Neural Induction. MS5 and S2 stromal cells were maintained in α-MEM medium containing 10% FBS and 2 mM l-glutamine (11). For some studies, transgenic MS5 cells were used that stably overexpress Wnt1 after transfection (Fugene-6) of a Wnt1 expression construct followed by G418 selection. Neural differentiation of hES cells was induced by means of coculture on MS5, MS5-Wnt, or S2 stroma at comparable efficiencies. hES cells were plated at 5–20 × 10³ cells on a confluent layer of irradiated (50 Gy) stromal cells in 6-cm cell culture plates in serum replacement medium containing DMEM, 15% knockout serum replacement (Invitrogen), 2 mM l-glutamine and 10 μM β-mercaptoethanol. After 16 days in serum replacement medium, cultures were switched to N2 medium modified according to ref. 19. Medium was changed every 2–3 days, and growth factors were added in various combinations and at various time points as described: 200 ng/ml SHH, 100 ng/ml FGF8, 20 ng/ml brain-derived neurotrophic factor (BDNF), 10–20 ng/ml glial cell line-derived neurotrophic factor, 1 ng/ml transforming growth factor type β3 (R & D Systems), 0.5–1.0 mM dibutyryl cAMP, and 0.2 mM ascorbic acid (AA) (Sigma–Aldrich). Rosettes structures were harvested mechanically from feeders at day 28 of differentiation and gently replated on 15 μg/ml polyornithine/1 μg/ml laminin-coated culture dishes in N2 medium supplemented with SHH, FGF8, AA, and BDNF (passage 1). After 7–9 days (≈80% confluency), cells were mechanically passaged after exposure to Ca²/Mg²-free Hanks' balanced salt solution for 1 h at room temperature and spun at 200 × g for 5 min. Cells were resuspended in N2 medium, replated again onto polyornithine/laminin-coated culture dishes (50–100 × 10³ cells per cm²) in the presence of SHH, FGF8, AA, and BDNF (passage 2). After an additional 7–9 days of culture, cells were differentiated in the absence of SHH and FGF8 but in the presence of BDNF, glial cell line-derived neurotrophic factor, transforming growth factor type β3, dibutyryl cAMP, and AA.

Immunocytochemistry. Cells were fixed in 4% paraformaldehyde/0.15% picric acid and stained with the following primary antibodies. Rabbit polyclonal antibodies included TH and vesicular monoamine transporter 2 (VMAT2, Pel-Freez Biologicals); nestin 130 (R. McKay, National Institutes of Health, Bethesda); glial fibrillary acidic protein and DA (Chemicon); GABA and serotonin (Sigma); Pax2, Pax6, and β-tubulin III (Covance); Pax5 (clone A2; M. Busslinger, Institute of Molecular Pathology, Vienna); aromatic l-amino acid decarboxylase (AADC, Protos Biotech, New York); synaptic vesicle 2 (SV2, Developmental Studies Hybridoma Bank, University of Iowa, Iowa City); and synapsin (Calbiochem). Mouse monoclonal antibodies (IgG) included Oct4 (Santa Cruz Biotechnology), TH (Sigma), class III β-tubulin (Tuj1, Covance, Princeton), En1, and Lmx (Developmental Studies Hybridoma Bank). Mouse monoclonal antibodies (IgM) included Ki67 (Sigma) and O4 (Chemicon). Appropriate Alexa488- and Alexa555-labeled secondary antibodies (Molecular Probes) and 4′,6-diamidino-2-phenylindole counterstain were used for visualization.

RT-PCR, Electrophysiology, and Electron Microscopy. Total RNA was extracted by using the RNeasy kit and DNase I treatment (Qiagen, Valencia, CA). Total RNA (2 μg each) was reverse transcribed (SuperScript, Invitrogen). PCR conditions were optimized and linear amplification range was determined for each primer by varying annealing temperature and cycle number. PCR products were identified by size, and identity was confirmed by DNA sequencing. MS5 cells were negative for all primers used in this study except for 18S ribosomal RNA. Under coculture conditions (days 0–28), hES cell progeny was mechanically separated from feeders to avoid cross contamination of the RNA pool by feeder cells. Primer sequences, cycle numbers, and annealing temperatures are provided upon request. For quantitation, gels were imaged by using a 12-bit charge-coupled device camera (Alpha Innotech, San Leandro, CA). Data are presented as means of normalized expression levels (18S ratios) obtained from three independent experiments and scaled such that the maximum level of expression during the observed time period was arbitrarily set at 1. Semiquantitative RT-PCR data were from H1 cells. Similar data were derived from H9 and HES-3 cells. Electrophysiological analyses were carried out as described recently for characterizing mouse ES-derived neurons (11). Freeze substitution-postembedding-ImmunoGold labeling for TH was performed according to ref. 20 (for details, see Supporting Materials and Methods, which is published as supporting information on the PNAS web site).

HPLC Analysis. Reversed-phase HPLC for the detection of DA in the supernatant was performed as described in ref. 21. Samples were collected at day 50 of differentiation, stabilized in orthophosphoric acid and metabisulfite, and extracted by aluminum adsorption (Chromosystems, Munich). Separation of the injected samples (ESA Autosampler 540) was achieved by isocratic elution in MD-TM mobile phase (ESA, Bedford, MA) at 0.5 ml/min. The oxidative potential of the analytical cell (ESA Mod. 5011, Coulochem II) was set at 350 mV. Results were validated by coelution with catecholamine standards under various buffer conditions and detector settings.

Semiquantitative RT-PCR analysis confirmed a time-dependent decrease in the expression of ES cell markers, such as Oct4, Cripto (TDGF1), and Nanog, upon stromal feeder-mediated differentiation (Fig. 1M). A delayed decrease in Oct4 expression was observed consistent with recent work in mouse ES cells, suggesting that transiently sustained Oct4 levels are required for neural induction (23). The decrease in ES cell markers was associated with increased expression of neural markers, including Pax6 and the neuronal marker MAP2 (Fig. 1N). No glial fibrillary acidic protein was detected, suggesting absence of astrocytic differentiation. As expected, control genes related to endodermal and mesodermal differentiation did not show consistent increases or decreases in gene expression during stromal feeder-mediated differentiation of hES cells (Fig. 1O).

At 4 weeks of differentiation, neural rosettes expressing Sox1 (Fig. 1P), NCAM (Fig. 1Q), and Pax6 were composed of proliferating Ki67-positive precursors (Fig. 1R). Cells within the rosettes were devoid of neuronal markers, such as Tuj1 or MAP2. However, cells emanating from the borders of individual rosettes were often positive for Tuj1 and exhibited neuronal morphologies (Fig. 1S). Comparable results were obtained with stromal feeder cells derived from the aorta-gonad-mesonephros region, such as the line S2 (11). The neural inducing effects were confirmed in multiple hES and nonhuman primate ES cell lines, including the hES cell lines H1, H9, and HES-3, and the monkey lines R366 and Cyno1.

Patterning and Differentiation of Neural Precursors from ES-Derived Rosettes. Previous work identified SHH and FGF8 as crucial factors in the specification of midbrain DA neurons (24). These studies in explant culture and subsequent work with mouse ES cells (2, 11, 22) demonstrated that the effect of SHH and FGF8 on dopaminergic differentiation is limited to distinct developmental windows that correspond to the establishment of the isthmic organizer and events controlling dorsoventral patterning during neural tube closure. Based on the work in mouse ES cells, we predicted that in hES cells SHH and FGF8 might act between days 12 and 20 of differentiation to specify human midbrain DA neuron fate. This time period corresponds to the onset of neurulation until the time of neural tube closure in human development, taking into account the presumptive age of hES cells at the time of isolation (5–7 days) (25).

Exposure to SHH and FGF8 from day 12 to day 20 of differentiation, followed by differentiation in the presence of AA and BDNF, resulted in a 3-fold increase in TH-positive cells (see Fig. 2H, P0), TH being the rate-limiting enzyme in the synthesis of DA. However, no expression of other midbrain-related markers, such as Pax2, Pax5, En1, or AADC was observed by immunocytochemistry, suggesting that early exposure to SHH and FGF8 is not sufficient to induce a full midbrain DA neuron phenotype. Interestingly, neural rosettes persisted in culture for several weeks, even in the absence of any exogenous mitogens or feeder cells, and remained devoid of cells expressing distinct region-specific transcription factors. This result suggested that neural rosettes may retain the capacity to respond to patterning cues beyond the predicted temporal window.

To test whether rosettes respond to midbrain patterning cues, clusters of rosettes were mechanically isolated at day 28 (passage 0), replated on polyornithine/laminin precoated dishes (passage 1), and cultured for 7–10 days in N2 medium containing SHH and FGF8. After the next passage (passage 2) cells were grown in SHH and FGF8 for an additional week before differentiation in N2 medium supplemented with glial cell line-derived neurotrophic factor, dibutyryl cAMP, and transforming growth factor type β3 [previously found to enhance En1 expression in the developing rodent midbrain (26)]. After differentiation of passage 2 cells (50 days in culture) 30–50% of the total cells derived from H1, H9, or HES-3 (Fig. 2 A–F) lines were Tuj1-positive neurons. All Tuj1-postitive cells were negative for Ki67. Virtually all cells negative for TH or Tuj1 expressed the intermediate filament nestin, suggesting neural precursor identity (Fig. 2I). Among the Tuj1-positive neuronal population 64–79% of the cells expressed TH (H1, 64 ± 2%; H9, 70 ± 2%; HES-3, 79 ± 7%) (Fig. 2G). Similar data were obtained with the two monkey ES cell lines R366 and Cyno1 (data not shown). The percentage of TH expressing neurons was highly dependent on exposure to SHH and FGF8 during passages 1 and 2 (Fig 2H). In addition to TH-positive neurons, ≈5% serotonin-positive (Fig. 2J) and 1–2% GABA-positive neurons (Fig. 2K) were detected. No coexpression of GABA and TH was observed, compatible with midbrain dopaminergic differentiation. GABA and TH coexpression is observed in olfactory DA neurons (27) and transiently in striatal GABAergic neurons (28). No cholinergic or glutamatergic neurons were detected at that stage. The neurotransmitter phenotype of the other Tuj1-positive neurons (29%) remains unknown. No glial fibrillary acidic protein-positive cells were observed up to passage 2. However, astrocytic differentiation could be readily induced in long-term cultures (>70 days) exposed to either serum or ciliary neurotrophic factor (Fig. 6 A and B, which is published as supporting information on the PNAS web site). A few O4-positive oligodendrocytes displaying characteristic star-like morphologies could also be detected in long-term serum-free cultures (Fig. 6C), confirming that neural rosettes can yield progeny in all three CNS lineages.

Gene expression analysis from day 28 (passage 0) to day 50 (passage 2) of differentiation showed a complete loss of ES markers, including Oct4 (Fig. 3A). Pax6 expression reached a plateau at the rosette stage, whereas MAP2 expression continued to increase up to passage 2 (Fig. 3A). This finding suggests that neural induction is largely completed by day 28, whereas neurogenesis continues at least up to passage 2. Induction of markers characteristic of midbrain DA neuron development was readily detected during this period (Fig. 3B).

Immunocytochemical data of passage 1 and passage 2 cells revealed that, unlike at day 28 of differentiation, hES-derived cells started to express key markers of midbrain DA neuron development. A developmental progression was observed with Pax2 appearing first in passage 1 cells in the absence of En1 expression (Fig. 3C). Cells at passages 1 and 2 continued to exhibit neural precursor characteristics expressing Ki67 and nestin (Fig. 3D). At passage 2, cells coexpressing Pax2 and En1 (Fig. 3E) and cells expressing Lmx (Fig. 3F) were observed. The total number of En1-positive cells was highly dependent on the developmental stage. The percentage of En1 cells in the total population varied from 0% (passage 0 and early passage 1) up to 30% (passage 2). At day 50, the few remaining rosettes were surrounded by TH-positive cells often radially oriented toward the center of the rosette (Fig. 3G). Coexpression studies showed that Pax2- and Pax5-expressing cells were negative for TH (Fig. 3 H and I). In contrast, cells positive for En1/TH were readily detected (Fig. 3J). Our data suggest that, similar to mouse in vivo development, human midbrain DA neurons in vitro are derived from proliferating progenitors that sequentially express Pax2 and Pax5 and, upon exiting the cell cycle, become positive for En1 and eventually TH. Single-cell lineage analysis will be required to fully confirm this finding. Additional markers expressed in TH-positive cells were AADC (Fig. 3K), VMAT2 (Fig. 3L), and synaptic markers, such as SV2 (Fig. 3M) and synapsin (Fig. 3N). Although 100% of the TH-positive cells coexpressed VMAT2, only 50% of the TH-positive cells coexpressed AADC (Fig. 3K), and 74% were immunoreactive for DA (data not shown). At day 50 of differentiation, 15% of TH-positive cells expressed SV2, confirming the immature state of synaptic development in these cells. No colocalization of TH and DA-β hydroxylase was observed at any stage of differentiation, confirming the absence of noradrenergic and adrenergic neurons (data not shown).

Stromal feeder-mediated neural induction was first described for mouse ES cells and has been proposed to specifically induce neurons of midbrain DA neuron phenotype (3). Later studies in mouse and nonhuman primate ES cells demonstrated that the neural-inducing effects are separate from the effects on patterning and that differentiation conditions can be adapted for a wide range of neuronal and glial subtypes (11, 14, 22). Here, we show that neural-inducing effects for hES cells can be provided by bone marrow and aorta-gonad-mesonephros-derived stromal feeder cells. In contrast to previous work on the differentiation of cynomolgus monkey ES cells on PA6 (14), we did not observe significant differentiation of hES cells into retinal pigment epithelial cells. However, the hES line HES-3 and the rhesus monkey ES cell line R366 yielded some (<5%) pigmented epithelial cells if maintained on MS5 beyond day 28 of differentiation (data not shown).

Based on their distinctive morphology, it has been suggested that hES derived rosettes might mimic the developing neural tube (8). Dorsoventral patterning characterized by the formation of distinct transcription factor expression domains occurs during neurulation and is largely completed at the time of neural tube closure (29). Expression of Sox1, Pax6, nestin, and NCAM in the absence of specific dorsoventral markers within rosettes suggests that these cells might correspond to a neural plate rather than a neural tube stage. This interpretation is compatible with the robust response of these cells to the patterning effects of SHH and FGF8.

Our study did not show a differentiation bias toward glutamatergic or GABAergic neurons as observed previously with hES-derived precursors maintained under neurosphere-like conditions (7, 8). Although the reasons for these differences remain to be elucidated, we speculate that the absence of FGF2-mediated proliferation might be an important component (11). Our results suggest that rosettes maintained in the absence of FGF2 remain amenable to both anterior–posterior and dorsoventral patterning and may serve as a universal source for neuronal subtype specific differentiation. Contamination of the dopaminergic population with serotonergic neurons is not surprising given the close developmental relationship between these two ventral neuron subtypes (24) and confirms the validity of the developmental model proposed here. Serotonergic differentiation might be further enhanced by using developmentally relevant factors, such as FGF4.

The sequence of transcription factor expression involved in midbrain DA neuron development in vitro is similar to the expression patterns in vivo with Pax2, aldehyde dehydrogenase 1, and Lmx1b expression preceding gradual increases in the expression of markers characteristic of early postmitotic neurons, such as En1, nuclear orphan receptor 1, Pitx3, and, finally, TH. Although robust En1 expression was observed during DA neuron differentiation, not all TH-positive neurons remained En1-positive, in contrast to rodent midbrain development (30). It will be essential to test whether loss of En1 reflects a feature of normal human midbrain development or an idiosyncrasy of our in vitro conditions. Not surprisingly, DA transporter immunoreactivity was absent in our TH-positive cell population. This finding is probably due to the relatively immature state of the TH-positive cell population, the serum-free culture conditions, and the absence of a striatal target, known to be essential for DA transporter expression in vivo (31). Exposure of hES-derived neural precursors to SHH/FGF8 dramatically increased the percentage of TH-positive neurons and subsequently the percentage of mature human DA neurons. The high DA neuron yield is illustrated by the fact that for every single undifferentiated hES cell plated at the start of the protocol, >100 TH-positive neurons can be harvested at day 50 of differentiation. Starting with a single 6-cm culture dish, the number of TH-positive cells generated by day 50 amounts to the total number of DA neurons present in the adult human substantia nigra (≈1 × 10⁶ cells). The time frame for midbrain DA neuron differentiation (7–8 weeks in vitro) may appear long. However, midbrain DA neurons are born in vivo at 6.5–8.5 weeks after conception (32), a time frame comparable to our in vitro data.