE cells cultured at low density, it resulted in the inhibition of Smad 1/5/8 phosphorylation (23). It remains to be elucidated whether this later effect represents an artifact of in vitro culture or a specific retinoid effect at the single cell level.

Other studies made use of RA combined with FGF ligands (FGF4, FGF7, FGF10) or of Noggin combined with FGF2 and EGF to induce pancreatic cells albeit at a lower efficiency than for combined Noggin and RA (Table 3) (13,14,16,17,19,23). Our unpublished observations suggest that FGF signals favor hepatic gene expression in the absence of BMP antagonism, but the addition of FGF ligands together with BMP antagonism (Noggin) and RA enhances pancreatic gene expression. Furthermore, the induction of PDX1⁺ cells by combined Noggin and RA requires at least a basal level of FGF activity (via the extracellular signal–regulated kinase/mitogen-activated protein kinase [ERK/MAPK] pathway [Fig. 3]) given that the addition of the MAPK-inhibitor U1026 prevented pancreatic cell differentiation (20). Because RA is known to act synergistically with FGF signaling during endoderm patterning, supplemented FGF can be seen as an enhancing factor in protocols that already integrate Noggin and RA (13,14,19). This view might resolve the nature of factors X suggested as involved in the indirect effects of RA in Xenopus dorsal endoderm (37). It was recently suggested that in the presence of FGF7, RA can efficiently induce PDX1+ progenitors from hES-derived DE cells seeded at low density (5–50,000 cells/cm²) and that this effect is paralleled by inhibition of Smad1/5/8 phosphorylation, therefore recapitulating BMP antagonism described above (23).

Recent analysis of the signaling network during early liver and pancreas development in 3–6 somites-stage mouse embryos further indicated that inhibition of the activin signaling significantly increases the proportion of cells fated toward PDX1 expression. These findings therefore suggest that the addition of activin inhibitors to the above-mentioned cocktails might still boost the occurrence of PDX1⁺ cells in hES cell cultures (13,24).

Despite the fact that combining BMP antagonism with retinoid signaling efficiently generates PDX1⁺ pancreatic progenitors from hES-derived definitive endoderm, the requirement for hedgehog antagonism with cyclopamine during pancreas induction from these cells has not been clarified by these studies. The initial report by D'Amour et al. (18) indicated the necessity to antagonize hedgehog signaling that is indeed activated during ES cell differentiation, and it was shown to also play a negative role ex vivo by limiting endocrine and exocrine genes expression in embryonic (E12.5) mouse pancreatic explants (38,39). However, additional studies using Noggin, FGF ligands, and/or RA after DE induction succeeded in generating PDX1⁺ cells without the use of hedgehog inhibitors (14,17,19,23). Owing to the previously described increase in the expression of hedgehog ligands and effectors upon DE induction, we added cyclopamine to the combination of Noggin and RA in our experimental setup, but it remains unclear whether this was mandatory for the efficient differentiation of PDX1⁺ cells that we observed (20,38,39). While waiting for conclusive data on this issue, we speculate that supplementation of hedgehog antagonists would become obsolete in settings where combined Noggin and RA treatment of DE cells essentially induces a “true pancreatic endoderm” cell type, which is normally devoid of hedgehog activity. Gli1 transcripts profiling in cells treated with Noggin and RA combination would help to clarify this issue.

Pancreatic progenitor cells are characterized by their expression of a subset of transcription factors. For instance, the combined expression of PDX1, FOXA2, SOX9, HNF6, NKX6.1, and PTF1a (Fig. 3) define the multipotent pancreas progenitor (9,33). In the endoderm, the last two transcription factors are more specific to the pancreas, whereas PDX1, FOXA2, SOX9, and HNF6 are also detected in the adjacent duodenum, stomach, and liver. Following the induction of pancreas progenitor cells from hES-derived DE cells, PDX1 co-expression with FOXA2, SOX9, and HNF6 was demonstrated in a context where hepatic (AFP, ALB) and intestinal (CDX1, CDX2) markers were not induced or only minimally induced (Fig. 2) (14,17,20). However, none of the above-mentioned studies have assessed the expression of all the six transcription factors by the pancreatic progenitors obtained from hES cells. More specifically, demonstration of PTF1a expression has remained elusive until now, possibly owing to the lack of an optimal antibody against this protein. The same holds true for NGN3, which has been detected at the protein level in only few studies. In our experimental setup, PTF1a transcripts were detected only at low levels, whereas a significant emergence of NKX6.1-positive cells was denoted. Nevertheless, when combined with the demonstration of INS and PDX1 co-expression at later stages, these findings point toward the pancreatic nature of the differentiated PDX1⁺ cells. Additional efforts are required to obtain their efficient differentiation toward pancreatic endocrine cells in vitro (Table 3) (14,20).

During pancreas development, the PDX1⁺ progenitor cells are amplified thanks to signals emanating from the pancreatic mesenchyme. FGF10 controls this mesenchymal-to-epithelial interaction, and its over-expression beyond the competence window (E11.5–E13.5 in the mouse) significantly induces progenitor cell proliferation and arrests their terminal differentiation (40,41). This signaling operates via activation of the Notch pathway, which temporally prevents progenitor differentiation toward the endocrine or exocrine fates (42). In agreement with these, FGF10 or EGF supplementation induces the proliferation of hES-derived PDX1⁺ cells, suggesting the contribution of ERK/MAPK and AKT activation in this effect (14,16,20). Furthermore, before the proliferation step, establishment of the PDX1⁺ progenitors from hES-derived definitive endoderm was also shown to be highly dependent on ERK/MAPK signaling and possibly mediated by FGF2 (20,43).

The initiation of endocrine differentiation from the PDX1⁺ progenitors relies on the activation of NGN3 expression, which should be paralleled or preceded by downregulation of Notch activity and further sustained by additional induction of PAX6 (Fig. 1) (44,45). Although the transcription factors cascade regulating pancreatic endocrine differentiation has been largely explored, to date the instructive extracellular signals that trigger their coordinated expression remain elusive. To this end, there is a growing list of proposed growth factors and chemicals whose effects will need to be extensively assessed in the endocrine commitment and maturation of PDX1⁺ progenitors obtained in vitro from hES-derived DE cells. These include but are not limited to Wortmannin and LY294002 that are PI3K inhibitors (46), VEGF (4,47), RA (26), Conophylline (48), Wnt, and Hedgehog ligands (49,50), and antagonists of the Notch pathway such as gamma secretase inhibitors, Notch ligands, and soluble Notch. Not only is the nature of these factors important, but their timely combination will also be of consideration given the demonstration of a competence window for endocrine patterning beyond which the endocrine differentiation arrest becomes irreversible (41). For instance in our experimental setup (Table 3), FGF10 was supplemented for controlling PDX1⁺ cell proliferation together with a gamma secretase inhibitor (Compound E) for Notch antagonism in view of endocrine induction (20). Considering the above-mentioned crosstalk between FGF10 and Notch activity in the proliferation of the PDX1⁺ progenitors, such a protocol therefore included two phenomenas that were actually antagonistic. This issue might explain the very limited progression toward endocrine cells that we obtained, a hypothesis that will need further evaluation by dissociating FGF10 treatment from Notch inhibition in this protocol.

The next step after endocrine commitment will be the further endocrine differentiation and maturation by use of old and new players such as betacellulin, activin, exendin-4, insulin-like growth factor, HGF, nicotinamide, bone morphogenetic protein, FGF, and RA (14,16,18,20,26,28). Similarly, three-dimensional cultures will merit further investigation in order to mimic the actual environment of the β-cells in islets as was recently described in differentiating hES cells (16). Some of these factors have been used singly or in combination on PDX1⁺ progenitors generated from hES cells. Because of the complex nature of pancreatic endocrine differentiation, which is presently not yet elucidated, the generation of mature and functional β-cells has not been efficiently achieved in vitro. The demonstration of human-specific β–cell functions in animal models following transplantation of pancreatic progenitors derived from hES cells (19,32,51) underscores this gap in our current knowledge. Based on these data, it has been proposed that diabetic patients could be grafted with pancreatic progenitors derived from hES cells, eliminating the need for finding the efficient strategy for terminal differentiation in vitro. However, this approach would be hampered by teratoma formation from cells that were not fully differentiated at the time of transplantation (see below).

Although the differentiation of specific cell types such as pancreatic cells can be established through their characterization with several markers, it remains obvious that no protocol currently offers a factual conversion of all cultured stem cells to the desired phenotype. Therefore, the possibility exists that other cell types or even multi- or pluripotent cells might be transplanted to the diabetic recipients, which raises safety issues. Indeed, despite the fact that functional β-cells developed in vivo after transplantation of hES-derived pancreatic cell preparations, teratoma formation was also observed in at least 15% of grafts (19,51,52). Several tools have been investigated in order to select out the undifferentiated cells that express surface markers (SSEA4, SSEA3, TRA-1–60, TRA-1–81) or to specifically isolate the desired cell type (cell-trapping) that have been modified to express an antibiotic resistance gene, a fluorescent or a bioluminescent reporter under CK19, PDX1, or INS promoter (53,–55). Whereas these techniques would allow for the selection of differentiated β-cells up to purity, transplantation of such cells is presently not approved given the genotoxic, the mutagenic, and the oncogenic risks associated with transgenic approaches. New transgenic designs are currently under development to specifically address these issues.

The in vitro differentiation of functional β-cells from human pluripotent cells (hES and iPS) represents a major challenge for diabetes cell therapy in the future. Progress toward this goal already encompasses the efficient generation of definitive endoderm and PDX1-expressing pancreas progenitors. The recent successful studies indicate that the differentiation of PDX1⁺ progenitors concomitantly requires at least two events after DE establishment in order to mimic the in vivo situation: blockade of hepatic induction by BMP antagonism and induction of pancreas progenitors by retinoid signaling (Fig. 3). Optimization of this step might still prove beneficial in order to unravel the role of FGF and hedgehog modulation and in order to definitely establish cells that undoubtedly show combined expression of the major transcription factors recognized for the multipotent pancreas progenitor, namely PDX1, PTF1a, and NKX6.1. In addition, the recently described temporal shift in the requirement for BMP antagonism to a need for BMP signaling during mouse pancreas development will merit implementation in hES cell differentiation. Despite the latest data indicating differentiation of the PDX1⁺ progenitors into insulin/C-peptide⁺ cells, we believe that additional work is needed for an optimal in vitro control of the progression toward NGN3-expressing endocrine progenitor cells and for the maturation of glucose-sensing and insulin-secretion functions in the β-like cells. A step forward toward these goals will certainly shorten the time frame for achieving our objective of using hES cells as alternative sources of functioning β-cells in diabetes cell therapy.

This work was supported by a Vrije Universiteit Brussel (VUB) research grant under the code GOA41-Vrije Universiteit Brussel. J.K.M. is a postdoctoral research fellow with the VUB.

No potential conflicts of interest relevant to this article were reported.

J.K.M. researched the data, contributed to the discussion, and wrote, reviewed, and edited the manuscript. B.C. contributed to the discussion. L.S. contributed to the discussion. K.S. contributed to the discussion and reviewed/edited the manuscript. L.B. contributed to the discussion and reviewed/edited the manuscript.