Figure 2. (A) Timelines of published multistep procedures that have been used to induce differentiation of hESCs toward insulin-secreting cells before cell transplantation. (B) Characteristics of hESC-derived pancreatic cells that have been differentiated toward insulin-secreting cells. ActA, activin A; ALK5i, ALK5 inhibitor; BSA, bovine serum albumin; CDM, chemically defined media; Cyc, cyclopamine; D/F12-B, DMEM/F12 with BSA; EGF, epidermal growth factor; FBS, fetal bovine serum; FCS, fetal calf serum; Hep, heparin; HrgB, heregulin 1b; Nic, nicotinamide; Nog, Noggin; RA, retinoic acid; RPMI-B, RPMI1640 with BSA; SFM, serum-free medium; TBI, TGF-βR1 kinase inhibitor; TT, TTNPB.
Download figure to PowerPoint
Formation of definitive endoderm
The first step in differentiating hPSCs toward pancreatic β cells is the formation of definitive endoderm. There is some evidence, both in vitro and in vivo,[5, 6] that definitive endoderm derives from a bipotential mesendoderm progenitor that is marked by the expression of the transcription factors Brachyury and Mixl1.[7-9] Specification of the definitive endoderm lineage is accompanied by the upregulation of a number of transcription factors, most importantly, Goosecoid, Sox17, and FoxA2.[10-13] This same progression of gene expression occurs during both mouse and human PSC differentiation into definitive endoderm.[4, 14, 15]
In the majority of human PSC differentiation protocols, definitive endoderm formation is induced by treating cells with a high concentration of activin A—a transforming growth factor β (TGF-β) family member used to mimic the activity of Nodal. Nodal proteins play an evolutionarily conserved role in initiating gastrulation and in driving endoderm formation in all studied vertebrate species. In fish, frogs, and mammals, Nodal proteins act in a concentration-dependent fashion to promote formation of the mesoderm lineage at low concentrations and endoderm at high concentrations. The levels of Nodal signaling have also been linked to embryonic and germ layer patterning along the anterior–posterior axis such that high levels of Nodal promote anterior mesoderm and endoderm. Consistent with these studies in the embryo, the nodal mimetic activin A has distinct concentration-dependent effects. It was seen by Kubo et al. that different concentrations induce different developmental outcomes: high concentrations of activin A favored dorsal mesoderm and endoderm fates, whereas low concentrations of activin A favored more ventral mesoderm fates.
Subsequent studies utilizing hESCs have demonstrated that PI3K signaling must be suppressed for cells to optimally respond to activin/Nodal. Compounds such as wortmannin, which inhibits PI3K signaling, have been found to promote definitive endoderm formation and have been used in combination with activin A to induce definitive endoderm from hESCs. To increase the robustness of differentiation and reduce costs, researchers have sought to discover small molecule alternatives that have the ability to direct hPSC differentiation into definitive endoderm. Two such molecules, IDE1 and IDE2, have been identified and have been shown to be able to induce definitive endoderm formation (in the presence of serum) with similar efficiencies to activin A. The specific target molecule for these compounds has not been identified, though experiments indicate that activation of TGF-β signaling may be involved.
Other signaling pathways appear to modify the activity of activin A during the definitive endoderm induction step. In the embryo, these signaling pathways function during gastrulation and act downstream of Nodal. One example is the TGF-β superfamily molecule, BMP4, which is expressed in the posterior primitive streak. Mouse embryos lacking BMP4 fail to express genes associated with mesoderm formation, such as Brachyury, and die during gastrulation (at approximately embryonic day (E) 6.5). A number of studies have reported that the inclusion of low concentrations of BMP4 along with activin A on the first day of differentiation improves the efficiency of definitive endoderm formation.[23-26] Other signaling pathways that are essential for gastrulation have been shown to play a role in activin-induced definitive endoderm formation. For example, Wnt signaling is essential for vertebrate gastrulation and mouse embryos lacking Wnt3 fail to gastrulate. If canonical Wnt signaling is blocked during ESC differentiation, mesendoderm fails to form,[28, 29] whereas activation of canonical Wnt signaling enhances differentiation of hESCs toward definitive endoderm.
Formation of posterior foregut and pancreatic endoderm from definitive endoderm
In the vertebrate embryo, a number of molecular and morphogenic changes take place between the initial formation of definitive endoderm and emergence of Pdx1-expressing pancreatic endoderm. The definitive endoderm transitions from a two-dimensional sheet into a primitive gut tube that becomes patterned along the anterior–posterior and dorsal–ventral axes. During gut tube morphogenesis, the presumptive pancreatic endoderm comes in contact with mesodermal cell types including the septum transversum mesenchyme (ventral) and notochord (dorsal).
These mesodermal cell types are the source of several signaling molecules that regulate the formation of Pdx1-expressing endoderm, including fibroblast growth factor (FGF), retinoic acid (RA), and hedgehog. FGF signaling is involved in many stages of pancreas development, including global patterning of the endoderm along the anterior–posterior axis as well as posterior foregut patterning. In chick and mouse, low levels of FGF signaling promote posterior foregut fate and expression of Pdx1, whereas higher levels promote more posterior endoderm fates.[32-34]
The notochord, which is located adjacent to the dorsal endoderm until around E8.0, is an important source of several of these signaling molecules and has been shown to play an essential role in dorsal pancreatic development. Removal of the notochord in chick embryos before pancreatic bud formation resulted in failure to initiate dorsal pancreas development, though ventral pancreas development was unaffected. Further studies found that signaling molecules from notochord, including activin βB and FGF2, were able to repress the expression of sonic hedgehog (Shh) in the adjacent prepancreatic endoderm, though this was inhibited in the presence of exogenous Shh, suggesting that Shh acts as an antipancreatic factor.
RA signaling plays an essential role in the morphogenesis and organogenesis of numerous foregut organs, including the pancreas. The distribution of RA in the embryo is tightly controlled by synthesis from circulating retinol in a two-step reaction involving specific retinol and retinaldehyde dehydrogenases (RALDHs), as well as by degradation by specific cytochrome P450s (CYP26s). A variety of gain and loss studies indicate that retinoid signaling is required for pancreatic specification in a number of species.[38-41] For example, in zebrafish, it has been shown that retinoid signaling is essential for both pancreas and liver specification, and that exogenous RA treatment induces ectopic expression of pancreatic and liver markers. In both Xenopus and quail, it has been shown that retinoid signaling is essential for the formation of the dorsal pancreas, though it has no effect on ventral pancreas genesis. Further studies in Xenopus have demonstrated that while RA was sufficient to induce pancreas-specific genes in the dorsal pancreas, it was unable to induce these same genes in the ventral pancreas. Furthermore, in mice, loss of RALDH activity results in broad foregut organ abnormalities, including dorsal pancreas agenesis. Moreover, it was shown that RA signaling was sufficient to induce Pdx1 expression in the anterior endoderm.[38, 43]
Following from these developmental studies, nearly every published ESC differentiation protocol requires the addition of exogenous RA and/or FGF to promote the transition of definitive endoderm to Pdx1+ endoderm in mouse[44-46] and human (Fig. 2).[23, 25, 47-53] Although most protocols use FGF and/or RA, the interpretation of how these pathways promote expression of Pdx1 expression remains poorly understood.
Pancreatic endoderm to endocrine precursor cells
The commitment of pancreatic endoderm to endocrine precursor cells is an obligate step during the formation of β cells, though most differentiation protocols do not incorporate factors specifically designed to promote or enhance this process. This is due in part to a paucity of information about the signaling pathways that initiate endocrine cell formation and guide subsequent differentiation into the β cell lineage. There is a significant amount of information about the transcription factors that regulate endocrine and β cell fate in vivo, and these factors serve as useful markers to identify cell types in ESC-derived pancreatic cultures. For example, the transcription factor neurogenin 3 (Neurog3) has been an important marker because it is expressed early in the specification of the pancreatic endocrine lineage. However, as it is only expressed transiently and marks all endocrine cells, this makes it problematic to use as a marker to identify cells committing to a β cell fate.
In the mouse, Neurog3 has a bimodal expression pattern, which is thought to mark the first and second waves of endocrine cell formation. The first wave of Neurog3 expression is between E8.5 and E11.5, and the second wave of expression is initiated at E12, with peak expression observed at E15.5.[56, 57] Studies in both chick and mouse have shown that Neurog3 is involved in the epithelial–mesenchymal transition that is associated with endocrine cell development (Fig. 1), during which delaminating endocrine cells are marked by the expression of EphB3. It is thought that only endocrine cells specified during the secondary transition (E12 onwards) form definitive single-hormone cells in the adult pancreas, while endocrine cells formed during the primary transition (E8.5 through E11.5) do not contribute to the adult pancreas, instead playing a role during embryonic pancreatic function. This paradigm, however, is not supported by all studies of pancreatic cell development, some of which have demonstrated that a small number of endocrine progenitor cells formed before the primary transition are able to form hormone cells during the secondary transition, and hence, can contribute to the adult pancreas.
Perhaps the best-studied developmental signaling pathway involved in controlling endocrine precursor formation is Notch signaling. In a number of reports, it has been found that impaired Notch receptor activation or signaling resulted in upregulation of Neurog3 expression, leading to premature endocrine cell differentiation at the expense of both pancreatic progenitor cell expansion and exocrine cell differentiation.[62-68] In contrast, active Notch signaling appears to maintain cells as undifferentiated progenitors that are able to contribute to both proliferation and later differentiation. In this sense, the function of Notch signaling during endocrine precursor specification is analogous to its function during early mammalian neurogenesis.[69, 70] Despite a clear role for Notch signaling in repressing endocrine cell fate in vivo, relatively few protocols utilize Notch inhibitors such as DAPT during ESC differentiation (Fig. 2). This suggests that the endogenous Notch signaling levels in these cultures is not inhibitory to endocrine differentiation.
FGF10 produced by the surrounding pancreatic mesenchyme is also involved in pancreatic progenitor cell proliferation and acts upstream of Notch signaling. The role of FGF10 was initially identified in mouse knockout studies, where FGF10−/− mice exhibit severe pancreatic hypoplasia, owing to a profound reduction in the proliferation of pancreatic epithelial progenitor cells. By contrast, in transgenic mice that overexpress FGF10, pancreatic hyperplasia is seen, which results from the decreased differentiation of endocrine progenitor cells and concurrent expansion of progenitor cell numbers.[72, 73] This proliferative effect of FGF10 has also been observed in vitro in both isolated mouse and rat pancreatic epithelia. A study using isolated E10.5 mouse pancreatic epithelia found that the exogenous FGF10 increased the proliferation of pancreatic progenitor cells and promoted the expression of Hes1, a downstream target of Notch signaling. Moreover, in FGF10-treated epithelium, inhibition of Notch signaling caused the loss of Hes1 expression and a decrease in pancreatic progenitor cell proliferation, suggesting that Notch is downstream of FGF10.
Other signaling ligands that are secreted by the pancreatic mesenchyme also play a role in endocrine cell development. For example, mesenchyme-derived TGF-β signals were shown to be involved in regulating cell fate choice between endocrine and exocrine lineages.[76-78] There was, however, some degree of uncertainty as to which lineages TGF-β signaling was affecting: endocrine, exocrine, or both lineages via a common progenitor cell. Recently, the TGF-β superfamily members, TGF-β2 and TGF-β3, have been identified to play key roles in controlling the balance between endocrine precursor proliferation and differentiation in hESC cultures. In their report, Guo et al. identified a set of signaling factors, including TGF-β ligands that are expressed by the pancreatic mesenchyme in the embryo and were able to promote the expansion of hESC-derived pancreatic progenitors through a mechanism that is dependent on the duration of ligand signaling. It was seen that while a short burst of TGF-β signaling promoted upregulation of both Neurog3 expression and endocrine differentiation, prolonged TGF-β signaling lead to expansion of the progenitor pool and attenuation of endocrine development.
During the formation of the pancreas in human embryogenesis, NEUROG3 expression is observed at week (W) 8 in development, where it is coexpressed with other pancreatic transcription factors such as PDX1, as well as hormones such as insulin and glucagon. As in the mouse, NEUROG3 expression levels diminish over time, though it was still observable at W21. In contrast to Neurog3 expression patterns reported in the mouse, NEUROG3 expression has not been seen to be biphasic during human development. This may indicate that distinct primary and secondary transition events do not occur during human pancreatic development. There is, in fact, some debate about the functional necessity of NEUROG3 in human endocrine pancreas development. Several published reports describe patients with homozygous mutations in NEUROG3[82, 83] who were born with varying levels of circulating C-peptide, although one patient described had severely reduced C-peptide levels. Despite all patients' eventually developing diabetes, the fact that these patients often have detectable C-peptide suggests that human NEUROG3 mutations may not result in complete loss of function, or that NEUROG3 is not required for embryonic specification of all β cells in humans.
Commitment of endocrine progenitor cells to β cells
The signals that control the transition from endocrine precursors into functional β cells are not yet defined. It is, therefore, not surprising that differentiation of PSC-derived pancreatic progenitors into mature, functional β cells in vitro has not yet been achieved. Historically, studies have used insulin expression to identify β cells; however, it is now clear from mouse experiments that early insulin-expressing cells do not go on to become definitive β cells. This was initially demonstrated in mice, using a Cre/loxP tagging strategy, in which the early endocrine cells were irreversibly tagged using the activity of Cre recombinase, allowing for cell-lineage tracing. This work showed that the definitive β cells did not arise from early endocrine cells that were specified during the primary transition, but instead arose from a Pdx1+ progenitor cell that had not previously expressed insulin. This work suggests that it may be prudent to use other markers in addition to insulin in order to identify PSC-derived β cells.
There have been multiple studies that describe the emergence of hormone-expressing cells during human pancreas development. One report found that insulin-expressing cells first appear at W8 before other hormone-expressing cell types, such as glucagon or somatostatin, which were detected at W8.5, followed by pancreatic polypeptide (PP)-expressing cells at W10. At these early stages, insulin-expressing cells coexpress glucagon at frequencies ranging from 10% to 92%[84-86] and coexpress somatostatin at frequencies ranging from 10% to 97%.[84, 86] The existence of tri-hormonal (insulin/glucagon/somatostain) cells has also been reported. In contrast, PP-expressing cells have only been observed as monohormonal cells during human fetal development. Polyhormonal cells were seen to decrease in frequency after W9 and were not detected in pancreata at W22.
The use of hormone (markers) as a readout of PSC-derived β cells has misdirected many researchers into focusing on primary transition insulin+ cells. As is observed during embryonic development, human PSC-derived insulin-expressing cells are generally polyhormonal, expressing glucagon and/or somatostatin. These in vitro–derived polyhormonal cells are functionally immature and lack the capacity for glucose-stimulated insulin secretion (GSIS),[47, 24] which is the key hallmark of a functional β cell. As yet, however, there have been no reports suggesting that polyhormonal endocrine cells give rise to functional monohormonal endocrine cells, which is consistent with lineage-tracing experiments previously performed in mice.
To better identify cell types that can give rise to mature β cells, the field has focused on transcription factors, including Nkx6.1, MafB, and MafA, which mark progenitors and maturing β cells in vivo. For example, Nkx6.1 is initially broadly expressed in the mouse pancreas at E10.5 but becomes restricted to the insulin-expressing cells and scattered ductal cells between E13 and E15.5. In the adult mouse, Nkx6.1 is exclusively seen in the β cells. A similar pattern of NKX6.1 expression has been observed during human development, where NKX6.1 is broadly expressed in the pancreatic epithelium from W9, with subsequent downregulation of NKX6.1 expression in noninsulin-expressing cells by W13. Similar to the mouse, NKX6.1 expression is restricted to β cells in the adult human pancreas. Loss-of-function studies in the mouse showed that Nkx6.1 is required for β cell specification and in vitro overexpression studies suggest that it may improve β cell function, although this was not true in vivo.
There are now several reports examining the expression of NKX6.1 in differentiating hPSC cultures. Two early studies identified expression of NKX6.1 RNA and protein following the expression of earlier transcription factors such as PDX1; however, neither report demonstrated coexpression of NKX6.1 with other markers of β cell differentiation, such as insulin. More recently, NKX6.1 was observed to be coexpressed with C-peptide, PDX1,[20, 52, 93, 94] or insulin. However, these cells either demonstrated limited functionality or functionality was not examined.[25, 93, 94] It is therefore unclear if coexpression of NKX6.1 with either PDX1 or insulin will be a useful identifier of the functional β cell lineage. A recent study sorted differentiated hESCs on the basis of high (∼80%) or low (∼25%) NKX6.1 expression, then transplanted these different populations into diabetic mice. It was seen that while diabetes was reversed within 3 months in NKX6.1-high recipient mice, NKX6.1-low recipient mice remained hyperglycemic. This would suggest that high NKX6.1 expression may be an indicator of the definitive β cell lineage.
MafA and MafB are transcription factors that are expressed in both the mouse and human pancreas in a temporospatially regulated fashion. In the mouse, MafB is first observed at E10.5 during the primary transition in Neurog3+ progenitors and in insulin+ and glucagon+ cells. MafB expression then becomes restricted to glucagon+ cells (α cells) in the adult.[96, 97] In contrast, MafA expression is not observed until later in development—at E13.5 in the insulin+ cells specified during the secondary transition. In addition to their different temporal and spatial distributions, MafB and MafA play different functional roles during β cell differentiation. Surprisingly, only MafB is required for β cell development in the mouse—β cell numbers are not affected in mice lacking pancreatic Mafa. However, it is possible that MafB is able to compensate for the loss of MafA in this context. In contrast, MafA has been shown to control the glucose-responsive transcription of insulin and other associated genes in the definitive β cells.[99, 100] Mafa−/− mice have impaired glucose tolerance and defects in insulin secretion, emphasizing the importance of MafA in the maintenance of an appropriate GSIS response.
Interestingly, the expression pattern of MAFA and MAFB differ in the human, both spatially and temporally, as compared to the mouse. MAFA has been seen to be expressed throughout the developing pancreatic epithelium, including in the nascent endocrine cells from W9. In insulin+ cells, a strong nuclear expression of MAFA was seen, in contrast to weaker expression throughout the remaining epithelium. From W13 onwards, MAFA expression was downregulated, with a low level of expression restricted to the insulin+ cells, which was followed by complete loss of expression by W21. However in the adult, nuclear-localized MAFA expression was observed in β cells, similar to that seen in mice, suggesting possible biphasic expression of MAFA in humans. A second study found that MAFA transcripts increased from W9 through to W23, albeit at levels lower than those observed in the adult. The expression of MAFB initially resembled that of MAFA, in that at W9, expression is seen in both the pancreatic epithelium and in insulin+ cells. However, by W14, expression of MAFB is seen to be restricted to both insulin+ and glucagon+ cells. This spatial patterning is maintained throughout development, with expression of MAFB in both the α and β cells of the adult. This differs from the mouse, where MafB expression has not been observed in the β cells of the adult islet.