Volume 236, Issue 12 p. 3255-3266
Special Issue Reviews–A Peer Reviewed Forum
Free Access

Directing the differentiation of embryonic stem cells to neural stem cells

Chunyu Cai

Chunyu Cai

Biology Department, Wesleyan University, Middletown, Connecticut

Search for more papers by this author
Laura Grabel

Corresponding Author

Laura Grabel

Biology Department, Wesleyan University, Middletown, Connecticut

Department of Biology, Hall-Atwater Laboratories, Lawn Avenue, Wesleyan University, Middletown, CT 06459-0170Search for more papers by this author
First published: 06 September 2007
Citations: 68


Embryonic stem cells (ESCs) are a potential source of neural derivatives that can be used in stem cell-based therapies designed to treat neurological disorders. The derivation of specific neuronal or glial cell types from ESCs invariably includes the production of neural stem cells (NSCs). We describe the basic mechanisms of neural induction during vertebrate embryogenesis and how this information helped formulate several protocols used to generate NSCs from ESCs. We highlight the advantages and disadvantages of each approach and review what has been learned about the intermediate stages in the transition from ESC to NSC. Recent data describing how specific growth factors and signaling molecules regulate production of NSCs are described and a model synthesizing this information is presented. Developmental Dynamics 236:3255–3266, 2007. © 2007 Wiley-Liss, Inc.


The generation of neurons or glia that can be used to replace lost or damaged tissue is a key step in stem cell-based therapies designed to treat neurodegenerative disease and neurological disorders. Several different candidate sources have been proposed for producing these cells. Based upon their proliferative and pluripotential properties, embryonic stem cells (ESCs) are an attractive alternative to fetal or adult-derived central nervous system (CNS) tissue. The derivation of specific neuronal or glial cell types from ESCs invariably includes the production of neural stem cells (NSCs), the self-renewing, pluripotential stem cell of the nervous system, capable of differentiating into neurons, astrocytes, and oligodenrocytes (Temple,2001; Doetsch,2003). The desired fully differentiated cell can then either be produced in vitro before transplant by using signals that mimic the appropriate in vivo environment, or by transplanting the ESC-derived NSCs and relying upon the host environment to provide the appropriate cues.

Neurons and glia were among the first differentiated derivatives generated from mouse ESCs after their isolation over 25 years ago (Evans and Kaufman,1981; Martin,1981) and have successfully been derived from human ESCs as well (Thomson et al.,1998). Although both mouse and human ESCs are pluripotent early embryo-like cells, able to differentiate into derivatives of all three primary germ layers, they have several differences, including distinct growth factor requirements: leukemia inhibitory factor (LIF) and bone morphogenetic proteins (BMPs) for mouse ESCs and activin and fibroblast growth factor (FGF) for human ESCs (Brons et al.,2007; Tesar et al.,2007). This discrepancy is explained by recent reports documenting the isolation of pluripotential cells from the epiblast stage mouse embryo (Brons et al.,2007; Tesar et al.,2007). These mouse EpiSCs are more similar than mouse ESCs to human ESCs, based upon colony morphology, cloning efficiency, in vivo potential in chimeras, growth factor requirements, and transcriptional profile. Protocols for producing neural derivatives from ESCs were initially designed for mouse ESCs and modified for application to human ESCs. The recent realization that human ESCs reflect a later stage pluripotential cell type than mouse ESCs, epiblast vs. inner cell mass, may help in the future refinement of these protocols using human ESCs.

Several recent reviews have examined the conditions required to produce specific neuronal and glial subtypes, including neurons with specific transmitter profiles, from mouse and human ESCs (Dang and Tropepe,2006; Wilson and Stice,2006; Zhang,2006). We will focus here on the early steps of ESC neurogenesis, the transition from ESC to NSC. Multiple protocols have been established that produce ESC-derived NSCs. We will examine these protocols and compare their strengths and weaknesses as approaches for generating material for transplant. Several criteria will be considered, including the efficiency and yield of NSCs, the purity of the final population, the ability of the NSCs to produce all subtypes of neurons and glia upon subsequent induction, the ease and reproducibility of the approach, and the time required to go from ESC to NSC. We will also note whether the approach has been applied to human ESCs.

These protocols are based upon the mechanisms governing both neural induction during embryogenesis and neurogenesis in the adult CNS. Radial glia cells are the NSCs of the embryonic CNS, and astrocyte-like stem cells the NSCs of the adult brain (Doetsch,2003). Recent studies suggest that ESC-derived NSCs can have the properties of embryonic radial glia (Liour and Yu,2003; Plachta et al.,2004; Liour et al.,2006; Nat et al.,2007) and are able to proliferate and generate all types of neural derivatives, neurons, astrocytes, and oligodendrocytes. What are the similarities and differences between ESC-derived NSCs and those derived from the embryonic or adult CNS? In a direct comparison, mouse ESC-derived NSCs proliferate more readily than mouse somatic NSCs and prefer adherent conditions over suspension culture conditions (Colombo et al.,2006). The ESC-derived NSCs also differentiated into neural cell types with a higher efficiency than somatic NSCs and were able to produce more cell types, demonstrating a broader potential (Colombo et al.,2006). Transcriptional profiling demonstrated a close similarity between NSCs of both origins, but also showed that ES-derived NSCs displayed positional molecular information that indicated a broader potential than somatic NSCs, specifically including rostral spinal cord-specific markers (Colombo et al.,2006). In contrast, a similar gene profiling analysis comparing human ESC-derived NSCs with fetally derived NSCs led to the conclusion that there was limited similarity between the two cell types (Shin et al.,2007).

NSCs and progenitors derived from the embryonic or adult CNS have restricted potential that reflects their site of origin along the rostral/caudal and dorsal/ventral axes and are characterized by the expression of region-specific transcription factors (Temple,2001; Zhang,2006). On the other hand, ESC neurogenesis mimics embryonic neurogenesis by first producing multipotent NSCs that express Sox2, Sox3, Otx2, and Pax6, which characterize anterior neurectoderm identity, that can subsequently differentiate into neural progenitors with more restricted identity based upon the expression of region-specific transcription factors (Wilson and Edlund,2001; Barberi et al.,2003; Xia et al.,2007). Highly enriched populations of motor neurons can be induced by treatment of early stage anterior neurectoderm-like progenitors, but not later stage progenitors with regional identity, with appropriate signaling molecules (Zhang et al.,2001). Addition of caudalizing signals, such as retinoic acid (RA) or FGF2 can promote a spinal cord identity and subsequent production of motor neurons (Wichterle et al.,2002), whereas addition of Wnt and Nodal antagonists promotes production of telencephalic progenitors (Watanabe et al.,2005). Treatment with FGF8 and Sonic Hedgehog promote production of ventral midbrain precursors and subsequent differentiation of dopaminergic neurons (Lee et al.,2000; Barberi et al.,2003).

In addition to comparing available protocols for NSC production, we will also identify the molecular signals required for differentiation, survival, and proliferation ESC-derived NSCs. First, we will describe the basic mechanisms of neural induction and how these principals are applied in the various protocols used to generate neurons and glia from ESCs. We will then summarize more recent studies that have used ESC neurogenesis to identify the conditions and signals required to promote the production of NSCs from ESCs. Finally, we will integrate the conclusions from these studies to synthesize a unified model for NSC production.


Two basic approaches have been identified: (1) mimicking the environment that produces neurectoderm in the embryo by providing appropriate cell–cell interactions and signals through embryoid body formation, or (2) depriving the ESCs of both cell–cell interactions and signals by low density culture in serum-free medium, evoking a default mechanism for NSC differentiation. Some protocols combine aspects of both of these two approaches, promoting cell–cell interactions to facilitate formation of all three primary germ layers followed by neural lineage-specific selection under defined conditions. For both mouse and human ESCs, the quality of the starting ESC population is critical for the successful production of neural derivatives. As is the case for many in vitro protocols, cell populations at any stage may be heterogeneous and additional purification steps, such as fluorescence-activated cell sorting (FACS), may be required to obtain the desired cell type. Table 1 documents the properties of the six protocols described below.

Table 1. Comparison of Protocols Using ESCs to Derive NSCsa
Protocols 1. RA induction (4−/4+) 2. MEDII CM induction 3. Serum-free selection 4. Stromal co-culture 5. Low density clonal neurosphere 6. Monolayer Serum-free
Mouse ESC lines used D3, CCE E14, D3 J1, CJ7, D3, R1 ESC: CJ7, AB2.2,E14,ESB5 ntES:C4, C15,C16,CN1,2, CT2 R1 46C (E14 derived) 15 clones of ESC
Coculture, CM, or other factors None HepG2 conditioned medium None Stromal cell (MS5,S17,PA6 etc) LIF none
Initial plating density Not quantified >1×105 cells/ml 1×105 cells/ml 2-2.5×104cells/cm2 50 cells/cm2 1-20 cells/microwell 0.5-1.5x104 cells/cm2
Culture type Suspension Suspension Suspension + adherent Adherent Suspension Adherent
EB formation Yes Modified Yes No No No
Serum or serum replacement 10% FBS+ 10% newborn calf serum 10% FBS 10% FBS for EB then no serum 15% serum replacement No serum No serum
Days to reach NSC peak 8 7 10-12 6 3 (4 hrs in PBS) 5
% NSC at peak 39% neuron-like Cellsb Nearly 100% 95.7% NCAM+ >80% High, not quantified 100% 75%
NSC marker βIII tubulin Sox1, Sox2, nestin nestin nestin, NCAM, Musashi nestin Sox1, nestin
NSC Regional identity NA Otx1 (fore- and midbrain) En1, En2 (midbrain) Otx1 (fore- and midbrain) En1 (midbrain) No specific regional identity Emx2 (forebrain) HoxB1 (hindbrain) NA
Differentiation potential of derived NSC NA Neurons, glia (>95%), neural crest Neurons (Map2) Astrocytes (GFAP) Oligodentrocytes (O4) Neurons (dopaminergic, serotonin, GABAergic and motor neurons with high efficiency), glia Neurons (Map2) Astrocytes (GFAP) Oligodendrocytes (O4) Neurons (GABA, TH) Astrocytes (GFAP) Oligodendrocytes (CNPase)
Other lineages Many other lineages present None Non-neural lineages selected against in serum-free medium None primitive endoderm present (GATA4), no mesoderm or definitive endoderm Non-neural cell types and Oct4+ cells present
Key references: Bain et al.,1995 Bibel et al.,2004 Rathjen et al.,2002 Okabe et al.1996 Barberi et al.,2003 Kawasaki et al.,2000 Tropepe et al.,2001 Smukler et al.,2006 Ying et al.,2003
  • a ESCs, embryonic stem cells; NSCs, neural stem cells; RA, retinoic acid; CM, conditioned medium; NA, not applicable; GFAP, glial fibrillary acidic protein; FBS, fetal bovine serum; NCAM, nerve cell adhesion molecule; En1-2, Engrailed 1-2; MAP2, microtubule associated protein 2; GABA, γ-aminobutyric acid.
  • b Stem cells or progenitors not assayed.

Mimic the Embryo Environment: Embryoid Body Intermediates Make Neurectoderm

Initial studies demonstrating the generation of neural cell types from mouse ESCs were based upon first forming embryoid body intermediates. ESC-derived embryoid bodies are similar to embryoid bodies derived from teratocarcinoma stem cells, first described by Stevens and Pierce (Pierce and Dixon,1959; Stevens,1959) and subsequently demonstrated to undergo a similar pattern of differentiation to isolated inner cell masses cultured in vitro (Martin et al.,1977).

When mouse ESCs are removed from their feeder layer, and placed in suspension culture in the absence of the growth factor LIF, which promotes maintenance of the pluripotential state (Smith et al.,1988), they form aggregates, which within 2–4 days consist of an outer layer of hypoblast-like cells (extraembryonic visceral endoderm) surrounding an epiblast-like core (primitive ectoderm; Martin et al.,1977; Rathjen et al.,2002; Maye et al.,2004). At this stage, the embryoid body resembles the anterior prestreak stage embryo, with the epiblast-like core able to generate derivatives of all three primary germ layers; definitive endoderm, mesoderm, and ectoderm (Keller,2005). The embryoid body core continues to express the ESC marker Oct4 and begins expressing the primitive ectoderm marker FGF5 (Hebert et al.,1991; Rathjen et al.,2002; Maye et al.,2004). Between days 6 and 8, the core undergoes cavitation and forms an inner epithelial layer (Coucouvanis and Martin,1995). Cells within this layer can be committed to definitive ectoderm, characterized by Sox2 and Otx2 expression (Rathjen et al.,2002; Maye et al.,2004). Subsequently, a morphological transformation to columnar epithelium, resembling the neural tube, is accompanied by the expression of neurectoderm-specific markers such as Sox1 and Six3 (Rathjen et al.,2002; Maye et al.,2004).

Conditions that direct the differentiation of the embryoid core toward neurectoderm, at the expense of mesoderm or definitive endoderm derivatives, increase the pool of cells that can subsequently differentiate into neurons or glia. In the embryo, signals from a region of the anterior extraembryonic visceral endoderm (AVE) are known to promote anterior neurectoderm differentiation and the visceral endoderm may serve a similar function in the embryoid body (Thomas and Beddington,1996). More posterior in the embryo, the notochord supplies signals that pattern the developing neural tube, but no comparable structure is present in embryoid bodies (Stern,2005). Work from our laboratory suggests that the signaling molecule Indian Hedgehog, secreted by the visceral endoderm, promotes neurectoderm differentiation in vitro. Embryoid bodies derived from ESCs carrying loss of function mutations for components of the Hedgehog signaling pathway are able to generate primitive ectoderm, but not neurectoderm or neurons (Maye et al.,2004). The Hedgehog signaling cascade is best known for its role in ventral patterning (Briscoe et al.,2000; Wichterle et al.,2002); however, several recent studies support an additional role in NSC survival (Britto et al.,2000; Machold et al.,2003; Cayuso et al.,2006) and proliferation (Machold et al.,2003; Ahn and Joyner,2005; Palma et al.,2005) in the embryonic and adult mouse brain.

Retinoic acid protocol.

The yield of neural lineage cells generated by embryoid bodies can be dramatically increased by the addition of RA (Guan et al.,2001; Gottlieb,2002; Table 1; Fig. 1, protocol 1). RA has a well-established patterning role during development, for example in posteriorizing CNS tissue (Durston et al.,1989; Li et al.,2005) and ESC derivatives (Zhang,2006). Numerous studies also support a role for RA in embryonic neurogenesis in particular regions at specific times, and a recent study examining the effect of RA deficiency on granule cell differentiation suggests that RA is required for adult hippocampal neurogenesis (Jacobs et al.,2006). RA promotes the differentiation of pluripotential teratocarcinoma cells into neural progenitors and neurons (Jones-Villeneuve et al.,1983; Bain and Gottlieb,1994), and this treatment was applied to ESCs to promote neural induction. ESC embryoid bodies cultured for 4 days were treated for an additional 4 days with RA (4−/4+) and then attached to adhesive substrates for 7 days, and the extent of neuronal differentiation was determined based upon cell morphology, expression of cell type specific markers, and electrophysiological measurements (Bain et al.,1995). Although no markers for NSCs were assayed, there was a yield of up to 40% neuron-like cells, vs. just a few percent observed in the cultures not treated with RA. Related protocols increased the yield of neural derivatives, including glutamatergic neurons and motor neurons, with RA treatment (Wichterle et al.,2002; Bibel et al.,2004; Li et al.,2005). Subsequent studies from other groups have established that the NSCs generated using the RA protocol may have restricted differentiation potential. For example, when neural progenitors generated by RA induction were transplanted to the embryonic chick neural tube, they showed limited capacity to differentiate, compared with ESC-derived neural progenitors generated in the absence of RA (Plachta et al,2004). Additional drawbacks of this approach are the presence of numerous cell types in the final product, including mesoderm derivatives, and the length of culture time required, up to 2 weeks. This approach has been applied to hESCs (Schuldiner et al.,2001).

Details are in the caption following the image

Protocols for producing embryonic stem cell (ESC) -derived neural stem cells (NSCs). Protocols 1–3 include an embryoid body intermediate, whereas 4–6 are embryoid body-independent. Protocols 1–5 begin with feeder-dependent ESCs, whereas protocol 6 is most efficient when applied to feeder-free lines.

MEDII conditioned medium induction.

Studies from the Rathjen laboratory have demonstrated that conditioned medium from a human hepataocellular carcinoma cell line Hep-G2 can promote the homogeneous differentiation of primitive ectoderm-like (EPL) cells from mouse ESCs (Rathjen et al.,1999). Embryoid bodies formed from these cells do not have an outer visceral endoderm layer, but the remaining cells are able to form an essentially pure neurectoderm-like epithelial layer. Most cells at this stage expressed the NSC markers Sox1 (Pevny and Lovell-Badge,1997) and nestin (Hockfield and McKay,1985; Table 1; Fig. 1, protocol 2). The conditioned medium appears to contain signals that are capable of inducing neurectoderm differentiation at the expense of mesodermal and endodermal derivatives. These NSCs do not appear to have a restricted regional identity and, when subjected to different signaling factors, can be further induced to generate neurons or glia, as well as neural crest (Rathjen et al.,2002). However, the molecules in the Hep-G2–conditioned medium responsible for this activity have not been identified. This protocol has not yet been applied to human ESCs.

Embryoid body selection in defined medium.

This protocol has been used extensively by many groups and actually combines use of an embryoid body intermediate, in which cells of all three primary germ layers arise, followed by a neural lineage-specific selection step (Okabe et al.,1996; Guan et al.,2001; Wobus et al.,2001). This approach begins with early stage embryoid bodies. These are then plated on adhesive substrates in a minimal serum-free medium, ITSFn, which contains insulin, transferrin, selenium, and fibronectin (Table 1; Fig. 1, protocol 3). During several days of culture, non-neurectoderm cells, including undifferentiated ESCs, extraembryonic endoderm and mesoderm derivatives, die, leaving a neurectoderm-derived neural progenitor population, based upon expression of nestin, Sox1, Sox2, and other NSC markers (Okabe et al.,1996; Carpentino and Grabel, unpublished observations). Subsequent plating on laminin substrates in the presence of FGF2 promotes NSC proliferation, and at this stage cultures may be 80–95% NSCs (Okabe et al.,1996). NSCs generated using this approach have been used to produce a variety of specific neural and glial derivatives, including oligodendrocytes (Brustle et al.,1999), dopaminergic neurons, and glutamatergic neurons (Lee et al.,2000; Bibel et al.,2004). The strength of this method is the relative purity of the final product and the ability to generate large numbers of NSCs upon FGF2 addition, whereas weaknesses include the extended time it takes to generate NSCs, 10 days to two weeks, and the variability in the quality of the final product, which can be contaminated with mesodermal derivatives, likely due to the multiple steps. An accelerated version of this approach, which combines the selection and proliferation steps, produces a similar end product (Colombo et al.,2006). The embryoid body selection protocol has also been applied to the derivation of neurons from human as well as mouse ESCs (Carpenter et al.,2001; Zhang et al.,2001).

The Default Pathway

Based on work in nonmammalian species, particularly Xenopus and chick, a default hypothesis emerged for neural specification, proposing that the absence of signals mediates the decision to become neural vs. non-neural. Several studies support this hypothesis for ESC neural differentiation as well, but as observed for Xenopus and chick, the story appears to be more complicated than a simple default mechanism (Stern,2005). Positive factors are also required, and once the neural transition is achieved, signals are necessary to promote the survival and proliferation of these committed cells. In addition, there appear to be multiple intermediate stages between ESC and NSC, based upon both the expression of stage-specific markers and the distinct response of cells at different stages to extracellular signals.

The experiments of Hilde Mangold and Hans Spemann identified a region of the gastrula stage amphibian embryo, the dorsal lip of the blastopore, with the ability to induce ectopic neural tissue in cells otherwise fated to become epidermis (Spemann and Mangold,1924). Similar organizer-like regions were subsequently identified in many other species: Hensen's node in chick (Waddington,1936), the shield in teleosts (Luther,1935; Oppenheimer,1936), and the node and AVE in the mouse (Waddington,1936; Beddington and Robertson,1998). The identification of the dorsal lip of the blastopore led to several decades of unsuccessful attempts to establish the molecular nature of the induction signal. Finally, in the 1990s, a story began to emerge identifying several inhibitors of transforming growth factor-beta (TGF-β) family members, particularly BMP4, as neural inducers. The spatial and temporal pattern of expression of these inhibitors, including chordin, noggin, and follistatin, was also consistent with their role in neural induction (Smith and Harland,1992; Sasai et al.,1994,1995). These data suggested that simply antagonizing BMP signaling could promote neural specification, and several lines of experimentation supported this notion, including the observation that isolated animal caps dissociated to single cells to minimize signaling events, could form neural tissue after re-aggregation (Born et al.,1989; Godsave and Slack,1989; Sato and Sargent,1989). The story is not that simple, however, with additional factors now determined to play positive, as well as negative, roles in promoting neural fate. Several lines of investigation have established a positive role for FGFs in neural induction, both by down-regulating BMP targets, and by a BMP-independent mechanism (De Robertis and Kuroda,2004; Linker and Stern,2004). Also, several studies demonstrated an inhibitory role for Wnt signaling and a positive inductive role for antagonists of this pathway (Wilson et al.,2001). In contrast, some studies suggest Wnt signaling promotes neural induction (Baker et al.,1999). Whether this pathway acts to inhibit or promote neural differentiation depends upon the time of exposure to the signal and concentration of the signal present.

Given the evidence for a default pathway of neural induction, in which the absence of signals promotes this lineage specification, several approaches for promoting neural differentiation of ESCs include a selection step of culture in a serum-free, nutrient-poor neurobasal medium. These media were initially developed for culturing pure populations of NSCs derived from embryonic or adult neurogenic brain regions (Reynolds and Weiss,1996). These conditions are successful at supporting enrichment for NSCs by promoting the selective apoptosis of non-neural derivatives and excluding inhibitory factors, such as BMPs, present in serum. In particular, culture at low density under LIF-free, serum-free conditions promotes apoptosis of Oct4-positive ESCs (Tropepe et al.,2001; Barberi et al.,2003; Yamane et al.,2005). NSCs are more resistant to these conditions, perhaps due to autocrine signals.

Direct differentiation: culture at very low density in defined medium.

A series of studies from the van der Kooy group have demonstrated the transition of ESCs to NSCs at very low clonal densities (1–20 cells/well) under feeder-free conditions in serum-free defined medium (Tropepe et al.,2001; Table 1; Fig. 1, protocol 5). This differentiation occurs directly without embryoid body formation and includes the differentiation of an intermediate cell type, the primitive NSC. These studies use a neurosphere-based suspension assay for production of NSCs. ESCs first generate LIF-dependent primitive neurospheres that contain nestin-positive primitive NSCs that continue to express Oct4. These mature into LIF-independent, FGF/endothelial growth factor (EGF) -dependent neurospheres, that contain nestin-positive, Oct4-negative definitive NSCs. LIF is required at early stages of primitive NSC differentiation, most likely for survival of differentiating ESCs and to prevent neural differentiation. In addition, primitive neural stem cell differentiation is inhibited by BMP4 and promoted by the BMP antagonist noggin (Smukler et al.,2006).

The best evidence for a default mechanism of NSC differentiation comes from the observation that 80% of ESCs plated at low cell densities in phosphate buffered saline (PBS) begin to express nestin and Sox-1 within 4 hours (Smukler et al.,2006). Most of these cells, however, die due to the absence of conditions to support their survival and proliferation. This protocol, therefore, provides an excellent system to assay putative survival and proliferation factors. Forebrain and hindbrain markers have been observed in neuronal derivatives of these NSCs and neurons, astrocytes, and oligodendrocytes have been generated from these neurospheres (Tropepe et al.,2001). The differentiation of primitive NSCs from ESCs, however, is a rare event under these conditions, with only 0.2% of ESCs plated going on to form sphere-forming colonies in defined medium. This result suggests that this method is not an efficient means of producing NSCs, or their derivatives.

Direct differentiation: feeder-independent ESCs cultured at moderate density.

Direct differentiation of ESCs to NSCs can also be obtained under serum-free conditions in monolayer culture at moderate cell densities, but is most effective using ESCs adapted to growth under feeder-free conditions (Table 1; Fig. 1, protocol 6, Fig. 2B). This differentiation protocol was established by the Smith laboratory using the mouse ESC line 46C, which expresses green fluorescent protein (GFP) under the control of the promoter for the neurectoderm and NSC marker gene Sox1, providing an easy readout for the NSC transition in live cultures. Cells are plated at around 1 × 104/cm2 on gelatin-coated dishes in the absence of LIF in serum-free defined medium (N2B27). By 5 days, up to 75% of the cells express GFP as well as nestin. The transition to NSCs is not homogeneous and is accompanied by the death of significant numbers of GFP-negative cells in the cultures, suggesting that these conditions are effective at NSC production, at least in part due to selective survival of these cells. Surviving GFP-positive cells form rosettes in which cells elongate and align radially, in a manner reminiscent of neural tube formation, as well as ectoderm and neurectoderm differentiation in ESC embryoid bodies.

Details are in the caption following the image

Growth factors and signaling molecules regulate the multistep transition from embryonic stem cell (ESC) to neural stem cell (NSC). A: Oct4-positive ESCs first differentiate into primitive ectoderm-like cells (prim ecto), which express fibroblast growth factor-5 (FGF5) as well as Oct4. These cells then differentiate into primitive NSCs (prim NSC), characterized by the expression of moderate levels of nestin and Sox1, and a rounded morphology. These cells differentiate into NSCs with an elongated-radial glia-like morphology that express high levels of nestin and Sox1. LIF, Wnt, bone morphogenetic protein (BMP), Notch, FGFs, Hedgehog (Hh), and vascular endothelial growth factor-A (VEGFA) influence this process as shown. B: Images of ESC, prim ecto, prim NSC, and NSCs generated using the direct defined medium monolayer culture approach (protocol 6) showing expression of FGF5 in prim ecto, Sox1-GFP (green fluorescent protein) in prim NSCs and NSCs. Note the elongated morphology of NSCs.

BMP signaling appears to inhibit, while FGFs support NSC production in this system. Addition of BMP4 inhibits the transition to Sox1-GFP-positive NSCs, whereas treatment with the BMP antagonist noggin promotes it. Several different neural derivatives have been generated using this method. The advantages of this protocol include the relative rapid direct emergence of NSCs (4–6 days), and the ability to isolate a pure NSC population using FACS (Chung et al.,2006; Fukuda et al.,2006). In addition, in 5 days, 10,000 ESCs can produce 1–5 million cells, 70% of which are NSCs, making this an efficient means of NSC production. Disadvantages include the presence of other cell types and reliance upon a feeder-independent ESC line for maximal efficiency of NSC production, but the purity of the NSC population can be improved through trypsin-based passage (Conti et al.,2005). The NSC lines derived in this manner retain pluripotency and are able to differentiate into neurons, astocytes, and oligodendrocytes (Glaser et al.,2007). This direct differentiation protocol has been adapted for use with human NSCs (Zhang et al.,2001).

Stromal cell co-culture.

The concept of the stem cell niche is based largely on studies of hematopoietic stem cells indicating that their survival, proliferation, and differentiation rely upon interaction with bone marrow stromal cells (Kaushansky,2006). Surprisingly, these cells also have the ability to direct the differentiation of NSCs from both mouse and human ESCs. ESCs are plated at clonal densities on stromal cell lines (Table 1; Fig. 1, protocol 4). A variety of different stromal lines have proven effective. By day 6, NSC differentiation is apparent, based on expression of a variety of markers, including neural cell adhesion (NCAM) molecule and nestin (Kawasaki et al.,2000; Barberi et al.,2003). It is key to the success of this co-culture approach that the cells be plated at low density to minimize cell–cell interactions and that serum be removed from the medium and serum replacement used in its place, suggesting that serum contains an inhibitor of the NSC transition, a conclusion supported by the observation that BMP4 suppresses neural differentiation under these conditions. The identity of the factor, or factors, provided by the stromal cells, be it permissive or instructive, has not yet been elucidated, but studies suggest that both a soluble factor and direct contact between ESCs and stromal cells may be required, because fixed feeder cells retain neural inducing activity (Kawasaki et al.,2000). Advantages of this protocol include the relatively pure population of multipotent neural derivatives that can be induced to produce neural subtypes with high efficiency, and the absence of an embryoid body intermediate, allowing generation of NSCs in a few days. NSCs generated using this approach have been used to produce many neural derivatives, including dopaminergic and γ-aminobutyric (GABA) acid-ergic neurons (Kawasaki et al.,2000; Barberi et al.,2003) by treatment with region-specific signals.


Despite the distinct approaches used to generate NSCs from ESCs (Fig. 1), some common threads emerge from this work. The transition involves multiple stages that roughly mimic the steps identified during embryogenesis for neural specification: primitive ectoderm, definitive ectoderm, neurectoderm. Cells comparable to these stages are generated by ESCs in culture, using embryoid body-dependent or -independent methods (Fig. 2A). Mouse ESCs are comparable to the late inner cell mass cells, whereas human ESCs may represent the epiblast stage (Brons et al.,2007; Tesar et al.,2007). On the path toward neural specification, mouse ESCs differentiate into primitive ectoderm-like cells, capable of producing ectoderm, mesoderm, and endoderm derivatives, whereas human ESCs are already at this stage. Mouse ESCs at this stage still express Oct4 and begin to express FGF5 (Fig. 2). Next, the cells become restricted to definitive ectoderm and subsequently neurectoderm fate. This transition is marked by the down-regulation of Oct4 and FGF5 while neurectoderm markers, including Otx2 (Ang et al.,1994) and Sox1, are up-regulated (Fig. 2). The neurectoderm-like cells are comparable to the primitive NSCs identified by van der Kooy (Tropepe et al.,2001), expressing intermediate levels of Sox1 and nestin, and still reliant upon LIF for survival. These cells mature into NSCs and this transition is accompanied by a morphological change, as cells elongate into a radial glia-like phenotype, forming rosettes in monolayer. These NSCs express high levels of nestin and Sox1, no longer require LIF, and are responsive to FGF2 or EGF. Some evidence suggests further maturation into Sox1-negative, nestin-positive, brain lipid-binding protein (BLBP)–positive radial glia-like cells (Lowell et al.,2006).

The existence of intermediate cell types complicates the elucidation of the functional role played by signaling molecules in the production of ESC-derived NSCs. Factors may influence more than a single cell type and some evidence suggests that the same molecule may have distinct roles at different stages. It is clear that serum serves as a source of inductive and inhibitory factors, but even in the defined medium, low density-based protocols, endogenously produced signaling centers are established (Ying and Smith,2003) and autocrine signaling can play a role in neural induction (Fig. 2A; Tropepe et al.,2001; Ying and Smith,2003; Smukler et al.,2006).

BMPs Inhibit Neural Specification

In support of a default mechanism, studies using several different protocols have established that BMP4 inhibits NSC differentiation in vitro (Kawasaki et al.,2000; Tropepe et al.,2001; Ying et al.,2003; Gossrau et al.,2007; Xia et al.,2007). Addition of BMP antagonists, such as noggin (Tropepe et al.,2001), or mutations in a downstream effector, Smad4, promote ESC neurogenesis under low density, defined medium conditions. In addition, increased levels of Nodal, another TGF-β family member, inhibit, whereas the Nodal antagonist LeftyA promotes neural differentiation (Vallier et al.,2004; Watanabe et al.,2005). ESC lines carrying mutations for either Smad4 or the Nodal co-receptor Cripto show enhanced neural differentiation using the embryoid body-based defined medium protocol. However, under these conditions, both mutants continue to produce other derivatives as well, including mesoderm, suggesting that TGF-β inhibition alone is insufficient to block differentiation of other lineages from ESCs (Minchiotti,2005; Sonntag et al.,2005). The observation that BMP4 treatment results in a switch from a neural to an epidermal fate suggests that BMP acts specifically upon the primitive ectoderm-like cell to alter its differentiation (Kawasaki et al.,2000; Ying et al.,2003). These results suggest that removing the inhibitory effects of BMP can promote neural differentiation, but they do not eliminate a role for positive factors in NSC production.

Wnts Can Inhibit or Promote Neural Specification

As established for neural induction during Xenopus or chick embryogenesis, neural differentiation of ESCs can either be inhibited or promoted by Wnt signaling (Fig. 2A). Activation of the Wnt cascade by expression of Wnt-1 or a dominant active form of β-catenin, inactivating APC, or addition of lithium chloride, which inhibits GSK-3 and thereby β-catenin degradation, inhibit neural differentiation, whereas treatment with the Wnt antagonists SFRP2 and Dkk1, stimulates production of ESC-derived neural progenitors (Aubert et al.,2002; Haegele et al.,2003; Watanabe et al.,2005). In contrast, Otero et al. (2004) report that overexpression of Wnt3a or expression of a dominant-negative form of E-cadherin that increases intracellular levels of β-catenin, promotes neural differentiation under high density conditions that ordinarily block this lineage decision. The effect of Wnt signaling perturbation on ESC neural differentiation likely depends on the specific conditions used to induce neural differentiation, the means of perturbing the Wnt pathway, and the precise stage at which the intervention takes place. In addition, Wnt signals may activate several different downstream cascades, some independent of β-catenin, and one cascade may promote, while another inhibits ESC neural differentiation (Montcouquiol et al.,2006). Interestingly, despite Wnts' well-established role as a mitogen, its effect on ESC neurogenesis appears to be on neural differentiation and not on progenitor proliferation (Aubert et al.,2002; Otero et al.,2004).

Notch for Initial Differentiation

Studies in Drosophila established a role for Notch signaling in neurogenesis, but as an inhibitor of neural specification in favor of epidermal differentiation (Artavanis-Tsakonas et al.,1999). In the vertebrate CNS, Notch maintains NSCs in an undifferentiated, proliferative state through the action of Hes genes (Ross et al.,2003). Notch- or Hes-deficient mice show premature neural differentiation (Yoon and Gaiano,2005). In addition, Notch can play a positive role in neural differentiation, promoting production of prosensory patches in the chick inner ear (Daudet and Lewis,2005). Recent evidence suggests that Notch promotes neural lineage specification of ESCs. Activation of the pathway in ESCs cultured in neurobasal medium, using expression of the constitutively active Notch intracellular domain (NotchIC), promotes NSC differentiation, and the transition from primitive ectoderm to neurectoderm (Lowell et al.,2006). Treatment with the γ-secretase inhibitor L-685,458, which acts by blocking the cleavage of Notch to its active intracellular domain, inhibits NSC differentiation. These effects of Notch pathway perturbation are consistently observed using the stromal co-culture or embryoid body-based defined medium protocols for mouse ESCs and also for human ESCs. Careful analysis suggests that Notch signaling helps to homogenize the expression of Sox1 within the population by a community effect mechanism, while inhibiting differentiation of other cell types (Lowell et al.,2006). This conclusion is consistent with the observation that mice deficient for the Notch effector gene RBPJk have substantially thinner neurepithelium than wild-type mice, although this phenotype could be attributed to altered proliferation levels as well (Oka et al.,1995).

Hedgehog's Role in Proliferation of Primitive Ectoderm-Like Intermediate and Survival of NSCs

In addition to its well-established role in conferring ventral identity to neural tube derivatives, the Hedgehog signal has been implicated in differentiation, survival, and proliferation of many cell types in the embryonic nervous system. Studies also indicate that Sonic Hedgehog promotes both the survival and proliferation of NSCs and neuroblasts during adult neurogenesis (Machold et al.,2003; Ahn and Joyner,2005; Palma et al.,2005). Our initial studies, using ESCs carrying mutations in Hedgehog signaling components, suggested that this signal is critical for the differentiation of primitive ectoderm to neurectoderm (Maye et al.,2004). Our more recent analysis, using Hedgehog antagonists and the Sox1-GFP reporter system, has established two roles for Hedgehog signaling in ESC neurogenesis (Fig. 2A):(1) promoting proliferation of a Sox1-GFP-negative, FGF5-positive primitive ectoderm precursor; and (2) supporting survival of the Sox1-GFP-positive NSC (Cai and Grabel, unpublished data). Of interest, the source of the Hedgehog signal, based upon reverse transcription-polymerase chain reaction (RT-PCR) analysis of FACS-sorted Sox1-GFP-negative and -positive populations, appears to be the differentiating NSCs themselves. This finding, although surprising, is consistent with the observation that NSCs isolated from the neonatal mouse subventricular zone also secrete Sonic Hedgehog (Shh; Rafuse et al.,2005). These Shh-secreting NSCs support the survival of dopaminergic neurons when co-cultured with ventral midbrain tissue (Rafuse et al.,2005) and when transplanted into a mouse model of Parkinson's disease (Ostenfeld et al.,1999). Shh may also provide a survival advantage to NSCs over ESCs under serum-free, low density conditions.

FGFs for Differentiation and Proliferation

FGF2 is a well-established mitogen for CNS-derived embryonic and adult NSCs. Along with EGF, FGF2 addition promotes the production and passage of CNS-derived neurospheres (Reynolds and Weiss,1996). FGF2 also plays a role in supporting proliferation of ES-derived NSCs once established (Okabe et al.,1996; Conti et al.,2005). FGF2 also promotes proliferation of undifferentiated human ESCs (Xu et al.,2005), consistent perhaps with their epiblast origin (Brons et al.,2007; Tesar et al.,2007). A role for FGF4 in the production of ESC-derived NSCs has also been demonstrated, although the specific mechanism used is under debate. Studies suggest that FGF4 produced by undifferentiated mouse ESCs acts in an autocrine manner to promote neural specification (Fig. 2A). This conclusion is supported by the observation that FGF-receptor 1-deficient ESCs produce significantly fewer neural colonies using the low density neurosphere protocol (Fig. 1, protocol 5), an outcome also observed with addition of a function-blocking FGF antibody to wild-type cultures (Tropepe et al.,2001). Using the direct defined medium protocol (Fig. 1, protocol 6), treatment with an inhibitor of FGF receptor kinases, SU5402, or expression of a dominant-negative FGF receptor inhibits production of nestin-positive NSCs without effecting their growth or survival, whereas addition of FGF4 increases the frequency of NSCs present in the cultures (Ying and Smith,2003). SU5402 also inhibits the increased level of NSC production observed with overexpression of NotchIC, suggesting that FGF4 may act downstream of Notch to promote neural specification (Smukler et al.,2006). These experiments are consistent with a role for Notch in the ESC differentiation. However, the nestin expression observed under the extreme conditions of ESC culture in PBS was not inhibited by SU5402 treatment, or by FGF-receptor-1 deficiency, suggesting that FGF signaling supports survival or proliferation of cells already committed to a neural fate, rather than their actual differentiation (Smukler et al.,2006).

In the mouse embryo, FGF5 is transiently expressed by primitive ectoderm before gastrulation, and confers competence to respond to subsequent inductive signals (Hebert et al.,1991). We, and others, have shown that FGF5 is also transiently expressed upon primitive ectoderm differentiation of mouse ESCs (Rathjen et al.,1999; Maye et al.,2004). Of interest, the new isolated mouse EpiSCs express FGF5 (Brons et al.,2007), consistent with their epiblast origin. Our observation that ESCs gain the competence to respond to a Hedgehog antagonist only after they begin to express FGF5, suggests this growth factor may mediate the transition to primitive ectoderm, which can then respond to additional inductive signals mediating neurectoderm differentiation.

Vascular Endothelial Growth Factor, Opposite Effects on Primitive NSCs and NSCs

Vascular endothelial growth factor (VEGF) has a well-established role in angiogenesis and hematopoiesis and has recently been implicated in neural development (Shalaby et al.,1995; Jin et al.,2002; Fabel et al.,2003; Sun et al.,2006). Undifferentiatend ESCs express mRNA for VEGF-A and its receptor Flk1. Using low density neurosphere culture to generate ESC-derived primitive and definitive NSCs (Fig. 1, protocol 5), VEGF-A appears to inhibit the production of primitive-NSCs, but promote the production of definitive NSCs (Fig. 2A; Wada et al.,2006). ESCs deficient for the VEGF receptor Flk1 produced significantly more primitive sphere colonies than wild-type ESCs, whereas addition of VEGF-A inhibits production of primitive NSCs. In contrast, Flk−/− ESCs produce significantly fewer definitive NSCs (Wada et al.,2006). Based on the levels of cell death observed in control and VEGF-A–deficient cultures, this ligand acts by inhibiting the survival of primitive NSCs and promoting the survival of definitive NSCs.


ESC neurogenesis provides a readily manipulated in vitro system that can be used to characterize the intermediate cell types involved, identify the active growth factors and signaling molecules, and decipher how these factors modulate NSC production. Studies performed thus far suggest that Wnts and BMPs inhibit ESC neural induction by inhibiting the differentiation of primitive ectoderm and primitive NSCs. LIF supports this transition, likely by supporting ESC and primitive ectoderm survival. Some data suggest that Notch and FGF4 promote commitment to the neural lineage, although some claim that FGF4 acts to promote proliferation of an already committed cell, perhaps the primitive NSC. A role for RA in ESC neural induction has been well established, but definitive evidence identifying its mode of action is lacking. Hedgehog signaling promotes the proliferation of the primitive ectoderm intermediate, and promotes the survival of NSCs. VEGFA inhibits the survival of the primitive NSC and promotes the survival of the mature NSC. FGF2 and EGF are mitogenic for ESC-derived NSCs. These growth factors and signals may well act at later stages to regulate differentiation, proliferation, or survival of specific subtypes of neurons or glia.

It is clear that we have just begun to understand the complex inductive interactions that regulate ESC neurogenesis, but this research is driven by a potential practical application that hopefully will be realized in the not too distant future. Addition of the appropriate factors, at the appropriate stage, will allow us to optimize the production of NSCs and their derivatives for use in transplantation therapies designed to treat neurodegenerative diseases and nervous system injury.