Understanding neuroectoderm formation and subsequent diversification to functional neural subtypes remains elusive. We show here that human embryonic stem cells (hESCs) differentiate to primitive neuroectoderm after 8–10 days. At this stage, cells uniformly exhibit columnar morphology and express neural markers, including anterior but not posterior homeodomain proteins. The anterior identity of these cells develops regardless of morphogens present during initial neuroectoderm specification. This anterior phenotype can be maintained or transformed to a caudal fate with specific morphogens over the next week, when cells become definitive neuroepithelia, marked by neural tube-like structures with distinct adhesion molecule expression, Sox1 expression, and a resistance to additional patterning signals. Thus, primitive neuroepithelia represents the earliest neural cells that possess the potential to differentiate to regionally specific neural progenitors. This finding offers insights into early human brain development and lays a foundation for generating neural cells with correct positional and transmitter profiles.
Disclosure of potential conflicts of interest is found at the end of this article.
Early vertebrate neural development is a highly coordinated process in which the inner cell mass (ICM) cells of the blastocyst give rise to the epiblast. These epiblast cells go on to generate the three embryonic germ layers (ectoderm, mesoderm, and endoderm), and it is the ectoderm that eventually forms epithelia, including neuroepithelia (or neuroectoderm). Neuroepithelia make up the neural plate, which serves as the source of the entire central nervous system. During the process of folding into the neural tube, neuroepithelia must be correctly patterned to generate distinct classes of neural progenitors that will make up the forebrain, midbrain, hindbrain, and spinal cord. It has been proposed that ectoderm cells are first activated to a forebrain stage and are subsequently transformed with caudalizing signals to mid-/hindbrain and spinal cord fates in a gradient fashion, the so-called activation/transformation model [1, 2]. This would imply that the initial neuroepithelia posses a broader differential potential than do cells at later stages. An alternative hypothesis is the specification of neuroectoderm cells with different rostrocaudal identities by multiple signaling centers [3, 4]. Although it is clear that the number of neural stem cells in the developing neural tube decreases with time, it is still not completely understood how these cells are initially specified and distributed, as well as how the competency of these cells to form more restricted neural progenitors shifts over time .
Neural induction and early patterning have traditionally been studied in vertebrates such as chick, zebrafish, and amphibian. Establishment of mouse and human embryonic stem cells (ESCs) and development of protocols for directed differentiation of ESCs now make it possible to investigate these early events in mammals. ESCs transform to neuroepithelial cells rapidly in serum-free conditions [6, 7] and, depending on the presence of morphogens (e.g., fibroblast growth factors [FGFs], Wnts, and retinoic acid [RA]), the generated neuroepithelial cells are fated to cells of various regional identities including telencephalic , mid-/hindbrain [9, , –12], and spinal cord [13, 14]. These results suggest that much of the developmental potential present in vivo is maintained in vitro.
To identify intrinsic determinants and extrinsic factors that govern neural differentiation from human ESCs, we established a chemically defined, monolayer colony, neural differentiation culture system . Cellular and gene expression profiling analyses indicated that major events, such as formation of neural tube-like rosettes and expression of definitive neuroectodermal genes such as SoxB family members, mirrors neuroectoderm induction in vivo. We also discovered an earlier developmental stage in which cells express most known neuroectoderm markers and exhibit a uniform expression of anterior neural patterning genes. We refer to these cells as primitive anterior neuroepithelia. These primitive anterior neuroepithelial cells can persist in their rostral phenotype unless a caudalizing signal such as RA is added in the subsequent differentiation.
Materials and Methods
Maintenance of Primate ESCs.
Human ESC lines H9 and H1 (passages 20–35), and the rhesus monkey ESC line R366.4 (passages 20–25), were expanded on a feeder layer of irradiated mouse embryonic fibroblasts (MEFs) in an ESC growth medium (ESCM) that consisted of Dulbecco's modified Eagle's medium (DMEM)/F12, 20% Knockout serum replacement (KSR), 0.1 mM β-mercaptoethanol, 1 mM l-glutamine, nonessential amino acids (Gibco, Grand Island, NY, http://www.invitrogen.com) and 4 ng/ml FGF-2 (R&D Systems Inc., Minneapolis, http://www.rndsystems.com). The undifferentiated state of ESCs was maintained by routinely removing morphologically identifiable differentiated colonies and periodic quality control screening with Oct4 and stage-specific embryonic antigen 4 (SSEA4) immunostaining. Cells were also routinely screened with VenorGeM (Sigma-Aldrich, St. Louis, http://www.sigmaaldrich.com) for mycoplasma contamination, which can dramatically shift differentiation of hESCs.
Neural Differentiation of Primate ESCs.
Neuroectodermal differentiation of hESCs was modified from the procedure described by Zhang et al. . ESCs, grown on an irradiated MEF layer (Fig. 1A), were enzymatically separated from the MEFs and grown as aggregates in suspension in ESCM for 4 days to initiate the differentiation process. The ESC aggregates were then conditioned in a chemically defined neural medium (NM): DMEM/F12, nonessential amino acids, 2 μg/ml heparin for stabilization of FGFs, and the neural cell supplement N2 (Gibco) for an additional 2 days before attachment to a plastic or glass culture surface (Fig. 1B). Two days later, aggregate attachment was induced with a laminin-treated substrate (20 μg/ml in culture medium at 37°C for 12 hours for plastics or polyornithine-coated glass coverslips). As an alternative, aggregate attachment can also be induced with a pulse application of 10% fetal bovine serum (FBS) for 12 hours without reducing the subsequent yield of neural cells. Adherent cultures were maintained in NM with or without 20 ng/ml FGF2. Attached aggregates flattened over 1–2 days, and columnar cells developed and formed neural tube-like structures with the timing described (Fig. 1C, 1D). For cultures older than 15 days, columnar cells in the center of colonies were enzymatically separated and grown as free-floating spheres in NM with FGF or RA and then reattached to a polyornithine/laminin substrate. To investigate whether different neural differentiation protocols affect neural gene expression, hESC aggregates were also cultured without induced attachment in ESCM or NM. Rhesus ESCs were differentiated into neuroepithelial cells with the same approach; however, ESC aggregates were kept in ESCM suspension for only 3 days.
For neuronal differentiation, neuroepithelia cells (14–16 days) were lifted and grown for several days as aggregates and then replated on a polyornithine/laminin substrate in neuronal differentiation medium consisting of neurobasal medium (Gibco), N2, nonessential amino acids, supplemented with brain-derived neurotrophic factor (10 ng/ml; Peprotech, Rocky Hill, NJ, http://www.peprotech.com), glial cell line-derived neurotrophic factor (10 ng/ml; R&D Systems), cAMP (1 μM; Sigma-Aldrich), ascorbic acid (200 μM; Sigma-Aldrich), and laminin (20 μg/ml; Sigma-Aldrich). Cells were stained after 4 weeks of differentiation from hESCs.
RNA Isolation and Reverse Transcriptase Polymerase Chain Reaction
RNA extraction, reverse transcription, and polymerase chain reaction (PCR) amplification are described elsewhere , for quantitative PCR (qPCR) iQ SYBR green Supermix and MyiQ (Bio-Rad, Hercules, CA, http://www.bio-rad.com) were used. All primer sets were tested with MEF cDNA to rule out the possibility of any amplification of mouse transcripts as the result of any small amount of potential carry-over of MEF cells, especially at the earlier time points. The positive control for mRNA not expressed in samples such as α-fetoprotein (AFP) was hESCs that were grown in the presence of 10% FBS without MEF to induce random differentiation. For quantitative PCR, melt and standard curves for each primer set were generated to confirm that only one amplicon was generated at the same efficiency as the housekeeping gene GAPDH. Relative Pax6 and Sox1 gene expression levels were determined with the comparative CT method with each cDNA sample run in triplicate. See supplemental online Table 1 for a complete primer list.
Immunocytochemistry and Microscopy
Adherent cells on glass coverslips were rinsed with phosphate-buffered saline (PBS) and fixed with 4% paraformaldehyde for 20 minutes. Cells for Sox1 staining were additionally fixed and permeabilized with 100% ice-cold methanol for 20 minutes. Cells were treated with 0.2% Triton X-100 and 10% normal goat serum (Gibco) for 30 minutes before they were incubated at room temperature for 1 hour or at 4°C overnight in primary antibodies: BF1 (1:5,000; gift from Lorenz Studer, Sloan-Kettering Institute), Pax6 (1:5,000), Pax7 (1:2,000), Lhx2 (1:20), Hoxb4 (1:20; Developmental Hybridoma Bank), FABP7 (AB9558; 1:1,000) and Sox1 (AB5768; 1:500; Chemicon, Temecula, CA, http://www.chemicon.com), Sox2 (MAB2018; 1:1,000) and Otx2 (AF1979; 1:2,000; R&D systems), N-Cadherin (sc-8424; 1:1,000; Santa Cruz Biotechnology Inc., Santa Cruz, CA, http://www.scbt.com). Secondary fluorescent antibodies were used at 1:1,000 for 1 hour at room temperature (AlexaFluor 488 and 594; Molecular Probes Inc., Eugene, OR, http://probes.invitrogen.com). Coverslips were mounted with mounting medium (Immunotech, Luminy, France, http://www.beckmancoulter.com/products/pr_immunology.asp) and visualized with a Nikon TE600 fluorescent scope (Nikon, Tokyo, http://www.nikon.com) equipped with a SPOT camera and software (Diagnostic Instruments, Sterling Heights, MI, http://www.diaginc.com). Specificity of the antibodies was first tested in embryonic brain and spinal cord tissues. Negative controls without primary antibody were performed in all experiments to monitor nonspecific staining.
Fluorescence-Activated Cell Sorting
Cells were trypsinized and washed with a fluorescence-activated cell sorting (FACS) buffer (PBS, 0.1% NaN3, 2% fetal bovine or normal donkey serum), fixed, and permeabilized with ice-cold 0.1% paraformaldehyde for 10 minutes and 90% methanol for 30 minutes. Cells were incubated overnight in primary antibodies including a normal mouse IgG control (Upstate, Charlottesville, VA, http://www.upstate.com) at 1 μg of antibody per 1 million cells. Cells were treated with immunomatched Alexa 488 secondary antibodies for 2 hours, washed, and sorted with a Becton Dickinson FACSCaliber and analyzed with CellQuest Pro (BD Biosciences, San Diego, http://www.bdbiosciences.com).
RNA samples collected from H9 and H1 were collected at the following times: hESCs (day 0), 6-day aggregates, and 10- and 17-day neuroepithelia. All samples were compared to Universal Human Reference RNA (Stratagene, La Jolla, CA, http://www.stratagene.com). All RNA amplification, fluorescent labeling, array hybridization, scanning, scoring, and cataloging online was performed by microarray specialists in Patrick Brown's lab in Stanford, CA (detailed protocols can be found at http://cmgm.stanford.edu/pbrown/). All array raw data are available at Stanford Microarray Database (http://smd.stanford.edu), and Gene Expression Omnibus (accession number GSE5609). The gene chips contained 43,000 oligonucleotide targets representing almost all of the 30,000 characterized and unknown genes that make up the human genome. Briefly, after manually marking and removing from analysis any compromised spots on the array, all elements with a fluorescent intensity 1.5× higher than background were subject to analysis, and all genes with 70% or more of the samples that fit these requirements were analyzed. After selecting genes with a threefold difference in expression between any individual samples, Microsoft Excel (Microsoft, Redmond, WA) was used to collapse fold change measurements across samples and rank the order of genes on the basis of the largest fold change between stages of interest. Standard derivation (SD) was calculated for each gene in repeated time points, and the genes with SD > 1 were removed. All subsequent cluster analysis and statistics was performed with TreeView (http://rana.lbl.gov/) Ingenuity software (Mountain View, CA) and Microsoft Excel.
Neuroectoderm Differentiation from hESCs Progresses Through Morphologically Distinct Stages
The process of neuroectodermal differentiation was examined under adherent colony culture, which permits direct live observation of cellular changes. After the hESCs (Fig. 1A) were separated from MEFs as free-floating aggregates (Fig. 1B), they were cultured for 6 days before attachment to a feeder-free surface. The attached aggregates formed individual colonies of monolayer cells 1–2 days later, with increased cell density and compaction at the center. After 10–11 total days of differentiation from ESCs, cells began to elongate and lined up radially to form distinct columns of cells, which were morphologically distinct from the peripheral flat cells that outlined the clusters of columnar cells (Fig. 1C). Continued differentiation for an additional 4–5 days (total, 14–16 days) resulted in the further compaction of cells and formation of defined ridges of columnar cells. These ridges of columnar cells often formed rings with a distinct inner lumen, a structure reminiscent of the neural tube (Fig. 1D). We thus refer to these cellular structures as “neural tube-like rosettes.” With each differentiation (15 passages), 50 colonies were counted on the basis of the morphology described herein, and 95%–99% of colonies took on these morphologies every differentiation.
To test whether exogenous FGF2 is necessary for neural cell formation, the entire afore-described differentiation was also carried out without addition of exogenous FGF2. The only noticeable morphological difference was a decrease in the overall colony size, with less piling up of cells in the colony center, making neural tube-like rosettes less distinct at 15 days in the absence of FGF2. We then tested whether adhesion to a substrate affects neural differentiation. Cultures in which aggregates were grown in suspension for 15 days and then plated on a laminin-coated surface resulted in colonies with the same morphology as in Figure 1D, suggesting that the formation of these neural tube-like rosettes is largely timing dependent. Thus, in vitro differentiation of neuroepithelia from hESCs follows a temporal course with clearly identifiable morphological features.
Downregulation of ICM Markers Precedes the Appearance of Columnar Epithelia
To track differentiation, reverse transcriptase (RT)-PCR analyses were conducted on RNA samples taken from the above described four stages (days 0, 6, 10, and 15). ICM markers such as FGF4 , LBP-9 , ZFP42 (or Rex1) , and TDGF1 (human homolog of Cripto) , and the traditional ES cell markers Oct4, Nanog, and Sox2, were readily detected in hESCs as well as hESC aggregates at day 6, but became undetectable after 10–11 days of differentiation (Fig. 1E). FGF5, a gene not expressed until the epiblast cell stage in mouse embryology studies , was specifically upregulated in ESC aggregates at 6 days of differentiation and was then turned off upon further differentiation (Fig. 1E). PCR analysis throughout differentiation also identified the absence of AFP (endoderm marker) and only slight Brachyury (mesoderm marker) expression (Fig. 1E), signifying that the culture conditions bias the cells toward differentiation to an ectoderm cell fate.
Neural Gene Expression Coincides with Appearance of Columnar Epithelia in Rosettes
Genes expressed in the developing neural plate and tube were analyzed to identify the onset of a neural fate in differentiating cells. The earliest pan-neuroectodermal transcription factor, Sox1, was robustly expressed after 15 days at the neural tube-like rosette stage and had only a slight detectable trace at day 10 (Fig. 1E, 1F). In contrast, Sox2, which plays important roles in both ICM cells  and in later neural differentiation  was highly expressed by cells throughout differentiation (Fig. 1E). Similarly, Sox3 was also present throughout differentiation but showed the highest expression at 10 days (Fig. 1E). Another early neural transcription factor, Pax6, was detected at a high level by day 10 and continued expression after 15 days of differentiation (Fig. 1E, 1F). This is somewhat surprising because in other vertebrate studies, Pax6 is usually expressed by neuroepithelial cells after neural tube formation (i.e., after Sox1 expression in the neural plate) [23, 24]. Quantitative PCR analysis confirmed the robust change in expression of Pax6 before Sox1 (Fig. 1F). Zic1, a zinc finger protein expressed initially throughout the neural plate and, at later stages, the dorsal portion of the entire neural tube , exhibited the same temporal expression pattern as Pax6 (Fig. 1E). Churchill, which is induced by FGFs and is important during neural induction for inhibiting cells from a mesodermal fate , also showed a peak in expression at 10 days. N-Cadherin, a cell adhesion molecule that plays a critical role in neural tube formation, was detectable in ESCs but upregulated at 10 days of differentiation in columnar cells (Fig. 1E). This temporal expression pattern of neural genes also remained the same regardless of the presence of exogenous FGF2.
In accordance with the PCR results, immunocytochemical examinations indicated that the neural transcription factors Pax6, Sox1, and Zic1 were undetectable in undifferentiated hESCs (Fig. 2A, 2B; not shown). Pax6 expression was first observed as early as day 8 in scattered individual cells at the same time when the vast majority of cells were positive for Otx2 (Fig. 2M), a homeodomain protein expressed in the epiblast and subsequently anterior regions of the neural tube . By day 10, Pax6 was expressed throughout the columnar cells in the center of colonies, but was diminished or absent in the surrounding flat cells (Fig. 2E, 2F). This widespread Pax6 expression in columnar cells persists in neural tube-like rosettes at 17 days (Fig. 2I) and beyond. In contrast, Sox1 protein did not reach detectable levels until the neural tube-like rosette stage. Like Pax6, Sox1 was expressed in the columnar epithelial cells in the neural tube-like rosettes but not the surrounding flat cells at day 17 (Fig. 2J). Sox2 immunostaining revealed high expression in neuroepithelia at 10 and 17 days, particularly in the columnar rosette cells (Fig. 2H, 2L). However, its expression in hESCs as well makes it an unsuitable stand-alone definitive marker of neuroepithelia (Fig. 2D). N-Cadherin demonstrated a shifting expression pattern with diffuse, punctate staining at the ES cell stage and a distinct membranous staining after 10 days (Fig. 2C, 2G). As cells begin to compact and form neural tube-like rosettes at 17 days, N-Cadherin expression became concentrated in cells surrounding the forming lumen, displaying the same type of polarity seen in vivo in the neural tube  (Fig. 2K, 2L). FACS analysis of cells immunostained for Pax6 revealed that 54% of cells already expressed Pax6 at 8 days, whereas 95% expressed Pax6 by 11 days (Fig. 2N, 2O). Thus, the temporal expression pattern of early neural-associated genes suggests that the earliest neural progenitors appear between days 6 and 10, when cells take on a columnar morphology. One notable aspect of this well-conserved timeline is the early Pax6 expression before Sox1, which is a reversal of what has been observed in chick and mouse embryology studies.
Expression of Early Neural Markers Is Time-Dependent
To determine whether expression of neuroectoderm transcription factors and the unique sequence of Pax6/Sox1 expression is the result of specific culture environments, we examined their expression when cells were cultured continually in suspension or in non-neural inductive conditions. Pax6 mRNA was detected in cells after 10 days regardless of whether ES aggregates were kept in ESCM or switched to NM (Fig. 1G). In ESCM, ESC aggregates attached to plastic cultureware after 10 days, and the resulting cells did not resemble the neural tube-like rosettes after 17 days, indicating a nonfavorable neural inductive condition. This was further supported by a weak expression of Pax6 and a lack of Sox1 at 17 days. In NM, Sox1 was detected at day 17 even when cells were grown continually as free-floating aggregates (Fig. 1G). The initial expression of Pax6 by 10 days is timing dependent, and the sequence of Pax6 and Sox1 expression may be inherent to primate neural development because differentiating rhesus ESCs show the same sequence in gene expression (Fig. 1H).
Global Gene Expression Changes Confirm the Onset of Neural Fate when Columnar Epithelia Appear
To further characterize cell stages and identify potential molecular pathways involved in the progression from undifferentiated hESCs to neuroepithelia, we used microarray analysis. Six hundred fifty-four unique sequences showed at least a threefold change in expression between at least two of the four cell stages (undifferentiated ESCs, 6-day ESC aggregates, columnar cells at 10 days, and neural tube-like rosettes at day 17). Cluster analysis confirmed that the largest shift in gene expression (147 induced, 251 repressed) occurred between the day-6 hESC aggregate stage and the day-10 columnar cells (Fig. 3A; full gene list appears in supplemental online Table 2). This robust shift in global gene expression matched earlier results with known ESC/ICM and early neural markers, signifying that neuroectoderm fate is specified during this period. The columnar epithelial cells that express Pax6 and other neural genes are most likely the earliest stage of neuroepithelia.
The most prominent genes up regulated in a pattern similar to Pax6 were a collection of forebrain-associated genes known to interact with Pax6 and serve crucial roles in eye development (Fig. 3B; Table 1). Six3 and Lhx2 are both transcription factors known to bind Pax6  and are expressed early in the anterior portion of the central nervous system [30, 31], while Dachshund and Meis1/2, like Pax6, are proposed members of a group of genes important in eye development [32, 33]. Finally, Fabp7 is abundantly expressed in human fetal brain, but not other tissues . In rat cortical development, Fabp7 can be induced by ectopic Pax6 expression, and a knockdown of Fabp7 results in a decrease in proliferation and an increase in neuronal differentiation of neuroepithelia . In addition to PCR confirmation of the unique temporal gene expression of many of the genes identified through array analysis (Fig. 3B), immunostaining was used to confirm that these proteins such as Fabp7 were uniformly expressed in the neuroepithelia, much like other consensus early neural genes. By 17 days, FABP7 is widely expressed in nearly all cells, including those outside the distinct rosette clusters (Fig. 3D). The high incidence of Pax6-associated gene induction suggests a potential role for this set of genes in the widespread specification of forebrain neuroepithelia from hESCs.
Table Table 1.. Genes upregulated as columnar epithelia form at 10 days
Along with forebrain-associated transcription factors, another class of genes with unique expression patterns at all cell stages were cadherins. Cadherin 1(E-Cadherin) and 3 (P-Cadherin) were expressed at their highest levels early at the ES and 6-day aggregate stage, whereas Cadherin 6 (K-Cadherin, type 2) was at its lowest levels in ES cells. After 10 days of differentiation, at the earliest neuroepithelia stage, there was an increase in Cadherin 11 (type 2) and protocadherins 9, 16, and 17. Protocadherin 19 was not significantly upregulated until the neural tube-like stage after 17 days of differentiation (Table 1). Immunostaining for N-Cadherin throughout differentiation demonstrated that, in addition to changes in overall cadherin levels, cadherin protein localization changes as cells take on unique morphologies at each stage (Fig. 2C, 2G, 2K, 2L).
Analysis with Ingenuity software designed to identify activation of canonical pathways in large array datasets highlighted the Wnt/β-Catenin signaling pathway. Wnt5B, Frizzled 1 and 4, TLE4, and TCF7L2 were all upregulated between 6 and 10 days of differentiation (Fig. 3B; Table 1). In addition, there were several other notable genes with the same expression pattern that have previously been implicated in early neural development, including FGF9 and Sox5 and -9 [36, –38]. The specific upregulation of Wnt pathway elements and FGF9 during this stage suggests the involvement of these elements in the initial phase of human neuroectoderm specification. Additionally, nonmetastatic gene A (NMA) also known as BMP and activin membrane-bound inhibitor (BAMBI) was upregulated through the first 10 days of differentiation (Table 1; Fig. 3B).
In addition to the upregulation of neural development associated genes noted previously herein, it is important to acknowledge that the changes in gene expression between 10- and 17-day samples were the smallest, reinforcing the notion that the neuroectodermal fate is specified at around day 10. However, the high incidence of tumor suppressor-related genes such as TU3A and ST18 upregulated at 17 days (Table 2; Fig. 3B) may point to key differences between these neural tube-like cells and the earlier neuroepithelia at day 10.
Table Table 2.. Genes upregulated between each distinct cell stage
Regional Identity of hESC-Derived Neuroepithelia Can Be Controlled
Newly Formed Neuroepithelia Exhibit Anterior Characteristics.
Global changes in gene expression suggested that differentiating cells expressing the earliest neural markers were predominantly expressing genes associated with the anterior portions of the developing nervous system. Immunocytochemical analyses indicated that Pax6, Lhx2, BF1 (FoxG1B), and Otx2 were expressed in nearly all the nuclei of columnar epithelial cells, but not the peripheral flat cells at day 10 (Fig. 4A–4C). At the same time, posterior regional markers such as Hoxc8 and Hoxb4 were not expressed (Fig. 4D), suggesting that these first-stage neuroepithelia are uniformly forebrain in nature.
Initial Anterior Nature of Primitive Neuroepithelia Is Independent of Morphogens.
To determine whether the widespread induction of anterior-associated patterning genes was the result of exogenous FGF2 treatment, we omitted FGF2 from NM during culture. The overall morphology, timing, and extent of expression of early neural markers Pax6 and Sox1 and anterior markers such as Otx2 and Lhx2 were the same when FGF2 was omitted from cultures, as was the absence of posterior markers (data not shown). Although application of FGF2 had little apparent influence on anterior-posterior (A-P) patterning, cultures exposed to 20 ng/ml FGF2 on days 6–11 of differentiation did display an increase in cells positive for the dorsal neural tube markers Pax7  (supplemental online Fig. 1A, 1B) and Pax3  (data not shown) across all colonies.
To determine whether initial regional identity of neuroepithelia can be altered, we added the posterioralizing signal RA to differentiating cultures. Addition of 0.1 μM RA to cultures on days 6–8, during the same time window in which cells normally first turn on Pax6, blocked the formation of columnar cells at day 10 (supplemental online Fig. 1C) and resulted in no neural tube-like rosettes at 15 days and beyond.
Primitive Forebrain Neuroepithelia Can Be Maintained or Respecified to a Caudal Fate.
Culture of cells without RA led to maintenance of an anterior fate marked by widespread Otx2 expression and a lack of posterioralized cells in all rosettes (Fig. 4E, 4F). To confirm whether the anterior fate is maintained with further differentiation, we cultured these anterior cells for an additional week in neuronal differentiation medium until progenitors formed neurons. The majority of neurons (βIII-tubulin+) retain Otx2 expression in the nuclei even though some more mature neurons began to lose Otx2 expression. Neural progenitors, not labeled for tubulin, remain positively labeled for Otx2 (Fig. 4G).
Because the primitive neuroepithelia are uniformly forebrain in identity and early addition of RA disrupts neural differentiation, we asked whether the regional identity can be respecified. Addition of RA at day 10 did not disrupt the characteristic columnar neuroepithelia morphology and neural tube-like rosettes formed normally as in control conditions (Fig. 4H, 4I). RA treatment for 8 days or longer resulted in most cells in all rosettes expressing the posterior neuroepithelia marker Hoxb4, whereas only a few cells maintained Otx2 expression (Fig. 4H, 4I). Continued differentiation in neuronal differentiation medium and RA resulted in neurons (βIII-tubulin not shown), some of which maintained posterior markers such as Hoxb4, but did not express the anterior marker Otx2 (Fig. 4J). Finally, we asked whether the anterior fate of the neural tube-like rosette cells can be similarly redirected to a more caudal fate. Addition of RA at day 15–17 (neural tube like stage) resulted in slight downregulation of Otx2 and mild increase in Hoxb4 expression (not show), similar to our previous report . Thus, the neuroepithelia that emerge first are anterior in nature and can be respecified to a caudal fate only if morphogens are provided in the correct time window.
In contrast to the vast majority of reports that mouse and human ESCs rarely differentiate to cells with a forebrain identity [41, 42], we have developed a chemically defined system in which the initial primitive neuroepithelial cells exhibit a uniform anterior identity. The anterior nature of the primitive neuroectodermal cells can be maintained or respecified to expected caudal positional identities with particular morphogens, and this respecification occurs only during differentiation of the primitive neuroepithelia to definitive neuroepithelia. This finding lays down a conceptual and technical foundation for neuroectodermal and neural subtype specification in mammals, particularly humans.
During embryonic human development, the inner cell mass of a blastocyst organizes into two distinct layers, the epiblast and hypoblast, which are established at the start of the second week of gestation. By the end of the third gestational week, the neural plate has formed and begun to fold to form the neural tube [43, –45]. These morphological hallmarks, which are conserved across mammalian development, have been faithfully recapitulated in our defined system if hESCs are considered equivalent to the inner cell mass (5–6-day-old embryo). In the first week of differentiation, cells express all the ESC/inner cell mass markers. However, epiblast markers, particularly FGF5, are turned on during this period but are then turned off right before cells begin to express neural markers, suggesting that cells at this stage are analogous to the epiblast. This interpretation is also bolstered by the generation of non-neural lineage cells when RA is added on days 6–8, indicating a wider developmental potential of cells at this stage. The formation of neural tube-like rosettes by Sox1-expressing columnar epithelial cells after 2 weeks of differentiation bears a striking resemblance to neural plate/tube formation in an embryo at the third gestational week (Fig. 5A). Using this system, we have discovered that neuroectodermal specification may, in fact, take place at the beginning of the second week of differentiation culture, equivalent to the beginning of the third week of gestation. This time window is presumed to be the point leading up to and including primitive streak formation or gastrulation, which has also been implicated as the earliest identified point of neural induction in chick development . This stage is marked by a relatively homogeneous population of columnar epithelia, expression of Pax6 and anterior patterning genes like Bf1, Otx2, and Lhx2, and a complete lack of pluripotent gene expression. Together with the difference in their differentiation potential from the Sox1-expressing neural-tube-like cells (discussed later herein), we propose to refer to this unique cell population as primitive anterior neuroectoderm/neuroepithelia (PAN).
Previous studies in mouse ESCs have also identified a distinct primitive neuroepithelial cell type. van der Kooy et al.  identified a very rare (0.2%) primitive neural cell type with distinct growth factor requirements and the potential to form colony-forming neural stem cells as well as contribute to chimeric mice when injected into morula-stage embryos. The most obvious difference between these mouse primitive neuroepithelial cell types and the human PAN identified here is the speed with which they form. Nestin, the earliest identified neural precursor marker in mice, is turned on in less than 24 hours after disassociation of mESCs  and Sox1 expression peaks between 3 and 5 days . Here, we demonstrate that Pax6 is one of the first neural markers expressed by differentiating human cells; similar to nestin expression in mouse cells, it precedes Sox1 expression. Whether the human PAN generated here also maintains the developmental potential to contribute to chimeric mice is ethically untestable; however, the loss of expression of the pluripotency genes Oct4 and Nanog in cells at this stage suggests that it is unlikely. The use of primitive neuroectoderm in this context is intended to define the first cells possessing morphological characteristics and expressing known neural markers from differentiating hESCs and that these cells have a wider neural lineage potential than the Sox1-expressing definitive neuroectodermal cells.
The widespread anterior phenotype of primitive neuroepithelia generated in this differentiation scheme is very striking in that neuroepithelial cells differentiated from mouse (reviewed in ) and human  ESCs using RA or stromal cell cocultures are generally fated to mid-/hindbrain or more caudal lineages. Sasai et al.  have generated telencephalic precursors from mouse ESCs with similar anterior characteristics (BF1+, Six3+) under serum-free embryoid body-like conditions. As high as 30% of the telencephalic precursors can be differentiated from mouse ESCs when Wnt signaling is blocked by DKK1. In contrast, virtually all of the initial neural precursors carry anterior markers in a similar serum-free condition without the presence of exogenous morphogens, suggesting an intrinsic difference in tendency to produce anterior versus posterior cells between the two species. The early neuroepithelia cells generated in our present study appear to be a transient population and become regionalized along further differentiation to definitive Sox1-expressing neuroepithelial cells (Fig. 5B). The formation of an early population of neuroepithelia uniformly expressing anterior transcription factors that can be either maintained or shifted largely to a caudal fate with RA supports Nieuwkoop's model of activation/transformation . This model of neuroectoderm induction/patterning has been the basis for our ability to efficiently direct hESCs to spinal motor neurons and midbrain dopamine neurons [12, 14], and we believe it will be instrumental in specifying many other neural subtypes.
The robust neural differentiation in the absence of exogenous signaling molecules, even FGFs, could be used as evidence for the “default” hypothesis of neural induction . There was an increase in BAMBI (NMA), which functions as a cell surface level BMP inhibitor . Perhaps equally important, there were significant increases in other pathways postulated to play an accessory role in neural formation, including the Wnt signaling elements Wnt5B, Frizzled 1 and 4, TLE4, and TCF7L2, as well as FGF9. Finally, the prevalence of cadherin molecules that were both negatively and positively expressed as cells move to a primitive neural fate are potential mediators of neural differentiation through a cell adhesion-based mechanism that has previously been implicated in mouse cells . Additional experiments are needed to determine whether the neural differentiation seen here is directly inductive or selective expansion.
Formation of neuroepithelia along a defined timeline with identifiable morphological stages that correspond to distinct changes in gene expression all under completely defined culture conditions make this system an attractive model for the study of human neural induction and patterning. Although many of the genes identified in PAN cells may be important for neural specification, they may also play an important role in destabilizing the pluripotent ESC state. There is significant overlap between the genes upregulated at 10 days identified here and the list of genes that are directly repressed in human ESCs , perhaps most notably Pax6 and Meis1/2, which are also upregulated in mouse neural precursors . Dissecting regulatory elements from global gene expression will shed light on how hESCs initially choose a neural fate. Along the same lines, further study of the genes specifically upregulated after 17 days of differentiation, including the numerous tumor suppressor genes, as well as the striking redistribution of adhesion molecules, might give insight into the mechanisms used to solidify neural patterning fate. Although previous work has suggested that definitive neuroepithelia progenitors can be expanded for at least eight passages in the presence of FGF2 , future work aimed at isolating and expanding stage-specific precursors, including plastic PAN cells, will be important for making hESC-derived neural cells viable for therapeutic applications.
Disclosures of Potential Conflicts of Interest
The authors indicate no potential conflicts of interest.
We thank Patrick Brown of Stanford University and L. Studer for the Bf1 antibody, and the staff at WiCell for preparation of MEF for ES culture. This work was supported by the National Institutes of Health (RO1 NS045926; S.-C.Z.) and the Waisman Center (M.T.P.). X.C. is currently affiliated with Biopharmaceutical Sciences, University of California-San Francisco, San Francisco.