Deriving excitatory neurons of the neocortex from pluripotent stem cells

Abstract

The human cerebral cortex is an immensely complex structure that subserves critical functions that can be disrupted in developmental and degenerative disorders. Recent innovations in cellular reprogramming and differentiation techniques have provided new ways to study the cellular components of the cerebral cortex. Here we discuss approaches to generate specific subtypes of excitatory cortical neurons from pluripotent stem cells. We review spatial and temporal aspects of cortical neuron specification that can guide efforts to produce excitatory neuron subtypes with increased resolution. Finally, we discuss distinguishing features of human cortical development and their translational ramifications for cortical stem cell technologies.

INTRODUCTION

Deriving excitatory neurons of the cortex in vitro from cultured stem cells has been an active field for roughly twenty years. Initial approaches primarily used prenatal cortical tissue as the source of cells, which were grown in vitro with growth factors and other molecules to make neurospheres (Laywell et al., 2000; Ostenfeld et al., 2002; Reynolds et al., 1992; Tropepe et al., 1999) or adherent stem cell cultures (Conti et al., 2005). While these approaches have been useful for studying neural stem cell biology (e.g., (Mira et al., 2010; Nagao et al., 2008)), it is uncertain whether these neural stem cells have the potential to generate all types of excitatory cortical neurons. Using embryonic or other pluripotent stem cells to produce neurons may offer a solution to this potential limitation.

The recent advent of induced pluripotent stem (iPS) cell technology offers researchers the opportunity to study the properties of any human cell type with any genetic background, including neurons predisposed to diseases of the nervous system. Pluripotent cells capable of differentiating into any cell type can be generated from somatic cells by inducing the expression of key transcription factors that define the embryonic stem cell state (Hanna et al., 2007; Okita et al., 2007; Park et al., 2008b; Takahashi et al., 2007; Takahashi and Yamanaka, 2006; Wernig et al., 2007; Yu et al., 2007). iPS cell lines have been generated from patients exhibiting a range of nervous system diseases, including amyotrophic lateral sclerosis (ALS, Lou Gehrig’s disease), spinal muscular atrophy, Parkinson’s disease, Huntington’s disease, Down’s syndrome, familial dysautonomia, Rett syndrome, and schizophrenia (Brennand et al., 2011; Dimos et al., 2008; Ebert et al., 2009; Hotta et al., 2009; Lee et al., 2009; Marchetto et al., 2010; Nguyen et al., 2011; Park et al., 2008a; Soldner et al., 2009). In some cases, researchers have used iPS-derived neurons from disease vs. control patients to study in vitro disease mechanisms and treatments (Brennand et al., 2011; Ebert et al., 2009; Lee et al., 2009; Marchetto et al., 2010; Nguyen et al., 2011).

To date, there are only a few examples of patient-derived iPS cell lines for neurological diseases whose etiology involves cerebrocortical dysfunction (Brennand et al., 2011; Hotta et al., 2009; Marchetto et al., 2010; Park et al., 2008a). Given the complexity of the nervous system, analyses of disease phenotypes of iPS-generated neurons can be challenging, particularly if specific types of neurons are differentially sensitive to the mutation. For in vitro modeling of cortical diseases to be meaningful, we suggest that researchers should methodically produce specific subtypes of nerve cells, or even neural circuits, that are most relevant to the disease of interest. .

In this Review, we provide an overview of recent progress in deriving cortical excitatory neurons from embryonic stem (ES) and iPS cells and discuss the developmental principles upon which cortical neuron derivation strategies can be based. Additionally, we will cover recent discoveries in human cortical development that impact our approaches to recapitulate human cortical neurogenesis in vitro.

CURRENT PROGRESS IN CORTICAL NEURON DERIVATION

A brief summary of how excitatory neurons are generated provides an essential context for understanding pluripotent cell in vitro differentiation. The neurons of the cerebral cortex can broadly be divided into two categories – projection neurons that transmit signals to other cortical regions or subcortical targets using the excitatory neurotransmitter glutamate, and interneurons that regulate local circuitry using the inhibitory neurotransmitter GABA. The inhibitory neurons are not generated locally, but instead originate in the subpallium (ventral telencephalon) (Wonders and Anderson, 2006). They then tangentially migrate into the dorsal telencephalon (the pallium), which mostly consists of the immature cortex. The excitatory neurons are produced from the cortical neuroepithelium, which consists of radial glial stem cells (RG) (Kriegstein and Alvarez-Buylla, 2009). During neurogenesis, RG undergo asymmetric divisions to produce self-renewed RG cells and neuronally committed daughter cells (Malatesta et al., 2000; Miyata et al., 2001; Noctor et al., 2001) (see Fig. 1d). Through successive rounds of cell division, RG produce the diverse subtypes of cortical excitatory neurons; deep layer neurons, that project to subcortical targets, are generated early, whereas upper layer neurons, that make intracortical projections, are generated later (Hevner et al., 2003; Shen et al., 2006; Takahashi et al., 1999). Newly generated neurons use RG cell fibers to radially migrate, from their place of origin at the ventricular surface past earlier born neurons to take up their cortical position (Noctor et al., 2001; Rakic, 1974). In this manner, cortical neurons assemble into their characteristic ‘inside-out’ laminar pattern of neuronal subtype distribution. Some of these features of cortical neurogenesis can be reproduced in vitro as pluripotent cells differentiate into neural stem cells and then to excitatory neurons.

Figure 1. Pathways for generating cortical excitatory neurons from pluripotent cells in vivo and in vitro.

a) Pluripotent cells of the inner cell mass in the embryonic blastocyst are thought to differentiate into cells of the anterior neuroectoderm in the absence of any instructive signals through a series of default fate decisions. Shown in red are morphogens that promote alternative differentiation fates. Shown in green are factors that inhibit those morphogens and facilitate the default pathway.

b) The anterior (rostral) neuroectoderm gives rise to the telencephalon upon closure of the neural tube. Dorsal-ventral (D–V) patterning is driven by dorsally expressed Wnts and ventrally expressed SHH, which converge on the activity of Gli3 to determine whether FGF signaling will be restrained or promote ventralization.

c) The division of the adult cortex into specialized functional areas develops in response to embryonic patterning signals, produced at discrete locations, that induce gradients of expression for key transcription factors. The combinatorial expression levels of these transcription factors – Emx2, Coup-TF1, Sp8, and Pax6 – define a cell's position in relation to the rostral-caudal and dorsomedial-ventrolateral (dm-vl) axes of the embryonic cortex. The cross-repressive and co-stimulatory interactions depicted on the left have been defined genetically, and in some cases biochemically.

d) Through a series of asymmetric, self-renewing divisions, radial glial cells (RG) give rise to all the subtypes of cortical excitatory neurons in a defined temporal sequence. This neurogenesis is typically indirect, since the daughter cell is often an intermediate progenitor cell (IP) that divides again to produce two neurons (N). The markers depicted represent some subclasses of glutamatergic neurons: Cajal-Retzius cells (layer I, Reelin); corticothalamic neurons (layer VI, Tbr1); subcerebral projection neurons (layer V, Ctip2); and callosal projection neurons (layers II–III, Cux1). After neurogenesis, RG convert to gliogenic cells that give rise to astrocytes (A).

e) Producing cortical excitatory neurons in vitro from ES or iPS cells occurs in phases that resemble the natural developmental sequence. i) ES cells differentiate into anterior neural cells through default mechanisms that are promoted using biological or chemical inhibitors of the Smad and Wnt pathways. *Although exogenous FGFs promote caudal CNS fates, FGF inhibitors should not be used to favor telencephalization since proliferation of primitive neural cells requires autogenous FGF signaling (Smukler et al., 2006; Zeng et al., 2010). ii) Dorsal fates can be induced in ES-derived telencephalic cells by Wnt stimulation and/or SHH inhibition, or sometimes through intrinsic mechanisms. iii) ES-derived dorsal telencephalic cells respond to the intrapallial patterning signals present in the embryonic telencephalon. Low-density cultures intrinsically adopt a caudal cortex identity. iv) ES-derived cortical progenitor cells produce excitatory neuron subtypes in the same sequence that occurs during cortical neurogenesis in the embryo, followed by glial cell production. DAPT, a chemical inhibitor of Notch, can be used to prevent stem cell self-renewal and promote synchronous differentiation of neural precursor cells.

Here, we will summarize the recent work of groups that have developed methods for producing cortical excitatory neurons from pluripotent stem cells. Several protocols have been devised to recapitulate corticogenesis (Eiraku et al., 2008; Gaspard et al., 2008; Ideguchi et al., 2010; Li et al., 2009; Zeng et al., 2010) using either aggregate cultures or low-density adherent cultures (summarized in Table 1, and depicted collectively in Fig. 1e). We will discuss these methods separately.

Table 1.

Methods for deriving cortical excitatory neurons from pluripotent stem cells.

Group	Cell types	Telencephalic induction method (Foxg1)	Dorsal specification (Emx1, Pax6)	Area specification	Neurons derived
Sasai	mouse ES	SFEBq aggregate sort cells with   FOXG1:GFP+	intrinsic	intrinsic: heterogeneous FGF8: rostral cortex, OB block FGF: caudal cortex Wnt: cortical hem Wnt+BMP: choroid plexus	Lyr1: Rln/Tbr1/CR Lyr6: Tbr1 Lyr5: Ctip2 Lyr2–4: Satb2/Cux1/Brn2
	human ES	SFEBq aggregate	intrinsic	NC	Lyr1: Rln/Tbr1 Lyr6: Tbr1 Lyr5: Ctip2
Vanderhaeghen	mouse ES	low-density,   adherent	cyclopamine	intrinsic: visual   (caudal) cortex	Lyr1:Rln/Tbr1/CR Lyr6: Tbr1 Lyr5: Ctip2 Lyr2–4: Satb2/Cux1
Zhang, Li, Xu	human ES human iPS	embryoid bodies;   harvest neural   rosettes	intrinsic	NC	Lyr6: Tbr1/Map2 Lyr5: Ctip2/Vglutl
Weimann	mouse ES	low-density,   adherent,   MS5 feeders	NC	NC in vitro specified by graft   site in transplants	Lyr5: Ctip2/Otx1

Abbreviations: Rln, reelin; CR, calretinin; Lyr, layer; OB, olfactory bulb; NC, not characterized. References: Eiraku et al., 2008 (Sasai); Gaspard et al., 2008 (Vanderhaeghen); Li et al., 2009 (Li, Zhang); Zeng et al., 2010 (Li, Xu); Ideguchi et al., 2010 (Weimann).

Aggregate cultures

Sasai and colleagues pioneered an aggregate culture method, termed SFEBq (serum-free, floating embryoid body-like, quick aggregation), in which dissociated mouse embryonic stem (mES) cells were placed in non-adherent, round-bottom wells to undergo spontaneous neural differentiation (Eiraku et al., 2008). The cells aggregated into a spheroid and within five days remarkably self-organized into a polarized neuroepithelium, with their apical ends facing an inner lumen, and a basal deposition of laminin around the outside. To promote rostral neuralization, the cells were treated with inhibitors of Wnt and Nodal signaling during the initial period of neural specification. Further culture (without Wnt inhibitor) allowed the majority of the cells to naturally adopt a dorsal telencephalic (pallial) fate, with the majority of cells expressing the telencephalic marker, Foxg1/BF-1, and nearly all of those expressing the cortical marker Emx1 (Eiraku et al., 2008). The self-assembled neuroepithelia collapsed within days into smaller rosette structures, but the rosettes maintained some features of developing cortex, with apically polarized Pax6⁺ progenitor cells in the rosettes’ centers producing neuron subtypes in the same sequence that occurs in the embryonic cortex (Molyneaux et al., 2007). Production of layer I neurons (Reelin⁺) occurred first, subcortical projection neurons (Tbr1⁺, Ctip2⁺) second, and callosal projection neurons (Brn2⁺, Satb2⁺, Cux1⁺) later (Eiraku et al., 2008). However, these neurons were disorganized and did not assume the ‘inside-out’ laminar organization achieved by embryonic cortex, that inversely corresponds to cellular birthdate (Angevine and Sidman, 1961; Rakic, 1974; Takahashi et al., 1999). The SFEBq rosettes apparently lack the elements required for radial migration and column formation. When SFEBq-derived, GFP-labeled neurons were grafted en bloc into post-natal frontal mouse cortex, axonal projections were observed in the corpus callosum, striatum, thalamus, pyramidal tract, and pontine nuclear regions after four weeks, confirming that SFEBq cultures produced a broad spectrum of cortical neuron subtypes (Eiraku et al., 2008).

Going beyond simple cortical specification, Sasai’s group investigated methods for subregionalizing the SFEBq cultures with additional morphogen treatments – the only example so far of directed intra-pallial patterning in ES-derived cells. Various manipulations of FGF, Wnt, and BMP pathway activity altered the cells’ pallial fates along rostral-caudal or medial-lateral axes, inducing regionally specific markers of rostral cortex, caudal cortex, olfactory bulb, cortical hem, or choroid plexus (Eiraku et al., 2008).

Importantly, Sasai’s group has adapted the SFEBq method of excitatory neuron production for use with human ES (hES) cells, including among other modifications a longer incubation period that reflects the protracted sequence of human development compared to the mouse (Eiraku et al., 2008; Watanabe et al., 2007). Critically, however, the human SFEBq cultures were not reported to produce any late neurons with markers of upper cortical layers, despite some being cultured for as long as 106 days (Eiraku et al., 2008).

More recently, similar results using hES and hiPS cells were obtained through a simpler embryoid body (EB)-based method, with a high efficiency of dorsal telencephalic specification (Li et al., 2009; Zeng et al., 2010). EBs were cultured without growth factors for two weeks until neural rosettes formed. Gene expression analysis showed that certain Wnt morphogens (dorsalizing signals) were strongly induced during the second week, and nearly all the neural rosette cells were Foxg1⁺/Pax6⁺ by the third week. The cells exhibited the same responsiveness to dorsoventral patterning cues (Wnt vs. SHH) that Sasai’s group originally described (Watanabe et al., 2005). The progenitor cells generated Tbr1⁺ and Ctip2⁺ glutamatergic neurons but again, the production of late cortical neurons with markers typical of upper layers was not reported.

Low-density, adherent cultures

A remarkably simple protocol for producing cortical neurons from mES cells was reported by Vanderhaeghen’s group (Gaspard et al., 2008). In this method, mES cells were plated at low density in default differentiation medium. The cells naturally adopted a telencephalic identity, but in contrast to aggregate cultures, a majority of telencephalic cells expressed ventral progenitor cell markers within two weeks and differentiated into GABAergic neurons. Noting that SHH expression was induced during the period of neural conversion, the authors treated the cells with a SHH antagonist, resulting in nearly complete suppression of ventral markers and yielding glutamatergic neurons with pyramidal morphology, indicating a dorsal fate shift. These cells also exhibited the known sequence of neuronal subtype production, with Reelin⁺ and Tbr1⁺ neurogenesis peaking first, followed by Ctip2⁺ and then Cux1 ⁺ and Satb2⁺ neurons. However, the authors also noted a large underrepresentation of Cux1⁺ and Satb2⁺ neurons when they analyzed the expected proportions of each subtype, suggesting that in vivo cues are important for the full generation of late neurons destined for upper cortical layers.

Surprisingly, the cortical cells derived by Gaspard et al. (2008) displayed specific areal identity upon transplantation into the frontal cortex of neonatal mice, extending axonal projections to a repertoire of subcortical targets that would be expected from neurons in the visual/occipital cortex. Prior to grafting, most of the mES-derived neurons expressed Coup-TF1, which is expressed in the caudal but not rostral cortex. This suggested that the cells have an innate differentiation program that requires neither intracortical (e.g., FGF, Wnt, BMP gradients) nor extracortical (e.g., thalamocortical afferents) patterning cues to acquire area-specific neuronal properties.

In another study, mES cells were plated at low density on a feeder layer of stromal cells, cultured for a week in differentiation medium with FGF2 the last two days, and then either replated for an additional week of differentiation or transplanted into various locations of postnatal mouse cortex (Ideguchi et al., 2010). Cortical specification was suggested by Otx1 expression in about half of the cells, and by mRNA detection for Foxg1 and Emx1. The differentiated cell population included Ctip2⁺ neurons; whether other glutamatergic subtypes were also produced was not addressed. The progenitor cell population was a heterogeneous mixture of cell types found throughout the rostrocaudal axis, since mRNAs for midbrain and posterior CNS markers (En1, Hoxc5, and HB9), were also detected. Compared to the feeder-free telencephalic induction performed by Gaspard et al. (2008), it seems that the FGF2 and/or stromal cells may have interfered with the cells’ innate tendency to assume forebrain identity, consistent with the known caudalizing activity of FGF2 (Cox and Hemmati-Brivanlou, 1995; Koch et al., 2009; Xu et al., 1997). Most intriguingly, however, the cells transplanted by Ideguchi et al. into various regions of the mouse cortex eventually extended axons to subcortical targets in a manner appropriate to their cortical site. This targeting plasticity contrasted with the fixed targeting potential reported by Gaspard et al. (2008), who observed projections typical of visual cortex despite the cells being grafted into the frontal cortex. We will discuss this disparity later in the section on areal plasticity.

Importantly, the low-density, adherent protocols for deriving cortical excitatory neurons have not yet been adapted for use with human ES or iPS cells. This will be a critical advance if the protocols are to become useful for understanding human cortical development and disease.

DEVELOPMENTAL GUIDEPOSTS AND CHALLENGES

The ability to study diseases of the cerebral cortex in vitro and to develop cell-based therapies will be greatly aided by the ability to produce specific neuronal subtypes from pluripotent stem cells. For example, ALS involves the degeneration of not only motor neurons in the spinal cord, but also corticospinal motor neurons (CSMNs) in layer V of the motor cortex. To obtain a pure population of wild-type or disease-background CSMNs from pluripotent cell lines will require several steps: 1) direct pluripotent cells to a telencephalic fate; 2) direct telencephalic cells to a pallial fate; 3) direct pallial cells to the subregional fate of primordial motor cortex; 4) direct motor cortex precursors to a deep laminar fate to generate and/or purify CSMNs and not other cortical projection neuron subtypes. Here we review some of the mechanisms that generate various subtypes of cortical neurons from pluripotent stem cells, drawing on developmental studies.

Telencephalic induction and Foxg1 expression

Given the default differentiation of pluripotent cells toward anterior neuroectoderm (Kamiya et al., 2011; Munoz-Sanjuan and Brivanlou, 2002; Smukler et al., 2006; Tropepe et al., 2001; Wilson and Houart, 2004; Ying et al., 2003), which gives rise to the prosencephalon or forebrain (Fig. 1a), it might seem that achieving forebrain identity from ES cells would be a trivial matter. However, factors that promote caudal CNS regional identity such as retinoic acid (RA) and FGF2 (Cox and Hemmati-Brivanlou, 1995; Durston et al., 1998; Muhr et al., 1999; Xu et al., 1997) are commonly included in neural induction protocols. Even insulin, a component of the KSR and N2 supplements widely used for ES cell culture and neural differentiation, has been reported to exert weak caudalizing activity (Wataya et al., 2008). Small amounts of these and other morphogens, whether from endogenous sources, the culture medium, or feeder cell layers, can profoundly affect neural induction and CNS regional fates. For example, Marchetto et al. (2010) derived a mixture of glutamatergic and GABAergic neurons from Rett syndrome (RTT) patient-derived iPS cells. RTT-iPS-derived neurons displayed fewer glutamatergic synapses, altered morphology, and decreased electrophysiological activity compared to neurons from control iPS cell lines. No regional markers were used to suggest whether these resembled neurons of the forebrain, midbrain, hindbrain, or spinal cord; however, the protocol utilized RA throughout weeks 2–4, which overlaps the time during which neuroectoderm derived from human pluripotent cells is posteriorized by RA to a spinal cord fate (Li et al., 2005). Given that the protocol did not include the ventralizing factor Sonic Hedgehog (SHH), the RTT-iPS-derived neurons may have most closely resembled the interneurons of the dorsal spinal cord, which include both glutamatergic and GABAergic subtypes (Kiehn, 2010). Yet the disease literature the authors referenced to corroborate their experimental observations consisted of analyses in the neocortex or hippocampus of RTT patients or animal models (Marchetto et al., 2010). Thus, the authors’ findings, while highly intriguing, are of uncertain relevance to the actual disease pathology in Rett syndrome patients.

Another matter to consider when deriving anterior neural cells from pluripotent cells is that the forebrain includes both the telencephalon and the diencephalon. The signaling pathways that promote telencephalic over diencephalic fates are not well understood, but do require induction and the function of the transcription factor Foxg1 (BF-1) (Tao and Lai, 1992; Xuan et al., 1995). Proof of Foxg1 expression can be considered a prerequisite for any claim of cortical neuron production, whether excitatory or inhibitory. The proxy use of other markers, such as Pax6 in progenitor cells or VGLUT1 in glutamergic neurons (Bibel et al., 2004), is not sufficient evidence to signify cortical specification since Pax6⁺ progenitor cells and glutamatergic neurons are present in multiple regions of the CNS. Additionally, Foxg1 itself is not strictly a telencephalic marker since it is also expressed at low levels at the mid-hindbrain boundary (Hebert and McConnell, 2000). Therefore, multiple markers are required to correctly determine CNS regional identity and exclude possible alternative fates in ES-derived neural precursor cells.

Studer’s group reported a remarkably simple method for telencephalic conversion of human ES or iPS cells (Chambers et al., 2009). hES cells were plated individually on Matrigel and cultured in conditioned ES medium with Y-27632 to prevent the death of isolated hES cells (Watanabe et al., 2007). After 3 days, the medium was switched to a differentiation medium, with Noggin and SB431542 (BMP and Activin/Nodal inhibitors) added to broadly inhibit receptor activation by ligands of the TGFβ superfamily, thus strongly preventing SMAD transcriptional activity. After just a week of differentiation, the cells were largely converted to Pax6⁺ neuroectodermal cells that were capable of neural rosette formation and expressed Foxg1 (Chambers et al., 2009). The authors did not report any attempts to produce forebrain neurons from these cells, but they did respecify the cells using established protocols to generate midbrain dopaminergic neurons, potentially of interest in the treatment of Parkinson’s disease, and spinal cord motoneurons, potentially useful for the study or treatment of ALS and spinal muscular atrophy, in a relatively short amount of time. The advantages to this method of neural differentiation are its speed, plasticity, the absence of feeder cells, the use of defined medium, the uniformity of cell fates compared to using embryoid bodies, and the total yield since cells are at high density when differentiation begins. Others have reported similar, high efficiency neural induction using the compound dorsomorphin in place of Noggin in both hES and hiPS cells (Kim et al., 2010; Morizane et al., 2011; Zhou et al., 2010).

Dorsal-ventral specification and the Wnt-SHH axis

The opposing roles of Wnts and BMPs vs. SHH in the dorsoventral specification of the telencephalon are well established (Campbell, 2003) (Fig. 1b). In the developing chick telencephalon, treating ventral explant cultures with soluble Wnt3a had a dorsalizing effect, inducing Pax6 and suppressing Nkx2.1. Using soluble Frizzled receptor to block Wnt signaling in dorsal explants did precisely the opposite, exerting a ventralizing effect (Gunhaga et al., 2003). Similar results have been demonstrated in the embryonic mouse telencephalon by manipulating the levels of cytoplasmic β-catenin, the downstream effector of the Wnt signaling pathway (Backman et al., 2005). Conditional elimination of β-catenin in neural progenitor cells caused a loss of Emx1, Emx2, and Ngn2 expression in pallial tissues that instead expressed the ventral determinants Dlx2, Ascl1, and Gsx2. These effects were only observed if β-catenin was removed before the onset of neurogenesis. Conversely, excess β-catenin expression in the subpallium repressed Dlx2, Ascl1, and Nkx2.1 and induced the expression of pallial determinants Pax6 and Ngn2 (Backman et al., 2005). Thus, in the early patterning stage, Wnt signaling is necessary and sufficient to specify dorsal fate in the telencephalon. BMP signaling is essential in specifying the most dorsomedial telencephalic structure, the choroid plexus (Hebert et al., 2002).

SHH is equally vital for ventral telencephalic specification and, in excess, can drive the expression of subpallial fate determinants in the dorsal telencephalon (Chiang et al., 1996; Corbin et al., 2000; Fuccillo et al., 2004; Gaiano et al., 1999; Kohtz et al., 1998; Shimamura and Rubenstein, 1997). At the crossroads between the Wnt and SHH pathways is Gli3, the transcription factor that represses SHH target genes in the absence of SHH (Ruiz i Altaba, 1999; von Mering and Basler, 1999; Wang et al., 2000) and is a direct target of activated β-catenin (Alvarez-Medina et al., 2008). Gli3 activity in the pallium is critical for repressing ventral fate determinants, defining the pallial-subpallial boundary, and enabling the production of dorsal organizing signals (Wnts and BMPs) from the cortical hem (Grove et al., 1998; Kuschel et al., 2003; Theil et al., 1999; Tole et al., 2000). The major requirement for SHH and its mediator Smoothened (Smo) in subpallial development is to antagonize the formation of Gli3 repressor so that pallial determinants like Pax6 that initially occupy the entire telencephalic neural tube are progressively displaced as the subpallium expands dorsolaterally from its ventromedial point of origin (Fuccillo et al., 2004; Rallu et al., 2002). This subpallial expansion depends critically on FGF signaling (Gutin et al., 2006; Storm et al., 2006), and Gli3 repressor prevents the inappropriate expansion of FGF8 expression into the pallium (Kuschel et al., 2003).

Multiple research groups have demonstrated that the mechanisms that regulate dorsoventral fate in the telencephalon similarly regulate the dorsoventral properties of ES-derived telencephalic cells (Danjo et al., 2011; Elkabetz et al., 2008; Gaspard et al., 2008; Li et al., 2009; Watanabe et al., 2005; Watanabe et al., 2007). The Foxg1⁺ cells derived from mES cells by Sasai’s group using the original SFEB method were a heterogeneous mixture of dorsally (Pax6⁺) and ventrally (Nkx2.1⁺ or Gsx2⁺) specified cells, but treatment with Wnt3a or SHH effectively enriched for one versus the other (Watanabe et al., 2005). The improved SFEBq method, designed to reduce variability between experiments, generated mES-derived Foxg1⁺ cells that almost all expressed the dorsal marker Emx1 (Eiraku et al., 2008). The biological reasons for this pronounced dorsalization are unknown, but the cells could easily be redirected to a subpallial fate using SHH or chemical agonists of the SHH pathway (Danjo et al., 2011). The low-density plating method of Gaspard et al. (2008), like the original SFEB method, produced a mixture of dorsal and ventral mES-derived telencephalic cells, but with the balance tilted toward the ventral. In this case, simply using cyclopamine to inhibit endogenous SHH signaling was sufficient to produce a dorsal fate shift.

Human ES-derived telencephalic cells, in contrast, seem to have a stricter tendency toward dorsal fates, whether derived using SFEB, SFEBq, or EB-based methods (Eiraku et al., 2008; Li et al., 2009; Watanabe et al., 2007). SHH treatment alone had a pronounced, although partial, effect on repressing Pax6 and inducing Nkx2.1 in the SFEB system for hES cells (Watanabe et al., 2007), whereas in the EB-based system the combined use of SHH and the Wnt inhibitor Dkk1 achieved a more complete shift toward ventral fates (Li et al., 2009). Li et al. (2009) began to address the molecular mechanisms behind the apparent predisposition of human pluripotent cells to adopt dorsal telencephalic fates. During the second week of differentiation, they reported robust induction of Wnt ligands and Gli3 repressor. Exogenous Wnt treatments tripled the expression of Pax6 and Emx1 and doubled the expression of Gli3 repressor. Treatment with SHH attenuated the post-translational processing of Gli3 to its repressor form. The combined effects of SHH treatment and Wnt inhibition using Dkk1 produced a nearly complete reversal from dorsal to ventral specification.

Cortical area patterning by morphogen gradients

Subregional patterning in the telencephalon develops in response to varying levels of morphogens that are secreted by signaling centers at discrete locations around the forebrain (Fig. 1c). Most notably, the cortical hem is positioned caudodorsally and secretes multiple Wnt and BMP ligands; the anterior neural ridge is positioned rostrally and secretes FGFs; and ventral aspects of the telencephalon produce SHH (Grove et al., 1998; Hebert and Fishell, 2008; Hoch et al., 2009; Ohkubo et al., 2002; Shimamura and Rubenstein, 1997; Shimogori et al., 2004; Sur and Rubenstein, 2005). Additionally, the region of the pallial-subpallial boundary expresses neuregulins, TGFα, the Wnt inhibitor Sfrp2, FGF15 and Spry2 (a repressor of Fgf-signaling) (Assimacopoulos et al., 2003; Faedo et al., 2010; Subramanian et al., 2009). FGF15 and Spry2 expression from this region are implicated in regulating patterning and proliferation of the ventral cortex through controlling CoupTF-I expression (Borello et al., 2008; Faedo et al., 2010).

In the cortex, areal patterning is established by the graded expression of key transcription factors in the dorsal telencephalon, with Emx2, Coup-TF1, Sp8, and Pax6 being most well characterized (O'Leary et al., 2007) (Fig. 1c). Emx2 is expressed in a high caudodorsal to low rostroventral gradient, and Pax6 expression is the opposite; Sp8 is expressed in a high rostrodorsal to low caudoventral gradient, and Coup-TF1 expression is the opposite. Thus, the level of expression for each transcription factor in a given progenitor cell can be used to approximate that cell’s position in the cortical grid that is defined by the rostral-caudal and dorsalmedial-ventrolateral axes. Underlying these transcription factor gradients are varying levels of the patterning morphogens – particularly Wnts, BMPs, and FGFs – secreted from the various signaling centers. For instance, Wnt and BMP signaling, through their respective effectors β-catenin and Smad proteins, induce the expression of Emx2 (Theil et al., 2002). FGF8 signaling induces Sp8 expression and represses Emx2 and Coup-TF1, and in turn, the transcription factors can regulate the abundance of the morphogens and of each other (Armentano et al., 2007; Faedo et al., 2008; Fukuchi-Shimogori and Grove, 2003; Garel et al., 2003; Mallamaci et al., 2000; Sahara et al., 2007; Storm et al., 2006; Zembrzycki et al., 2007). FGF15 opposes the effects of FGF8 (Borello et al., 2008).

Sasai’s group has made use of these developmental principles to generate cortical neurons from mES cells in a subregionally specified manner (Eiraku et al., 2008). The cortical cells produced using the SFEBq method were a heterogeneous mixture of rostral (Coup-TF1⁻) and caudal (Coup-TF1⁺) cells but could be directed to more exclusive rostral or caudal fates using FGF8 or FGF antagonists, respectively. Wnt3a and BMP4 were used to induce the expression of dorsomedial markers of the cortical hem (Otx2⁺, Lmx1⁺) and choroid plexus (TTR⁺). These experiments have pioneered the way for future efforts toward more precise control over cortical subregionalization. For instance, some of the FGF8-induced cells expressed Tbx21, a marker of olfactory bulb projection neurons, derived from the rostral-most cortex. Perhaps an intermediate FGF8 concentration could effectively rostralize the cells for motor or somatosensory cortex formation without inducing non-cortical fates. Perhaps lower levels of Wnt and BMP signaling in conjunction with FGF antagonism could produce Emx2⁺/Coup-TF1⁺ cells without inducing cortical hem markers. Testing these patterning factors over a range of concentrations and in different combinations could produce cells that are characterized not in terms of whether they express Coup-TF1, Emx2, Sp8, or Pax6, but instead in terms of how much of each factor they express, and whether these amounts correspond with known cell positions in the grid defined by the rostral-caudal and dorsomedial-ventrolateral axes of the primordial cortex. Finally, the areal identity of these cells could be characterized after neuronal differentiation in vitro and in vivo.

Many of the markers that distinguish cortical layers vary from area to area, and neurons that project to subcortical targets do so in an area-specific manner (Molyneaux et al., 2007). Such criteria may be used to assay the areal identity of ES-derived neurons. For example, in contrast to the SFEBq method that produced a rostral-caudal mixture of cortical cell types (Eiraku et al., 2008), the low-density plating method of Gaspard et al. (2008) yielded mostly caudal (Coup-TF1⁺) cortical cells. When transplanted into frontal cortex, these cells extended axonal projections to subcortical targets that would be expected of cells from the visual (caudal or occipital) cortex, rather than cells from frontal cortex. This suggested that visual cortex is the default areal identity assumed by differentiating cortical cells in the absence of extrinsic patterning signals. Although the caudal cortex fate observed by Gaspard et al. was not intentional or directed, it seems likely that these cells would be amenable to the same morphogen-driven areal patterning techniques performed with the SFEBq method. The relative uniformity of areal identity adopted by these cells suggests that low-density plating methods may be superior for precise areal specification since all cells are likely to receive equal patterning signals, whereas the cells in SFEBq or other aggregate cultures may be differentially influenced by paracrine or cell-to-cell signals from other cells within the aggregate.

The ability to generate cortical neurons with areal specificity has not yet been reported using human pluripotent cells. Creating neurons with regional identity could be very helpful for modeling or potentially treating neurodegenerative or neurodevelopmental diseases which often target specific neuron subtypes. For example, cortical neurons with a frontal lobe identity could be useful for studying diseases like schizophrenia, or fronto-temporal dementia, and creating frontal lobe cortical motorneurons could be helpful for modeling or possibly treating ALS, while temporal lobe neurons would be helpful for studying Alzheimer’s disease and other disorders of memory.

Areal plasticity of cortical progenitor cells

The need to generate cortical neurons with subregional specificity would be unnecessary for transplanted cells if environmental cues prompted the cells to assume the areal identity of the transplant site. Such plasticity was reported by Ideguchi et al. (2010), who found that transplanted cortical cells derived from mES cells eventually extended axons to subcortical targets depending on their placement, with cells placed in the motor cortex projecting to motor cortex targets, visual to visual, etc. This targeting plasticity was not reported by Gaspard et al. (2008), who observed that the cells in their transplants projected to targets typical for visual cortex neurons, despite the cells’ being grafted into frontal cortex. The reason for this difference has not been investigated, but the plasticity reported by Ideguchi et al. may relate to the cells’ age at the time of grafting, rather than being a phenotype conferred by the stromal cells used for neural induction as the authors proposed. The transplants of Ideguchi et al. were performed after only seven days of differentiation – which may be roughly equivalent to mouse embryonic day 11.5 (E11.5) since mES cells are derived from the inner cell mass of the blastocyst at E4.5 – and probably consisted mostly of neural progenitor cells. Gaspard et al. transplanted cells after 12–17 days of differentiation (equivalent to mouse ~E17 or later) when cell fates are largely determined and many cells are postmitotic. This difference of 5–10 days is probably critical since heterotopic grafts of rat E12 cortex target subcortical regions defined by the recipient graft site, whereas the targets of rat E14 grafts are defined by the cortical area from which the donor cells originated (Gaillard et al., 2003; Pinaudeau et al., 2000). Whether the respecification occurred at the level of progenitor cells or the neurons produced by them was not determined, but it seems likely that rat E12 neural progenitor cells are still capable of responding to the morphogen gradients present within in the developing cortex by adjusting their transcription factor levels, whereas the areal identity of rat E14 progenitor cells is fixed. E12 and E14 in the rat are equivalent to ~E10.5 and ~E12.5 in the mouse (Schneider and Norton, 1979), and mouse subcortical projection neurons are not produced until after E12.5 (Polleux et al., 1997; Takahashi et al., 1999). Thus, the areal identity of mouse cortical progenitor cells is probably fixed by E12.5, and the transplanted cells of Ideguchi et al. (2010) presumably had not yet reached this stage. More detailed analyses will be needed to precisely determine the stage of neural differentiation at which targeting potential becomes fixed, and to learn the molecular changes responsible for this loss of plasticity.

The plasticity of early cortical neuroepithelial cells may present an opportunity to circumvent the requirement for areal specification in vitro if cells are transplanted after dorsal telencephalic fate is fixed, but while areal identity is still plastic. However, this strategy would entail losing the ability to transplant a single neuronal subtype since early cortical progenitors will likely proceed through the known temporal sequence of neuronal subtype production – a drawback in some situations. There may also be less control over the final dose of transplanted cells since proliferation will occur after transplantation. Finally, the less differentiated and more proliferative the cells are at the time of transplantation, the greater the risk of neural overgrowth (Elkabetz et al., 2008), so the stage of neural differentiation and the expected amount of proliferation would have to be very precisely controlled and accounted for.

The challenge of generating specific subtypes of cortical excitatory neurons

While progress is being made on elucidating the transcriptional regulation of fate specification of cortical excitatory neurons (Table 2) (Arlotta et al., 2005; Leone et al., 2008; Molyneaux et al., 2009), little is known about the molecular mechanisms that govern which subtype of cortical neuron is produced by a radial glial (RG) cell division at different times during neurogenesis (Figure 1d). Here we will focus on the feasibility of producing a single subtype of neuron from progenitor cells that are programmed to produce several cell types in a defined sequence. Is there a way to fix the neurogenic potential of a RG cell at a particular point in its progression, thus allowing the cell to continuously produce a single subtype of neuron through repeated asymmetric divisions?

Table 2.

Major classes of glutamatergic neurons in the mammalian neocortex.

Neuronal subtype	Function	Location	Markers	Fate determinants
Cajal-Retzius	Regulate radial migration	Layer I	Calretinin*, Reelm*, Tbr1	Tbr1
Callosal projection neurons	Connect cerebral hemispheres. Subsets send dual projections to striatum or to specific cortical regions.	Layer II Layer III Layer V	Brn1&2, Lhx2, Cux2*, Inhba*, EphA3*	AP2γ, Bm1&2 Cux1&2 Satb2
Stellate thalamorecipient neurons	Receive projections from thalamus	Layer IV	RorB
Subcerebral projection neurons, e.g. corticotectal, corticopontine, corticospinal motor	Project to tectum, pons, spinal cord, and other targets	Layer V	Ctip2, Fezf2, Otx1, Scip*, Sox5	Fezf2 Ctip2
Corticothalamic and subplate projection neurons	Project to thalamus	Layer VI	FoxP2*, Otx1, Tbr1, Sox5	Sox5 Tbr1

^(*)Note: many markers are expressed in multiple layers; expression specific to a layer is denoted with an asterisk.

It has been observed that β-catenin signaling activity progressively declines over a spatiotemporal gradient during the neurogenic period, suggesting that differential β-catenin activation might govern the neurogenic potential of a RG cell (Machon et al., 2007; Mutch et al., 2009). Cells electroporated with constitutively active β-catenin as late as E15.5 generated neurons that occupied deeper positions and expressed early subtype markers, and cells expressing dominant negative β-catenin at E13.5 produced neurons showing the opposite effects (Mutch et al., 2009). While these experiments are difficult to interpret given the multifaceted roles of β-catenin in RG biology, they suggest that the precise regulation of β-catenin signaling activity might be one approach for regulating excitatory neuron subtype production.

Early cortical precursor cells transplanted into later stage cortex can adapt to the host environment and switch to production of upper layer neurons (McConnell, 1988; McConnell and Kaznowski, 1991). However, late cortical precursors transplanted into earlier stage cortex do not regain their competence to produce early stage neurons (Frantz and McConnell, 1996), indicating that cell-intrinsic changes in competence make the neurogenic plasticity unidirectional – the “progressive restriction” model. In terms of decreasing β-catenin activity (see above), these observations could be interpreted to suggest that, as neurogenesis proceeds, the environmental signals that stimulate β-catenin signaling decline. At the same time, the progenitor cells also become less competent to respond to higher signal levels, placing a ceiling on their potential range of β-catenin activity, which ceiling progressively lowers until neurogenesis is extinguished. Besides forced expression of constitutively active β-catenin, another molecular perturbation that can reset or ‘rewind’ the progenitor cell’s neurogenic competence involves a temporary reduction in Foxg1 expression (Shen et al., 2006).

Until we have a better understanding of the transcriptional and signaling circuitries that determine the output of RG cell divisions, we will have to use alternative approaches to achieve single neuron subtype production. The timed application of the Notch pathway inhibitor DAPT has been used to force the differentiation of all progenitor cells at a given time (Eiraku et al., 2008). This is an effective means to obtain a pure population of layer I neurons, or a mixed population of layer I and layer VI neurons, or of layers I, VI, and V, etc. depending on the timing of DAPT application. Eiraku et al. (2008) combined this method with FACS-sorting of FOXG1∷GFP⁺ progenitor cells after the production of layer I (Reelin⁺) neurons, before applying DAPT to induce differentiation into an enriched population of Ctip2⁺ neurons. This strategy could not be employed for later stages since layer I Reelin⁺ neurons are the only ones that turn off Foxg1, but the principle of sorting and induced differentiation was nicely demonstrated.

For therapeutic human cells, sorting using a transgenic cell line is problematic since non-transgenic cells are highly preferred for patient transplants. But it might be possible to prevent neurogenesis while still allowing neurogenic competence to advance, and then removing the neurogenesis barrier and driving synchronous differentiation with DAPT. When constitutively active Notch (ca-Notch) was transfected into E13.5 radial glia, the affected cells were prevented from generating neurons and instead multiplied as RG cells. When the ca-Notch was excised at E15.5 by Cre transfection, the cells then produced neurons that went to the E15.5-appropriate upper layer position, rather than resuming where they had left off at E13.5 (Mizutani and Saito, 2005). Of course, they were in the later environment after ca-Notch excision, which would allow their competence to respond to the E15.5 environment, so the authors transplanted the double-transfected cells back into E13.5 animals and observed that they still produced upper layer neurons. This proved that neurogenic competence advanced even while Notch activity was maximal, and the generation of subtypes from the E13.5–15.5 window was completely skipped. (However, it did not address whether this advance was cell-autonomous or in response to changing environmental signals between E13.5 and E15.5.) Therefore, once dorsal telencephalic identity is established in differentiating ES cell cultures, it may be possible to overload the cells with Notch ligand to prevent neurogenesis while the cells’ neurogenic competence advances. At the desired time, DAPT can be added to drive differentiation to the desired laminar subtype.

Temporal and other challenges of generating upper layer neurons in vitro

To our knowledge, excitatory neurons of upper cortical layers have not been produced from human pluripotent cell lines by directed differentiation, although this has been accomplished with mouse ES cells (Eiraku et al., 2008; Gaspard et al., 2008). As the expanded upper layers of the cortex are among the most distinguishing features of human cortex, the generation of these neurons from human pluripotent cells has potential for revealing human-specific aspects of cortical circuitry. In addition, the neurons of mid to upper layers provide intracortical circuitry that is implicated in a variety of diseases of cortical function including schizophrenia, autism, learning disabilities, and mental retardation.

When early bona fide or ES-derived mouse cortical progenitor cells were plated at low density to observe the sequential production of cortical neuron subtypes, most of the neurons produced were early-born (layers I, VI, V) subtypes (Gaspard et al., 2008; Shen et al., 2006). Only a small percentage expressed the markers of later-born (layers II–IV) subtypes, indicating that most progenitor cells did not complete the entire sequence of cortical neurogenesis, despite the presence of FGF2 to spur proliferation. These cultures did, however, proceed normally to become gliogenic after the phase of neurogenesis had ended. In contrast, upper layer neurons seemed comparatively well represented (roughly half, by our qualitative assessment of the data) among differentiated mES cells after being cultured by the SFEBq method without any growth factors (Eiraku et al., 2008). These observations suggest that some features of aggregate culture are more permissive for upper layer neuron production, while low-density culture is somewhat prohibitive.

The removal of neural stem cells from their neuroepithelial environment likely results in less efficient Notch and β-catenin signaling, which are facilitated through apically localized proteins in radial glial cells (Bultje et al., 2009; Zhang et al., 2010). Ectopic FGF2 can compensate for both of these deficiencies (Shimizu et al., 2008; Yoon et al., 2004), but not without tradeoffs. FGF2 can act as a caudalizing agent for cells whose telencephalic identities are not yet fixed (Cox and Hemmati-Brivanlou, 1995), and a ventralizing agent for those whose identities are (Abematsu et al., 2006; Bithell et al., 2008). The effects of FGF2 on patterning can take place over multiple cell cycles (Koch et al., 2009), possibly explaining why early-born neurons were correctly specified but later-born subtypes were poorly represented in the experiments of Gaspard et al. (2008) and Shen et al. (2006). It may be possible to use other combinations of mitogens and morphogens, including Notch and Wnt ligands, to maintain cortical progenitor identity in low-density cultures through the duration of the neurogenic sequence. SFEBq aggregates appear to autonomously produce the right factors in the right combinations and levels to mimic the developing cortical neuroepithelium.

Although mouse SFEBq aggregates successfully produced upper layer neurons, human SFEBq aggregates apparently did not (Eiraku et al., 2008). If human SFEBq aggregates follow a natural developmental timecourse, when might we expect upper layer neurons to be produced? By immunostaining fixed sections from human fetal cortex, we have observed the emergence of Satb2⁺ neurons in the proliferative zone by gestational week 14 (GW14), and their arrival in the cortical plate begins by GW15 (unpublished data). The clinical term ‘gestational week’ is defined by the female patient’s last menses, so GW14 actually refers to roughly the twelfth week of fetal development. Thus, going from the blastocyst embryo (the stage at which hES cells are harvested) to upper layer neuron production in the cortex requires about 75 days of differentiation. The data shown by Eiraku et al. (2008) was obtained after 45–60 days of SFEBq culture, which could explain why they did not report upper layer neurogenesis. However, they reportedly cultured some human SFEBq aggregates for as long as 106 days, and still no upper layer neurons were reported. This may reflect a deficiency in the production of intermediate progenitor (Tbr2⁺) cells, which were noticeably scarce in human SFEBq aggregates compared to mouse. In humans and in mice, Tbr2 deletion causes microcephaly (Baala et al., 2007; Sessa et al., 2008), and the deficiency in neurogenesis is most pronounced in upper layers (Arnold et al., 2008). Beyond its well-appreciated role in transit amplifying cells, Tbr2 is also required for the proper differentiation of upper layer neurons (Arnold et al., 2008).

What elements do telencephalic SFEBq aggregates lack that might impact the scarcity of Tbr2⁺ cells? Tbr2⁺ cells produce chemokines that recruit migrating interneurons from the ventral telencephalon (Sessa et al., 2010), a mechanism for balancing excitatory and inhibitory neuron numbers that may also regulate Tbr2⁺ cell numbers. Hippocampal transit amplifying cells receive GABAergic and peptidergic inputs that regulate their proliferation and differentiation (Tozuka et al., 2005; Zaben et al., 2009); the cortex may very well employ similar mechanisms. Tbr2⁺ cells also interact with the vasculature in embryonic mouse cortex (Javaherian and Kriegstein, 2009; Stubbs et al., 2009). These interactions between Tbr2⁺ cells and their environment may be more acutely required in the human cortex, which takes several weeks to accomplish neurogenesis, compared to the mouse cortex, which takes only days. In addition, there are fundamental differences in the cellular mechanisms by which human and mouse cortices produce upper layer neurons, to which we will now turn our attention.

DISTINCT MECHANISMS OF CORTICAL NEUROGENESIS IN THE HUMAN BRAIN

The developmental guideposts we have discussed for differentiating pluripotent cells to cortical neurons have been established mainly in mouse models of cortical development. The human cortex, however, is structurally more complex and thousands of times larger than the mouse. As our knowledge of human brain development increases, we should expect to encounter distinct cellular mechanisms, reflected at the level of neural progenitor cells, that facilitate the development of a larger cortex with more complex circuitry. Here we will discuss recently characterized progenitor cell populations that are thought to account for the enormous increase in cell numbers that underlies the expansion of the human cortex, and the prospects for generating these cell types from pluripotent stem cells.

oRG cells, a novel type of neural stem cell and a predominant source of excitatory neurons

In the embryonic mouse cortex, neurogenesis occurs only in the periventricular region. The radial glia (RG) that function as neural stem cells divide at the ventricular surface, producing neuronal progeny that often divide again in the subventricular zone before migrating radially to the cortical plate (Haubensak et al., 2004; Noctor et al., 2004). This neurogenic scheme imposes a limit on the number of neural stem cells that can be accommodated based on the ventricular surface area.

In primates the scheme is very different. The labeling of dividing cells in fetal monkey cortex indicated that, unlike in the rodent, significant cell division occurs in a much larger germinal region at a distance from the periventricular germinal zones (Lukaszewicz et al., 2005; Rakic and Sidman, 1968; Smart et al., 2002). This expanded germinal region is histologically distinct from other cell layers and has been called the outer subventricular zone (OSVZ) (Smart et al., 2002). Correlating the gestational time of OSVZ proliferation with classic neuron birth-dating studies suggested that OSVZ progenitor cells might contribute to cortical neurogenesis (Lukaszewicz et al., 2005; Rakic, 1974), but the nature of these progenitor cells and the cell types produced by them were not understood until recently.

We recently characterized the progenitor cells in the human OSVZ and discovered a diversity of cell types that represent distinct stages in a lineal progression from neural stem cell to transit amplifying cell to committed neuronal progenitor (Hansen et al., 2010). The founder cell for this lineage is a distinct subtype of radial glial cells, termed oRG cells for OSVZ radial glia. In many respects, oRG cells are similar to traditional radial glia, functioning as neural stem cells that give rise to neuronal progenitor cells through self-renewing, asymmetric cell divisions. oRG cells also possess a long fiber that contacts the basal lamina at the cortical surface and presumably supports neuronal migration. However, oRG cells are morphologically unipolar and have no association with the ventricular surface. Therefore, unlike ventricular radial glia (vRG), oRG cells do not exhibit the apical-basal polarity that is characteristic of bipolar neuroepithelial cells (Fietz et al., 2010). In fact, oRG cells translocate further away from the ventricle with each cell division, with the basal fiber being the key cellular feature whose retention defines which cell daughter will remain a neural stem cell (Hansen et al., 2010).

The human OSVZ appears around gestational week 12 (corresponding to post-conception week 10) and within two weeks has expanded to become the predominant germinal zone (Hansen et al., 2010). Quantifying the relative numbers of OSVZ vs. VZ cell divisions at various gestational ages in macaque (Lukaszewicz et al., 2005), and correlating those ratios with the known gestational ages when neurons for each cortical layer are produced (Rakic, 1974), suggests that the contribution of OSVZ-derived neurons is modest in layer VI, substantial in layer V, predominant in layer IV, and almost total in layers II–III. Thus, the neural stem cells that give rise to most upper layer neurons in the primate cortex appear to be the oRG cells in the OSVZ, rather than the vRG cells near the ventricle.

The differences in progenitor cell biology and modes of cell division between oRG and vRG cells could be a potential source of neuronal diversity, if neurons derived from different RG cell types possess distinctive qualities that can be traced back to the cell of origin. It is also possible that neurons derived from VZ vs. OSVZ radial glia are indistinguishable, with a neuron’s timing of origin and migration to the cortical plate being the main determinant of what subtype it will become. Whether oRG cells give rise to distinct neuron subtypes or simply represent a cellular means to amplify neurogenesis is a pivotal question with obvious ramifications for producing specific subtypes of cortical neurons in vitro from human pluripotent stem cells.

The OSVZ, a germinal niche for oRG cells

If we assume that having the correct type of progenitor cell is important for producing neurons of a desired subtype, then this introduces a new challenge for producing excitatory neurons of upper cortical layers from hES cells. As mentioned earlier, upper layer neurons may be particularly involved in human diseases of cognitive function including dementia, retardation, schizophrenia, and autism. The ability to generate cells with the correct upper layer identity may be required in order to study the pathophysiology of these disorders. The oRG cells from which most human upper layer neurons originate represent a new target cell type for hES cell derivation.

Researchers commonly use the presence of neural rosettes in differentiating cell cultures as an indication that neural stem cells with properties of neuroepithelial/radial glial cells are abundant and actively producing neurons. In ventricular RG cells, the polarity protein Par3 and the engagement of N-cadherin at apical junctions are required to promote Notch and β-catenin signaling, respectively, and thus maintain progenitor status (Bultje et al., 2009; Zhang et al., 2010). Thus, apical-basal polarity is an essential property by which neuroepithelial cells comprise a self-supportive niche that does not require distinct supporting cells. The self-organization of dissociated, differentiating ES cells into a polarized neuroepithelium attests to the intrinsic value of cell-cell interactions for neural stem cell maintenance (Eiraku et al., 2008).

Lacking cell-cell interactions of this type, what are the mechanisms by which oRG cells persist in the OSVZ? We presume that there are compensatory mechanisms in the OSVZ that activate some of the same intracellular signaling pathways through alternate means to prevent oRG cell differentiation. The basal fibers of oRG cells are likely critical for receiving signals from the environment that restrain differentiation, induce proliferation, and promote survival. We demonstrated that Notch signaling is required for oRG cell maintenance (Hansen et al., 2010), though why apical polarity is not required for this pathway in oRG cells as it is in ventricular RG is unknown. Integrin binding is also required to maintain Pax6⁺ OSVZ progenitor cells (Fietz et al., 2010). Perhaps integrin-linked kinase in oRG cells provides an alternate means to activate AKT and inhibit GSK3 (Delcommenne et al., 1998), the non-canonical pathway by which N-cadherin engagement activates β-catenin signaling in ventricular RG (Zhang et al., 2010). At any rate, it is clear that signals from extracellular sources are indispensable for oRG cell maintenance. These signals may be features that define the OSVZ as a germinal niche for oRG cells.

Evidence in both human and ferret cortex indicates that oRG cells sometimes undergo symmetric proliferative divisions, resulting in two oRG cells (Hansen et al., 2010; Reillo et al., 2010). This manner of expanding the oRG cell population requires the newly generated oRG cell to grow a basal fiber de novo, which we have observed directly (Hansen et al., 2010). It has been proposed that contact with the basal lamina at the pial surface is essential for oRG cell maintenance (Fietz and Huttner, 2011; Fietz et al., 2010). However, it is unlikely that all OSVZ-derived oRG cells are required to extend their newly grown fibers over such a great distance to maintain their identity. We propose that elements within the OSVZ are sufficient to support oRG cell function, including ligands that activate Notch and integrins. The oRG cell population has an outer limit, approximately halfway through the cortical wall, that demarcates the boundary of the OSVZ germinal region. oRG cells either cannot translocate beyond this limit or else they lose their neurogenic capacity in so doing.

What is the likelihood of reconstituting in vitro the aspects of OSVZ cytoarchitecture that are required to sustain oRG cell-driven neurogenesis? Might the OSVZ arise spontaneously within human ES cell-derived SFEBq aggregates if they can be cultured for long enough periods of time? The self-organized neuroepithelia from SFEBq-cultured hES cells, unlike those from mES cells, show a remarkable proclivity to retain an extended laminar organization rather than collapsing into smaller rosettes, even after eight weeks in culture (Eiraku et al., 2008). This suggests that they might be amenable for longer-term culture and the development of more complex cytoarchitecture. However, two structural elements of the OSVZ – thalamocortical projections and the vasculature – have extra-telencephalic origins and thus cannot be generated from within telencephalic SFEBq aggregates. Clues suggest that these OSVZ features are important for supporting the oRG cell population.

The structural framework of the OSVZ is a complex matrix of vertically and horizontally oriented cell fibers. The vertical fibers derive from ventricular and OSVZ radial glial cells. As for the horizontal fibers, the OSVZ is identical with the lower strata of the “stratified transitional field” through which thalamocortical afferents (TCAs) traverse (Altman and Bayer, 2002; Altman and Bayer, 2005). Although TCAs have been well studied for their involvement in cortical area specification (O'Leary et al., 2007), their effects on cortical progenitor cells have received little attention. Reillo et al. (2010) performed binocular enucleation of newborn ferrets to induce hypoplasia in the lateral geniculate nucleus (LGN), the portion of the thalamus that projects to the visual cortex. The next day, they observed a lower rate of proliferation in OSVZ radial glia in the visual cortex, and several weeks later, a 35–40% reduction in the size of area 17 (Reillo et al., 2010). The mechanism by which TCAs support oRG cell proliferation are unknown, although the association between β1 integrin and L1 cell adhesion molecule (Ruppert et al., 1995) is a potential means by which oRG cells and TCAs may interact.

The developing vasculature is also probably an important component of the oRG cell niche in the OSVZ. Years ago, Golgi stains showed several examples of radial glial fibers that terminate on blood vessels within the cortical wall, and some of these fibers were traced to “displaced radial glial cells” outside the ventricular zone (Schmechel and Rakic, 1979) – likely oRG cells. In similar fashion, the adult neural stem cells of the mouse lateral SVZ, which are also derived from radial glia (Merkle et al., 2004), extend basal processes to contact blood vessels in the adult brain (Mirzadeh et al., 2008; Shen et al., 2008; Tavazoie et al., 2008). The basal lamina surrounding endothelial cells is another potential substrate within the OSVZ that may engage integrins on oRG cell fibers. The vasculature may also provide soluble factors that help maintain and expand the oRG cell population, as shown for embryonic mouse radial glia (Shen et al., 2004). Finally, the vasculature may support the organization and proliferation of Tbr2⁺ intermediate progenitor cells in the OSVZ, as described in the rodent embryonic SVZ (Javaherian and Kriegstein, 2009; Stubbs et al., 2009).

The probable requirements for thalamocortical projections and the vasculature in supporting the oRG cell niche do not mean that oRG cells could not be maintained in SFEBq aggregates, but the signaling pathways involved may need to be deciphered so that exogenous supplements could substitute. One might even imagine introducing ES-derived endothelial cells or ES-derived thalamic cell aggregates into the cortical SFEBq environment to support OSVZ development. Alternatively, it may be that the complex tissue organization of the OSVZ is entirely unnecessary to support oRG cell function. Small numbers of oRG-like cells have been observed in developing mouse cortex, which lacks the OSVZ as a distinct germinal region (Shitamukai et al., 2011; Wang et al., 2011). Neurospheres derived from human fetal cortical cells have been cultured for several weeks with FGF2 and EGF, followed by plating and observing their behavior in vitro. Examples of RG-like cells with unipolar morphology and the distinctive mitotic behavior of oRG cells were observed (Keenan et al., 2010), suggesting that the combination of growth factors and neurosphere culture may be sufficient to maintain oRG cells. However, there is still the issue of how to generate oRG cells, and their mechanism of origin in vivo is not well understood.

Disorders of cortical development, such as lissencephaly and microcephaly, as well as a variety of malformations associated with abnormal patterns of gyrification may well correspond to functional abnormality of specific neural stem or progenitor cells such as ventricular radial glia, oRG cells, intermediate progenitor cells, or transit amplifying cells. The ability to study these disorders in human cells using iPS technology will depend on our ability to generate specific human cortical progenitor cell types in vitro. We have much to learn before successfully recapitulating the complex features of human cortical neurogenesis in a differentiating pluripotent cell-based system. But simple techniques or principles may emerge that, like the remarkable self-organization of cortical tissue that occurs in SFEBq cultures (Eiraku et al., 2008), will permit differentiating hES cells to recapitulate their full developmental program.

CONCLUSION

Our knowledge of pluripotent cell differentiation, cellular reprogramming, human brain development, and neurological diseases is rapidly expanding. Neurological diseases may be among the most challenging to treat with cell-based cell therapies due to the extraordinary complexity of the nervous system, but may also afford the greatest therapeutic opportunity since adult neurogenesis is limited or non-existent in most regions of the CNS. There is enormous diversity of cell subtypes in the central nervous system, and most neurodegenerative diseases, including Parkinson’s, ALS, Huntington’s, Alzheimer’s, and a range of other disorders of motor and cognitive function each target a very specific subset of neurons. In order to treat these disorders with cell replacement therapy, or to model their pathogenesis using iPS technology, we will need to generate the specific nerve cells targeted by the disease.

Our ability to direct the differentiation of mouse pluripotent cells toward specific subtypes of cortical excitatory neurons has vastly improved in recent years. However, our ability to do the same with human ES and iPS cells has lagged behind. There are multiple differences between human and mouse cortical development that contribute to the difficulty of deriving cortical neurons from human pluripotent cells, not the least of which is the hugely protracted time course of human development compared to the mouse. Events that take place over days or weeks in developing mouse brain may take months or years in human brain development.

One method that may accelerate the differentiation process may be to adopt direct reprogramming techniques. Every neuron is likely to rely on a key number of ‘terminal selector’ genes that specify its particular subtype and function (Hobert, 2008). Recent studies have used transcription factors to reprogram glial cells, inhibitory neurons, or even fibroblasts into excitatory neurons (Blum et al., 2011; Rouaux and Arlotta, 2010; Vierbuchen et al., 2010). Though the field is now in its infancy, the possibility to transdifferentiate various cell types into specific neurons may come of age in the future. Understanding the hierarchies of transcriptional circuitry for each neuronal subtype will be key to determining which transcription factors and regulatory RNAs will most successfully effect the desired reprogramming events for a given subtype. Genomic expression profiling efforts are making progress in this direction (Arlotta et al., 2005; Molyneaux et al., 2009; Nelson et al., 2006; Willi-Monnerat et al., 2008).

The incredible complexity of the adult nervous system, along with the intricate mechanisms by which neural circuits develop and refine over long periods of time, raise doubts over whether naïve neurons produced in vitro and transplanted into the dense, mature parenchyma of the adult CNS can interpret their environment in a way that allows them to integrate appropriately into existing circuits. Cells from embryonic mouse cortex transplanted into the damaged motor cortex of adult mice extend area-appropriate projections, form synaptic contacts, and become myelinated (Gaillard et al., 2007), raising hopes for our ability to repair or reconstruct damaged or diseased cortical circuitry. As our understanding in these fields continues to mature, more opportunities for therapeutic intervention and technological improvement will continue to unfold.