Shaping our Minds: Stem and Progenitor Cell Diversity in the Mammalian Neocortex

Abstract

The neural circuits of the mammalian neocortex are crucial for perception, complex thought, cognition and consciousness. This circuitry is assembled from many different neuronal subtypes with divergent properties and functions. Here, we review recent studies that have begun to clarify the mechanisms of cell-type specification in the neocortex, focusing on the lineage relationships between neocortical progenitors and subclasses of excitatory projection neurons. These studies reveal an unanticipated diversity in the progenitor pool that requires a revised view of prevailing models of cell-type specification in the neocortex. We propose a “sequential progenitor-diversification model” that integrates current knowledge to explain how projection neuron diversity is achieved by mechanisms acting on proliferating progenitors and their postmitotic offspring. We discuss the implications of this model for our understanding of brain evolution and pathological states of the neocortex.

Introduction

The mammalian neocortex contains hundreds of cell types that are assembled into neural circuits dedicated to complex tasks such as sensory perception, control of motor function, learning and, at least in some species, reasoning, conscious thought and language. Neurons and glia are the principal neocortical cell types, and neurons can generally be further classified as either projection neurons or interneurons. Projection neurons are glutamatergic excitatory neurons that primarily extend long-range axons to other cortical areas or to subcortical and subcerebral target regions (Fig. 1). Interneurons are GABAergic inhibitory neurons that usually extend shorter projections locally within the same cortical area. Other classes of neurons, such as Cajal-Retzius cells and subplate neurons, exist only transiently to guide neocortical development. Two major types of macroglial cells, oligodendrocytes and astrocytes, also contribute to neocortical function by myelinating axons and by providing regulatory functions within the neuronal networks, respectively.

Figure 1. Major subtypes of neocortical projection neurons, classified by projection pattern and laminar position.

(A) Sagittal view of corticofugal projection neurons in lower layers. Corticothalamic neurons are found mostly in layer VI and project to the thalamus. Subcerebral projection neurons are located in the lower parts of layer V and send primary projections to the spinal cord, pons and superior colliculus. (B) Coronal view of corticocortical projection neurons in upper layers. Most ipsilateral and callosal projection neurons are located in layers II-III and project within the same hemisphere or to the contralateral hemisphere via the corpus callosum, respectively. A subset of callosal neurons is also found in the superficial parts of layer V (not shown). Columnar projection neurons are found in layer IV and send short axons locally within a neocortical column. (C) Magnified view of projection neuron subtypes represented in A and B, detailing the relationships between laminar positions and projection patterns.

Within these principle classes of neocortical cell types, numerous subclasses can be defined. This has been particularly well documented for interneurons and projection neurons, but is less established for glial cells. Interneurons have been divided into four major subclasses comprising at least 19 distinct subtypes based on molecular, morphological and physiological distinctions (Miyoshi et al., 2007; 2010). The classification of excitatory projection neurons is equally complex. Because the laminar architecture of the neocortex is one of its most prominent anatomical features (Fig. 1), projection neurons have been categorized according to their radial position within the six neocortical cell layers (named layers I – VI). This concept is useful since neurons with similar connectivity patterns tend to occupy the same layer (Fig. 1). Most layer VI neurons form corticothalamic connections, whereas neurons connecting to basal ganglia, midbrain, hindbrain and spinal cord are typically found in layer V (Molyneaux et al., 2007). Layer IV spiny stellate neurons project locally within a neocortical column (Gilbert and Wiesel, 1979; Lund et al., 1979; Martin and Whitteridge, 1984; Anderson et al., 1994), and neurons in layers II and III connect the two cerebral hemispheres through projections across the corpus callosum (Fame et al., 2011). It is becoming increasingly clear, however, that these hodological and laminar criteria are not sufficient to define all subtypes of excitatory projection neurons, as neurons with similar projection patterns are often dispersed across multiple layers (Fame et al., 2011). Moreover, diverse subsets of neurons with different molecular characteristics are often found within a single layer (Molyneaux et al., 2007). We are therefore likely only in the initial stages of uncovering the full diversity within the vast numbers of neocortical projection neurons.

Significant progress has been made over the past decade in defining some of the mechanisms that establish cellular diversity in the neocortex. Perhaps one of the most important observations is that the developing central nervous system contains many spatially segregated germinal zones that generate distinct cell types. For example, the majority of neocortical interneurons are generated in the ganglionic eminences in the ventral telencephalon, from where they migrate into the neocortex along tangential routes (Welagen and Anderson, 2011). Many oligodendrocyte populations also originate outside the neocortex in several regions of the ventral forebrain (Kessaris et al., 2006). The generation of different cell types from distinct progenitors is, at first approximation, an elegant solution to a complex problem. However, excitatory projection neurons and astrocytes, which constitute the vast majority of neocortical cells, are derived locally from one germinal zone within the neocortical primordium (Hebert and Fishell, 2008; Mérot et al., 2009). The mechanisms that lead to the generation of multiple neocortical cell types from one progenitor domain are only just beginning to be understood and are the focus of this review.

Progenitor domains and radial units: integrating lineage-related neurons into functional units

The identification and characterization of progenitor cells that generate neocortical cell types has been a complex task that has puzzled scientist for centuries. Starting in the late 19^th century, classical studies by histologists such as Golgi, Magni, His and Cajal shaped the concept that the development of the neocortex depends on proliferating cells that line the lateral ventricles. These researchers studied the morphologies of cells in the developing neocortex, and His was the first to demonstrate that mitotic figures are abundant close to the ventricular surface, but relatively sparse elsewhere (for a historical perspective, see recent reviews by (Breunig et al., 2011) and (Lui et al., 2011)). In 1970, the Boulder Committee formulated the view that the ventricular zone (VZ) contains a single type of multipotent progenitor cell that generates all neocortical cell types (The Boulder Committee, 1970). They named this progenitor the “ventricular cell”. A second progenitor domain just outside the VZ was also recognized from the late 19^th century on. The Boulder Committee proposed to name this domain the subventricular zone (SVZ), and held the view that the SVZ generates a special class of neurons and all macroglia (The Boulder Committee, 1970). However, other researchers including Smart, who studied mitosis in the developing neocortex quantitatively, suggested that at least some SVZ cells are daughters of VZ progenitors and that many of them generate neurons (Smart, 1973). Others supported the view that the SVZ predominantly contains the progenitors for glial cells (Takahashi et al., 1995).

The characterization of neocortical progenitors was limited by the tools available at the time and initially relied on static images; histological snapshots of cortical development at different time points. Significant technological advances in the field included the introduction of tracers like [³H]-thymidine, which facilitated birth-dating studies, and identification of molecular markers that distinguish different cell types. These new tools led to two important extensions of the original concepts. First, birth-dating studies in rodents and primates demonstrated that, at first approximation, neurons with a similar laminar fate are born at the same time (Angevine and Sidman, 1961; Rakic, 1974). Second, studies in the monkey neocortex revealed that the VZ is a mosaic of cells expressing or devoid of glial fibrillary acidic protein (GFAP), prompting Rakic and colleagues to postulate that GFAP⁺ progenitors produce astrocytes and radial glial cells (RGCs), the latter serving as a scaffold for migrating cells, whereas GFAP⁻ progenitors generate neurons (Levitt et al., 1981; 1983).

These and other studies led to the influential radial unit hypothesis (Rakic, 1988; 1995). This hypothesis integrates the findings from many researchers and proposes that the neocortex consists of ontogenetic columns that are generated from progenitor cells near the ventricle. The daughter cells of these progenitors migrate radially along radial glial fibers into the neocortical wall, such that neurons of the same ontogeny tend to form a radial unit with related function. These proliferative units form a proto-map that is subsequently refined by thalamic inputs to establish cortical areas with distinct sizes, cellular compositions and functionalities. The number of ontogenetic columns determines the size of the cortical surface area, while cortical thickness is determined by the output of neurons from the progenitors within a column.

Achieving diversity: the progressive restriction model of cell-type specification

The identification of the cortical VZ (and SVZ) as the germinal zone for the generation of projection neurons and astrocytes has raised important questions with regard to the composition of the progenitor pool. One important clue came from birth-dating studies, which demonstrated that cortical projection neurons and astrocytes are generated in a defined temporal sequence. At early stages of neocortical development, a preplate that consists of the earliest-born neurons and possibly other cell types forms between the VZ and the meninges at the brain surface. This preplate is subsequently split into the marginal zone and subplate by waves of migrating neurons that are born in an inside-out order: lower layer VI and V neurons are born first, followed by layer IV, III, and II neurons. Finally, towards the end of neurogenesis, progenitors generate astrocytes (reviewed in (Qian et al., 2000) and (Pinto and Götz, 2007)). Two alternative models could explain the mechanism by which this temporal order is established (Fig. 2). In one model, the fate potential of a common progenitor might change over time to generate the different subtypes of projection neurons and astrocytes in a defined temporal order (Fig. 2A). Alternatively, multiple progenitor types may co-exist, each of which is intrinsically programmed or extrinsically triggered to generate a specific subclass of neurons or astrocytes on a progenitor-specific time line (Fig. 2B).

Figure 2. Different models of projection neuron cell-type specification in the developing neocortex.

(A) A single kind of RGC progenitor sequentially generates all subtypes of projection neurons and macroglia. The fate potential of the common RGC is progressively restricted over time; thus, projection neuron subtype identity is specified by birth date. (B) Distinct subtypes of RGC progenitors co-exist and are pre-specified to generate different subtypes of projection neurons and macroglia. Projection neuron subtype fate is therefore specified by progenitor type, rather than by birth date. VZ, ventricular zone; SVZ, subventricular zone; LL, lower layers; UL, upper layers; MZ, marginal zone.

Several experimental strategies have been employed to distinguish between these models. Landmark heterochronic transplantation studies in ferrets provided evidence for a common progenitor whose fate potential is restricted over time, such that it sequentially generates the different types of projection neurons in order (McConnell and Kaznowski, 1991; Frantz and McConnell, 1996; Desai and McConnell, 2000). Early progenitors, which normally produce lower-layer neurons, are capable of producing upper-layer neurons when transplanted into older host animals. Older progenitors, however, are restricted in their competence and can only produce upper-layer neurons, even in a younger host environment. These studies also indicated that by a time a neuron has progressed through its final mitotic division and initiates migration, it has acquired the information necessary to migrate to the layer typical of its birth date.

In vitro studies with cells isolated from the developing neocortex of mice lend additional support to the concept of a common progenitor that generates projection neurons in a defined temporal order. When clonal relationships were analyzed between single neocortical progenitors and their daughters, lower-layer neurons were most commonly generated after fewer cell divisions than upper-layer neurons. Progenitors from older mice also appeared to be more restricted in their ability to generate earlier-born neuronal subtypes (Shen et al., 2006). With the advent of stem cell technology, attempts have been made to generate neocortical neurons in cultures from human stem cells. The in vivo temporal order by which stem cells generate different neocortical neurons was maintained in vitro, with the caveat that neuronal subtypes for upper cortical layers were largely not detectable (Eiraku et al., 2008; Gaspard et al., 2008).

Additional support for a common neocortical progenitor stems from lineage-tracing studies using retroviral vectors. In these studies, proliferating cells in the VZ of developing mice were infected with replication-incompetent retroviruses. The viral genome, containing a marker gene such as LacZ, only integrates into the genome of the actively replicating progenitor cells. The virus and marker gene then serve as heritable tracers to allow analysis of lineage relationships between the progenitor and its offspring. These studies have provided evidence that single VZ progenitors can generate neurons destined for multiple neocortical cell layers. Clones generated from late progenitors are confined to increasingly superficial layers, indicative of progressive restriction of progenitor fate potential (Luskin et al., 1988; Price and Thurlow, 1988; Walsh and Cepko, 1988; Reid et al., 1995; 1997). Occasionally, but rarely, progenitors generate both neurons and astrocytes (Luskin et al., 1988; Price and Thurlow, 1988; Parnavelas et al., 1991; Grove et al., 1993; Luskin et al., 1993; Reid et al., 1995; McCarthy et al., 2001). Similar conclusions were reached from cell-lineage analysis using chimeric mice generated from genetically marked embryonic stem cells, in which sparse labeling of neocortical progenitor cells permitted the tracing of the fates of their offspring (Tan et al., 1998).

Retrovirus lineage tracing studies also demonstrated that neurons derived from a single progenitor can spread over substantially greater territory in tangential directions than predicted by the radial unit hypothesis (Walsh and Cepko, 1992; 1993). This suggests that radial columns, which are thought to be the primary information processing units of the neocortex, are assembled from daughter neurons generated from more than one progenitor cell. Genetic studies in mice indicate that the integration of neurons into radial columns is under control of EphA/ephrin signaling (Torii et al., 2009), thus providing a mechanism for the mixing of ontogenically distinct neurons.

Finally, Doe and colleagues have provided evidence that the concept of progressive restriction of a common progenitor applies to the nervous system of Drosophila. The competence of Drosophila neuroblasts to generate different neuronal subtypes changes over time as the neuroblasts sequentially express the transcription factors hunchback, Krupple, Pdm and Castor, which specify the production of different neurons in temporal order (Isshiki et al., 2001; Pearson and Doe, 2003; Grosskortenhaus et al., 2005).

The results from these studies in vertebrates and invertebrates are compelling and have been interpreted as support for an evolutionarily conserved mechanism for temporal restriction of the fate potential of a common progenitor type (Fig. 2A). However, recent studies provide evidence that the germinal zone of the neocortex is far more complex than initially anticipated and actually consists of several different progenitor types. In the following, we will summarize these studies and then revisit the interpretations of the results from transplantation experiments, clonal analyses, and retrovirus lineage-tracing studies in light of these recent findings.

Progenitor diversity: a panoply of progenitor types in the neocortical germinal zone

The cellular composition of the neocortical germinal zone has been studied in various species ranging from rodents to man. Here, we will focus on rodents to establish general principles, but will also refer to the far more complex primate brain. The origins of all neocortical neurons and macroglia can ultimately be traced to neural stem cells derived from the anterior neuroectoderm. These cells display many epithelial characteristics and are known as neuroepithelial cells (NECs). During neurulation, beginning around embryonic day (E) 8 in the mouse, anterior NECs undergo rapid proliferation by symmetric cell division to expand the neural stem cell population that will give rise to the forebrain (Smart, 1973). By E9, the anterior neural tube closes to form the lateral ventricles and the NECs therefore line the ventricles as a pseudostratified neuroepithelium. NECs maintain a characteristic apicobasal polarity and are anchored to one another at the ventricular surface by tight junctions and adherens junctions (Aaku-Saraste et al., 1996; Zhadanov et al., 1999; Manabe et al., 2002), and to the basal lamina at the pial surface by integrins (Graus-Porta et al., 2001; Radakovits et al., 2009).

Between E9 and E10, near the onset of neurogenesis, NECs begin to transform into a distinct progenitor type: RGCs (Fig. 3). During this transformation, NECs lose some of their epithelial properties in favor of certain glial characteristics, but retain contacts with the ventricular and pial surfaces that give them their radial morphology; hence the term RGC. Among the changes characterizing the NEC-to-RGC transition are the loss of tight junctions (Aaku-Saraste et al., 1996), the acquisition of glycogen storage granules (Brückner and Biesold, 1981; Gadisseux and Evrard, 1985), and the expression of astroglial genes such as brain lipid-binding protein (BLBP), astrocyte-specific glutamate transporter (GLAST) and tenascin-C (Hartfuss et al., 2001; Heins et al., 2002; Noctor et al., 2002). RGCs still retain many NEC characteristics, however, and the two cell types likely co-exist for some time (Götz and Huttner, 2005).

Figure 3. Subtypes of stem and progenitor cells in the developing neocortex.

Early in neocortical development, neuroepithelial cells divide symmetrically to expand the progenitor pool before transforming into radial glial cells (RGCs). RGCs typically divide asymmetrically to self-renew and produce either neurons or intermediate progenitor cells (IPCs). IPCs divide symmetrically to generate pairs of neurons, or in some cases additional IPCs that then make neurons. Short neural precursors are similar to IPCs in that they undergo terminal symmetric divisions to make neurons, but like RGCs they maintain an apical end foot and are located in the ventricular zone. Basal RGCs (bRGCs) have a basal attachment at the pial basement membrane similar to RGCs, but do not maintain an apical process and thus have their cell bodies located in the outer margins of the SVZ. bRGCs self-renew (not shown) and generate IPCs and neurons. At the end of neurogenesis, RGCs and bRGCs transform into astrocyte progenitors. For abbreviations see legend to Fig. 2.

Although it was suggested over a century ago that neurons and astrocytes are generated from mitotic cells in the VZ, it was thought that these progenitors were distinct from RGCs, which were believed to only provide a scaffold for migrating neurons (see (Noctor et al., 2002) for a historical perspective). Only recently was it discovered that RGCs are the progenitors of most neurons and macroglia in the neocortex (Malatesta et al., 2000; Hartfuss et al., 2001; Miyata et al., 2001; Noctor et al., 2001; Tamamaki et al., 2001; Noctor et al., 2002) and in other regions of the CNS (Malatesta et al., 2003; Anthony et al., 2004; Casper and McCarthy, 2006). These discoveries were facilitated by the advent of new technologies, such as live imaging to follow individual GFP-labeled progenitors and their progeny over time. Although the overlapping properties and lineal relationship of NECs and RGCs make it difficult to distinguish which cells derive directly from each progenitor type, it appears that only small populations of postmitotic cells are generated from NECs before they transform into RGCs (Noctor et al., 2002; Malatesta et al., 2003; Anthony et al., 2004; Attardo et al., 2008; Kowalczyk et al., 2009).

Unlike NECs, which typically divide symmetrically to expand the progenitor pool, RGCs tend to divide asymmetrically to self-renew and generate a non-RGC daughter cell (Iacopetti et al., 1999; Miyata et al., 2001; Noctor et al., 2001; Haubensak et al., 2004; Miyata et al., 2004; Noctor et al., 2004). Although time-lapse imaging studies have demonstrated that some of these daughter cells are postmitotic neurons (Miyata et al., 2001; Noctor et al., 2001; Miyata et al., 2004; Noctor et al., 2004), only 10–20% of asymmetrically dividing RGCs generate neurons directly (Attardo et al., 2008; Kowalczyk et al., 2009). Instead, most RGC divisions produce an RGC and yet another type of neural progenitor known as an intermediate progenitor cell (IPC) (Fig. 3) (Haubensak et al., 2004; Miyata et al., 2004; Noctor et al., 2004). IPCs are distinct from NECs and RGCs in several important ways, perhaps the most functionally relevant difference being that IPCs primarily undergo symmetric terminal divisions to produce pairs of neurons (Haubensak et al., 2004; Miyata et al., 2004; Noctor et al., 2004). Some IPCs can also undergo a limited (1–3) number of additional symmetric divisions to generate more IPCs before making neurons (Noctor et al., 2004; Wu et al., 2005), thus functioning as a class of neuron-restricted transient amplifying cells.

IPCs can be further distinguished from RGCs by their expression of the transcription factor Tbr2 and the downregulation of the RGC-specific transcription factor Pax6 (Englund et al., 2005). Additionally, IPCs lose contact with the ventricular surface and migrate to more basal positions before undergoing further rounds of mitosis, and in the process switch from the radial morphology of RGCs to a multipolar morphology (Miyata et al., 2004; Noctor et al., 2004). These basally dividing IPCs increase in number as neocortical development proceeds, thus creating an anatomically distinct proliferative region, the SVZ.

Another type of neocortical progenitor has been termed short neural precursor (SNP) on the basis of its unique morphology (Fig. 3) (Gal et al., 2006). Similar to RGCs, SNPs divide in the VZ and have a radial morphology with an apical process in contact with the ventricular surface. Unlike RGCs, however, the basal processes of SNPs do not reach the basal lamina. SNPs and RGCs differ in additional ways, including their abilities to utilize the GLAST and Tα1 promoters (Gal et al., 2006; Mizutani et al., 2007), their use of Notch downstream signaling (Mizutani et al., 2007), their cell-cycle kinetics, and the behaviors of their immediate progeny (Stancik et al., 2010). However, based on their similar neurogenic properties, it has been proposed that SNPs may represent a subset of IPCs that have not lost contact with the ventricular surface (Kowalczyk et al., 2009). Indeed, single-cell molecular profiling of neocortical cells revealed a population of progenitors in the VZ that more closely resembled IPCs than RGCs (Kawaguchi et al., 2008). Nevertheless, SNPs express Pax6 but not Tbr2 (Stancik et al., 2010), indicating some molecular differences between them and IPCs.

In contrast to the apically attached SNPs in the VZ, studies using Golgi staining (Schmechel and Rakic, 1979), DiI labeling (Voigt, 1989; deAzevedo et al., 2003), or retroviral labeling (Noctor et al., 2001; 2004) have identified a population of RGC-like cells in the outer margins of the SVZ that maintain only basal attachments to the pial surface. Recent characterizations of these cells in several species (Fietz et al., 2010; Hansen et al., 2010; Reillo et al., 2011; Shitamukai et al., 2011; Wang et al., 2011; García-Moreno et al., 2012; Kelava et al., 2012) have demonstrated that they are indeed mitotic cells that express several progenitor- and RGC-specific markers and undergo asymmetric divisions to self-renew and produce IPCs, neurons and astrocytes. Thus, these cells appear to the have the characteristics of basally displaced RGCs and have therefore been named outer, intermediate or basal RGCs (bRGCs) (Fig. 3). bRGCs are vastly more abundant in the primate brain compared to rodents and occupy a specialized subdomain of the SVZ, the so-called outer SVZ (oSVZ). The massive amplification of bRGCs has been proposed to be important for generating the expanded numbers of neocortical neurons in primates, especially those occupying the enlarged upper layers (Lui et al., 2011; Lamonica et al., 2012).

In summary, although NECs were once thought to be the primary progenitor of all neurons and macroglia in the developing neocortex, recent technical advances have facilitated the identification of several distinct progenitor types that may represent a continuum in the lineage from NECs to postmitotic neurons and glia (Fig. 3). The primary role of NECs appears to be to expand the progenitor pool before transitioning into RGCs in the VZ. RGCs, in turn, rarely generate neurons directly, but instead self-renew while producing IPCs to amplify neuronal output from the SVZ. Intermediate states also exist within the lineage; IPC-like SNPs produce neurons from the VZ, whereas RGC-like bRGCs give rise to neurons and glia from the oSVZ. Thus, cell-fate specification in the neocortex is more complex than previously appreciated, as excitatory neuron subtypes and astrocytes are generated within a framework of diverse progenitor populations.

Functional diversity: neuronal and glial progenitors

Although several progenitor types have been identified in the VZ and SVZ, we are only just beginning to understand their fate potentials and linage relationships to distinct neocortical cell types. The first evidence that linage-restricted progenitors might exist came from studies using clonal analysis of progenitors and their progeny, which hinted at the existence of distinct progenitors for neurons and glia. The majority of progenitor clones cultured in vitro (Luskin et al., 1988; Williams and Price, 1995; Qian et al., 1998; Malatesta et al., 2000; Heins et al., 2002; Malatesta et al., 2003; Shen et al., 2006) or labeled by retroviral vectors in vivo (Luskin et al., 1988; Walsh and Cepko, 1988; 1992; Grove et al., 1993; Walsh and Cepko, 1993; Parnavelas et al., 1995; McCarthy et al.,2001) produce either neurons or macroglia, but not both. The glia-specific progenitors are further restricted and typically produce either astrocytes or oligodendrocytes (McCarthy et al., 2001; Malatesta et al., 2003). Only 10–20% of cortical progenitors isolated at early embryonic stages appear to be multipotent in terms of generating neurons and glia, pointing toward a model of progenitor diversity with respect to the neuronal and macroglial lineages. This lineage restriction is in place already at the beginning stages of neurogenesis and is maintained in vitro (Malatesta et al., 2000; McCarthy et al., 2001; Malatesta et al., 2003), thus raising the possibility that an early specification event may initiate an intrinsic molecular program for progenitor fate restriction.

Several studies indicate possible molecular heterogeneity between multipotent and lineage-restricted progenitors. When RGCs from an hGFAP-GFP transgenic mouse line were sorted by fluorescence and cultured at clonal density, the majority of these progenitors produced only neurons or macroglia (Malatesta et al., 2000), whereas a similar strategy using BLBP-GFP mice primarily isolated RGCs that were multipotent, giving rise to both neurons and glia (Anthony et al., 2004). Subsets of RGCs differentially express the markers RC2, GLAST and BLBP, prompting the suggestion that these molecular expression profiles may correlate to lineage restriction of neuronal versus glial fate (Hartfuss et al., 2001). However, fate-mapping studies indicate that this heterogeneity may reflect spatiotemporal differences caused by a developmental gradient, rather than true lineage restrictions (Anthony et al., 2004; Anthony and Heintz, 2008). In support of temporal changes in progenitor properties, several studies have shown that extrinsic signals such as Notch and BMPs induce neocortical progenitor cells to become gliogenic (Gaiano and Fishell, 2002; Rowitch and Kriegstein, 2010). The competence to respond to these signals is acquired during late stages of embryogenesis and depends on MEK/ERK/MAPK signaling (Li, Snider et al. 2012). It should be noted, however, that little is known about the functional diversification within the macroglial lineage. Perhaps one neocortical progenitor type generates both neurons and astrocytes, whereas a second progenitor type is lineage-restricted to generate functionally distinct astrocytes.

Neuronal progenitor diversity: distinct progenitors for excitatory neurons of lower and upper neocortical layers

Recent studies have provided compelling evidence of heterogeneity within the neurogenic fraction of the progenitor pool as well, in particular with respect to the mechanisms governing the production of early-born lower-layer neurons versus late-born upper-layer neurons. For example, the bHLH transcription factors Ngn1 and Ngn2 are required for specifying the regional, laminar and neurotransmitter fates of excitatory neurons specifically during lower-layer neurogenesis, but not later during upper-layer formation (Schuurmans et al., 2004). Instead, specification of these fates in upper-layer neurons requires the transcription factors Pax6 and Tlx (Schuurmans et al., 2004). Thus, distinct molecular pathways may control the basic differentiation programs of lower- and upper-layer excitatory neurons. Consistent with this idea, a number of genes that are restricted to either upper- or lower-layer neurons in the mature neocortex are also enriched in subsets of progenitors during development. RGCs in the early cortex express markers of lower-layer neurons, including Emx2, Fezf2, Otx1 and Sox2 (Frantz Franco and Müller et al., 1994; Leingärtner et al., 2003; Chen et al., 2005; Bani-Yaghoub et al., 2006). In contrast, a number of genes specific for upper-layer neurons, such as Cux1, Cux2, Satb2 and Svet1/Unc5D, are expressed at high levels in IPCs in the SVZ during middle and late stages of neocortical development (Tarabykin et al., 2001; Nieto et al., 2004; Zimmer et al., 2004; Britanova et al., 2005). In addition, the SVZ is preferentially expanded specifically during the time of upper-layer neurogenesis (Takahashi et al., 1999; Lukaszewicz et al., 2006; Martínez-Cerdeño et al., 2006; Dehay and Kennedy, 2007). Together, these studies led to the hypothesis that IPCs in the SVZ generate upper-layer neurons, whereas lower-layer neurons are derived directly from RGCs in the VZ (Tarabykin et al., 2001; Nieto et al., 2004; Zimmer et al., 2004). This interpretation is supported by molecular profiling studies that suggest RGC subtypes with distinct gene expression programs differentially generate IPCs versus neurons (Pinto et al., 2008) through a mechanism that depends on the transcription factor AP2γ/Tcfap2c, at least in the occipital cortex (Pinto et al., 2009).

However, IPCs are present throughout neocortical development, even during lower-layer neurogenesis (Smart, 1973; Haubensak et al., 2004; Miyata et al., 2004; Noctor et al., 2004; Wu et al., 2005), and the majority of neurons in all layers of the neocortex derive from IPCs, with only ~10% of excitatory projection neurons coming directly from RGCs (Kowalczyk et al., 2009). Because IPCs are derived from RGCs (Haubensak et al., 2004; Miyata et al., 2004; Noctor et al., 2004), these two progenitor types therefore likely represent different states of progenitor differentiation rather than separate fate-restricted lineages. Intriguingly, some of the aforementioned upper-layer specific genes that are expressed at high levels in the SVZ actually start to be expressed in the VZ. For example, low levels of Cux1 and Cux2 mRNAs can be detected in subsets of cells in the VZ at E13.5 (Nieto et al., 2004; Franco et al., 2012), the same developmental time point when lower-layer markers such as Fezf2 and Otx1 are also found in the VZ (Frantz et al., 1994; Molyneaux et al., 2005). These correlations suggest that some molecular differences that distinguish lower- and upper-layer neurons are established in subsets of RGCs that coexist developmentally, thus raising the possibility that RGCs could be a heterogeneous progenitor population.

To test this hypothesis, genetic lineage-tracing experiments have been employed to determine the developmental stage at which the fates of lower- and upper-layer neurons diverge (Fig. 4) (Franco et al., 2012). A mouse line in which Cre recombinase is expressed from the Cux2 locus accurately fate-maps the entire lineage of upper-layer, but not lower-layer, neurons (Franco et al., 2011; 2012). In addition to upper-layer neurons in the mature cortex, a subset of RGCs in the embryonic VZ also belong to the Cux2 lineage (Franco et al., 2012). These Cux2⁺ RGCs can be identified even before the onset of neurogenesis and co-exist with Cux2⁻ RGCs throughout neocortical development. Moreover, temporal lineage-tracing analysis using inducible Cux2-CreERT2 mice have demonstrated that these Cux2⁺ RGCs give rise specifically to upper-layer neurons, whereas lower-layer neurons are generated from Cux2⁻ RGCs (Franco et al., 2012). Thus, a subset of RGCs is specified early in development to produce upper-layer neurons (Fig. 4).

Figure 4. Distinct lineages of fate-restricted RGCs generate lower- versus upper-layer projection neurons.

(A) Corticofugal projection neurons in lower layers V-VI are generated from Cux2⁻ RGCs that are neurogenic during early stages of neocortical development. Corticocortical projection neurons in upper layers II-IV are made from Cux2⁺ RGCs that co-exist with Cux2⁻ RGCs during early stages of neocortical development, but are initially proliferative and only become neurogenic at later developmental stages. (B-C) Hypothetical lineage trees for the different subtypes of neocortical projection neurons. In both models, neuroepithelial cells transform into RGCs, which are then specified into the Cux2⁺ and Cux2⁻ lineages. In (B), Cux2+ and Cux2− RGCs are then progressively restricted over time to sequentially generate the different projection neuron subtypes based on birth date. In (C), progenitors of the two lineages undergo further specification at the progenitor state to generate additional lineages of fate-restricted RGCs that produce a specific projection neuron subtype. For abbreviations see legend to Fig. 2.

Cux2⁺ progenitors are primarily proliferative early in development, during lower-layer neurogenesis, and only generate significant numbers of neurons at later stages (Franco et al., 2012). This delayed neurogenesis by Cux2⁺ progenitors compared to their Cux2⁻ counterparts provides important insight into the well-documented relationship between cell birth date and laminar fate, indicating that neuronal birth date is more likely a consequence of cell fate-specification, rather than a cause. Further supporting this notion is the observation that even the minor fraction of Cux2⁺ neurons that settles in lower layers are molecularly most similar to upper-layer callosal projection neurons (Franco et al., 2012), indicating that lineage determines the molecular and functional identity of a neuron regardless of its birth date or laminar position. To further test this idea of fate-restriction, Cux2⁺ RGCs were forced to generate neurons prematurely, during the peak of lower-layer neurogenesis (Franco et al., 2012). Prematurely born neurons belonging to the Cux2⁺ lineage still maintain their normal laminar and molecular identities, in agreement with an intrinsic mechanism for neuronal fate-specification that is independent of birth date. Together these data demonstrate that a lineage of RGCs is restricted in its fate potential even before the onset of neurogenesis, and provide evidence for a model in which fate-specification instructs the temporal order of projection neuron production rather than vice versa, at least for upper- versus lower-layer neurons. This concept that molecular fate determination is independent of birth date is also supported by studies on the generation of Tbr2⁺ lower-layer neurons (Hevner et al., 2003). BrdU pulse-labeling experiments demonstrated that Tbr1⁺ neurons settle in lower layers regardless of birth date, even when generated later in development, thus indicating that neuronal fate is primarily molecular and secondarily temporal (Hevner et al., 2003).

Constructing lineage trees: progressive restriction versus lineage restriction

How can the recent identification of lineage-restricted RGCs be reconciled with previous studies supporting the progressive restriction model of fate-specification? In fact, the available experimental evidence is surprisingly consistent and can be incorporated into a coherent model of neocortical neurogenesis. In this section we re-evaluate the evidence from transplantation experiments, in vitro clonal analyses and retrovirus lineage-tracing studies in the context of these recently identified lineage-restricted progenitors.

Transplantation experiments form the foundation of the progressive restriction model. These studies have shown that progenitor pools isolated during the earliest stages of neocortical development are competent to generate neurons destined for all neocortical cell layers, whereas progenitor pools isolated at later stages are more restricted in their fate potential (McConnell and Kaznowski, 1991; Frantz and McConnell, 1996; Desai and McConnell, 2000). As pointed out by Desai and McConnell, these results can be equally explained either by progressive restriction of a common progenitor, or by the co-existence of multiple fate-restricted progenitors (Desai and McConnell, 2000). In this latter model, the early neocortex contains progenitors for both early-born lower-layer neurons and late-born upper-layer neurons, but lower-layer progenitors soon get depleted through neurogenic divisions. Thus, the late neocortex predominantly contains progenitors for late-born upper-layer neurons and can therefore only generate a restricted subset of neurons.

One intriguing additional observation from the transplantation experiments is that when cells from the early germinal zone are heterochronically transplanted into older animals, they primarily generate upper-layer neurons, but not the lower-layer neurons appropriate for the donor age (McConnell and Kaznowski, 1991; Desai and McConnell, 2000). This could suggest that the new host environment instructs the transplanted progenitors to switch their competence. An alternative interpretation, however, is that signals in the germinal niche preferentially promote survival and differentiation of age-matched progenitors. In this model, the early environment is suitable for all progenitors, but late progenitors preferentially thrive in the late environment; thus, it is not the progenitor but the environment that undergoes progressive restriction in its potential to support progenitor subtypes. Evidence for such a model has been provided by recent studies using samples of cerebral spinal fluid (CSF) to culture neocortical progenitors in vitro. These studies have demonstrated that progenitor survival and proliferation are maximal when the developmental age of the CSF is matched to that of the progenitor cells (Lehtinen et al., 2011). To further test this model, it will be important to repeat the transplantation experiments while taking advantage of molecular markers for progenitor and neuronal subtypes that are now available.

In vitro studies that suggested a specified temporal birth order for lower- and upper-layer neurons can also be interpreted in light of the existence of distinct progenitors. Clonal analyses performed in vitro have shown that lower-layer neurons are generated after fewer progenitor divisions than are upper-layer neurons (Shen et al., 2006). Some clones contained Cajal Retzius cells and projection neurons, which is somewhat surprising because the two cell types are generated in vivo from different progenitors that reside in distinct brain regions. Importantly, however, this study did not demonstrate that a single progenitor could sequentially generate neurons of lower layers V/VI followed by upper layers II/III/IV. In fact, careful examination of the lineage trees shows that each individual tree arising from a single progenitor always contained only one subtype (either lower-layer or upper-layer) of excitatory projection neuron (Shen et al., 2006). Thus, these data are consistent with the existence of separate progenitors for lower- and upper-layer neurons. The timing difference observed in vitro could be due to lineage-specific differences in the cell cycle program; lower-layer progenitors may go through fewer rounds of division than upper-layer progenitors before exiting the cell cycle. This model is supported by recent evidence. When Cux2⁺ and Cux2⁻ progenitors are cultured in vitro, the former tend to undergo proliferative divisions before generating upper-layer neurons, whereas the latter differentiate rapidly into lower-layer neurons (Franco et al., 2012). Because upper-layer progenitors initially represent less than 5% of all RGCs (Franco et al., 2012), these additional rounds of cell division might be necessary to generate sufficient numbers of upper-layer neurons to form the outer layers of the neocortex, which occupy a larger territory than lower layers.

Finally, retrovirus lineage-tracing studies are remarkably consistent with a multiple-progenitor model. Although these studies have revealed clones that populated both upper and lower layers, several quantitative studies have indicated that clones predominantly consisted of cells residing in either lower or upper layers (Luskin et al., 1993; Reid et al., 1995). This is precisely the result expected based on the Cux2 lineage-tracing studies; individually labeled Cux2⁺ progenitors should mostly generate clones in upper layers, whereas neurons derived from Cux2⁻ progenitors should largely settle in lower layers. Interestingly, a minor fraction of neurons derived from Cux2⁺ progenitors can be found in lower layers, even though they are molecularly more similar to their upper-layer siblings (Franco et al., 2012). Thus, just like the pattern observed in the retroviral studies, at least some Cux2⁺ progenitors are expected to generate clones that primarily reside in upper layers but also contribute a small number of cells to lower layers. Notably, although the retrovirus lineage-tracing experiments are frequently cited in support of the progressive-restriction model, Walsh and Cepko discussed that it appears from their studies and from the studies of others that “…upper and lower cortical layers tend to have different precursors” (Walsh and Cepko, 1992).

The available experimental evidence is therefore consistent with a model in which progenitors are specified to generate lower- versus upper-layer neurons. However, it is important to point out that lower and upper layers are further diversified into neuronal subtypes that are distinguished by their gene expression profiles, morphologies, projection patterns and functions (Molyneaux et al., 2007; Leone et al., 2008; Fame et al., 2011). Thus, an unanswered question remains how this vast diversity in neuronal phenotype is established. As one possibility, distinct progenitors for lower- versus upper-layer neurons might be initially specified during early stages of neocortical development and then further diversified through other mechanisms, such as progressive restriction (Fig. 4B) or subsequent specification events (Fig. 4C). For example, a Cux2⁺ progenitor may first generate layer IV neurons before being progressively restricted to generate layer II/III neurons (Fig. 4B). The fully differentiated phenotypes of the distinct subclasses would then be executed by postmitotic transcriptional cascades that were initiated in dividing progenitors (Arlotta et al., 2005; Chen et al., 2005; Molyneaux et al., 2005; Chen et al., 2008; Kwan et al., 2008; Lai et al., 2008; Molyneaux et al., 2009; Bedogni et al., 2010; Han et al., 2011; McKenna et al., 2011; Miyoshi and Fishell, 2012). These models can be tested with the molecular markers for subtypes of progenitors and differentiated neurons that are now available.

Managing diversity locally: signals derived from within the progenitor domain

The identification of RGC subtypes that are functionally distinct raises a number of important questions regarding their molecular and cell biological differences. In particular, it will be of great interest to identify the signals that establish the temporal order of neurogenesis from upper- and lower-layer progenitors (Fig. 5). In this regard, the wealth of data available on the mechanisms by which asymmetric cell division affects progenitor self-renewal and differentiation might reveal important clues. Drosophila neuroblasts provide a classic example of asymmetric progenitor cell division. In this example, the decision between self-renewal and differentiation is regulated by the orientation of the mitotic spindle and cleavage plane in relation to polarized cell fate determinants (Siller and Doe, 2009; Tajbakhsh et al., 2009). This model appears to hold true for vertebrate neurogenesis in the neocortex. In mice, asymmetric cell division of polarized RGCs results in unequal inheritance of their basal and apical processes by their progeny (Miyata et al., 2001; Kosodo et al., 2004). Perturbations in mitotic spindle orientation during RGC divisions shift the normal proportions of RGCs, IPCs and neurons generated from RGCs (Sanada and Tsai, 2005; Konno et al., 2008; Postiglione et al., 2011; Shitamukai et al., 2011). In Drosophila, asymmetric cell division and cell fate are linked through the unequal segregation of polarity proteins such as the Stau/Mir/Prospero/Brat basal complex and the Par3/Par6/aPKC apical complex (Siller and Doe, 2009). Likewise, genetic studies in mice have confirmed the essential roles of many polarity proteins for proper maintenance, self-renewal and differentiation of RGCs in vertebrates (Cappello et al., 2006; Costa et al., 2007; Schwamborn et al., 2009; Kim et al., 2010; Vessey et al., 2012).

Figure 5. Signals that control RGC proliferative behavior.

A number of signaling molecules that are secreted by the meninges or that are present in the cerebrospinal fluid can act on RGCs to regulate their proliferative and neurogenic behaviors. Similarly, IPCs and neurons secrete signals that provide a feedback mechanism for controlling the balance between proliferation and neurogenesis in RGCs.

The primary mechanism by which polarity complexes execute their function is to control the localization of the fate determinants Numb and Numb-like. In the mouse neocortex, asymmetric segregation of the apical complex protein Par3 in dividing RGCs promotes unequal inheritance of Numb and Numb-like by their progeny, which in turn determines which daughter cell remains an RGC and which one becomes an IPC or neuron (Bultje et al., 2009). Numb/Numb-like drives differential daughter cell fates by inhibiting Notch signaling, possibly by controlling endocytosis and degradation of Notch (Tajbakhsh et al., 2009). Notch signaling represses proneural genes and is important for maintenance and self-renewal of RGCs during neurogenic stages of neocortical development (Gaiano et al., 2000; Nakamura et al., 2000; Hitoshi, 2002; Mizutani, 2005; Mizutani et al., 2007; Imayoshi et al., 2010). In the canonical Notch signaling pathway, Notch ligands bind to transmembrane Notch to promote release of the Notch intracellular domain (NICD), which then translocates into the nucleus to form a complex with the DNA-binding protein Rbpj/CBF1. The NICD-Rbpj complex subsequently activates expression of transcription factors Hes1 and Hes5, which in turn repress proneural genes. Rbpj-mediated signaling is attenuated in IPCs compared to RGCs (Mizutani et al., 2007), and knockout of Rbpj in mice abolishes Notch signaling and leads to premature depletion of RGCs in favor of IPCs and neurons (Imayoshi et al., 2010). Taken together, these studies support a model by which polarization of apical complex proteins promotes segregation of Numb/Numb-like during RGC divisions, resulting in high Notch signaling in one daughter cell that becomes the self-renewing RGC and low Notch signaling in the daughter that adopts an IPC or neuronal fate (Bultje et al., 2009).

Although Notch signaling is not obviously different between upper- versus lower-layer progenitors at mid-neurogenesis (Franco et al., 2012), it will be interesting to test whether differential Notch signaling earlier in development could limit neurogenesis from upper-layer progenitors during lower-layer formation. Alternatively, Notch signaling in early corticogenesis could serve to enhance intrinsic differences between lower-layer RGCs that tend to undergo neurogenic divisions and upper-layer RGCs that preferentially undergo symmetric proliferative divisions (Franco et al., 2012). Because the Notch ligands Deltalike1 and Deltalike3 are expressed by IPCs and neurons, but not RGCs (Campos et al., 2001), one intriguing possibility is that postmitotic lower-layer neurons may signal back to symmetrically dividing upper-layer RGCs to promote the progenitor state in both of their daughter cells (Fig. 5). Evidence for such a feedback mechanism has been obtained in mice (Yoon et al., 2008) and zebrafish (Dong et al., 2012). These studies show that the ubiquitin ligase Mib1, which is required for generating functional Notch ligands (Koo et al., 2005), is preferentially inherited by IPCs and neurons in a Par3-dependent manner (Dong et al., 2012) and is required for RGC maintenance during asymmetric and symmetric divisions (Yoon et al., 2008; Dong et al., 2012). Such a model could account for the expansion of the upper-layer progenitor pool during lower-layer neurogenesis (Franco et al., 2012).

Additional feedback signaling mechanisms have been proposed to regulate RGC behavior (Fig. 5). The transcriptional repressor Sip1 acts in newborn neurons to regulate levels of neurotrophin 3 (Ntf3) and fibroblast growth factor 9 (Fgf9), which serve as signals from neurons to progenitors (Seuntjens et al., 2009). Deletion of Sip1 in neurons leads to their overexpression of Ntf3 and premature production of upper-layer neurons (Seuntjens et al., 2009). Although it was proposed that Sip1/Ntf3 signaling may regulate a switch from lower- to upper-layer fate (Seuntjens et al., 2009), an alternative hypothesis is that this signal may be preferentially received by upper-layer progenitors as a means to control the timing of upper-layer neurogenesis. Such a mechanism could be integrated with a Notch pathway described above, such that Notch signaling first promotes expansion of the upper-layer progenitor pool early in development, and then once a critical number of lower-layer neurons are generated they could produce sufficiently high levels of Ntf3 to initiate neurogenesis from the expanded upper-layer progenitor population. In this way, an inter-lineage feedback mechanism would ensure that correct numbers of projection neuron subtypes are generated at appropriate times.

Managing diversity from afar: signals that originate outside the progenitor domain

RGCs are also positioned to receive instructional information from sources other than IPCs and neurons (Fig. 5), most notably from the CSF. The apical domains of RGCs line the CSF-filled lateral ventricles, into which RGCs extend their primary cilia (Cohen et al., 1988; Dubreuil et al., 2007). Many signaling molecules are secreted into the CSF, including FGFs, insulin-like growth factors (Igfs), retinoic acid (RA), sonic hedgehog (Shh), transforming growth-factor beta/bone morphogenetic proteins (TGFβ/BMPs) and Wnts (Lehtinen and Walsh, 2011). CSF samples from mouse or rat lateral ventricles are sufficient to stimulate progenitor proliferation and maintenance in neocortical explants and neurosphere cultures (Lehtinen et al., 2011). The proliferative effects of CSF on progenitors depends in part on Igf2, which is secreted into the CSF by the choroid plexus (McKelvie et al., 1992) and binds to the primary cilia in the apical domain of RGCs (Lehtinen et al., 2011). Igf2 is both necessary and sufficient to promote progenitor proliferation in culture (Lehtinen et al., 2011), and genetic perturbations of Igf2 or its receptor Igf1R lead to microcephaly in mice (Kappeler et al., 2008; Liu et al., 2009; Lehtinen et al., 2011). These studies indicate that Igf2 in the CSF binds to Igf1R on the primary cilia of RGCs to regulate progenitor proliferation. Interestingly, both the levels and effects of Igf2 are highest during late neocortical development and more modest at earlier stages (Lehtinen et al., 2011), raising the possibility that Igf2 may preferentially regulate late neurogenesis. In support of this model, Igf2-deficient mice have a specific decrease in upper-layer neurons (Lehtinen et al., 2011). In future studies it will be interesting to test whether RGCs committed to upper-layer fate may be particularly responsive to Igf2 signaling compared to their lower-layer counterparts.

CSF-derived Wnts are additional candidate signaling molecules that might regulate RGC behavior, since Wnt signaling via β-catenin is required for early neocortical specification and maintenance of progenitor cells (Machon et al., 2003; Junghans et al., 2005; Zhou et al., 2006). Consistent with this idea, constitutive activation of Wnt/β-catenin signaling forces RGCs to undergo excessive proliferative divisions, resulting in expansion of the RGC pool at the expense of neurogenesis (Chenn and Walsh, 2002). Conversely, blocking β-catenin-dependent transcription forces premature differentiation of RGCs into neurons (Woodhead et al., 2006). β-catenin signaling is reduced in IPCs compared to RGCs (Mutch et al., 2010), and deletion of β-catenin in RGCs causes increased production of IPCs prior to increased neurogenesis. Together these data suggest that activation of β-catenin signaling by Wnts in the CSF could be involved in controlling the decision between self-renewal and differentiation in RGCs. The Wnt/β-catenin signaling pathway may be integrated with the aforementioned apicobasal polarity mechanism that determines RGC proliferative behavior. Since β-catenin is a component of cadherin-based adherens junctions between RGCs, these specialized cell contacts could play a pivotal role in progenitor behavior by maintaining contact of the apical domains of RGCs with the CSF, regulating β-catenin localization and function, and serving as docking sites for Par proteins (Manabe et al., 2002). Whether Wnt/β-catenin signaling or cadherin-based adherens junctions are differentially regulated in upper- versus lower-layer RGCs remains to be explored.

Finally, the basal domain of a highly polarized RGC may also be important for receiving and organizing progenitor fate signals (Fig. 5). For example, it has been suggested that RGC self-renewal is assured only in the daughter cell that inherits both the apical and basal domains (Konno et al., 2008; Shitamukai et al., 2011). Importantly, attachment of RGC basal processes to the meningeal basement membrane is required for efficient progenitor maintenance (Radakovits et al., 2009). These long basal processes are anchored by β1-integrins (Itgb1) attached to the extracellular matrix (ECM) secreted by the meninges. RGCs in which Itgb1 is knocked out detach their radial processes from the meninges (Graus-Porta et al., 2001) and die by apoptosis (Radakovits et al., 2009), suggesting that the basal processes of RGCs may receive trophic signals from the meninges. In support of this idea, Itgb1 knockout mice have significantly smaller brains compared to controls (Radakovits et al., 2009). One potential meningeal-derived trophic signal is Bmp7. Loss of Bmp7 causes reduced proliferation and survival of RGCs (Segklia et al., 2012), whereas its over-expression promotes premature RGC differentiation (Ortega and Alcántara, 2010). Meningeal-derived RA is another important regulator of RGC proliferative behavior. Failure to form forebrain meninges in Foxc1 mutant mice causes a dramatic expansion of RGCs at the expense of IPCs and neurons (Siegenthaler et al., 2009), a phenotype that appears to be caused at least in part by loss of meningeal-derived RA. Intriguingly, recent studies in primates suggest that integrins and their ECM ligands might have important roles in the oSVZ, where bRGCs reside (Fietz et al., 2010; 2012), raising the possibility that cell-ECM interactions affect specific progenitor subpopulations. It will be interesting to test whether integrins and their ligands might affect progenitors for upper and lower neocortical cell layers in different ways.

The sequential progenitor-diversification model: implications for brain evolution and psychiatric disorders

Scientist have been fascinated with the complexity of the neocortex for centuries and have appreciated since the late 19^th century that it consists of a multitude of cell types with diverse shapes, connectivity patterns and functions. Recent large-scale gene expression profiling projects and in situ hybridization screens have revealed that even cell types with similar morphologies, laminar positions and projection patterns can be subdivided into neuronal subtypes with distinct molecular signatures (Lein et al., 2007; Ng et al., 2009; Hawrylycz et al., 2010; Belgard et al., 2011; Bernard et al., 2012; Hawrylycz et al., 2012; Zeng et al., 2012). One major challenge is to understand how this diversity is achieved and how evolutionary forces have expanded the cellular and functional complexity within the neocortex.

One solution to the problem of cell type diversification has been the establishment of different germinal zones for distinct cell types, as is the case for interneurons versus projection neurons. However, this does not apply to all the different neuronal subtypes of the neocortex, such as the different subclasses of projection neurons that are derived from one germinal zone. Thus, additional mechanisms are necessary to generate the full complement of neuronal subtypes. Based on current data that we have summarized in this review, we propose a “sequential progenitor-diversification model” to explain how diversity among neocortical projection neurons is achieved. This model involves at least three sequential steps of cell-type diversification: (1) the initial specification of lineage-restricted progenitors for lower- and upper-layer neurons; (2) further diversification of lineage-restricted progenitors by means of progressive restriction, additional lineage-restriction events, or a combination of both; (3) execution of the final differentiation programs at the stage of the postmitotic neuron. Importantly, each step provides a point of control for determining the precise timing and numbers of the different subclasses of excitatory projection neurons.

The identification of distinct progenitor cells for neurons of upper and lower cortical layers has interesting implications for the generation of radial columns, which are thought to be the primary information processing units of the neocortex. Since radial columns span both lower and upper layers, each column is necessarily assembled from daughter cells from at least two different progenitors. Thus, mechanisms must exist that allow for the integration of appropriate numbers and subtypes of layer-specific neurons into functional columns. Previous studies have provided evidence that Eph/ephrin signaling is important for the integration of neuronal clones into radial columns (Torii et al., 2009). It will be important to address the extent to which Eph/ephrin signals might coordinate the behavior of neurons derived from the Cux2⁺ and Cux2⁻ lineages.

Specification of lineage-restricted progenitors has so far been documented conclusively only for two distinct progenitors, those that generates neurons for lower versus upper neocortical layers. While additional lineage-restricted progenitors might exist, the two identified progenitor subtypes might be of particular evolutionary importance. Upper layers are thought to be a more recent addition to the neocortex and have been massively expanded during primate evolution. It has been proposed that the enlargement of the SVZ and especially its outer domain, the oSVZ that contains bRGCs, was critical for the increased output of upper-layer neurons in primates (Fietz et al., 2010; Hansen et al., 2010). It is tempting to speculate that two consecutive evolutionary events might have enabled neocortical expansion. In the first event, separate progenitors for lower and upper layers were established. Upper-layer progenitors then underwent further diversification to generate additional progenitor subtypes, such as bRGCs, to increase the output of upper-layer neocortical neurons. This model can be tested experimentally, as it predicts a direct lineage relationship between Cux2⁺ RGCs and bRGCs.

Notably, neurons of upper cortical layers form projections within the two neocortical hemispheres and connect them across the corpus callosum. Upper neocortical neurons are also critical for higher associative brain functions and complex thought, and they are prominently affected in psychiatric disorders such as schizophrenia and autism. These disorders are considered to arise from complex interactions between genetic and environmental factors. It will be important to define to what extend disease mechanisms can be linked to defects that manifest already at the progenitor state, leading to functional perturbations in the neocortical circuits that are then targets for environmental influences. Notably, mutations in genes that are linked to psychiatric disorders, such as the DISC gene in schizophrenia, have profound effects on the development and function of neocortical progenitors (Mao et al., 2009), suggesting that at least some disease aspects might manifest at the stem and progenitor cell state.