Neural Subtype Specification from Embryonic Stem Cells

Abstract

One of the keys to using embryonic stem cells (ESCs) in brain research and potential application in neurological diseases is directed differentiation of neuronal and glial subtypes. This may be achieved by application of developmental principles in guiding cell lineage specification from naïve stem cells. Establishment of defined ESC differentiation models that recapitulate in vivo development, especially from human ESCs, will most likely provide a dynamic tool for dissecting molecular mechanisms underlying early embryonic development that is otherwise not readily obtainable. This is also a rational and realistic way of producing enriched populations of functional neurons and glia for pathological analyses as well as possible therapeutic applications.

Embryonic stem cells (ESCs), derived from the inner cell mass of the early embryo, are capable of producing all cell types that make up an organism. Generation of specialized cell types from ESCs in vitro as well as in vivo (producing teratoma) is essentially a recapitulation of early embryonic developmental processes. The ESC differentiation system, especially with human ESCs, offers a window to mysteries underlying early development, such as the complex brain development that might otherwise be unattainable in other systems. Similarly, naturally occurring or genetically manipulated mutant ESCs may be exploited to unveil developmental disorders and/or certain pathological processes. The differentiated progenies such as neural cells may also potentially be used for drug screening and therapeutic applications in neurological injuries and diseases.

A key step to unlocking the potential applications of ESCs is directed differentiation of functional target cells such as subclasses of neurons and glial cells. How and how well are we differentiating functional neural cell types from mouse and human ESCs? Do in vitro differentiation systems thus far devised recapitulate early neural development? Are ESC‐differentiated neuronal and glial subtypes equivalent to their counterparts in the brain? Here I intend to summarize the success and failure of our ability to direct ESCs to neurons and glial cells with a focus on a few extensively pursued cell types and differentiation systems.

DEVELOPMENTAL PRINCIPLES AS GUIDELINES FOR NEURAL SUBTYPE SPECIFICATION FROM ESCS

Induction of the neuroectoderm.  Neurons and glia are born shortly after induction of the neuroectoderm. The neuroectoderm, in the form of the neural plate (a sheet of neuroepithelium) and neural tube, appears at the beginning of the second week of mouse embryonic development. In humans, it is easily discernable by the end of the third week of gestation. The biochemical events that lead to neuroectoderm differentiation must occur prior to the cellular differentiation, most likely during or prior to gastrulation, in which cells move to form the three primary germ layers. This timeline forms the basis for directing stem cells toward their neural fate. The molecular mechanism underlying neural induction, inferred largely from studies using xenopus, chick, and other lower vertebrates, appear to involve multiple classic pathways such as activation of fibroblast growth factors (FGFs) and/or inhibition of bone morphogenetic proteins (BMPs) and Wnts (54, 66). FGFs may instruct a “pro” neural state at an early stage whereas BMP antagonists may subsequently solidify the neural identity. Inactivation of Wnts appears to be prerequisite for orchestrating FGF and anti‐BMP signals in neuroectodermal specification, and the presence of Wnts limits or forms the boundary of the neural plate. Thus, these signaling pathways play a spatially and temporarily differential role in specifying the neuroectoderm. Limited studies using mouse (2, 60, 70) and human (19, 26, 43, 77) ESC as a model system also indicate the involvement of these pathways in mammalian neuroepithelial differentiation. It will be important to apply appropriate soluble factors at the right time in order to induce a correct type of neuroepithelial cells.

Specification of neurons and glia from neuroepithelia.  Each neuron in a given location of the brain and spinal cord carries a unique set of transmitter(s) and makes connections with its own target(s). At the same time, neurons with the same transmitter phenotypes reside in different regions of the central nervous system (CNS) exerting distinct functions. It is thus likely that assignment of the positional identity and specification of the transmitter phenotype is operated by separate processes. Little is known about how these two parallel pathways are coupled in the specification of neural subtypes. This presents a challenge for differentiating naïve stem cells to neuronal subtypes with correct positional and transmitter phenotypes such as midbrain dopaminergic neurons or perhaps the substantia nigra type of dopaminergic neurons. At present, the principle used for directing stem cell differentiation is largely based on the well‐characterized positional specification, or neural patterning.

Because of differential temporal and spatial exposure to morphogens, the process of neuroectoderm formation does not occur homogenously and simultaneously. Neuroepithelium forms first in the head region and gradually extends caudally to form the entire neural plate. This suggests that the regional identity of the initially specified rostral neuroepithelia (or prospective forebrain neuroectoderm) is not fixed and can be respecified to a more caudal fate. Meanwhile, the neural plate begins to fold and fuse dorsally at the future neck area, which extends both rostrally and caudally to form the complete neural tube. Thus, neuroepithelial cells are temporally and spatially different from each other at the time when the neural plate and neural tube are formed. Accordingly, neuroepithelial or progenitor cells generated from stem cells in a Petri dish are likely to be different from each other depending on the morphogens used. Even the same morphogen when applied at a different time or in a different amount will result in production of different types of progenitors. Given the differential roles of these morphogens in neuroepithelial induction and continued use of these signaling molecules in subsequent specification of positional identity and neuronal and glial fate, one can appreciate the importance of selecting these soluble factors for differentiating ESCs to neuroepithelial cells in order to achieve subtype neural cell differentiation.

Identification of neural subtypes that are generated in vitro.  Identity crisis is the major contributor to confusion in the field of stem cell research. When tissues are disaggregated and placed in a Petri dish, the easiest and the most important criterion for defining a cell, the positional identity, is lost. This leads to the reliance on the use of molecular markers in defining a cell type. However, dependence on these markers is not without caveats. It is particularly problematic for ESC‐derived progenies, as most molecular markers are not limited to only neural cells but are also expressed by nonneural cells. Nestin, for example, is a good neural progenitor marker. However, it is expressed by human ESCs as well as differentiated nonneural cells. It thus requires a set of markers in combination with morphological and functional indicators to define a cell type. The absence of certain markers is equally important. Subcellular localization of the markers is often indicative of the specificity of the reagents and in some cases suggestive of functional attributes. Cells under stress conditions can exhibit binding to many antibodies against structurally distinct antigens, thus giving false positive staining. The characteristic neuronal and glial morphology and the stereotypic neuronal electrochemical property such as a typical resting membrane potential and sodium‐gated action potential can significantly aid in the identification of neurons and glial cells (57, 72).

Determination of neuronal subtypes can be achieved by expression of homeodomain transcription factors and/or secretion of transmitters. Most neurotransmitters are simple amino acids or small peptides. Most cells possess these essential amino acids and/or their metabolites or can simply take them up from the culture medium. Therefore, the simple presence of these amino acids or peptides does not necessarily indicate they are transmitters. A whole package is required to define the transmitter phenotypes, such as synthesizing and metabolizing enzymes, transporters, etc.

DIFFERENTIATION OF NEUROEPITHELIAL CELLS

Differentiation of neuroepithelial cells from ESCs is the prerequisite step and gatekeeper toward the generation of neuronal and glial classes. In the developing embryo, neuroepithelial cells are segregated into rostro‐caudal and dorso‐ventral domains by the time of neural tube closure, where they are fated to different progenitor pools. This means that naïve neuroepithelial cells in the early neural plate, in response to locally derived signals, differentiate into region‐specific progenitors. In analogy, neural progenitors differentiated from ESCs in a Petri dish will adopt a regional identity depending upon the presence of morphogens at the time the precursors are responsive. Hence, the neuroepithelial cells may represent fundamentally different groups of progenitors even though they can equally produce neurons and glial cells.

Position identity and differentiation potentials of ESC‐derived neural progenitors depend upon the morphogens used

Forebrain progenitors.  In mammals, especially in primates, the forebrain is the largest part of the CNS. During development, the neuroectoderm at the head region forms first. Accordingly, one would expect that neuroepithelial cells with forebrain identity can be readily differentiated from ESCs. Indeed, neuroepithelial cells differentiated from mouse or human ESCs initially express forebrain homeodomain transcription factors such as Otx1/2 and brain factor 1 (Bf1) but not hox genes that are generally present in the hindbrain and spinal cord (26, 65). These cells are likely primitive neuroepithelial cells, equivalent to the prospective neuroectodermal cells that initially form during gastrulation (37, 54). With the presence of mitogens such as FGF2, which itself can caudalize neuroepithelial cells, it is not surprising that fewer forebrain progenitors are present in most of the neural differentiation cultures. This may explain why few reports on differentiation of forebrain neuroepithelial cells from ESCs are available and why generation of forebrain progenitors is often viewed as a difficult task. It has been suggested that neuroepithelial cells generated from mouse ESCs after the removal of leukemia inhibitory factor (LIF) and addition of FGF2 are likely forebrain progenitors (60), although it was not confirmed whether these progenitors carry forebrain but not hindbrain and spinal cord homeodomain markers. Sasai and colleagues have recently shown that the proportion of mouse ESC‐derived progenitors with forebrain characters, as indicated by the expression of Bf1, can be increased to 30% when the Wnt signaling is blocked by dkk1 (62). Wnts, as well as retinoic acid (RA) and FGFs, are implicated in the caudalization of the neuroectoderm during development. Ying and colleagues have developed a simple neural differentiation protocol by removal of LIF (70). The neuroepithelial cells generated in this way appear to possess some of the forebrain progenitor characteristics and produce gamma‐aminobutyric acid (GABA) neurons (12) (see below). We have shown that human ESC‐derived neuroepithelial cells in a chemically defined condition exhibit almost exclusively the forebrain identity such as expression of Otx2 and Bf1 but not hox genes, at least at the primitive neuroepithelial stage (26). Manipulation of these forebrain precursors will open an avenue for specification of a rich array of neuronal subtypes that are present in the forebrain.

Posterior progenitors.  Neural progenitors differentiated from ESCs by treatment with RA (0.1–1 M) display hindbrain and spinal cord phenotypes assessed based on expression of a host of hox genes (62, 65). Treatment of ESCs with RA has been the most commonly used approach for neural differentiation from mouse ESCs (4, 14, 55) because of its simplicity and high efficiency. RA does not appear to act as a neural inducer in this system. Instead, it acts mainly to caudalize the neuroepithelial cells. This is because the neuroectodermal fate, based on the expression of the definitive neuroectoderm transcription factor Sox1 and nestin, is rapidly induced 1–2 days following the removal of the self‐renewing factor LIF from ESCs (60, 70), whereas RA is usually added to the differentiation culture 2–4 days later (31). Neural differentiation through treatment with RA is particularly useful for generation of motor neurons and other cell types that are located in the brainstem and the spinal cord. RA promotes the expression of Hox genes but suppresses forebrain genes such as Otx2 and Bf‐1 in a dose‐dependent manner (26, 65). By adjusting the concentration of RA, neural progenitors may be biased toward hindbrain or spinal cord fate. However, RA alone does not appear to induce further caudal (thoracic and lumbar) spinal cord fate. The motor neurons generated from mouse ESCs in the presence of RA mostly manifest upper cervical spinal cord phenotypes by expressing HoxC5 (65). Those derived from human ESCs following FGF2 and subsequent RA treatment display lower cervical spinal cord phenotypes by expressing HoxC8 (26). This is reminiscent of the finding from embryologic studies that RA promotes the upper cervical but represses thoracic homeodomain gene expression (29). Thus, RA alone will not be sufficient to specify progenitors within all the spinal cord segments. If cells with more posterior spinal cord phenotypes are desired, other morphogens such as FGFs, GDF11 or these morphogens in combination with RA may be needed. By the same token, RA may need to be avoided or its concentration substantially reduced in the culture if cell types with forebrain phenotypes are the differentiated target.

Progenitors with mid/hind brain characters.  The fact that differentiation of progenitors with the seemingly uniform spinal cord characteristics requires discrete sets of signaling implies that it will not be trivial to specify a more restricted mid/hind brain fate from ESCs. Interestingly, specification of midbrain progenitors from ESCs has been a major focus of ESC neural differentiation. It is well established that patterning of mid/hind brain region is controlled by signals released in the isthmus area, mainly FGF8. FGF8a is preferentially localized to the midbrain region whereas FGF8b mainly in the hindbrain region. Both can bind to an FGF receptor but the affinity of the b isoform is about 100 times higher than that of the a isoform (21, 27, 28). Treatment of mouse ESCs with FGF8 b, that is commercially available, indeed promotes the differentiation of progenitors with mid/hindbrain phenotype, as determined by the expression of Engrailed 1 (En‐1) and Pax2 (5, 25, 44, 68). Given the expression of En‐1 in mid/hind brain region and Pax2 preferentially in the hindbrain region, it is conceivable that the neural progenitors induced by FGF8 treatment may represent more hindbrain progenitors than midbrain cells. This may explain why most of the dopamine neurons differentiated from mouse and human ESCs using this “widely accepted standard protocol” do not possess critical midbrain dopamine neuron phenotypes, such as expression of Lmx1a/b, Nurr1 and ptx3. This illustrates the necessity of inducing correct types of neural progenitors in an attempt to achieve target neuronal differentiation. It also points to a need to modify the existing protocols based on the developmental finding described above in order to specify ESCs to ventral midbrain progenitors that will ultimately give rise to dopamine neurons that are lost in Parkinson’s disease.

From the above brief account on differentiation of neuroepithelial cells and neural progenitors from ESCs using various inducing factors, it becomes obvious that neural progenitors differ from each other significantly despite a similarity in morphology, expression of common progenitor markers, and their ability to differentiate into neurons and glial cells. RA treatment tends to generate neural cells that are fated to hindbrain and spinal cord identities. FGF8 appears to favor the differentiation of neural progenitors with mid/hind brain characteristics, albeit with limited potency by itself. FGF2‐induced neural progenitors appear to be a mixed population exhibiting phenotypes ranging from the forebrain to spinal cord characteristics. Other morphogens such as noggin, an antagonist of transforming growth factor β (TGFβ) family, have also been shown to efficiently promote neuroepithelial differentiation (19, 43). Similarly, mouse ESCs readily differentiate to neuroepithelial cells when the ESCs are dissociated into single cells and cultured at a low density. This culture system is designed to remove the inhibitory signaling of TGFβ family thus promoting neural differentiation (60). It is not clear at present what positional identity these neural progenitors exhibit.

Dorsal‐ventral phenotypes of neural progenitors.  I have so far only summarized the neural precursors that exhibit broad rostral–caudal positional characteristics. In order to specify subtypes of neurons and glia, precursor cells need to be further restricted to a narrower rostral–caudal region such as a certain segment of the spinal cord. This can be potentially achieved by adjusting the concentrations of a morphogen or combination of morphogens, as in the case of progenitors of different sub‐regions of the spinal cord induced by RA or FGFs and GDF11. Further restriction by dorsal–ventral morphogens such as BMPs, Wnts, and sonic hedgehog (SHH) may limit progenitors to a state that is equivalent to a specific dorsal ventral domain of the neural tube. RA and SHH at an appropriate concentration enrich for progenitors that express Olig2 (26, 65), a transcription factor expressed by motor neuron progenitors at the pMN domain of the ventral neural tube (58, 79). Similar to what we have learned from chick embryo studies, the ventralizing morphogen SHH activates the transcription of Nkx6.1 and Nkx2.2 but represses Irx3 and Pax6, which coordinate to activate Olig2 expression. In practice, both the activated and repressed genes may be good indicators of whether the choice of morphogens is appropriate for induction of a particular neural progenitor. Similarly, FGF8 and SHH, at a correct isoform and concentration, may favor the induction of progenitors that are normally residing in the ventral midbrain or hindbrain, thus facilitating the generation of midbrain dopaminergic neurons or hindbrain serotonin neurons (5). FGF8 and SHH are also required for specifying ventral forebrain progenitors (15, 47, 49). This would mean that additional factors such as timing, space, quantity of the same factors or other molecules, are needed to differentiate subtypes of progenitors. In contrast to SHH, BMP, applied right following neuroepithelial induction, results in differentiation of neural crest‐like cells (35). Tuning the correct combination of morphogens at the appropriate concentration and at the right time is not trivial and often is only determined via trial and error.

Can subtype progenitors be maintained and expanded?  If the neuroepithelial cells, especially regionalized neural progenitors, can be maintained and expanded, a large population of progenitors with similar characteristics and differentiation potential can be obtained. This has significant application value as a homogeneous population of neural stem/progenitors provides a convenient tool to study the fundamental biology of tissue‐specific stem/progenitors. It also allows production of a homogeneous population of subclasses of neural cells from the progenitors over an extended period. This is particularly useful for neural transplantation purposes as progenitors are easier to manipulate and more responsive to environmental cues for differentiation and integration than are postmitotic neurons.

Neural stem/progenitor cells have been expanded in culture in the presence of mitogens such as epidermal growth factor (EGF) and/or FGF2 (63). However, the expansion is accompanied by diminished potential of these progenitors to generate neurons over glial cells (59). This trend is in general agreement with the shift from neurogenesis to gliogenesis during normal development, suggesting the preservation of intrinsic cellular program in governing cell fate in vitro. In this regard, a recent report by Conti et al was quite interesting, in that mouse ESC‐derived neural stem cells were expanded for over 100 passages in the presence of FGF2 and EGF yet retained the capability to produce a large proportion of β_III‐tubulin positive neurons (12). The expanded neuroepithelial cells display a bipolar morphology and express Sox2, Pax6, nestin, and RC2, suggestive of a radial glia‐like phenotype. During development, radial glia functions as neural stem cells (16, 39). The authors postulate that the maintenance of radial glial property endows the cells with a “niche” that traps the cells in a state of symmetric cell division. How a simple treatment with FGF2 and EGF creates the special niche remains peculiar. FGF2 and EGF have been widely used for expanding neural stem/progenitors isolated from rodent and human brain and, with no exception, fail to preserve a high degree of neurogenesis over a long period. Temple and colleagues have shown that factors released from vascular endothelial cells can prolong the neurogenic potential of neuroepithelial cells (50), suggesting the importance of a neurogenic niche in maintaining the stem/progenitor cell state. The suggestion that the ESC‐derived neuroepithelial population contains endothelia‐like niche cells is also plausible given the tendency of ESCs to give rise to many cell types other than the neural lineage. On the other hand, it is peculiar why the neuronal cells, differentiated from expanded progenitors, regardless of whether they are derived from cortex, striatum, or ESCs, exhibit an almost uniform GABAergic phenotype (12). If the neural stem/progenitors can differentiate to neurons with multiple transmitter phenotypes (other than GABAergic) before expansion, it would suggest that FGF2 and EGF selectively promote the proliferation of GABAergic progenitors. Alternatively, prolonged expansion with the growth factors may alter the phenotypes of progenitors, thus influencing their differentiation potential.

It is encouraging that the neurogenic potential of neural stem cells can be maintained after extensive expansion, even though there are many unanswered questions. A related question is whether the positional identity and/or differentiation potential of subtype neural progenitors can be maintained after in vitro expansion. Limited studies indicate that expanded neural stem/progenitors retain positional information associated with their origins of brain regions (17, 40). This, however, does not imply the maintenance of the differentiation potential. Embryonic ventral mesencephalic progenitors, which produce robust dopaminergic neurons at the time of isolation, lose their dopaminergic potential shortly after expansion in the presence of FGF2 (56). Similarly, human ESC‐derived neural progenitors retain their positional identity, based on homeodomain transcription factor expression, and a high degree of neurogenic potential even after months of expansion (73, Zhang). However, the potential to produce large projection neurons such as midbrain dopamine neurons and spinal cord motor neurons fades in two to four passages and is replaced by other neuronal populations. Nevertheless, neuroepithelial cells expanded in the form of neural tube‐like rosettes continuously generate new dopaminergic neurons for a much longer period (Zhang, unpub. obs.). This suggests that neural tube‐like rosettes may act as an embryonic niche to maintain the progenitor state, giving hopes for extending the neurogenic potential of a subtype neuronal progenitor.

SPECIFICATION OF NEURONAL SUBTYPES FROM ESCS

The birth of a neuronal type is the consequence of the interplay between intrinsic program of precursor cells and extracellular signals at a given time and place (20, 30, 41). At an early stage, extracellular factors appear to dictate the fate choice of precursors whereas following specification of a neuronal progenitor, extrinsic factors have little influence on lineage choice as the progenitors essentially mature following their intrinsic process. Therefore, the key to making the right type of neurons is to specify a correct type of neuronal progenitor. Although extrinsic morphogens play a major role in specifying stem cells to neuronal progenitors, intrinsic cellular programs are always in play. Neurons are born first, followed by glial cells. Within the neuronal lineage, large projection neurons appear prior to smaller interneurons. Thus, steps for specifying large projection neurons from ESCs are relatively straightforward as compared with generation of interneurons and glial cells. This, however, is not the case for neural stem/progenitors expanded from embryonic brain tissues, because the intrinsic clock for the generation of large projection neurons has elapsed. Large projection neurons are often targets of neurodegenerative diseases such as striatal projection GABA neurons that are affected in Huntington’s disease, midbrain dopamine neurons that are degenerated in Parkinson’s patients and motor neurons that are lost in amyotrophic lateral sclerosis (ALS) patients. Developmentally, molecular pathways that lead to the specification of neuronal types in the structurally relatively simple spinal cord region are better understood than those governing cell birth in more complex structures such as the brainstem and particularly the forebrain. Consequently, differentiation of spinal cord cells such as motor neurons is better achieved than those of brainstem and forebrain.

Coupling the intrinsic cellular program with extrinsic factors is critical for specifying neuronal subtypes.  Specification of spinal cord motor neurons during development is one of the best‐characterized pathways among neural cells. Not only have the transcriptional codes been elucidated for progenitors and postmitotic motor neurons but the inductive extracellular molecules are also defined. Specification of motor neuron progenitors requires activation of class II (induced by SHH, eg, Nkx6.1 and Nkx2.2) and suppression of class I (inactivated by SHH, eg, Pax6 and Irx3) homeodomain transcription factors. Activation of Nkx6.1 and suppression of Irx3 defines the dorsal boarder of the motor neuron progenitor domain. Activation of Nkx2.2 and suppression of Pax6 defines the ventral edge of the progenitor domain. Together, the combinatorial actions of these mutual repressive homeodomain proteins restrict the motor neuron progenitor domain, resulting in the appearance of Olig2‐expressing motor neuron progenitors (20). Remarkably, such a complicated transcription network can be orchestrated by a rather simple soluble factor SHH at a specific concentration. With the continued action of RA, the motor neuron progenitors become postmitotic motor neurons by expressing Lim transcription factors such as Lim3, Isl1/2, and HB9. This molecular pathway learned from chick embryo studies has now been largely recapitulated using mouse ESCs. By treatment of mouse ESCs with RA and SHH, Wichterle and colleagues have successfully differentiated stem cells to spinal motor neurons with up to 20% efficiency (65). The effect of RA and SHH is transduced through the regulation of homeodomain transcription factors that are necessary for the birth of spinal cord motor neurons. The mouse ESC‐derived motor neurons can innervate muscles following transplantation into chick embryos.

Using the same principle, we have established a chemically defined system for efficient differentiation of spinal motor neurons from human ESCs. Human ESCs are first differentiated to neuroepithelial cells in the presence of FGF2 (77) followed by treatment of the enriched neuroepithelial cells with RA and SHH (26). In this study, we have identified two sequential developmental stages during neuroepithelial differentiation, an early primitive neuroepithelial stage that is characterized by columnar epithelial cells expressing most of the neuroectodermal transcription factors but not the definitive neuroectodermal transcription factor Sox1, and a later definitive neuroepithelial stage that is characterized by Sox1‐expressing columnar epithelial cells forming neural tube‐like rosettes (Figure 1A). The primitive neuroepithelial cells can be readily caudalized by RA and these posteriorized progenitors can be efficiently differentiated into HB9‐expressing motor neurons in the presence of SHH, with about 20%–50% of the total differentiated progenies. In contrast, FGF2‐induced definitive neuroepithelial cells, although they can be caudalized by RA, generate fewer motor neurons (<5%) under the same culture condition (Figure 1B). Analyses of homeodomain transcription factors regulated by the morphogen treatments at each of the developmental stages indicate that RA (at 0.1–1 µM) eliminates rostral homeodomain transcription factor Otx2 and elicits hox gene expression in the primitive neuroepithelial cells in a dose‐dependent manner. SHH subsequently coordinates with the caudalized neuroepithelial cells in regulating the expression of Pax6, Nkx6.1, and Irx3, which activates the transcription of Olig2 and subsequently HB9 (26). In contrast, RA has less caudalizing effect on FGF2‐induced definitive neuroepithelial cells. Addition of SHH does not create a transcription network that is necessary for activating Olig2 and HB9 (Figure 1B). Thus, specification of spinal motor neurons from human ESCs requires sequential activation/inhibition of transcription factors similar to those learned from chick embryo studies. Importantly, activation of the motor neuron specification machinery is contingent upon the specific neuroectoderm induction process, i.e., RA is required before neuroepithelial fate is determined. This translates into an early application of RA between primitive and definitive neuroepithelial cells in the differentiation cultures. Similarly, early application of FGF8 appears to induce midbrain type of dopaminergic neurons more efficiently, whereas later application of FGF8 has little effect even though tyrosine hydroxylase (TH)‐expressing dopamine neurons can be readily differentiated under both conditions (68) (Figure 1C). Therefore, it is crucial to couple the intrinsic program of precursors at a particular developmental stage with a specific set of extrinsic morphogens in order to efficiently differentiate ESCs to a subclass of neurons.

Figure 1.

Stereotypic neuronal subtype specification from human embryonic stem cells (ESCs). A. Human ESCs were differentiated to primitive neuroepithelial cells (NE) at around day 10 and then NE that exhibit neural tube‐like rosettes in 14–17 days of differentiation in a chemically defined neural medium (77) without the presence of morphogens. B. Treatment of the primitive NE (green arrows) with retinoic acid (RA) and sonic hedgehog (SHH) resulted in efficient generation of HB9+ spinal motor neurons. Similarly, FGF8 and SHH treatment at this stage resulted in differentiation of TH+ dopamine neurons some of which also expressed the midbrain transcription factor engrailed 1 (En1). C. Treatment of more advanced NE (red arrows) with RA and SHH resulted in few HB9+ motor neurons although similar proportion of Isl1+ neurons, mostly likely spinal interneurons. Similarly, FGF8 and SHH treatment at a later stage resulted in differentiation of TH+ dopamine neurons, many of which expressed Bf1, a transcription factor expressed by forebrain cells. Bar = 50 µm. (B) and (C) are reproduced partially from Li et al (26) and Yan et al (68), respectively, with permission.

Coordination of positional patterning and transmitter specification may be necessary for generation of functional neuron subtypes.  Although RA‐induced mouse and human motor neurons acquire cholinergic phenotypes along differentiation, it does not necessarily mean that regionally patterned progenitors will automatically acquire a correct type of transmitter phenotypes or vice versa. This debatable concept appears to manifest in an effort to induce ESCs to various subtypes of neurons that are present in the mid/hind brain region. The brainstem is structurally similar to the spinal cord and is also segmented during development, although changes from one segment to another in the brainstem are more drastic than those in the spinal cord. This would imply that each segment of the brainstem may need to be patterned somewhat independently. Signaling in the mid/hind brain boundary, eg, by FGFs, plays a key role in patterning the mid/hind brain (18, 21, 47). If explants taken from the anterior neural plate region are cultured in the presence of FGF8 and SHH, a large number of progenitors will adopt a dopaminergic fate (69). This has been interpreted as the requirement of FGF8 and SHH for specification of midbrain dopamine neurons.

By treatment of mouse ESC‐derived neuroepithelial cells with FGF8 and SHH, a large proportion of TH+ dopamine neurons (~30% of total differentiated progenies) can be differentiated (5, 25). These cells are indeed dopamine neurons as they secrete dopamine in an activity‐dependent manner and reverse locomotive functional deficit in Parkinson’s rodents following transplantation into the lesion‐side striatum. Thus, FGF8 and SHH treatment appears to activate the dopaminergic pathway. Many of these dopamine neurons express a mid/hind brain transcription factor En1, when FGF8 is added at an early stage (5) or even after the expansion of neuroepithelial cells (25), suggesting that FGF8 can promote the patterning of mid/hind brain cells. For human ESCs, we found that a simple shift of the timing of FGF8 application significantly alters the phenotype of dopamine neurons (68). Treatment of human ESC‐derived primitive neuroepithelial cells with FGF8 and SHH results in differentiation of dopamine neurons, a significant population of which express En1. Addition of FGF8 and SHH to definitive neuroepithelial cells results in generation of TH neurons, most of which do not coexpress the midbrain transcription factor En1. Instead, many of them coexpress Bf1 (Figure 1C) or GABA, suggesting that many of these dopamine neurons may represent forebrain/olfactory dopamine neurons. Some do not express Bf‐1 or En‐1, resembling those in the hypothalamus (80). Thus, FGF8 and SHH can promote the differentiation of dopamine neurons but may not be sufficient to restrict cells to the midbrain dopamine neuronal fate, particularly in humans. Additional factors must come into play in order to further restrict precursors to the midbrain fate.

Coculture with mesoderm‐derived stromal cells, such as PA6 cells (9, 22, 35) and MS5 cells (5, 44), is at present the most widely used approach for differentiating mouse and human ESCs to dopaminergic neurons. Although FGF8 and SHH have been added to the culture system, the (dopaminergic) neural inducing effect comes from direct contact with stromal cells (22, 44) and the inducing activity remains even after the PA6 cells are fixed with formaldehyde (22). This suggests that factors other than FGF8 and SHH play a critical role in midbrain dopaminergic neuron specification. It has also been shown that neural progenitors that overexpress Nurr1 can differentiate to dopamine neurons when cocultured with mesencephalic astrocytes (61). One of the signals associated with astrocytes may be Wnts (10). Wnt1 has been found to be critical for the morphogenesis of the midbrain and the genesis of dopaminergic neurons (45), partly by maintaining expression of En1, a midbrain transcription factor (13). Null mutation in Wnt1 results in loss of major parts of the midbrain and enlargement of the forebrain (32). Overexpression of Wnt1, however, increases the size of the midbrain (42). Other Wnts, including Wnt3a, and Wnt5a are involved in the proliferation, survival and maturation of midbrain dopamine neurons (11). Thus, more sophisticated approaches such as coordinated action of FGF8, SHH, and Wnts, are required for generation of midbrain dopamine neurons from ESCs, particularly those of human origin.

Are the dopamine neurons generated from mouse and human ESCs with combinations of FGF8/SHH or by treatment with stromal signals the midbrain dopaminergic neurons? By reevaluating the phenotypes of dopaminergic neurons generated from mouse and human ESCs using the so far published FGF/SHH or stromal cell inducing protocols, we found that only a small fraction (10%–30%) of mouse ESC‐derived and hardly any of the human ESC‐derived dopamine neurons express Lmx1b and ptx3 (Zhang, unpublished), the latter of which is exclusively expressed by midbrain dopamine neurons and is required for the survival of postmitotic dopamine neurons (52, 53). Using a green fluorescent protein (GFP) reporter in the ptx3 locus, Li and colleagues had a similar finding (78). The dopaminergic transmitter phenotype appears to be efficiently specified whereas the midbrain specification is much less effective. By expressing Lmx1a, a transcription factor expressed by ventral midbrain precursors as well as a much larger population of dorsal progenitors (34, 34), under the control of the nestin enhancer Ericson, Perlmann and colleagues can efficiently differentiate mouse ESCs into dopamine neurons that exhibit correct phenotypes of midbrain dopaminergic neurons (1). These in vitro produced mouse dopamine neurons possess the whole set of genes in the same TH‐expressing neurons, including Lmx1a, Lmx1b, En‐1, Nurr1, ptx3, and dopamine transporter. This will be a “gold standard” for defining the identity of midbrain dopaminergic neurons that are produced in a Petri dish. This study also points to the necessity of combining SHH/FGF8 with Lmx1a when inducing midbrain dopamine neurons. Thus, the molecular pathways for specification of transmitter phenotypes and for ventral midbrain positional identity need to be coupled in order to produce functional midbrain dopamine neurons.

The above discussion also points to an important technical issue in characterizing neuronal subtypes generated in vitro. Most of the analyses are limited to the demonstration of TH expression and dopamine release. Expression of midbrain dopamine neuron phenotypes is generally performed using RT‐PCR. As neural progenitors with various regional specificities and transmitter properties are present in most differentiation culture conditions, the mere expression of midbrain neural transcription factor mRNA in bulk cultures is not sufficient to demonstrate that the TH+ neurons are the midbrain type of dopamine neurons. Colocalization of TH with nuclear midbrain transcription factors will be necessary to confirm the midbrain dopamine neuron identity.

Specification of forebrain neuronal subtypes is hardly explored.  The forebrain is evolutionarily the newest part of the CNS and contains the richest array of subtypes of neurons that organize into the most complex structure. The patterning of the forebrain is barely understood and specification of neuronal subtypes in the forebrain is largely unknown. The poor understanding of the molecular mechanisms underlying forebrain development is one of the main reasons why hardly any report is presented on efficient differentiation of a subtype of forebrain neurons. Technically, commonly used mitogens such as FGF2 often possess caudalizing effect on neuroepithelial cells, especially at an early stage. Although preventing caudalization by dkk1 increases the pool of forebrain neural progenitors derived from mouse ESCs (62), the progenitor population is still too small to produce a meaningful proportion of subtype forebrain neurons unless the differentiated progenies can be significantly enriched or purified.

Neuroepithelial cells differentiated from human ESCs initially are almost exclusively of forebrain identity (26). This indicates a possible strategy for generating forebrain neural cells by simply preventing caudalization of the primitive forebrain neuroepithelial cells. By using factors other than FGF2 in our culture system, these primitive forebrain neuroepithelial cells can be maintained to a large extent (Li and Zhang, unpublished). This has laid down a solid foundation for the generation of forebrain neuronal subtypes. Still, neuronal subtypes are usually located in a small anterior‐posterior forebrain domain and at a particular cortical layer or a subcortical area, suggesting that a delicate patterning would be necessary. An added difficulty is the almost total lack of molecular markers for progenitors of forebrain neuronal subtypes. Careful and creative strategies will be needed in order to direct ESCs to the vast array of neuronal subtypes that are harbored in the primate forebrain.

SPECIFICATION OF GLIAL SUBTYPES FROM ESCS

Understanding the molecular switch from neurogenesis to gliogenesis may be a key to specification of oligodendroglial lineage.  During embryonic development, glial cells, including astrocytes and myelinating oligodendrocytes, are generated following the birth of major neuronal types. The same neurogenesis to gliogenesis sequence is preserved when early neuroectoderm is cultured (46) or ESCs are differentiated along the neural lineage (71). In our chemically defined culture system, for example, human ESCs generate neuroepithelial cells in 2–3 weeks. These neuroepithelial cells differentiate predominantly into βIII‐tubulin+ neurons in the next 2–3 weeks. Glial fibrillary acidic protein (GFAP+) astrocytes appear thereafter, 6–9 weeks after human ESCs are differentiated. O4+ oligodendrocytes arise in a much longer period of differentiation (77). This temporal sequence of neuronal and glial differentiation roughly corresponds to the timeline observed from limited samples of fetal tissues (51). Thus, the intrinsic program governing neuronal and glial lineage development is retained in vitro.

Myelinating oligodendrocytes have been efficiently generated from mouse ESCs (6, 8, 31). The general procedure is to expand the ESC‐derived neuroepithelial cells with FGF2, EGF, or both until the gliogenic phase. The gliogenic progenitors are then expanded in a glial restricted medium in the presence of FGF2 and platelet‐derived growth factor (PDGF), both of which are known to promote proliferation of oligodendrocyte progenitors (33). Oligodendrocytes generated in this way are bona fide myelinating cells as they produce myelin sheaths following transplantation into the dysmyelinating or injured rat spinal cord (8, 31). Nevertheless, these procedures do not tell us how oligodendrocytes are specified from ESCs as the large proportion of oligodendrocytes is mainly achieved through expansion of the progenitors that had spontaneously differentiated. Rodent oligodendrocytes can be cultured to near pure populations through expansion of their progenitors by growth factors such as FGF2 and PDGF or conditioned medium from the B104 neuroblastoma cells (3, 75, 76). We have devised a protocol for oligodendrocyte specification from mouse ESCs based on the understanding that oligodendrocytes are derived from a progenitor that expresses the transcription factor Olig2 (48). Using a similar procedure described for generating motor neurons (see previous section), we are able to induce mouse ESCs to Olig2‐expressing neural progenitors. In the absence of RA but presence of SHH, these Olig2 progenitors differentiated mainly to oligodendrocyte progenitors during the gliogenic phase (Du and Zhang, in preparation). Thus, oligodendrocytes can be specified from mouse ESCs by SHH.

Differentiation of oligodendrocytes from human ESCs has been observed (19, 77). However, the proportion of oligodendrocytes among the differentiated progeny is generally very low. Attempts to expand the progenitors with known growth factors that are effective for rodent oligodendroglial progenitors are not successful, similar to what is seen using brain‐derived progenitors (67, 74). Keirstead and colleagues have recently reported the generation of pure cultures of oligodendrocytes using a rather simple “expansion protocol” (38). Human ESCs are cultured in a “glial restricted medium,” similar to the Sato medium (7), with the presence of 2–4 ng/mL of FGF2 and 20 ng/mL of EGF for 6 weeks. The culture is exposed to RA (10 µM) for 8 days after ESCs are differentiated for 2 days, which results in the formation of “yellow spheres.” Upon removal of EGF, these cells become multipolar cells with branched processes (38). Oligodendrocytes generated in this way, however, do not exhibit typical morphology that is generally unambiguous regardless whether they are cultured from brain tissues or from mouse and human ESCs (Figure 2A). It will be important that such unprecedented oligodendrocyte differentiation efficiency from human ESCs be reproduced by independent laboratories.

Figure 2.

Oligodendrocytes from human neural stem/progenitors and human embryonic stem cells (ESCs). A. Neurospheres, generated from embryonic human (18–20 gestation weeks) cerebral tissues, were differentiated for 7 days. Immunostaining with O4 revealed differentiating oligodendroglial cells with typical tripolar to multipolar morphologies. “NS” indicates neurosphere. B. Human ESCs were differentiated to Olig2 progenitors in 4 weeks according to Li et al (26) and then differentiated to neurons and glial cells for additional 8 weeks. Oligodendrocytes with a multipolar morphology appeared. Immunostaining with O4 at this stage revealed characteristic membrane and process staining (C). Blue: Hoechst labeled nuclei. Bar = 50 µm. (A) is reproduced partially from Zhang et al (74), with permission.

We have shown that the Olig2 neural progenitors can be readily differentiated from human ESCs in response to RA and SHH (26). These Olig2 progenitors generate mostly motor neurons during the neurogenic period. However, Olig2 progenitors persist after neurogenesis. Oligodendrocytes with typical morphology and sequential myelin protein expression appear during the gliogenic phase several weeks later (Zhang, Figure 2B,C). This suggests that the Olig2 progenitors may differentiate into oligodendrocytes. Attempts to promote the proliferation of the Olig2 progenitors with similar strategies used for mouse Olig2 progenitors have not yet been successful, again suggesting a potentially different molecular mechanism underlying the expansion of human oligodendroglial progenitors. The key to the production of oligodendrocytes from human ESCs is perhaps the signaling pathway that switches the neurogenic potential of Olig2 progenitors to the gliogenic capability. Unraveling the molecular switch from neurogenesis to oligodendrogliogenesis in human cells will be a challenging task given the extremely long stretch between neurogenesis and gliogenesis in humans. Manipulating the switch might also speed up gliogenesis.

Astrocytes: too common to be ignored?  Astrocytes are the most abundant cell type and are distributed throughout the brain and spinal cord. Unlike their cousin oligodendrocytes, which have a unique myelinating capability, astrocytes participate in almost every aspect of physiology and pathology of the CNS. The generation of astrocytes is poorly understood, partly because of lack of markers for their progenitors (48, 72). Neural progenitors isolated from brain tissues or derived from ESCs generate mainly astrocytes during gliogenic stages without special treatment. A recent finding by Rowitch and colleagues, however, indicates that some astrocytes may be specified from a specific progenitor pool in a particular CNS location through a similar molecular mechanism as for generation of neurons and oligodendrocytes. In the spinal cord, progenitors in a small domain just dorsal to the oligodendrocyte progenitor domain express a transcription factor stem cell leukemia (scl), which appears to be critical for the specification of astrocytes. Knockout of scl results in the enlargement of the ventral oligodendrocyte progenitor domain, suggesting that the specification of the two major types of glia may be adjusted through mutual repression at the transcription level (36). How this finding may be applied to astrocyte differentiation from ESCs remains to be seen. This finding may also help generation of oligodendrocytes from stem cells.

Although astrocytes appear to distribute uniformly throughout the CNS, astrocytes in different parts of the brain and spinal cord possess differential functional attributes. Astrocytes in the ventral mesencephalon, but not the dorsal mesencephalon or other parts of the brain, can promote the specification of neural progenitors to adopt the midbrain dopaminergic neuronal fate (61). Such functional difference may be attributed to the molecular profiles produced by region‐specific astrocytes. Ventral mesencephalic astrocytes express a high level of Wnts (10) which has been shown to be important in the specification of midbrain dopamine neurons (11). Can region‐specific astrocytes be differentiated from naïve ESCs? More importantly, will in vitro produced astrocytes possess the functional property associated with region‐specific astrocytes? If the answers are yes, the implication including therapeutic application will likely be far reaching.

ARE WE THERE YET?

Neural differentiation from both mouse and human ESCs is generally robust judging from differentiated cells that possess neuronal morphology and express general neural markers. This picture is less bright if one is to ask whether a subclass of neurons or glial cells can be selectively differentiated. Hence, directed differentiation of neural subtypes from ESCs will remain a major focus of efforts in the next few years toward applying ESC technology in studies of developmental neurobiology, neurodegeneration, and regeneration. Along these efforts, we may find answers to many of the questions we have been trying to answer.

Are ESC‐derived neuronal subtypes equivalent to their counterparts in the brain?  This question goes back to the heart of how a neural subtype can be directed from ESCs. Motor neurons differentiated from mouse and human ESCs based on developmental principles correspond to the cervical somatic spinal motor neurons judging from series of molecular markers and their interactions with target muscles. The identity of ESC‐generated dopaminergic neurons is less clear as limited molecular markers have been employed, which is not sufficient to define dopaminergic subtypes [except (1)]. Most of the cultures including ours comprise many different subtypes of dopamine neurons. This is because a simple application of FGF8 and SHH or in combination with stroma signals is not sufficient to restrict most ESC‐derived neuroepithelia to a ventral midbrain fate, at least for primate cells. More sophisticated procedures will be required to guide differentiation of dopamine neuronal subtypes. One may ask further which subtype of midbrain dopamine neurons, substantia nigra vs. ventral tegmental area, is produced from ESCs in response to a specific set of signals. Similarly, which pool of upper limb muscles are target of the cervical spinal motor neurons? Information concerning the development of some of these neuronal subtypes is available, which may be used to guide neuronal differentiation from ESCs. More importantly, ESC neural differentiation system provides a dynamic model to tease out molecular mechanisms underlying decision‐making process along each individual lineage branch.

Does the positional identity of neuronal subtypes matter?  Much of the discussion points to the critical role of positional information precursor cells possess in order to make a fate decision. The final position of a neuron also links to its function. Dopaminergic neurons in the substantia nigra and ventral tegmental area in the ventral midbrain possess distinct molecular characteristics, project to diverse brain regions, and execute unique functions. Developmentally, motor neuron progenitors always innervate the correct target muscles and do not connect to mismatched muscles even when a segment of spinal cord is deleted or is rotated upside down (23, 24). Whether this matters from a therapeutic standpoint is debatable. Transplanted neural progenitors into ectopic brain regions can mature and form functional synapses even though these grafted cells retain their donor regional identity (64). Of course, it is not known whether such synapses will yield appropriate function. This issue will become clearer when transplant studies in both developing and diseased environments are carried out.

Can neural differentiation from human ESCs be simply scaled up from mouse ESCs?  For structures such as the spinal cord that are evolutionally conserved, fundamental principles and even techniques learned from mouse ESCs may be readily applied to humans. A good example is the differentiation of cervical/brachial spinal motor neurons in which a similar set of morphogens at equivalent concentrations induces comparable populations of motor neurons from mouse and human ESCs with similar characteristics (26, 65). Even under such a condition, nuances exist, which involves significant trial and error (26). Differentiation of midbrain dopamine neurons can be readily achieved from mouse ESCs with a simple application of FGF8 and SHH at the correct time in an appropriate concentration (Zhang). This is achieved through induction of endogenous factors such as Wnts, which act in concert with the exogenous factors in a relatively short developmental window to specify a cell fate. In human ESCs, a simple FGF8/SHH treatment has much less effect in inducing a ventral midbrain fate. Additional steps are required to supplement the deficit in the long stretched developmental period in order to ultimately activate the midbrain dopaminergic pathway. The forebrain is quite different between rodents and humans. The motor cortex, for example, is located in the forefront of the forebrain in rodents and birds whereas in humans, it is a narrow band in the middle part of the cerebral hemisphere. This would suggest that even the initial patterning of cortical motor neuron progenitors may well be significantly different between rodents and humans. Not only neuronal types, but also glial cells may be generated through different mechanisms. Myelinating oligodendrocytes can be readily differentiated and expanded from mouse ESCs, which is not so in human ESCs. We do not even have a strategy for expanding isolated human oligodendrocyte progenitors in culture yet, although we have known appropriate procedures for mouse cells for over two decades. Thus, we may have to learn and uncover mysteries of human cell biology by working on human cells directly. The dismissive attitude toward the originality of human cell studies discourages investigators from exploring basic biological questions that are unique to human cells and fuels the unrealistic rush for clinical application. The net consequence is in fact slowing down the arrival of stem cell application in therapy.