Programming embryonic stem cells to neuronal subtypes

Abstract

Richness of neural circuits and specificity of neuronal connectivity depends on the diversification of nerve cells into functionally and molecularly distinct subtypes. While efficient methods for directed differentiation of embryonic stem cells (ESCs) into multiple principal neuronal classes have been established, only a few studies systematically examined the subtype diversity of in vitro derived nerve cells. Here we review evidence based on molecular and in vivo transplantation studies that ESC-derived spinal motor neurons and cortical layer V pyramidal neurons acquire subtype specific functional properties. We discuss similarities and differences in the role of cell intrinsic transcriptional programs, extrinsic signals and cell-cell interactions during subtype diversification of the two classes of nerve cells. We conclude that the high degree of fidelity with which differentiating ESCs recapitulate normal embryonic development provides a unique opportunity to explore developmental processes underlying specification of mammalian neuronal diversity in a simplified and experimentally accessible system.

Introduction

One of the tantalizing features of the central nervous system (CNS) is the precision with which nerve cells establish stereotypical neural circuits. Ongoing characterization of the developing CNS is revealing an unprecedented degree of neuronal diversity that corresponds with and likely underlies the richness and specificity of neuronal connectivity and function. Subtype diversification of spinal motor neurons, retinal amacrine cells, cortical layer V pyramidal neurons or olfactory sensory neurons underlies functionally relevant differences in axonal projections, dendritic arborization, and molecular and electrophysiological properties of these neurons [1-5]. How such neuronal diversity is established during mammalian development remains poorly understood in part due to technical challenges inherent to the analysis of complex and heterogeneous CNS tissue under poorly accessible experimental conditions. Recapitulation of normal neural development with pluripotent embryonic stem cells (ESCs) in vitro might therefore become a powerful and indispensable tool in the field of developmental neurobiology. It provides an unparalleled experimental access to the developing mammalian CNS previously enjoyed only by researchers studying oviparous vertebrates and lower organisms.

Major advances have been reported in directed differentiation of pluripotent cells into distinct classes of nerve cells (reviewed in [6-9]). These studies firmly established that differentiating pluripotent cells respond correctly to developmental signals that pattern the developing CNS along the rostro-caudal or dorso-ventral axes and give rise to nerve cells expressing congruent region specific markers and neurotransmitter phenotypes. In contrast, less attention has been paid to whether in vitro generated neurons also acquire refined subtype specific properties. This is largely due to the lack of understanding of mechanisms underlying subtype diversification of mammalian nerve cells in vivo. Modeling the acquisition of subtype specific nerve cell identities in a simplified in vitro system might therefore provide a powerful tool to dissect the roles of cell intrinsic genetic programs, extrinsic diffusible signals, and cell-cell interactions in the fine-grained diversification of developing nerve cells.

Motor neuron subtype diversity

Subtype diversification of neurons is particularly apparent in the motor system. Coherent transmission of motor signals relies on the existence of hundreds of spinal motor neuron subtypes, each communicating with a different muscle group in the periphery (Figure 1A-C). Developmentally encoded motor neuron subtype diversity prescribes not only the connectivity between motor neurons and their muscle targets, but also between motor neuron subtypes and efferent connections from cognate proprioceptive sensory neurons and from descending motor pathways transmitting commands originating from the motor cortex [4,10]. The presence of more than five hundred muscle groups in the mammalian body implies the existence of similar diversity of motor neuron subtypes.

Figure 1. Functional diversity of cortical projection neurons and spinal motor neurons.

A) Spinal cord motor neurons acquire distinct columnar identities depending on their rostro-caudal position and Hox gene expression profile. Motor neuron columnar identity determines the cell body position and general muscle target specificity of motor neurons: medial motor column (MMC) neurons innervate axial muscles, hypaxial motor column (HMC) cells innervate hypaxial muscles, lateral motor column (LMC) neurons innervate limb muscles and preganglionic motor column (PGC) neurons innervate sympathetic ganglia.

B) LMC and MMC neurons can be distinguished not only by the differential cell body position (lateral vs medial) and axonal projections (limb vs. axial muscles), but also by differential expression of FoxP1 and Lhx3 transcription factors.

C) LMC neurons are further subdivided into the Lhx1⁺ lateral (l) and Isl1⁺ medial (m) divisions, harboring motor neurons innervating dorsal and ventral limb muscles, respectively. Each division is further parceled into motor pools that innervate distinct limb muscles. For example, LMCm contains a Scip expressing motor pool innervating flexor carpi ulnaris (FCU) muscle and Pea3 expressing pool innervating cutaneous maximus muscle. LMCl also contains a Pea3 expressing motor pool, but this one is dedicated to innervate latissimus dorsi (LD) muscle.

D) Cortical projection neurons found within layer V of the neocortex target distinct regions of the CNS. Cortical motor neurons (blue) project axons to pons and spinal cord. Neurons from the auditory cortex (grey) project axons to inferior colliculus while neurons from the visual cortex (orange) project to pons and superior colliculus. Schematic adapted from [6].

E) Layer V cortical neurons are subdivided into Scip, Ctip2, and Fez1 expressing corticofugal and Lmo4 expressing callosal projection neurons. Corticofugal pyramidal neurons acquire diverse areal subtype identities that define and stabilize their general axonal trajectories (D). Finally, neurons in individual cortical areas are further functionally diversified. For example, corticospinal layer Vb neurons in the motor cortex innervate distinct spinal cord segments (e.g. cervical vs. thoracic), forming a refined somatotopic map within the motor cortex.

Specification of somatic motor neuron subtypes can be deconstructed into several developmental steps (Figure 1A-C, Figure 2). First, a subset of spinal motor neurons is instructed by non-canonical Wnt signals to acquire the identity of axial muscle innervating median motor column (MMC) neurons [11,12]. Second, spinal motor neurons are diversified along the rostro-caudal axis by retinoic acid (RA), Wnt and Fgf signals that induce differential patterns of Hox gene expression along the spinal cord [13,14]. Non-MMC neurons at limb level spinal cord express Hox6 and Hox10 genes, respectively, and acquire the identity of limb innervating lateral motor column (LMC) neurons. Non-MMC neurons in the thoracic and cervical spinal cord acquire the identity of hypaxial motor column (HMC) neurons that innervate body wall muscles [4]. While subtype diversity of MMC and HMC neurons has not been systematically examined, diversification of LMC neurons into distinct subtypes is well-documented. In response to a local RA source, LMC neurons are subdivided into medial and lateral LMC divisions that innervate ventral and dorsal limb muscles, respectively [15,16]. Finally, LMC neurons within each division are organized into motor pools, groups of neurons dedicated to innervating a single muscle group in the limb [17]. Recent studies indicate that the establishment of motor pool diversity relies on the Hox-based transcriptional network that operates intrinsically in nascent motor neurons [4].

Figure 2. Generation of neuronal subtype diversity from ESCs.

A) Specification of diverse classes of nerve cells from ESCs depends on the spatial and temporal patterning of differentiating cells. Spatial patterning of the neuroepithelium along the rostro-caudal (R-C) and dorso-ventral (D-V) axes in response to diffusible patterning signals (RA, Wnt, FGF, Shh, BMPs) results in the formation of molecularly distinct neural progenitor domains, including neocortical progenitors (rostral, dorsal identity) and motor neurons progenitors (caudal, ventral identity). Temporal patterning further increases the diversity of nerve cells generated from each progenitor domain (e.g. early born layer V cortical neurons and spinal motor neurons vs. late born oligodendrocytes and layer II/III pyramidal neurons).

B) Principal classes of nerve cells are diversified into numerous neuronal subtypes endowed with specialized functional properties. Spinal motor neurons acquire distinct columnar identities controlling their general axonal projections (in part controlled by Wnt4/5 signals), limb innervating LMC neurons acquire divisional identities (LMCm vs. LMCl, in response to paracrine RA signaling) and motor pool identities (controlled by intrinsic Hox regulatory network). Analogously, cortical layer V projection neurons can be subdivided into Va callosally and Vb corticofugally projecting neurons. Layer Vb neurons are further diversified according to their rostro-caudal and medio-lateral position within the neocortex (in response to FGF gradient and afferent inputs). Finally, layer Vb neurons within each cortical area are further diversified to achieve the fine topographic functional organization characteristic of the mature neocortex.

Specification of subtype identity in ESC-derived motor neurons

Detailed understanding of spinal motor neuron subtype diversity in vivo provides a platform to examine whether similar diversification of motor neurons can be achieved in vitro from differentiating ESCs. Initial methods for differentiation of ESCs to motor neurons relied on two principal patterning signals –RA that induces neuralization and caudalization of ESCs and sonic hedgehog (Hh) that directs ventralization of the spinal neural progenitor cells [18]. ESC-derived motor neurons express bona fide motor neuron markers such as Hb9, Isl1/2, and Lhx3 transcription factors, biosynthetic enzyme choline acetyltransferase (ChAT), and vesicular acetylcholine transporter (VAChT) indicative of their cholinergic identity [18]. In addition, ESC-derived motor neurons develop electrophysiological properties typical of spinal motor neurons and form synaptic terminals that elicit clustering of acetylcholine receptors when co-cultured with myotubes [19,20].

Motor neurons derived under RA/Hh conditions acquire the identity of cervical level spinal motor neurons characterized by the expression of Hoxc4 and Hoxa5 genes [18,21,22] (Figure 1A). Motor neurons within the cervical spinal cord acquire MMC and HMC identities that can be distinguished by the persistent expression of Lhx3 in MMC neurons while this transient motor neuron marker is rapidly downregulated in HMC cells [12,23]. The majority of ESC-derived motor neurons expresses Lhx3 [18,24] indicating that in vitro derived motor neurons are of uniform MMC-like subtype identity. Interestingly, when transplanted into the developing chick neural tube, ESC-derived motor neurons innervated axial, hypaxial as well as limb muscles [18]. This result has been interpreted as a failure of MMC motor axons to navigate to their proper axial muscle target via a nerve branch overloaded by supernumerary transplanted axons [24,25]. Consistent with this view, transplantation of a limited number of motor neurons resulted in a more uniform projection of axons along the axial nerve branch [24]. As in vivo, the choice of ESC-derived motor neuron axons to innervate the axial nerve branch appears to depend on increased level of FGFR1 expression by MMC neurons [26,27].

Despite the uniform expression of Lhx3 in nascent ESC-derived motor neurons, we recently observed that a significant proportion of motor neurons downregulates Lhx3 expression and acquires expression profile characteristic of HMC neurons when cultured following dissociation (S. Nedelec and H. Wichterle, unpublished observation). Consistent with this observation, retrograde tracing of limb or hypaxially projecting engrafted ESC-derived motor neurons confirmed their HMC-like identity and lack of Lhx3 expression [21]. Therefore, we favor the conclusion that, similar to cervical spinal cord motor neurons, ESC-derived neurons acquire both MMC and HMC identity, but HMC neurons are initially masked by an extended duration of Lhx3 expression.

Examination of RA/Hh differentiated motor neurons failed to reveal whether in vitro generated cells are capable of acquiring highly specialized subtype identities that would enable them to engage in functional motor circuits. The general lack of molecular markers that would differentiate between cervical motor neurons innervating different muscle groups precludes this level of analysis. To overcome this limitation, retinoid independent differentiation conditions were recently developed that give rise to more caudally positioned motor neurons of brachial and thoracic identities [21]. In contrast to RA dependent generation of Hox4/5 expressing cervical motor neurons (likely by direct regulation of these Hox genes by RA), differentiation of caudal motor neurons relies on the patterning events driven solely by endogenous signaling centers formed in differentiating embryoid bodies. Embryoid bodies cultured in the absence of extrinsic factors express caudalizing signals of the Fgf and Wnt families as well as the ventralizing signal Hh providing a milieu conducive for the induction of Hox6, Hox8 and Hox9 genes and for the specification of brachio-thoracic motor neurons (Figure 1A, 2A).

Consistent with their Hox expression profile, a subset of ESC-derived motor neurons acquires LMC identity of limb innervating motor neurons marked by high-level expression of transcription factor FoxP1 [21,23]. Moreover, ESC-derived FoxP1 expressing motor neurons migrate into the lateral spinal territory occupied by the endogenous LMC neurons and project axons correctly into the limb when transplanted into the developing spinal cord. Finally, ESC-derived LMC neurons acquire subtype identity of Pea3 expressing cutaneous maximus and Scip expressing flexor carpi ulnaris motor pools (Figure 1C) and, as in vivo, expression of Pea3 depends on a permissive GDNF signal [21,28]. Together these studies provide compelling evidence that ESCs can be differentiated into nerve cells that acquire not only generic but also some of the most refined subtype specific properties characteristic of neurons in vivo.

Differentiation of ESCs in vitro does not result in a single preferred or “default” neuronal subtype, but in a mixture of subtypes, not unlike the intrasegmental diversity of motor neuron subtypes observed in vivo. Such diversity could result from an intrinsic genetic program similar to the one controlling specification of diverse neural cells from a common progenitor in the developing Drosophila [29,30], or it could be an outcome of a diversification of a uniform population of “naïve” motor neurons in response to cell-cell interactions and paracrine signals. ESC differentiation in vitro provides a convenient system in which normal neighbor-neighbor relationships and extrinsic cues can be easily disrupted. Dissociation and mixing experiments demonstrate that specification of motor pools does not rely on extrinsic cues [21] and therefore is likely driven by an intrinsic genetic program [31]. In contrast, divisional identity of mouse ESC-derived motor neurons can be influenced by the extrinsic RA signal [21] analogous to the specification of divisional identity in the developing chick spinal cord [15].

Considering that motor pool diversification is controlled by a cell-intrinsic program, it is unlikely that differentiation protocols relying solely on extrinsic signals will ever produce a homogenous neuronal population. It remains to be determined whether combination of extrinsic programming with intrinsic modification of the transcriptional network might overcome this limitation and yield more uniform population of motor neurons. Such findings will have a positive impact on the utilization of ESC-derived neurons for biochemical and clinical applications in which defined and uniform populations of neurons are desirable. In future, it will be critical to determine whether human ESCs (hESCs) follow similar developmental principles and whether they can be coerced to differentiate into diverse and well-defined motor neuron subtypes. In contrast to mouse cells, many hESC-derived motor neurons express caudal brachial marker Hoxc8 under the RA/Hh differentiation protocol raising the possibility that some of them might be of LMC character [32,33].

Neocortical neuronal subtype diversity

While a strict developmental control of neuronal subtype diversity is a sensible way to ensure reproducible and reliable transmission of signals between the CNS and periphery, central connectivity might benefit from a greater degree of plasticity. Accordingly, specification of neuronal subtype diversity in the developing neocortex does not rely only on developmental programs and local paracrine signals, but also on patterns of innervation by efferent axons [34].

Neocortex develops in the rostral and dorsal aspect of the neural tube. Principal classes of neocortical neurons are generated in a stereotypic temporal sequence and settle according to their birth-date and molecular identity within distinct cortical layers [35]. The first born neurons are Cajal-Retzius cells occupying the marginal zone, corticothalamic neurons that settle in layer VI are generated next, followed by layer V pyramidal neurons, layer IV neurons, and Layer II/III pyramidal neurons. Neurons within each layer are further diversified. Based on their projections, layer V pyramidal neurons are subdivided into layer Va callosally projecting neurons (expressing Lmo4) and layer Vb corticofugal neurons (expressing Scip, Ctip2, Fez1) [1,36] (Figure 1E). Ongoing molecular studies indicate a further degree of intralaminar diversity among layer V neurons that yet remains to be correlated with defined functional attributes [37].

Furthermore, layer V pyramidal neurons are diversified into distinct cortical area subtypes according to rostro-caudal and medio-lateral position within the developing neocortex (Figure 1D). For example, corticofugal layer V pyramidal neurons in the visual (occipital) cortex express Coup-TF1 transcription factor and project axons to pontine nuclei and optic tectum (superior colliculus), while layer V pyramidal neurons in the motor cortex express Diap3, Igfbp4, and Crim1 and project axons to pontine nuclei, red nucleus, and the spinal cord [1]. The diversity of pyramidal neurons likely extends further, as distinct somatotopic areas of motor cortex contain corticospinal motor neurons targeting different spinal cord levels (Figure 1E), motor columns, and motor pools [38-43].

Programs controlling diversification of layer V neurons are not well understood (Figure 2B). Arealization of the developing neocortex is initiated by locally secreted patterning signals, including rostrally secreted Fgfs [44]. However, unlike in the developing spinal cord, genetic programs initiated by these patterning gradients are not the ultimate determinants of cortical neuron subtype identity. Postmitotic cortical neurons retain a great degree of plasticity and their ultimate specification critically depends on thalamocortical and other efferent inputs, as revealed by lesion and transplantation studies (reviewed in [45,46]). For example, grafts of late embryonic occipital cortex to postnatal somatosensory area acquire the architecture and molecular characteristics of barrel cortex and grafts into motor area result in the retention of corticospinal axonal projections that are normally eliminated in visual layer V neurons (reviewed in [34]). How intrinsic genetic programs are integrated with extrinsic functional inputs to generate the extraordinary level of subtype diversity of cortical neurons remains poorly understood.

Specification of principal classes of neocortical neurons from ESCs

Neural tissue induced in the absence of caudalizing signals in the developing embryo acquires rostral forebrain identity [47,48]. The same principle applies to the patterning of ESC-derived neurons - in the absence of exogenous factors, differentiating ESCs preferentially acquire rostral neural identity [49-57]. Neuralization of ESCs under these conditions critically depend on endogenously expressed Fgf5 signal [58] and can be improved by blocking residual BMP, Nodal, and Wnt signaling pathways [51,59] or by controlling the initial size of ESC aggregates [56].

Retinoid free differentiation protocols most closely conform to the principles of neural induction derived from studies of developing vertebrate embryos. Accordingly, ESCs differentiated as free floating embryoid bodies [51], adherent monolayers [50,60] or cocultures with stromal cells [61] under serum free conditions generate primarily forebrain cells expressing marker FoxG1 (also known as BF1). ESC-derived forebrain progenitors are patternable along the dorso-ventral axis by application of Wnt3a, Shh antagonist cyclopamine, or Shh (Figure 2A) and give rise to neocortical progenitors expressing Pax6 and Emx1 or basal forebrain progenitors expressing Nkx2.1, respectively [51,55,57,60,62]. The patterned progenitors differentiate into classes of forebrain neurons that display characteristics of neurons found in the rostral, caudal and dorsal regions of the cortex [51,56,57,60], ventral basal ganglia [51,57,62], hypothalamus [55], olfactory bulb [56], and optic vesicle [63,64].

Neocortical progenitors derived in vitro from ESCs not only express correct combination of molecular markers, but are also capable of arrangement into three-dimensional neural rosettes, whose cytoarchitecture is analogous to the developing neuroepithelium [56]. When differentiated into neurons, ESC-derived cortical progenitors generate all principal classes of cortical neurons found in vivo and their birth-date sequence conforms to the temporally progressive specification of cortical plate neurons observed in vivo [56,60,65]. ESC-derived cortical progenitors differentiate first into Reelin and calretinin positive layer I Cajal Retzius neurons; followed by Tbr1 positive layer VI neurons; Ctip2 and Emx1 expressing layer V neurons; and finally to layers II and III neurons expressing Brn2 and Satb2 [56,60]. Although neurons of all cortical layers are generated in vitro, the proportion of early born deep layer neurons is significantly greater compared to the later born more superficial neurons. Considering how similar is the program of ESC differentiation to cortical development, it might help to define intrinsic genetic programs activated in maturing progenitors and to identify paracrine signals contributing to the temporal patterning of the developing neocortex [35,66].

Regional subtype diversity of in vitro derived cortical pyramidal neurons

Efficient derivation of deep layer cortical nerve cells, many of which exhibit properties of corticofugal layer V pyramidal neurons, raised a question whether in vitro generated pyramidal neurons are largely of one particular subtype identity, a mixture of different identities, or whether they retain plasticity and will acquire their final identities only in response to local cues and efferent innervation as has been demonstrated for primary embryonic cortical tissue.

To examine this question, Gaspard and colleagues implanted ESC -derived neurons into frontal (∼ motor) cortex area in neonatal mice. One month later, engrafted ESC-derived neurons were observed to innervate targets that are typically innervated by visual and limbic cortices (visual and limbic thalamic nuclei and superior colliculus), whereas no innervation was detected in the pyramidal tract innervated by motor cortex [60]. These axonal pathfinding preferences indicate that the majority of in vitro generated corticofugal neurons acquire the identity of occipital and limbic cortical neurons and that their identity was not respecified in response to local cues and efferent inputs upon transplantation into the motor cortex. In light of the regional plasticity of late embryonic visual cortex grafted into the somatosensory area [67] this is a surprising result. Unfortunately, control grafts using primary cortical tissue were not performed by Gaspard et al. and therefore it remains possible that subtle technical differences might underlie the different outcome of the two studies. Alternatively, in vitro derived cortical neurons may fail to acquire the ability to respond properly to efferent inputs due to a defect in their programming. Or in vitro conditions provide signals that initiate the specification of visual and limbic cortical identities and terminate the critical period during which cortical identity can be respecified in heterotopic grafts. Resolving between these possibilities is of great interest as it may provide mechanistically important insights into the process of cortical neuron subtype specification.

Consistent with their predominantly visual cortical identity, in vitro generated cells express Coup-TF1, a marker of the occipital cortex [56,60]. Interestingly, this regional molecular identity can be altered by modulating FGF signaling, implicated in the anterior patterning of the developing neocortex [44]. Treatment of ES cells differentiating as embryoid bodies with Fgf8 results in nearly complete suppression of Coup-TF1 expression while blocking of endogenous Fgf signaling increases the number of Coup-TF1 positive cells [56]. Although these results demonstrate that pre-patterning of the neocortex into different regions can be recapitulated in vitro, neurons generated under the different conditions were not subjected to homotopic or heterotopic transplantations. Thus, it remains to be seen whether neurons derived under these conditions would innervate targets corresponding to their pre-patterned identity or whether they would exhibit subtype plasticity and innervate targets according to the site of transplantation.

Most recently, ES cells were efficiently converted into cortical neurons by co-culture with MS5 stromal cell line [61]. Majority of neurons generated under this condition express Ctip2 and Otx1, markers of deep layer V and VI corticofugal neurons. Interestingly, transplantation of these pyramidal neurons into motor, somatosensory and visual cortex resulted in striking differences in axonal pathfinding preference. Grafts in visual cortex did not extend axons to the pyramidal tract or spinal cord while a substantial number of axons from ES cell derived neurons grafted into the motor cortex were detected in both of these target areas of endogenous cortical motor neurons [61]. These observations are consistent with the idea that regional subtype identity of cortical pyramidal neurons is at least partially plastic and can be influenced by the site of transplantation. It remains to be determined whether differences in the culture conditions (stromal co-cultures vs. monolayer differentiation in the presence of cyclopamine) might underlie differential plasticity of in vitro derived cortical cells. In this regard it is of interest to note that developmental plasticity of neural progenitors can be maintained by activation of Hh signaling pathway [49], raising the possibility that inhibition of endogenous Hh signaling by cyclopamine in Gaspard's study might cause a loss of a residual pool of patternable progenitors.

The areal diversity of layer V pyramidal neurons and of their axonal projections can be compared to the diversity and axonal selectivity of motor neuron columnar subtypes. Analogously, the next important step will be to determine how pyramidal neurons acquire distinct subregional identities, akin to spinal motor pool subtypes, that correlate with refined patterns of cortical neuronal connectivity and that underlie the formation of precise topographic maps evident in cortical areas.

Conclusions and Future Directions

The qualitative leap forward in our ability to differentiate ESCs under conditions that recapitulate normal embryogenesis provides an opportunity to study mammalian neural development in an accessible and convenient system capable of producing virtually unlimited supply of neural cells. This review focused on spinal motor neurons and cortical pyramidal neurons, two classes of nerve cells that were recently demonstrated to acquire refined subtype specific identities in vitro. However, advances in programming numerous other classes of nerve cells exhibiting aspects of subtype diversity, including retinal cells [63,64,68], forebrain interneurons [51,57,62]; hypothalamic neurons [55], midbrain dopaminergic neurons [69-71], inner ear hair cells [72,73] and cerebellar neurons [74-77], indicate that this area of research is rapidly expanding. Characterization and experimental manipulation of class and subtype diversity of nerve cells generated under these protocols will advance our understanding of neuronal differentiation and will help to define molecular mechanisms controlling the complex and precise neuronal connectivity established during CNS development.