Rapid induction and long-term self-renewal of primitive neural precursors from human embryonic stem cells by small molecule inhibitors

Abstract

Human embryonic stem cells (hESCs) hold enormous promise for regenerative medicine. Typically, hESC-based applications would require their in vitro differentiation into a desirable homogenous cell population. A major challenge of the current hESC differentiation paradigm is the inability to effectively capture and, in the long-term, stably expand primitive lineage-specific stem/precursor cells that retain broad differentiation potential and, more importantly, developmental stage-specific differentiation propensity. Here, we report synergistic inhibition of glycogen synthase kinase 3 (GSK3), transforming growth factor β (TGF-β), and Notch signaling pathways by small molecules can efficiently convert monolayer cultured hESCs into homogenous primitive neuroepithelium within 1 wk under chemically defined condition. These primitive neuroepithelia can stably self-renew in the presence of leukemia inhibitory factor, GSK3 inhibitor (CHIR99021), and TGF-β receptor inhibitor (SB431542); retain high neurogenic potential and responsiveness to instructive neural patterning cues toward midbrain and hindbrain neuronal subtypes; and exhibit in vivo integration. Our work uniformly captures and maintains primitive neural stem cells from hESCs.

Human embryonic stem cells (hESCs) hold enormous promise for regenerative medicine (1). Typically, hESC-based applications require in vitro differentiation of hESCs into a desirable homogenous cell population. Despite the enormous progresses made in differentiating hESCs into various functional cells, a major challenge of the current hESC differentiation paradigm is the inability to effectively capture and stably expand primitive lineage-specific stem/precursor cells. These cells would ideally retain broad differentiation potentials (e.g., have the ability to serially repopulate the entire specific tissue) and, perhaps more importantly, the developmental stage-specific differentiation propensity, and would be devoid of tumorigenicity concerns. In the case of neural induction of hESCs by various advanced methods (2–5), there is still a lack of robust, chemically defined conditions for the long-term maintenance of primitive neural epithelial precursor cells, which are highly neurogenic and can be patterned/regionalized by specific morphogens (6, 7). Under typically used growth factor conditions (including bFGF, EGF), neural stem cells (NSCs) “transition” in a few passages into a more glial-restricted precursor state (8), which is significantly less neurogenic. In addition, in vitro cultured NSCs respond poorly to patterning cues and exhibit a narrow repertoire for generating specific neuronal subtypes. Previous studies in murine ESCs (mESCs) have suggested the existence of leukemia inhibitory factor (LIF)-responsive primitive NSCs (6). However, these cells could not be maintained in culture. Recent studies in neural induction of hESCs have identified rosette-type NSCs that represent neural tube-stage precursor cells. These rosette NSCs were capable of responding to patterning cues that direct differentiation toward region-specific neuronal fates, but still could not be stably maintained (4). Recently, Koch et al. reported long-term expansion of hESC-derived rosette-type NSCs (9). However, the study used the conventional and undefined embryoid body (EB) differentiation strategy and required tedious mechanical isolation of the overgrown neural rosettes from replated EBs. In addition, under these conditions, NSCs could not maintain stable spatial properties and switch from forebrain to hindbrain identity after prolonged expansion.

In our attempts to convert conventional hESCs to a mESC-like naïve state by small molecules, we fortuitously created a homogenously converted cell population by combined treatment of human LIF (hLIF) and two small molecules, CHIR99021 and SB431542, for about 10 d under chemically defined conditions. Remarkably, this population of cells, growing in colonies, appeared to self-renew and stably maintain their characteristics over numerous passages under these defined conditions. CHIR99021 (referred to hereafter as CHIR) is a small molecule inhibitor of glycogen synthase kinase 3 (GSK3) and can activate canonical Wnt signaling (10), which has been implicated in ES cell self-renewal (11). SB431542 (referred to hereafter as SB) is a small molecule inhibitor of transforming growth factor β (TGF-β) and Activin receptors, and has been implicated in the mesenchymal-to-epithelial transition and reprogramming (12, 13). Interestingly, these converted cells did not express the pluripotency markers Oct4 and Nanog, but were positive for Sox2 and alkaline phosphatase (ALP). Subsequent studies revealed that this expandable cell population has features of primitive neuroepithelium (and hereafter we refer them as primitive neural stem cells/pNSCs). Interestingly, the self-renewal of pNSCs is dependent on LIF, which has been implicated in the self-renewal of mESC-derived primitive NSCs (6, 14). Previous in vivo developmental studies have shown that bFGF-responsive definitive NSCs first appear on embryonic day 8.5 (ED 8.5) in mouse embryos (15, 16). However, at an earlier stage (ED 5.5–7.5), primitive NSCs are LIF-dependent, and the in vivo generation of primitive NSCs was independent of Notch signaling. We reasoned that if pNSCs are analogous to primitive NSCs during development, temporarily inhibiting Notch signaling should not block the induction of pNSCs. Indeed, temporal treatment by another small molecule inhibitor of γ-secretase, Compound E (referred to hereafter as C-E; ref. 17), further accelerated neural induction and generated the homogenous self-renewing pNSC population within 1 wk. Even after long-term expansion and repeated passaging in the presence of hLIF, CHIR, and SB, pNSCs retain remarkably high neurogenic propensity, broad differentiation potential, responsiveness to extrinsic morphogens for subsequent development into subtype-specific neuronal identities, and the ability to integrate in vivo. Our work uniformly captures and stably maintains primitive neural stem cells from hESCs.

Results

Synergistic Inhibition of GSK3, TGF-β, and Notch Signaling Pathways Converts hESCs into Homogenous pNSCs.

hESCs were cultured on X-ray inactivated CF-1 mouse embryonic fibroblasts (MEFs) in hESC growth media (DMEM/F12 containing 20% KSR and 10 ng/mL bFGF) or on Matrigel under feeder-free and chemically defined conditions as described (18). Primitive neuroepithelium was induced by switching from hESC growth media to neural induction media (1:1 Advanced DMEM/F-12:Neurobasal media supplemented with N2, B27, hLIF), supplemented with CHIR and SB, with or without C-E, for 7 d (a schematic representation of the differentiation process is shown in Fig. 1A). hESC differentiation was monitored by immunocytochemistry, flow cytometry and real-time PCR. As shown in images of the same visual field, the differentiated cells exhibited homogenous epithelial morphology during the entire differentiation process (Fig. S1A 1–4). Real-time PCR analysis revealed that combined treatment with hLIF, SB, and CHIR (with or without C-E) induced a rapid loss of Oct4 and Nanog expression (Fig. 1B). However, the expression of Sox2, a pluripotency marker that is also a persistent marker of NSCs (19), remained largely unchanged. Pax6, an early marker of neural induction, was significantly up-regulated after 5 d in the presence of C-E (0.1 μM), whereas its up-regulation was first detected at the sixth day in the absence of C-E treatment (Fig. 1B). Consistent with this observation, immunocytochemistry confirmed the faster induction of Pax6 protein on the sixth day in the presence of C-E, as Pax6 protein only became detectable from the seventh day onwards in the absence of C-E (Fig. S1B 7–12). In contrast, only a small fraction of cells were positive for Pax6 on day 7 when hESCs were treated with SB, C-E, and hLIF (Fig. S1C1). Similarly, no Pax6 positive cells could be detected at the same time point when hESCs were treated with CHIR, C-E, and hLIF (Fig. S1C2). These data suggest that inhibition of Notch signaling can enhance early neural induction. Interestingly, real-time PCR analysis showed that the induction of the Pax6 gene occurred in parallel with the suppression of BMP4 gene expression as well as induction of Noggin (BMP antagonist) expression (Fig. 1B), suggesting that endogenous mechanisms of BMP signaling inhibition may contribute to neural induction. Real-time PCR analysis also demonstrated that the differentiation is highly specific toward the neural lineage. Along with the induction of Pax6, epiblast-associated nonneural genes such as Brachyury, Eomes, and Sox17, were repressed synchronously with pluripotency markers Oct4 and Nanog (Fig. 1B), suggesting the presence of an intermediate cell type resembling differentiating epiblast cells before hESC neuralization. This highly directed neural induction was further confirmed by immunocytochemistry. Double staining of Oct4 and Nestin showed that Oct4 expression gradually diminished and was almost undetectable after 5 d of treatment with hLIF, SB, CHIR, and C-E, whereas Nestin-expressing cells became the predominant population, comprising ∼99% of the population on day 7 (Fig. S1B 1–6). To further quantify the efficiency of the neural induction, the expression of Oct4, Sox2 and CD133, was analyzed by flow cytometry. In development, the neural plate and neural tube exhibit CD133 (Prominin-1) immunoreactivity (20, 21). In vertebrate embryos, Sox2 is one of the earliest markers for the neural plate. During hESC differentiation, the earliest Oct4-negative, but Sox2/CD133-positive cell population would represent the primitive neuroepithelium. FACS analysis showed that more than 96% of undifferentiated hESCs were positive for both Oct4 and Sox2 (Fig. 1C). After treatment, FACS confirmed the rapid loss of Oct4 expression. Especially Oct4-positive cell number dropped substantially on day 5, when Pax6 was first induced, suggesting that day 5 was the turning point of neural induction. In addition, FACS analysis further showed that the addition of C-E induced a much more rapid loss of Oct4 expression and consequent neural conversion. At day 5, only 13% of cells were still positive for Oct4 in the presence of C-E, whereas 33.9% were positive in its absence. Despite the loss of Oct4 expression, cells persistently maintained a high level of Sox2 expression (>96%) at all time points examined during differentiation, and >97% of cells were only positive for Sox2 at day 7 with C-E treatment (Fig. 1C). In addition, FACS analysis showed that 98% of undifferentiated hESCs were positive for CD133 and that small molecule treatment initially induced the loss of CD133. However, along with the induction of Pax6 from day 5 onwards, the CD133-positive cells increased significantly, with >98% cells being CD133-positive on day 9 (Fig. 1D). These homogenously differentiated neural cells could be stably expanded on MEF feeder cells or Matrigel coating in the presence of hLIF, CHIR, and SB, and are referred to hereafter as pNSCs. In the present study, pNSCs were regularly expanded on Matrigel. Taken together, these data suggested that the combination of hLIF/CHIR/SB/C-E directs the specific induction of primitive neuroepithelium within 7 d that can long-term homogenously self-renew under hLIF/CHIR/SB conditions without the need for any cell purification. Chambers et al. (2) recently demonstrated that dual inhibition of Smad signals by Noggin and SB431542 could convert >80% hESCs to neural fate in 13 d. However, the Noggin/SB431542 condition (which also contains undefined serum products) generated heterogeneous neural populations containing cells of different developmental stages (e.g., nonpolarized neuroepithelia and polarized rosette-like structures). Most importantly, the dual Smad inhibition protocol cannot capture the NSCs and maintain their self-renewal. Our described neural induction process is much faster, more specific, and more efficient, representing a chemically defined single-step strategy for obtaining self-renewing homogenous primitive NSCs from hESCs cultured in a monolayer. The results of our strategy are highly reproducible in multiple different hESC lines, including H1 (Fig. 1), HUES9, and HUES1 (Fig. S2A), under both feeder and feeder-free (Matrigel) culture conditions.

Fig. 1.

Real-time PCR and flow cytometry analysis of neural induction from hESCs treated with LIF, CHIR, and SB (with or without C-E). (A) Schematic representation of the neural induction process. (B) The expression of Pax6, Sox2, Nanog, Oct4, BMP4, Noggin, Eomes, Brachyury, and Sox17 was analyzed by real-time PCR. (C and D) Flow cytometry analysis was used to quantify cells expressing Oct4, Sox2, or CD133 during neural induction. CDM, chemically defined medium.

pNSCs Can Long-Term Self-Renew and Represent the Pre-Rosette Stage NSCs.

pNSCs can long-term self-renew over serial passages on Matrigel with SB, CHIR, and hLIF. pNSCs generated from HUES9 and H1 hESCs were routinely passaged 1:10 and have been cultured for >30 passages without obviously losing proliferative capacity, which is equivalent to at least 94 population doublings. However, individual omission of hLIF, SB, or CHIR from the media compromised pNSCs’ long-term self-renewal. Single pNSC is clonogenic on Matrigel in the presence of hLIF/SB/CHIR (Fig. 2A). However, no colonies were observed under conditions including C-E, suggesting that Notch signaling is critical to pNSC self-renewal. Consistently, treatment with C-E for 48 h rapidly induced pNSCs to differentiate into Doublecortin (DCX)-positive neuronal precursors (Fig. S2B 1 and 2). Despite their highly proliferative and clonogenic capacity, pNSCs are not tumorigenic in SCID beige mice. We transplanted the early-passage (passage 6, about 30 d in serial culture) and late-passage (about passage 27) of HUES9- and H1-derived pNSCs (2 × 10⁶ cells suspended in Matrigel) into 24 SCID beige mice s.c. These mice have been observed for as long as 6 mo with no sign of neoplasm formation, whereas the control animals transplanted with the parental hESCs produced teratomas within 6 wk.

Fig. 2.

pNSCs stably self-renew and maintain a homogenous primitive NSC phenotype after long-term cultures. The pNSCs cultured on Matrigel exhibited characteristic epithelial morphology. (A) A single cell-derived pNSC colony on Matrigel. (Inset) pNSCs were positive for ALP. (B–I) Immunocytochemistry showed that pNSCs (passage 6) expressed genes recently identified as rosette-type NSC markers, including PLZF, ZO-1, and N-cad; CNS neural stem cell makers such as Nestin, Pax6, and Sox2; the cell proliferation marker Ki-67; the anterior neural markers Forse1 and Otx2; and the midbrain marker Nurr1. (J) Flow cytometry analysis showed that pNSCs stably expressed NSC and cell proliferation markers after long-term in vitro expansion, including Nestin, Pax6, CD133, Forse1, and Ki-67. (K and L) Nonneural lineage markers and genes associated with midbrain were analyzed by RT-PCR.

Remarkably, the long-term expanded pNSCs maintain a stable pNSC phenotype. The pNSCs cultured on Matrigel exhibited typical epithelial morphology and positive ALP staining (Fig. 2A). Immunostaining showed that both the early-passage (passage 6) and late-passage (passage 27) pNSCs stably expressed genes recently identified as rosette-type NSC markers (4), including PLZF (promyelocytic leukemia zinc finger), ZO-1, and N-cad (N-cadherin); CNS (central nervous system) neural stem cell markers, such as Nestin, Pax6, and Sox2; anterior neural markers Forse1 and Otx2; and the midbrain marker Nurr1 (Fig. 2 B–I and Fig. S3A 1–8). Expression analysis by microarray confirmed the dramatic up-regulation of neural lineage genes such as Ascl1, Pax6, Dach1, N-cad, and Nestin, and down-regulation of pluripotency gene Oct4 in pNSCs in comparision to hESCs. However, both hESCs and pNSCs express ZO-1 and Sox2 at similar level. Even after long-term passaging, pNSCs uniformly expressed a panel of primitive neuroepithelial genes, including Sox2, N-cad, PLAZ, Dach1, ZO-1, Pax6, and proneuronal gene Ascl1, and both early- and late-passage pNSCs demonstrated highly similar transcriptome profile (Fig. S3B 1 and 2). Notably, N-cad and the tight junction protein ZO-1 were expressed evenly on the surface of both early- and late-passage pNSCs, suggesting that pNSCs are primitive, nonpolarized prerosette NSCs (2). Indeed, pNSCs gained rosette-like structures with apical N-cad expression and interkinetic nuclear migration after being cultured in neural induction media with 20 ng/mL bFGF for 4 d (Fig. S3C 1 and 2). Consistent with their highly proliferative capacity, pNSCs uniformly expressed Ki-67 (Fig. 2F and Fig. S3A5). The stable phenotype of pNSCs after extensive passaging was further confirmed by flow cytometry. Both early-passage and late-passage pNSCs exhibited nearly identical expression patterns for a set of NSC-specific markers, such as Nestin (98.2% positive), Pax6 (95.4% positive), CD133 (93.9% positive), and the cell proliferation marker Ki-67 (98.3% positive; Fig. 2J), whose uniform expression confirmed that pNSCs were a homogenous, expandable NSC population. Indeed, genes associated with non-neural lineages, such as Eomes, Brachyury, Sox17, or K15 were undetectable in pNSCs by RT-PCR (Fig. 2K). FACS analysis with propidium iodide revealed a very similar cell cycle profile for both early- and late-passage of pNSCs. The cell cycle distribution (G₁, S, and G₂/M) of early-passage pNSCs (P7) is 52.7%, 26.6%, and 16.5%; and the cycle distribution of late-passage pNSCs (P28) is 51.7%, 32.4%, and 12.4%, respectively. Interestingly, FACS analysis showed that the expression of Forse1, an anterior NSC marker, was not homogenous (53.6% of pNSCs were positive for Forse1; Fig. 2J). Whether Forse1-negative pNSCs have a more posterior identity needs to be further characterized. In addition, pNSCs did not express neural crest cell markers such as HNK1, Sox10, or p75. However, we did detect a small percentage (∼3%) of pNSCs positive for AP2, a premigratory neural crest gene initially expressed throughout the neural plate border (22). To rule out the possibility that pNSC cultures contained a separate (parallel or unrelated) neural crest cell population, we preformed clonal analysis by immunostaining of single pNSC-derived colonies. AP2-positive cells representing 2–5% of cells in each colony were detected in all examined colonies (n = 16), and they were mostly seen at the border of the colonies (Fig. S4), suggesting that they were derivatives of pNSCs. However, whether these AP2-positive cells possess neural crest potential remains to be confirmed.

To examine the multipotency of long-term expanded pNSCs, both early- and late-passage pNSCs were plated at ultra-low density in six-well plates (200 cells per well) and cultured in differentiation media for 2 wk. Among the single pNSC-derived cell clusters (n = 29), 100% contained both MAP2-positive neurons and GFAP-positive astrocytes (Fig. S5 A–C), but no cells positive for the oligodendrocyte marker O4 or the neural crest lineage markers peripherin and α-SMA were detected at this time point. Previous studies have shown that bFGF and/or EGF-expanded NSCs lose neurogenic propensity and become more gliogenic after long-term culture (23). However, pNSCs expanded under our described conditions retained high neuronal differentiation propensity. Flow cytometry analysis showed that pNSCs at passage 8 and passage 25 could give rise 73.9% and 77.6% MAP2-positive, or 71.4% and 74.4% NeuN-positive neurons, respectively (Fig. 3A 1 and 2). During CNS development, neurogenesis largely precedes gliogenesis. NSCs from earlier stages generate more neurons and have a lower propensity to produce glia than those from later stages. The remarkably high neurogenic potential and propensity of these long-term expanded pNSCs is consistent with their self-renewal in the primitive state. Importantly, pNSCs could effectively differentiate and generate mature neurons that fired action potentials (5 cells in 7 tested cells; Fig. 3B), and produced fast inactivating inward Na⁺ currents (n = 8 of 8 cells recorded) that were sensitive to the Na⁺ channel blocker Tetrodotoxin (TTX; Fig. 3C). Furthermore, these differentiated neurons manifested spontaneous excitatory postsynaptic currents (sEPSCs) and/or inhibitory postsynaptic currents (sIPSCs) in 4 of 6 cells recorded (Fig. 3D), indicating that they can form functional synapses. Next, to further examine pNSCs’ potential in vivo, they were transplanted into the lateral ventricle of neonatal mice (P2-3). Histological analysis of GFP-expressing pNSC (passage 27) grafts one month after transplantation revealed that engrafted cells were distributed in many brain areas, including the corpus callosum, the subcallosal zone, the caudate-putamen (Fig. S6 A and B), and the hindbrain (Fig. S6 C–J). Most engrafted cells (>50% in the forebrain, and >80% in the hindbrain) expressed differentiated neuronal markers such as MAP2 (Fig. S6A 1–4). In addition, we also detected DCX-positive engrafted cells in the subcallosal zone (Fig. S6B 1–4), where endogenous adult neural progenitor cells reside (24), but not in non-neurogenic environments such as the hindbrain, suggesting that their neuronal differentiation was influenced by the host environment. Although we failed to detect the mature neuronal marker NeuN in the subcallosal zone or caudate-putamen, a subset of GFP-expressing cells in the clusters near the aqueduct exhibited NeuN expression (Fig. S6C 1–4). We also failed to detect spontaneously differentiated tyrosine hydroxylase (TH)-positive dopaminergic (DA) neurons, but some engrafted cells (∼10%) appeared to have differentiated into GABA-expressing inhibitory neurons (Fig. S6D 1–4). GFP-expressing cells in the hindbrain, closely associated with presynaptic puncta labeled by synaptophysin, were also observed, indicating the synaptic contacts of the transplanted cells with the host mouse neurons (Fig. S6E 1–4). All GFP positive cells also exhibited human nucleus antigen immunoreactivity (Fig. S6F 1–4), further confirming their human cell identity. In addition, some GFP-expressing hindbrain neurons also exhibited c-fos, a marker for neuronal excitation (Fig. S6 G–I). On the other hand, we did not find any GFAP-positive engrafted cells (Fig. S6J), suggesting that pNSCs preferentially differentiate into the neuronal lineage in vivo.

Fig. 3.

pNSCs retain high neurogenic potential during long-term culture. (A) Flow cytometry analysis showed pNSCs at passage 8 and passage 25 could give rise to 73.9% and 77.6% MAP2-positive, or 71.4% and 74.4% NeuN-positive neurons, respectively. (B) Representative traces of evoked action potentials (whole-cell recording, current-clamp mode) generated by neurons after 4 wk of differentiation from pNSCs. Traces of Tetrodotoxin (TTX)-sensitive whole-cell currents recorded in voltage-clamp mode. (C) Cells were hyperpolarized to −90 mV for 300 ms before applying depolarizing pulses to elicit Na⁺ and K⁺ currents. (D) Traces of spontaneous excitatory postsynaptic currents (sEPSCs) and spontaneous inhibitory postsynaptic currents (sIPSCs), both recorded at a holding potential of −60 mV, indicated synapse formation. B–D represent the data recorded from pNSCs at passage 25 that had spontaneously differentiated to display neuronal properties.

pNSCs Possess Mesencephalic Regional Identity and Can Be Respecified Toward Caudal Cell Fates.

It is worthwhile to note that pNSCs express the forebrain/midbrain gene Otx2 and the midbrain gene Nurr1 by immunostaining (Fig. 2 H and I). RT-PCR analysis confirmed the expression of Otx2 and Nurr1, and showed that pNSCs also express other midbrain genes, such as En-1, Lmx1b, Pax2, and Pitx3 (Fig. 2L). In contrast, the forebrain-restricted transcription factors FoxG1 and Emx2 were barely detectable, and anterior hindbrain transcription factors, such as Gbx2, HoxB2, and HoxA2, were expressed at low levels as indicated by RT-PCR (Fig. S7A). pNSCs expressed the dorsal neural tube gene Pax3, the ventral neural tube genes Nkx2.2 and Nkx6.1, the NSC marker Dach1 and the Notch effector Hes5 (Fig. S7A). These observations suggested that in vitro-expanded pNSCs may possess a mesencephalic regional identity. To confirm this, two different differentiation protocols were used to examine the potential of pNSCs to generate midbrain DA neurons: in the “induced” protocol, pNSCs were first treated with 100 ng/mL SHH (Sonic Hedgehog) and 100 ng/mL FGF8b for 10 d and were then further differentiated in the presence of 10 ng/mL BDNF, 10 ng/mL GDNF, 10 ng/mL IGF1, 1 ng/mL TGF-β3, and 0.5 mM dibutyryl-cAMP (dbcAMP) for another ∼2-3 wk (Fig. 4A). In the “default” protocol, pNSCs were directly terminally differentiated in the presence of BDNF, GDNF, IGF1, TGF-β3, and dbcAMP for 3 wk without pre-patterning by morphogens (Fig. 4B). Both the “induced” and “default” differentiation conditions produced >50% TH-positive neurons that also exhibited aromatic l-amino acid decarboxylase (AADC), En-1, Lmx1a, Nurr1, FoxA2, and Pitx3 immunoreactivity (Fig. 4 A 1–7 and B 1–7). Notably, Pitx3 is a homeobox gene uniquely expressed in midbrain DA neurons (25). Real-time PCR confirmed the significant up-regulation of TH, AADC, En-1, Nurr1, and Pitx3 (Fig. 4 A8 and B8). Flow cytometry quantification demonstrated that the “induced” and “default” differentiation protocols produced 54.8% and 62.7% TH and MAP2 double positive neurons, respectively (Fig. 4 A9 and B9). These data demonstrate that pNSCs possess mesencephalic regional identity and can differentiate into DA neurons with very high efficiency.

Fig. 4.

pNSCs possess mesencephalic regional identity and can differentiate into DA neurons with high efficency. (A) In the “induced” protocol, pNSCs were treated with SHH and FGF8b for 10 d before they were terminally differentiated in the presence of BDNF, GDNF, IGF1, TGF-β3, and dbcAMP for another ∼2–3 wk. (B) In the “default” protocol, pNSCs were directly terminally differentiated in the presence of BDNF, GDNF, IGF1, TGF-β3, and dbcAMP for 3 wk. Under both protocols, pNSCs gave rise to TH positive neurons that also exhibited AADC, En-1, Lmx1a, Nurr1, FoxA2, and Pitx3 immunoreactivity (A 1–7 and B 1–7). Real-time PCR further confirmed the significant up-regulation of TH, AADC, En-1, Nurr1, and Pitx3 (A8 and B8). Flow cytometry quantification demonstrated that the two differentiation protocols produced 54.8% and 62.7% TH and MAP2 double-positive neurons, respectively (A9 and B9).

Because they exhibit features of pre-rosette primitive NSCs, pNSCs were further examined for their responsiveness to instructive regional patterning cues. pNSCs were sequentially treated with caudalizing retinoic acid (RA, 1 μM) for 7 d, 100 ng/mL SHH and 0.1 μM RA for another 7 d, and then 50 ng/mL SHH and 0.1 μM RA for an additional 7 d. The cells were then terminally differentiated in the presence of 10 ng/mL BDNF and 10 ng/mL GDNF in differentiation media for about 7 d. Real-time PCR assays demonstrated significant induction of posterior genes, including HoxB4, HoxA5, and HoxC5 after treatment with 1 μM RA for 1 wk (Fig. S8A), suggesting that pNSCs are responsive to the caudalizing effect of RA. Under such conditions, immunocytochemistry showed that pNSCs could differentiate into choline acetyltransferase (ChAT)-positive neurons that are also positive for MAP2 and Isl-1 (Fig. S8 B and C). Flow cytometry analysis showed that 53.7% cells were double-positive for Isl-1 and MAP2 (Fig. S8D). Real-time PCR assays confirmed the significant induction of ChAT, HB9, Isl-1, and Lim3 after terminal differentiation (Fig. S8A), suggesting an induction of motor neurons. These data indicated that pNSCs retain responsiveness to instructive cues promoting the induction of hindbrain neuronal subtypes.

Discussion

To realize the potential of cell-based therapy for treating injuries and degenerative diseases, renewable sources of stem/progenitor cells need to be developed. Although hESCs indefinitely self-renew and have the differentiation potential to become any cell type, they are practically inferior to lineage-restricted cells as they are prone to causing teratomas and do not repopulate host tissues in vivo. However, significant challenges also remain in terms of the isolation and long-term expansion of most tissue-specific stem/progenitor cells from adults (e.g., even for the arguably most studied hematopoietic stem cells). Consequently, differentiation of hESCs into renewable tissue-specific cell types is highly desirable for various biomedical applications. If achieved, cell populations could be carefully quality controlled and serve as starting materials, skipping hESCs that cannot be used directly. Furthermore, despite significant advances in development of various neural induction conditions for hESCs, most differentiation protocols use poorly defined culture conditions (e.g., going through EB formation, using undefined medium supplements/KSR), and usually yield mixed populations containing neural cells at different developmental stages, or even other embryonic germ layer lineages and undifferentiated hESCs. In the present study, our serendipitous observation led us to develop a robust chemically defined condition using specific small molecules that rapidly and uniformly converts hESCs into pNSCs, and, most importantly, enables their long-term expansion without a loss of high neurogenic propensity and regionalizable plasticity. To our knowledge, this is the fastest and most efficient method so far to produce neural stem cells from hESCs. In addition, pNSCs differ from previously reported hESC-derived NSCs in that they represent the primitive pre-rosette neuroepithelium that has never been long-term expanded in vitro before. Interestingly, pNSCs possess features of mesencephalic precursor cells and can differentiate into DA neurons spontaneously with high efficiency in the absence of pre-patterning. Real-time PCR analysis showed the up-regulation of endogenous SHH, FGF8, and the ventral patterning gene Nkx6.1 under both “induced” and “default” differentiation protocols (Fig. S7B), suggesting the cells could be specified into DA neurons by an endogenous mechanism. These observations are reminiscent of the previous in vivo studies that showed DA neurons originated from SHH-expressing domains of the ventral midbrain (26). In addition, a mouse study demonstrated the antagonistic interaction between the activation of Wnt/β-catenin and SHH (27). The activation of β-catenin in the ventral midbrain promoted the expansion of early DA progenitors, but led to a reduced expression of SHH. The removal of the GSK3 inhibitor (CHIR) during pNSC differentiation may lead to down-regulation of Wnt/β-catenin signaling and facilitate the up-regulated SHH expression in turn. Considering the significance of developing renewable sources of DA neurons, it would be useful to examine whether pNSC transplantation could attenuate the Parkinson's symptoms in animal models in the future.

Recent studies suggest that GSK3 plays key roles in many fundamental processes, including mediating signaling downstream of Wnt, FGF, Hh, and Notch during neural development (28–30). In our neural induction protocol, however, replacement of CHIR with Wnt3a induced significant spontaneous differentiation and could not generate a homogenous NSC population, suggesting that GSK3 inhibition may coordinate multiple signals besides canonical Wnt activation in the context of neural induction under this condition. One possible explanation for this specific neural induction is that inhibition of TGF-β/Nodal signaling by SB431542 not only blocks the formation of mesendoderm, but also engages in cross-talk with GSK3-mediated signaling (for example FGF signaling) to enhance neural induction, possibly by modulating a downstream component of endogenous BMP signaling (2, 31). In addition, very recent studies showed that GSK3 is a master regulator of in vivo neural progenitor homeostasis (28, 29). It is possible that neural induction is also coupled with the capture/maintenance of primitive NSCs through GSK3 inhibition. Specifically, the combination of GSK3 inhibitor, TGF-β receptor inhibitor, and hLIF is uniquely required for long-term self-renewal of pNSCs under chemically defined conditions. Recent in vivo studies demonstrated that TGF-β pathway activation counteracts canonical Wnt and negatively regulates self-renewal of midbrain neuroepithelial stem cells in the developing mouse brain (32). Loss of TGF-β signaling results in neuroepithelial expansion in the midbrain, but not the forebrain (32). The use of GSK3 inhibitor (which can activate canonical Wnt) and TGF-β receptor inhibitor may partly recapitulate such in vivo self-renewal signals of midbrain NSCs. With an improved understanding of the signaling mechanisms involved in lineage specification and maintenance of tissue-specific stem cells, this strategy could also be generalized and applied to the capture of self-renewing stem cells from other germ layers, such as endoderm or mesoderm. Finally, this protocol also provides a valuable tool with which to study the early molecular events initiating human neural induction.

Materials and Methods

For further details of cell cultures, neuronal differentiation, immunocytochemistry, flow cytometry, quantitative and semiquantitative RT-PCR, electrophysiological analysis, microarray analysis, teratoma assays, in vivo transplantation, and histology, see SI Materials and Methods. The antibodies used in this study are shown in Table S1. For the primers of quantitative and semi-quantitative RT-PCR, see Table S2.