Molecular Neurobiology of Development

Flora M. Vaccarino and James F. Leckman

1. Introduction. For most species, the nervous system is the most complex part of  the organism. In mammalian species, over one third of the entire set of genes is expressed in the central nervous system (CNS). The reason for this abundance in gene expression and protein products lays in the intrinsic complexity, both structural and functional, of the CNS. It is reasonable to assume that CNS-specific transcripts are necessary not only to encode the wide variety of cellular and molecular components of the mature CNS, but that a substantial percent guide CNS morphogenesis and are primarily expressed during development. In this chapter, we to introduce the reader to an emerging understanding of some of the basic genetic rules that control CNS morphogenesis and to discuss how the workings of this genetic machinery is dependent on cell-to-cell communication and variations in the cellular micro-environment. Throughout this chapter, we will draw parallels between invertebrate and vertebrate development to show conserved and divergent aspects of CNS development during the course of evolution. The vastness of this topic prevents us from examining it thoroughly. We have deliberately chosen examples to illustrate specific principles of CNS development. For more complete and systematic accounts, the reader is encouraged to consult one of a growing set of volumes on this topic (Kandel et al.,1991; Bayer and Altman, 1991; Lawrence, 1992; Cowan et al.,1997; Wolpert et al.,1998)

2.  Gradients of morphogens form the pattern. Development involves the proliferation of cells to generate a multicellular organism and differentiation and sorting of these cells into appropriate patterns. Pattern formation involves a series of cell fate decisions and the reciprocal positioning of the progenitor and differentiated cells. These decisions are arranged in a hierarchy of choices, where the simplest (i.e., axis formation) prefigures and regulates the more complex, such as the location and connectivity of different neuronal types.

During CNS patterning, cells coordinate their replication rate, fate and relative position by reciprocal interactions. These interactions, mediated by direct cell to cell contacts, by the local action of secreted morphogens and by hormonal influences from the larger macro-environment, influence morphogenetic programs by inducing and regulating  the expression of developmental control genes.  Direct cell-to-cell contact mediate the lateral specification between groups of cells, whereas gradients of morphogens organize growth and pattern at a distance.

Morphogens are "form generating" substances, whose configuration within a tissue "prefigures" the pattern (Meinhardt, 1983; Lawrence and Struhl, 1996). These substances set up an extracellular concentration gradient.  The concentration gradient orchestrates a coherent set of cellular behaviors that will eventually result in the proportionate growth of an organ, including the finest details. For example, different scalar concentrations may specify the type of cells and their relative position within the field; the slope of the gradient may be correlated to the degree of growth of the intervening cells, and the direction of the gradient with respect to the compartment may determine polarity.  All this information (cell fate, position, polarity and growth) is in principle specified by the induction of target genes by the morphogen. Different genes whose activation depends on different thresholds of morphogen will be turned on at different distances from the source. Thus morphogens act by altering the pattern of gene expression in target cells. The morphogen theory is largely based on the early development of the fruit fly, the cellular blastoderm stage, when there are no membrane barriers to the free diffusion of molecules between cell nuclei. In vertebrate species, morphogens may not be able to diffuse between cells over a long range. A modification of this theory thus proposes that an intercellular relay of secondary short-range morphogens guides morphogenesis(Reilly and Melton, 1996).

 Morphogen gradients provide a stimulus that is "interpreted" in different ways according to the intrinsic genetic program of a particular body compartment (Papalopulu and Kintner, 1994; Gurdon, 1987). The new information conveyed by morphogens is differentially “read” according to the developmental history of target cells. This would explain how a common signal can induce different types of cell behaviors in different compartments of the CNS at different stages of development. 

 Specific examples can be found from the earliest stages of development. In Drosophila, before egg fertilization, localized signals present in the egg cytoplasm- the products of "maternal effect genes"- direct early pattern formation. Four maternal derived signals are responsible for the formation of the principal body axes, the anterior-posterior, the dorso-ventral, and the terminal. These are established independently, and provide a pattern for subsequent development (St Johnston and Nusslein-Volhard, 1992).

One of the maternal signals is the mRNA of a gene called bicoid, that is present only at the future anterior pole of the egg. Bicoid mRNA is translated after fertilization to produce an anterior to posterior gradient of bicoid protein, which despite not being a secreted protein can freely distribute itself  within the syncitial blastoderm of Drosophila's embryo. Loss of bicoid precludes head formation and injection of bicoid protein in ectopic locations results in ectopic heads, with the most anterior end closest to the site of injection (St Johnston and Nusslein-Volhard, 1992). Thus the shape of the bicoid concentration gradient directly defines the anterior-posterior polarity. Bicoid protein is a homeodomain transcription factor that binds the enhancer region of zygotic genes named gap genes. These genes, in turn, control the development of the head and anterior thorax.  Since bicoid binding sites on different genes have different threshold affinities for bicoid, the concentration gradient defines different positional values along the embryo (Hulskamp et al.,1990).

3. Hierarchical genetic programs control the orderly progression of morphogenesis. The ontogenetic development of the central nervous system is driven by hierarchical sets of evolutionary conserved genes. With exquisite timing, this genetic program progressively unfolds with each phase of development. Although the sequence of gene expression that controls neural development is preset and stereotyped, it is also driven to a large extent by local interactions among developing cells. The evolutionary advantage of these interactions is that complex organisms are able to re-assess cellular identities and make appropriate adjustments to environmental perturbations.  This is particularly true at the earliest stages of embryonic development, when there is the need to reciprocally coordinate the layout of different tissue layers and compartments.  Developmental plasticity allows higher organism to set up and use alternative genetic pathways to implement adjustments to environmental fluctuations; the system thus becomes  more flexible and less prone to failure (Capecchi, 1997). As the genetic blueprint unfolds, cell fates are increasingly autonomous from cellular interactions and becomes less sensitive to variations in the environment.

In Drosophila, maternal effect genes such as bicoid and the other signaling systems involved in setting up the anteroposterior and dorsoventral axes trigger the expression of zygotic gap genes. Gap genes are activated in large, discrete blocks, which segmentally organize the body plan. Gap genes are involved in the specification of the earliest and most fundamental pattern, the formation of parasegments, that is the foundation for all subsequent morphogenesis. Reciprocal interactions among gap genes regulate their correct order and positioning within the embryo. The positional information provided by the overlapping series of gap protein gradients directs the expression of pair rule genes and segment polarity genes, which appear as a series repeating stripes along the Drosophila embryo (Pankratz and Jackle, 1990).  This mechanism involves the direct specification of each stripe by a particular combination of gap genes, some activating transcription in a broad domain and others setting the borders. Thus, the invertebrate body plan unfolds in a rigorously specified hierarchy of events. 

Even though a segmental organization is much less obvious in the vertebrate CNS, gap genes are evolutionarily conserved and exert a similar function in higher species. This is especially evident by comparing "loss of function" mutants of gap genes in different species. For example, orthodenticle (otd) and empty spiracles (ems), are activated by high concentrations of bicoid in partially overlapping regions of the insect head (Cohen and Jurgens, 1990; Finkelstein and Perrimon, 1990).  Otd or ems are required for the specification of the first or the second and third neuromeres, respectively, which are the segments of the insect brain (Hirth et al.,1995).  The mammalian homologues of otd and ems,  Otx1/Otx2 and Emx1/Emx2,  are expressed in partially overlapping domains of the forebrain (Simeone et al.,1993; Simeone et al.,1992; Boncinelli et al.,1993). The knock-out of Otx2 by homologous recombination results in a failure of development of the mammalian prosencephalon (Acampora et al.,1995; Ang et al.,1996; Pannese et al.,1995). This is the result of an apparent failure of induction or specification, suggesting an evolutionary conserved role of gap genes in the primary regionalization of the CNS (Acampora et al.,1995).

4. The specification of regional fate: increased complexity in body plan is associated with the accumulation  of variants of developmental genes.  Once the allocation of sets of cells in the embryonic parasegments and segments of the fly embryo is completed by the orderly sequence of expression of gap, pair rule and segment polarity genes, these sets of cells become the founders of the so called compartments (Lawrence and Struhl, 1996).  Compartments are cell assemblies acting as autonomous units of development and lineage restriction.  Cells belonging to one compartment do not mix with cells of other compartments and cell lineages do not cross compartment boundaries (see section 5). The founder cells of one compartment share a "genetic address" given by a unique combination of homeodomain selector genes (Lawrence and Struhl, 1996; Morata and Lawrence, 1977; McGinnis and Krumlauf, 1992). The cooperative action of selector genes gives to the cells of one compartment, and to their descendants, a unique set of instructions that define their identity. This assumption is justified by the fact that mutations in homeotic selector genes in Drosophila change the phenotype of one body part to that of another. These homeotic mutations led to the discovery of conserved sets of homeobox genes clustered in the genome of several species.  As in the case of gap genes, homeodomain proteins  bind specific segments of DNA, called enhancers, involved in the spatial  and temporal control of gene transcription  (Gehring et al.,1994). Little progress has been made in identifying the targets of homeobox genes.  Direct targets of homeodomain genes are not likely to directly control cellular functions, but other transcription factors that are part of a genetic cascade that progressive refine the body pattern. The ultimate targets of this cascade are genes with more restricted or "executive" roles, including molecules involved in axon guidance, cell growth or neurotransmitter phenotypes. 

In the mouse, four Hox clusters are present, characterized by sequence similarity to genes of the single HOM-C complex of Drosophila (Figure 1). Similar genes are colinearly arranged  on different clusters, and share identical direction of transcription, suggesting that the four Hox clusters arose by duplication of an ancestral complex. These and other observations gave origin to the speculation that duplications of these homeobox gene clusters may have contributed to the evolution of the vertebrate body plan (Krumlauf, 1994; Ruddle et al.,1994).  Loss of function mutations of Hox genes in mice result in deletions of rhombomeres and other growth abnormalities that are not obviously consistent with their hypothesized role of conferring a specific identity on a compartment (Chisaka et al.,1992; Chisaka and Capecchi, 1991). However, it is possible that Hox genes have multiple roles, both in segmentation and in segment specification (Lumsden and Krumlauf, 1996; Capecchi, 1997). An intriguing feature of the Hox clusters is that the order of the genes in the chromosome recapitulates their anteroposterior pattern of expression within the embryos, such that genes at 3' are the most cranial and genes at 5’ progressively are more caudal. Since no Hox genes are expressed in vertebrates above the midbrain, it was long suspected that different sets of homeobox genes might be present in the most anterior parts of the neuraxis.

 Several families of homeobox genes expressed in the prosencephalon have been described (Xue et al.,1991; Murtha et al.,1991; Price, 1993; Simeone et al.,1992). These prosencephalic, homeodomain-containing genes are divergent homeobox genes which, in contrast to Hox genes, do not exist in multigene clusters. The reason underlying this lack of correspondence remains to be elucidated, but suggests the idea that the morphogenesis of the anteriormost regions of the neuraxis follows different rules. Similar to Hox genes, each of these divergent homeobox gene families comprises several homologous members that presumably arose by gene duplications. Does this represent gene diversification, a protective redundancy, or both?  An intriguing feature is that members of these homeodomain gene families overlap extensively in their expression pattern and either cooperate or partially compensate each other in function (Yoshida et al.,1997; Suda et al.,1997). One of the most striking examples is represented by the Dlx family. Dlx genes comprise six members all expressed within the ventral telencephalon (basal ganglia) and ventral diencephalon, with a sharp border between dorsal and ventral thalamus. Null mutation by germline homologous recombination of a member of the Dlx family, Dlx2, does not lead to any apparent dysfunction within the CNS (Qiu et al.,1995). However, an abrupt decrease in Dlx2 expression triggered by antisense oligonucleotide treatment causes abnormalities of proliferation and differentiation in cells from the ventral telencephalon in vitro (Ding et al.,1997). Thus, each Dlx gene may exert unique functions during development, but Dlx family members may be able to compensate for one another. Indeed, mice carrying a double knockout of Dlx1 and Dlx2 show abnormal differentiation of the basal ganglia (Anderson et al.,1997b). The impact of a combinatorial loss of pairs or sets of genes may give an indication on how these family members interact in normal development.

5. Regional specification of the CNS is genetically "coded" but also sensitive to cell to cell interactions.  One important aspect of the phenotype conferred by homeodomain selector genes is the establishment of a specific surface affinity that make "like" cells stick to one another. Cells of a given compartment not only share aspects of their fate, but they tend not to mix with cells of adjacent compartments. Thus, proliferation and cell lineages are delimited by compartment boundaries (Lawrence and Struhl, 1996). These boundaries are not fixed, but dynamically change with further development.

Signaling centers at compartment boundaries and in other specialized structures are a major force driving morphogenesis and cell fate.  Boundary regions are not just partitions that restrict cell mixing, but areas where cells interact and release diffusible factors. For example, ligands of the FGF system, including FGF3 and FGF8, are expressed in the rhombomere boundaries (Mahmood et al.,1995; Crossley and Martin, 1995). In the spinal cord, two ligands acting in contrasting manner on homeobox gene expression are generated at opposite poles of the developing neural tube. Sonic Hedgehog (Shh), a member of a family of secreted proteins structurally similar to Drosophila's segment polarity gene Hedgehog (Hh), is synthesized by the ventralmost regions, the notochord and floor plate. Shh diffuses within the ventral tube and exerts distinct functions through different concentration thresholds: it specifies floor plate, motor neurons, and ventral interneuron fates in this region (Tanabe and Jessel, 1997). Conversely, TGFß-like bone morphogenetic proteins (BMP) are present in the ectoderm at the boundary with the neural folds, and in the dorsal midline of the cord after neural tube closure. BMPs may induce subsets of sensory interneurons in the dorsal spinal cord (Liem et al.,1995). It is likely that these diffusible ligands specify cell fate by modifying the expression of specific transcription factors. For example, Shh up-increases the expression of HNF3ß and nkx 2.1 and down-regulates Pax3 and Msx1, resulting in a rapid disappearance of these genes from ventral regions. This regional downregulation is necessary for the generation of ventral neuron types (Tanabe and Jessel, 1997).

In the vertebrate brainstem, two signaling systems, Shh secreted from cells along the basal plate (ventral midline) and FGF8 which diffuses from the isthmus and anterior neural ridge (see below) cooperate to induce dopaminergic and serotoninergic cell fate (Ye et al.,1998).  Subtle variations in these interactions may alter the number of these cell types, which are involved in many aspects of neuropsychiatric disorders.  For example, dopaminergic transmission has been involved in schizophrenia, Tourette Syndrome and drug addiction, and an increase in serotonin is one of the most consistent abnormalities found in autism. Future studies linking gene knockout for these secreted factors with precise aspects of CNS development are eagerly awaited.

A segmental organization is much less obvious in the vertebrate forebrain, due to the growth of the cerebral cortex, basal ganglia and associated regions. Nonetheless, even within the morphologically complex organization of the forebrain, regional boundaries may be delimited based on the pattern of expression of homeodomain genes, gap genes and other transcriptional regulators during embryogenesis. The dynamic expression patterns of these transcription factors are thought to reflect the acquisition of regional identity. A major boundary with restricted cell mixing appears to exist between the basal ganglia and the dorsal telencephalon (the future cerebral cortex) (Fishell et al.,1993). Different sets of homeobox genes are expressed within these two regions, Dlx and Gbx2 genes being restricted to the basal region, and Emx1/2, Pax 6 and Tbr-1 to dorsal regions (Porteus et al.,1991; Simeone et al.,1992; Simeone et al.,1994; Bulfone et al.,1995) (Figure 2). Interestingly, the null mutation of Pax 6 causes a profound disruption of this boundary (Stoykova et al.,1996; Stoykova et al.,1997) and the lack of two Dlx genes appears to disrupt neuronal migration across this boundary (Anderson et al.,1997a).  

Elegant fate mapping studies in avian embryos and differential patterns of gene expression have suggested the existence of a major longitudinal boundary within the forebrain, dividing the ganglionic eminence (basal telencephalon) from the future cerebral cortex (Couly and Le Douarin, 1987; Shimamura et al.,1995).  Further divisions of the forebrain in the transversal plane into segmental units called  "prosomeres" has also been proposed (Bulfone et al.,1993) (Figure 2). According to this map, the cranialmost regions of the CNS are hypothalamus and pituitary, and the cerebral cortex develops as a dorsolateral overgrowth that successively, because of its greater mass, overlays, dorsally and anteriorly, the anteriormost regions of the CNS.  However, the location of lineage restriction boundaries in the forebrain is still an open question (Walsh and Cepko, 1992; Walsh and Cepko, 1993; Anderson et al.,1997b), and more experiments are needed to understand forebrain patterning.  Another area of intensive investigation is the identification of signaling centers that regulate forebrain morphogenesis and cell identity, similarly to those described for the spinal cord (Rubenstein and Shimamura, 1997; Shimamura and Rubenstein, 1997; Ye et al.,1998).

In principle, if the regional and cellular identity of a set of cells was given only by the expression of a combination of selector genes, then these decisions should be cell-autonomous, that is, independent from the influence of the cells' neighbors. This theory can be tested by transplantation experiments: cells allocated to a given region should not change their fate when transplanted to a different environment. However, contrary to this hypothesis, progenitor cells from the mouse basal ganglia can integrate themselves into the neuroepithelium of the dorsal cortex and, in time, become indistinguishable from cortical neurons, both morphologically and in projection patterns  (Fishell, 1995).

The rhomboencephalon of the vertebrate brain has been an ideal region for testing the autonomy of cell fate decisions. This structure is organized in a metameric manner, similar to the segments of the invertebrate CNS. Each rhombomere expresses a unique combination of Hox genes, called a Hox code, which specifies each rhombomere's phenotype, including the descendant neuronal types, their location and their peripheral connections (Kessel and Gruss, 1990; Lawrence and Morata, 1994; Keynes and Krumlauf, 1994).  Transplantation studies have confirmed that there is a direct correlation between commitment to a rhombomere-specific fate and Hox gene expression. Emerging rhombomeres maintain their original Hox code and do not change their identity when transplanted to a more anterior location (Guthrie et al.,1992; Guthrie et al.,1993; Guthrie and Lumsden, 1992). However, in other cases, depending on the stage at which the grafting is performed, neural tube translocations, inversions, and transplantation of individual neurons to a new location lead to a respecification of motor neuron identity, as defined by their peripheral projection and by their Hox gene expression (Appel et al.,1995; Eisen, 1991; Matise and Lance-Jones, 1996; Tanabe and Jessel, 1997). For example, rhombomeres transplanted to a more posterior location are respecified to a Hox code and a fate corresponding to that of the new location (Grapin-Botton et al.,1995). These changes in fate may be due to a "posteriorizing" signal, expressed within the neuroectoderm that can transform anterior tissue into posterior (Marshall et al.,1992; Kessel, 1993; Grapin-Botton et al.,1995). Such observations confirm the existence of rostrocaudally and dorsoventrally restricted signals that control cell identity by acting on the expression homedomain proteins in progenitor cells. The posterior inducer is possibly a retinoid. The major source of retinoic acid in the vertebrate CNS is the regressing Hensen’s node, a region at the boundary between rhombencephalon and spinal cord. This region is the source of a gradient of active retinoids that diffuse within the plane of the neuroectoderm excluding the head (Balkan et al.,1992). Systemic administration of retinoic acid has profound influences in CNS development. In vivo retinoic acid treatment causes anterior to posterior transformation resulting, in the most extreme case, in an almost complete absence of the forebrain (Tabin, 1991; Marshall et al.,1992; Simeone et al.,1995). Thus retinoic acid may contribute to the early distinction between trunk and head by activating trunk-specific regulatory genes, and by suppressing anteriorly-expressed genes, specifically, Otx2, which is necessary for rostral CNS development (Simeone et al.,1991; Simeone et al.,1990; Acampora et al.,1995; Ang et al.,1996; Simeone, 1998).

In summary, regional specification as exemplified by the Hox code and by the spatially restricted expression of other homeobox genes is not completely autonomous, but dynamically interacts with morphogenetic signals emanating from cell-to-cell interactions at the boundaries.

6. Boundary regions may contain specialized cells that act as "organizers". Organizers are specialized groups of cells that are capable of inducing in adjacent cells a coherent set of behaviors, including changes in fate, growth, and migration.  Thus organizers are thought to influence, either directly or by a chain of indirect reactions, cascades of gene expression in distant cells, leading to the generation of a new structure in surrounding tissue.  Typically, the transplantation of cells from an organizer to a different location induces the growth of an ectopic organ. Spemann and Mangold in 1921 found that the transplantation of the dorsal blastopore lip (the tissue at the end of the primitive streak) of an amphibian early gastrula in the ventral ectoderm could induce a secondary CNS and body axis (as reviewed by (Kintner, 1992)). They named this tissue an "organizer" and postulated that it induced in the host tissue an appropriate pattern of nervous tissue differentiation. In the mammalian CNS, the isthmus, the constriction that delimits mesencephalon from rhomboencephalon, contains cells that have similar properties. Transplantation of isthmic tissue into the caudal diencephalon causes the ectopic growth of a secondary midbrain, which is a mirror image of the first. It appears that a factor normally expressed in the isthmus, fibroblast growth factor 8 (FGF8), is largely responsible for the morphogenetic action of isthmic tissue. A localized source of FGF8 transplanted in the caudal diencephalon is capable of inducing a second midbrain in this forebrain location, which is also in a mirror image orientation with respect to the normal midbrain (Crossley et al.,1996). Furthermore, FGF8 induces in the forebrain the ectopic expression of FGF8 itself and of genes normally involved in the morphogenesis of the midbrain, including Wnt-1 and En  (Crossley et al.,1996).  These observations suggest that FGF8 is involved in the normal morphogenesis of the midbrain, possibly forming a gradient peaking at the isthmus, and that an ectopic source of FGF8 cranial to the isthmus is able to form a new isthmus-like signaling center. The ectopic gradient of FGF8 (and possibly of other factors) is "interpreted" in a graded way by the new compartment and is sufficient to induce a transformation of the caudal diencephalon to midbrain. Recent data have shown that FGF8 is necessary for the formation of the caudal midbrain and of rostral hindbrain, as these regions are absent in mouse embryos that carry a deficient FGF8 gene (Meyers et al.,1998). Interestingly, at an earlier stage of development, FGF8 is one of the factors to be expressed within the primitive streak and the node, the mouse homologue of the amphibian organizer.

7. The anteriormost CNS region, the cerebral cortex, has undergone a progressive expansion during the course of evolution.

 After the initial regional specification of the CNS, each region has evolved mechanisms for the control of its overall growth and for the differentiation of appropriate numbers of the indigenous cell types. The cerebral cortex has undergone a considerable expansion in its surface area during phylogenesis. The surface area of the human cortex is 1000-fold larger than that of the mouse and 100-fold larger than that of the monkey. This increase in surface area is not matched by a corresponding increase in thickness (the human cortex is only 3-fold thicker than the mouse) (Rakic, 1988; Rakic, 1995). Furthermore, cell and synaptic density are not substantially different in the mouse and human cerebral cortex (Changeaux, 1985). Two general conclusions can be drawn from these considerations: (1) the fundamental unit of the cortex (i.e., cortical column) has remained substantially the same during the evolution of the mammalian species, but the number of these units has increased; (2) there should be a mechanism operative during ontogenesis that leads to an increase in cortical surface area while maintaining the correct proportion of the different neuronal types.

Cortical neurons are generated, in virtually all mammalian species, during embryogenesis, in a layer of cells situated around the cerebral ventricles, the pseudostratified ventricular epithelium (PVE). Progenitor cells within this layer proliferate, undergoing up and down movements correlated with the cell cycle  (see below) and, after their final mitosis, leave the PVE and start migrating toward the primordial cerebral cortex  (cortical plate) (Sidman and Rakic, 1973). Although cortical areas differ, the pattern of neurogenesis appears to be substantially similar throughout the cerebral cortex. The mechanisms that control the emergence of diversity within cortical areas may be intrinsic to the progenitors  (i.e., the idea of a protomap present within the ventricular neuroepithelium) (Rakic, 1988). Alternatively, the incoming afferent population may account for the generation of these regional differences while maintaining both a fundamental identity in the cellular components of the cortex and their regularly arrayed manner (Shatz et al.,1990; O'Leary et al.,1994).

8. Neuronal migration in the cerebral cortex is guided by specialized glial cells and by extracellular molecules.

An integral component of one cell phenotype is the capability of reaching the appropriate location. Very little is known about the molecular signals that guide migrating neurons to the appropriate place in the cortical plate. In laminar structures such as the cerebral and cerebellar cortices, glial cells of a specialized nature, the Bergmann glia and the radial glia for the cerebellum and cerebral cortex, respectively, are thought to guide young neurons in their radial migratory path. A smaller proportion of neurons migrates non-radially through unknown routes.

(See Figure 3)

Furthermore, the settling and aggregation of cells in a particular layer is coupled to their generation time.  The cerebral cortex begins a simple cell layer called the preplate (The Boulder Committee, 1970). This primitive layer is split by incoming young neurons forming the first two cortical layers, the marginal zone (layer I) and the subplate (Marin-Padilla, 1978). Cell labeling studies with radioactive thymidine have shown that subsequently cells pile up underneath the marginal zone to form the other cortical layers, in the sequence 6-5-4-3-2 (Bayer and Altman, 1991). Since the youngest cells always migrate past the last and occupy the area nearest to the marginal zone, it has been hypothesized that marginal zone cells (the Cajal-Retzius cells) secrete a chemoattractant or guidance molecule that controls the migration of cortical cells.

Several mouse and human mutations exist that disrupt neuronal migration and cortical layer morphogenesis. The analysis of these mutations has given important insights into the complex signaling mechanisms that guide migrating neurons and has emphasized the importance of genetic components in the mechanism of cell migration . The reeler,scrambler and yotari mutations in mouse cause an identical phenotype, in which younger migrating neurons are unable to penetrate the layer of older neurons. As a result, the preplate is never split into marginal zone and subplate and there is an inversion of cortical layers. Thus, the mutation may be viewed as blocking development at the preplate stage, with a consequent disorganization of the cerebral cortex and other laminated structures (Caviness, 1982; Sheppard and Pearlman, 1997). The reeler genes encode for Reelin, a secreted protein with structural homology to other extracellular matrix proteins (D'Arcangelo et al.,1995). Reelin is produced by Cajal-Retzius cells during and after the phase of migration and is densest in the marginal zone (D'Arcangelo et al.,1995). The scrambler and yotari mutations involve the gene mouse disabled homolog 1 (mdab1) (Sheldon et al.,1997). Conversely, the disruption of the mdab1 gene in mouse causes a scrambler-type phenotype (Howell et al.,1997). Mdab1 encodes for an intracellular phosphoprotein that is present throughout the developing cerebral cortical wall and controls cell migration in the cells that express it.  However, mdab1 has no catalytic activity, so it may function by interacting with other proteins through its PTB domain and tyrosine-phosphorylated motifs. For example, mdab1 interacts with Src, Abl and other nonreceptor tyrosine kinases. It is interesting to speculate that reelin may be a potential ligand for non-receptor tyosine kinase pathways involved in the control of migration within young neuronal cells. However, the story is more complex since, in addition to reelin, there is a host of other mutations that disrupt cortical migrations. For example, the disruption of cyclin-dependent kinase 5 (Cdk5) and its activator, p35/Cdk5r, also cause an inversion of the inside-out neurogenic gradient in the cortex (Gilmore et al.,1998). Cdk5 is a neurofilament kinase expressed in postmitotic neurons. Potentially, all these molecules could operate in a common signaling pathway.

Two genetic mutations in human, X-linked lissencephaly and “double cortex”, are phenotypically similar to reeler and scrambler mutations. These disorders result from a severe arrest of the migration of cortical neurons, and produces mental retardation and seizures (Dobyns et al.,1996). X-linked lissencephaly and “double cortex” result from the mutation of doublecortin, a protein expressed in the fetal brain (Gleeson et al.,1998; des Portes et al.,1998). Doublecortin is a likely substrate for protein kinases of the Abl and MAPK families and thus, like mdab, may be involved in signal transduction (Gleeson et al.,1998). This emphasize the idea that the different genes that produce abormalities of cell migration may be part of the same pathway, and that signaling through non-receptor tyrosine kinases may be critically involved  in the regulation of neuronal migration. Other genes and pathways are likely to be involved in the regulation of neuronal migration through different mechanisms. For example, LIS1, a gene on chromosome 17p13, is associated with human lissencephaly and with Miller-Dieker syndrome. LIS1 codes for a regulatory subunit of platelet activating factor (PAF) acetylhydrolase (Dobyns et al.,1993; Hirotsune et al.,1998). Neural cell adhesion molecules N-cadherin and N-CAM also control neuronal migration. In most cases, the precise mechanisms whereby these factors  control neuronal migration is still undetermined.

9.  Most cortical progenitors are restricted early to a specialized fate, but a few pluripotent cells persist in the adult forebrain.

There is little doubt that the process of neurogenesis, that is the specification of a wide variety of cells in appropriate numbers at appropriate times, involves a progressive restriction of the fate of pluripotential cells. But when this restriction occurs, and the precise mechanisms that guide it, are still open to question. A method that has been used to identify cell lineages in the cortex is the microinjection of replication defective retroviruses that express lacZ in the cerebral ventricles during neurogenesis. These recombinant viruses integrate their genome into a host cell during the cell’s DNA replication and are faithfully copied into the descendants (Price et al.,1987; Cepko et al.,1990).  At least in principle, the whole lineage of cells that originates from the infected progenitor will be marked by lacZ, which is detected histochemically. The in vivo retrovirus technique has shown that, early in neurogenesis, the great majority of progenitors are lineage-restricted, that is, clones of cells are composed either of a type of neurons or of glia. Mixed clones only infrequently occur in the telencephalon in vivo  (Luskin et al.,1988; Luskin et al.,1993). This suggests that progenitors labeled at E14-16 (the beginning of rat cortical neurogenesis) are already fated to generate specific types of neurons or glia.

Recent studies have shown that the periventricular region of the telencephalon contains pluripotent precursor cells (stem cells) capable, at least in vitro, of differentiating into both neurons and glia (Weiss et al.,1996). The failure to detect these cells with retroviral techniques may be due to their relatively rare occurrence in vivo. Interestingly, a small number of these pluripotential cells persist in the subependimal zone of the adult brain in virtually every mammal that has been examined (Reynolds and Weiss, 1992). These cells, if placed in vitro in the presence of growth factors (see below), will give rise to a mixed population of neurons and glia. However, in vivo, the majority of their progeny differentiates into glia or undergoes apoptosis, except for a small percent of neurons that migrate into the olfactory bulb (Morshead et al.,1994; Kuhn et al.,1997). The adult hippocampus also contains stem cells that are capable of differentiating into neurons, the hippocampal granule cells. The capability of the hippocampal stem cells to give rise to new neurons in the adult has been documented in rodents as well as in primate species, including the human brain (Kempermann et al.,1997; Eriksson et al.,1998; Gage et al.,1998). The conditions that promote adult hippocampal neurogenesis and the functional significance of adding extra neuronal cells to the adult synaptic circuit is presently not clear. Nevertheless, these finding have generated considerable interest since the theoretical possibility exists that new neurons could be generated to replace those lost to disease or degeneration (Weiss et al.,1996). 

  How can a single progenitor cell give rise to daughter cells with distinct fates? The molecular mechanisms of lineage restrictions are likely to be similar to those described in Drosophila's neuroblasts. They reflect what has already been discussed regarding the formation of compartments, that is, the differential inheritance and activation of selector genes in daughter cells (Jan and Jan, 1995; Rhyu et al.,1994). Genes that regulate the orientation of the mitotic spindle at mitosis influence the asymmetric segregation of molecules in daughter cells and may have effects on cell fate (Kraut et al.,1996; Spana et al.,1995; Zhong et al.,1996; Jan and Jan, 1998). In other instances, daughter cells are initially equivalent, but secondary cell-to-cell interactions modify their fate. These interactions in some cases require cell contact.  A classical example in Drosophila is the singling out of neuronal precursors among competent cells in the neurogenic cluster, involving competitive cell surface interactions through Notch/Delta (Jan and Jan, 1995; Artavanis-Tsakonas et al.,1995). In other cases, cell interactions result in the local release of factors, as previously discussed for Shh  in the neural tube. In the brain, Shh, members of neurotrophins, platelet derived growth factor (PDGF) and fibroblast growth factor (FGF) families have all been implicated in influencing either neuronal or glial cell fates (Vaccarino et al.,1995; Vescovi et al.,1993; Jessel and Melton, 1992; Jessel and Lumsden, 1997; Hynes et al.,1995; Vicario-Abejon et al.,1995; Raff et al.,1988).  A recent report shows that the generation of dopamine and serotonin neurons requires the presence of at least two families of factors, Shh and FGFs, and that each controls cell fates along either the antero-posterior or the dorso-ventral axis in the neural tube (Ye et al.,1998).

An important aspect of cell fate is the determination of the final position of cortical neurons within the cortical laminar architecture. To establish whether intrinsic or extrinsic factors regulate this phenomenon, cortical progenitors from a neurogenetic period when layer 5/6 are generated have been transplanted to the neuroepithelium of older animals, that is generating layer 2/3 neurons.  If the progenitors differentiate and migrate according to their own schedule, lineage determination is intrinsic. If, however, they differentiate according to the new environment, that is, they  migrate to layer 2/3, one cell’s lineage can be influenced by surrounding cells. The results of these experiments have supported a mixed theory. Cortical progenitors for layer 5/6 neurons generally migrate according to their own schedule. However, if these cells are transplanted early, so that they are present in the new environment during the first 4 hours after their last mitotic division, their final position can is determined by the new environment.  After this time, their fate is fixed and no longer can be influenced by the surrounding cells. Thus, there seems to be an interaction between phases of the cell cycle and receptivity of the cell to environmental influences that are important for cell specification (McConnell and Kaznowski, 1991). Interestingly, the fate of older dividing cells, the progenitors for layer 2/3 neurons, is not as flexible and cannot be changed upon transplantation to that of an earlier progenitor cell (Frantz and McConnell, 1996).  This experiment again suggests that cell to cell interactions may be important for the specification of cell fate, but only for early progenitor cells (Frantz and McConnell, 1996).  Once cells have acquired a state of determination, they cannot  "revert" to an earlier fate, and developmental choices may not be undone.

10. The number of neurons generated in the cerebral cortex is determined by the rate of proliferation and the number of mitotic cycles.  The rate and mode of divisions of progenitor cells and the number of cell cycles they undergo before differentiation have a direct impact on o the quantity of cells generated.

A factor that likely contributes to the regulation of neurogenesis is the length of the cell cycle.  Cumulative labeling of the proliferative population with bromodeoxyuridine, an S-phase marker, has established that the murine cerebral cortex is generated through 11 cell cycles and that the length of the cell cycle increases from approximately 8 hours at E11 to 20 hours by the end of the neurogenesis  (E17) (Takahashi et al.,1995).

Interestingly, it is during the last three and slowest mitotic cycles that the vast majority of neurons of the cortex are generated. Takahashi et al (1996) measured the fraction of progenitor cells that exit the cycle through mouse neurogenesis (the Q fraction) (Figure 4a).  Q progressively increases in a nonlinear fashion, from 0 at the beginning of neurogenesis (when all the cells are proliferative), to 1 at the end (when all the progenitor cells exit the cycle). When Q is equal 0.5, the number of cells that leave the cycle is equal to those that re-enter it; this point is reached at E14 in the mouse PVE. Thus, for the first 7 cell cycles (the interval E11-E14) the PVE increases in size, and there is still a net amplification of the proliferative population. After Q exceeds 0.5, however, the rate of cells leaving the proliferative population to become neurons exponentially increases, resulting in the contraction of the PVE (Figure 4a). As a consequence, most cortical neurons, populating the dense upper cortical layers, are formed during the last two to three days of neurogenesis, and during the last integer cell cycles (Figure 4b) (Takahashi et al.,1994; Takahashi et al.,1996; Caviness et al.,1995).

It is likely that the progressive increase in the Q fraction is due to a change in the mode of division of PVE progenitor cells. For example, it has been postulated that the progenitor population expands through symmetric divisions before neurogenesis begins (Rakic, 1995), whereas a predominant asymmetric mode of division underlies the generation of neurons. Neurons can be also generated through symmetric terminal divisions (Figure 4c). Morphologically identified symmetric and asymmetric divisions have been recently imaged in the developing ferret neocortex. In symmetric divisions, the two daughter cells continue to reside in the PVE and remain closely associated, whereas in asymmetric divisions one daughter starts migrating to the upper layers (Chenn and McConnell, 1995)(Figure 4c). These imaging studies have confirmed that symmetric divisions predominate before neurogenesis and gradually decrease during cortical development. However, the PVE is a mosaic of different types of progenitor cells dividing in symmetric or asymmetric fashion, and single progenitors may switch from a mode a division into the other (Halliday and Cepko, 1992; Cai et al.,1997). The factors that regulate these decisions are unknown, but they are likely to be similar to those that promote the decrease in Q and the increase in P fractions. It is also possible that cell death may contribute to the depletion of progenitors, shrinkage of the PVE and the eventual end of neurogenesis (Blaschke et al.,1996; Thomaidou et al.,1997). However, mice lacking cell death effector molecules through germline homologous recombination have profound abnormalities in cortical anatomy, suggesting that these molecules may have a role during earlier phases of cortical morphogenesis (Kuida et al.,1996; Kuida et al.,1998). The time course of morphogenetic cell death in the developing cortex is the subject of intense investigation (see section 12).

 The factors that control the decision of one cell to exit or re-enter the cell cycle ultimately produce variations in the total number of mitotic cycles for a given proliferative population and affect the total number of cells generated. The progression of the cell cycle is regulated through a checkpoint in early G1. Exposure to mitotic growth factors during this phase leads to an early commitment of the cell to divide again, whereas growth factor deprivation in G1 leads to the degradation of D cyclins and to a failure to re-enter the cycle (Ross, 1996). Both epidermal growth factor (EGF) and basic fibroblast growth factors (FGF2) are present in the developing cerebral cortex, have predominantly mitogenic actions and sustain the proliferation of progenitor cells (Weiss et al.,1996). Whereas EGF acts relatively late, FGF2 acts before and during neurogenesis (Drago et al.,1991; Nurcombe et al.,1993; Vaccarino et al.,1995; Robel et al.,1995). FGF receptors are abundant in the PVE at early phases of neurogenesis (Vaccarino et al.,1999a; Vaccarino et al.,1999b). Several lines of evidence indicate that FGF2 extends the proliferative potential of both neuronal and glial progenitor cells in vitro by increasing the number of cycles these cells undergo before finally differentiating (McKinnon et al.,1990; Bogler et al.,1990; DeHamer et al.,1994). This action is relevant also in vivo. The microinjection of FGF2 in the embryonic cerebral ventricles increases the generation of cortical neurons; conversely, mice lacking a functional FGF2 gene have a decrease in the number of cortical cells and a decrease in the number of neuronal progenitors early in neurogenesis (Vaccarino et al.,1999a). Thus, FGF2 is required for the attainment of an appropriate cell number in the cortex. 

While FGFs promote cycling and self-renewal of progenitor cells, factors of the TGFß family such as bone morphogenetic proteins (BMP) promote cell cycle exit and neuronal differentiation, both in the spinal cord and in the telencephalon (Liem et al.,1995; Li et al.,1998). In addition to BMPs, several other secreted factors appear to promote neuronal differentiation, including pituitary adenylate cyclase-activating peptide (PACAP) and amino acid neurotransmitters (see section 11). PACAP and amino acids such as glutamate and GABA act (at least in vitro) by inhibiting precursors from re-entering the cell cycle (LoTurco et al.,1995; Antonopoulos et al.,1997; Lu and DiCicco-Bloom, 1997).

In summary, multiple factors interact within microdomains of the PVE to regulate the timing of exit from the cell cycle, the number of progenitor cell divisions, and cell differentiation. Other molecules may modulate the length of the cell cycle and cell death among progenitor cells.

11. Classical neurotransmitters have a role during development.  The specialized role of classical neurotransmitters as mediators of synaptic communication between neurons may have evolved from a basic function in primitive organisms, in which these substances  were used in intra- and intercellular signaling. For example, acetylcholine, GABA and the monoamines regulate basic functions such as growth, movement, cell cleavage and metamorphosis in Tetrahymena, flatworms and other primitive organisms (Lauder, 1993). These phylogenetically old functions are reiterated during development in the nervous system. Classical neurotransmitters are present in the nervous system long before synaptic connections are formed, and experimental increase or decrease in transmitter levels affect CNS morphogenesis at many different levels, including cell proliferation, neurite outgrowth and cell migration. Furthermore, these effects are mediated by the same intracellular transduction pathways activated by synaptic stimulation in neurons (adenyl cyclase, phosphatidylinositol turnover, calcium influx, etc.), which suggests that they are receptor mediated (Lauder, 1993). Often, receptors for these transmitters are present during CNS development at higher levels than the adult, and furthermore, specific receptor subunits may be exclusively present during the developmental period (Poulter et al.,1992; Meinecke and Rakic, 1992; Bovolin et al.,1992), suggesting that they subserve developmentally specific functions.

Cells containing gamma aminobutyric acid (GABA), the major inhibitory transmitter of the CNS, are present at the earliest stages of neurogenesis in the cerebral cortex both in rodents (E14) and in monkey (E41).  GABA immunoreactivity can be visualized in neurons of the cortical plate, in migrating neurons of the intermediate zone and even in morphologically mature, neuron-like cells within the ventricular zone (Schwartz and Meinecke, 1992; Meinecke and Rakic, 1992).  Electrically coupled progenitor cells within the ventricular zone express bicuculline-sensitive GABA_A receptors (Lo Turco and Kriegstein, 1991), and studies in culture have suggested that these receptors mediate the depolarization of progenitor cells, instead of causing hyperpolarization as in the mature CNS (Serafini et al.,1995). In fact, both glutamate and GABA depolarize neuroepithelial progenitor cells of the neural tube by an action on sodium/calcium channels, and this depolarization may cause a decrease in their proliferation (Serafini et al.,1995; LoTurco et al.,1995).

Brainstem neurons containing biogenic amines, such as noradrenaline, serotonin and dopamine, are "born" early and their projections reach the cerebral cortex during a time (E14) when cortical neurogenesis is beginning  (Levitt et al.,1997). However, aminergic terminals do not invade the cortical plate but form synaptic contacts in the subplate, a layer of cells situated under the cortex that provides a temporary scaffolding for the incoming cortical afferents. Subsequently, the afferents progressively grow into the cortex, whereas the subplate undergoes cell death during the postnatal period (Shatz et al.,1990; Ghosh et al.,1990; Friauf et al.,1990). The early exposure of cortical neurons to biogenic amines raises the question of their role in neuronal differentiation. Numerous reports, especially utilizing cells in culture, have described both stimulatory and inhibitory effects of monoamines and acetylcholine on neurite elongation, depending on the state of the cell and the presence of non-neuronal cells. Growth inhibitory signals provided by neurotransmitters may represent "stop" signals for the growth cone preluding to synapse formation. For example, treatment of experimental animals during embryogenesis with cocaine, which increases the availability of monoamines by blocking their reuptake, causes permanent morphometric alterations in the cingulate cortex, specifically excessive growth of apical dendrites and disorganization of cortical cytoarchitecture (Levitt et al.,1997). Cocaine exposed animals were found to have a decreased coupling of D1 dopamine receptors to Gs proteins in the cingulate cortex. These animals were also selectively impaired in dopamine-mediated behaviors, suggesting that cocaine treatment resulted in a permanent and selective dysfunction of the D1-receptor system.  Thus, the in vivo experiments with cocaine are consistent with the evidence in primary cultures, suggesting that dopamine, via the activation of D1 receptors, decreases axonal and dendritic outgrowth (Swarzenski et al.,1994; Reinoso et al.,1996).  These studies also reveal the potent (and apparently irreversible) effects of cocaine on the developing CNS.

12. Large numbers of neurons naturally die during the early phase of CNS maturation.  Neurons are produced in excess and are later eliminated by a process of natural cell death called "apoptosis" (Pettmann and Henderson, 1998). Apoptosis is the death of a cell carried out by a program encoded by its own genome, and is descriptively characterized by a stereotyped sequence beginning with shrinkage and breakage of the chromatin in the nucleus (Fraser and Evan, 1996; Johnson and Deckwerth, 1993). This death program involves a common set of molecules conserved throughout evolution, beginning with primitive unicellular eukaryoriotes (Vaux and Strasser, 1996).  Thus apoptosis may be as old as the cell itself. The reason why cell death programs have been conserved through animal evolution may be that the ability of an organism (or of a colony of cells) to kill part of self may provide competitive advantage to the remaining cells. In general, cells eliminated by apoptosis are abnormal and/or potentially dangerous (i.e., cells that fail to follow the appropriate programs of division or differentiation; cancer cells; autoreactive lymphocytes; or cells infected with a virus).

Apoptosis plays an essential role during development, such as tissue sculpting during morphogenesis, and the maturation of neuronal circuitry in the CNS. The common mechanism is that excess cells undergo apoptosis under conditions of scarcity of trophic factors. For example, in sympathetic ganglia, neurons undergo dramatic cell death in early embryogenesis unless they are able to make connection with their target. The target is source of nerve growth factor (NGF) for which several neurons compete. The characterization of NGF has led to the discovery of large families of growth factors present in many different areas of the peripheral and central nervous system, factors that in many cases perform functions similar to those of NGF. However, in the CNS, deprivation of a target does not always lead to death of the afferent neuronal population. In this location, growth factors may be delivered to a neuronal cell body not only retrogradely by the target, but also anterogradely by the afferents. Furthermore, neuronal activity may regulate the synthesis of growth factors, and pattern of activity in the CNS are not only triggered by afferent stimulation, but are often a characteristic of the network (for example, re-entrant circuitry), or arise as a result of the spontaneous activity of isolated groups of neurons.

The time course of developmentally regulated cell death has not been defined rigorously.  It is known, however, that the phase of cell death is temporally defined and that different cell populations die at different times. For example, in rodents, sympathetic neurons are pruned down to reach the adult number by midgestation, whereas for cranial regions of the neuraxis the phase of cell death extends into the perinatal period (Ferrer et al.,1990; Ferrer et al.,1992). In the mammalian cerebral cortex, it has been estimated that approximately 25% to 30% of the neurons die and this process is completed in rat by the third postnatal week (see Figure 5).  However, the actual number of cells that die during this period and the mechanisms involved are largely unknown. An intriguing line of speculation has focused on the extent to which these mechanism are similar to those involved in the elimination of autoreactive lymphocytes in the immune system (Kuida et al.,1996; Raff, 1992).

13. Inborn genetic programs control the formation of synaptic connections, whereas activity regulates their remodeling. One of the most formidable tasks for neurons within the developing CNS is to successfully "find" the appropriate target to establish synaptic connections. In the adult human CNS, over a trillion neurons each connect with, an average, a thousand target cells according to precise patterns essential for proper functioning (Tessier-Lavigne and Goodman, 1996). Rather than establishing random connections that are later on shaped into the normal pattern of connectivity, from the very beginning neurons are endowed with the capability to "choose" appropriate targets. Different molecular mechanisms underlie the directed growth of the axon, the recognition of the target, and the transformation of the growth cone into a synapse; these mechanisms appear to be independent of electrical activity (Goodman and Shatz, 1993). The growth of axons toward the target is orchestrated by master control homeodomain proteins which, in turn, control the transcription of subordinate genes that regulate axonal pathfinding, differential cell adhesion and synapse formation.  The resulting pattern of connectivity is largely accurate but diffuse and will be later refined by mechanisms influenced by electrical activity.

Growing axons "navigate" in the developing neuropil with the help of a sophisticated structure, the growth cone, and make few errors of navigation.  The philopodia, dynamic web-like structures at the end of the growth cone of immature neurons, are able to "sense" or explore variations in the surrounding environment. These cues result in either axon growth (attractive cues) or in withdrawal and abrupt turn in a different direction (repulsive cues). Attractive and repulsive cues are either diffusible (long range) or nondiffusible (short range); short range cues generally involve contact of axons with cell surface and extracellular matrix proteins  (Tessier-Lavigne and Goodman, 1996). Once the target is finally reached, the growth cone stops and transforms into a synapse.

Short range guidance cues are expressed along defined pathways either on cells (including other axons) or on the extracellular matrix. Some of these pathways provide a permissive substrate that favors the adhesion of the fiber and allows the exertion of considerable tensile force on the axon, which in turn favors growth. These are substrates used by many different sets of axons and thus provide broad “highways” of growth. The molecules involved belong to the category of  "cell adhesion molecules" (CAMs), which are expressed throughout the CNS and support axonal elongation and fasciculation (growth of several neurites in bundles).  CAMs belong to three families, the cadherins, the integrins, and the immunoglobulin (Ig) superfamily. These transmembrane proteins possess a large extracellular moiety mediating adhesion and an intracellular portion that is linked to the cytoskeleton and participates in growth factor signaling. For example, the integrin family comprises several a and b subunits that are expressed by cells in distinct combinations, giving rise to heterogeneous binding specificities for ligands in the extracellular matrix, such as the glycoproteins laminin, vinculin and fibronectin (Reichardt and Tomaselli, 1991).

In contrast, members of the Ig superfamily promote adhesions between cells by homophilic (like-like) or heterophilic (like-unlike) interactions among different members such as N-CAMs, L1 and axonin-1, expressed on the surface of adjacent cells. In this manner, Ig-like molecules may regulate axon fasciculation by pulling axons together. Their reciprocal binding is also influenced by their degree of glycosylation and their sialic acid content, both of which are highly regulated (Rutishauser and Jessel, 1988). For example, a local increase in sialic acid content in Ig-like molecules expressed on the surface of axons promotes their defasciculationand and sorting out into smaller bundles as they reach the target.  Axon fasciculation can also be regulated by repulsive forces that push axons together by providing an unfavorable substrate. This mechanism explains why antibodies that block the function of Sema I, a transmembrane repulsive protein belonging to the Semaphorin family, trigger the defasciculation of axons in the grasshopper limb bud (Tessier-Lavigne and Goodman, 1996). Some glycoproteins of the Ig superfamily have a specific regional distribution, such as Telencephalin, expressed exclusively in neurons of the telencephalon (Yoshihara et al.,1994). This agrees with the suggestion that CAMs may be downstream targets of homeodomain proteins and, as such, may be involved in organizing the regionalization of the CNS (Edelman and Jones, 1993).

Diffusible factors act as short or long range soluble chemoattractants or chemorepulsants for growth cones. These molecules are distributed in concentration gradients and aid selective growth in a particular direction by specific sets of neurons. Netrin, for example, are a family of diffusible protein related to laminins that act as long range chemoattractants or, occasionally, as chemorepulsants. Cells at the ventral midline of the CNS express netrins and attract axons toward the midline in the formation of commissural connections. This guidance role of netrins is conserved from fruit flies to vertebrates (Serafini et al.,1996; Kennedy et al.,1994; Serafini et al.,1994; Harris et al.,1996; Ishii et al.,1992). Other long range cues are provided by ligands for receptor tyrosine kinases (RTKs), including fibroblast growth factor (FGF) and neurotrophin families (McFarlane et al.,1996; McFarlane et al.,1995; Riddle et al.,1995; ElShamy et al.,1996; Barbacid, 1995; Cabelli et al.,1995; Cohen-Cory and Fraser, 1995).

Selective recognition of cells within a target area is regulated by both soluble and cell-associated factors that organize the arrangements of axons at the cellular and subcellular level. First, axons have to recognize and invade the target area, a function that is regulated by both pathway and target-derived cues. For example, axons express members of the RTKs that recognize gradients of ligands secreted by the target. Some evidence suggests that retinal axons follow a downward gradient of FGF to enter the tectum (McFarlane et al.,1996; McFarlane et al.,1995), whereas an increasing gradient of NGF or NT3 is required for target recognition in other areas (ElShamy et al.,1996).

 After reaching the target area, axons may have to be sorted according to a particular cell type, or a dendritic layer to which they project, or they must be arranged according to the original topographical information that they carry, such as in maps.  A classical example is the arrangement of retinal axons in the tectum, which must reproduce a map of the retina. Experiments have shown that the retinotopic order is preserved even when size disparities are introduced, suggesting that selective affinity for molecular gradients are involved (Patterson and Hall, 1992).  Among the specific sets of molecules that precisely orchestrate these cellular behaviors, there is a family of RTKs, the Eph family. Their membrane-anchored ligands RAGS and ELF-1 are expressed as overlapping anterior-to-posterior gradients in the tectum (Tessier-Lavigne and Goodman, 1996).  These ligands act as selective contact repellents, with ELF-1 repelling temporal axons and RAGS repelling all retinal axons with a smooth gradient of sensitivity across the anteroposterior axis  (Zhang et al.,1996; Drescher et al.,1995; Cheng et al.,1995; Nakamoto et al.,1995).

The initial growth of synaptic connections is appropriate with regard to the target, but diffuse. In a second phase, these connections are rendered more precise by elimination of axons and synapses within inappropriate areas. Considerable evidence suggests that axons and synapses are formed in excess and are pruned down in large number during the early postnatal development (Innocenti, 1981). In rodents, this process extends during the first weeks after birth, partially overlapping with postnatal cell death (see Figure 5). In the fetal rhesus monkey, new axons start growing into the two major cerebral commissures (the corpus callosum and the anterior commissure) in the last weeks of gestation, peaking at birth. Axons are subsequently eliminated in the first two or three postnatal months at a precipitous rate. For example, during the first 3 postnatal weeks, axons are eliminated from the anterior commissure and from the corpus callosum at an average rate of, respectively, 1 and 50 axons/sec (LaMantia and Rakic, 1994; LaMantia and Rakic, 1990). The processes regulating this wholesale elimination are of considerable neurobiological interest, since the morphometry of the cerebral commissures seems to correlate with a variety of behavioral differences, including sex, sexual orientation, and handedness, although the reliability of these measures has been questioned (Weis et al.,1988; Allen and Gorski, 1991; Allen and Gorski, 1992).

Similar to axons, synapses are initially overproduced in the infant primate, reaching their maximum number simultaneously in all cortical areas, as well as in all cortical layers, during infancy (2-4 months of age) (Rakic et al.,1986). Cortical synapses are eventually pruned down to a density of approximately 15 to 20 synapses per 100 µm² of neuropil. Since neurons in different cortical layers and areas are born at different times, the rapid increase in the rate of synaptogenesis is not linked to the time of neurogenesis. The commonality in the rate of synaptogenesis contradicts the idea that afferents have a role in this process, and suggests that the cortex develops, at least initially, in a unitary manner that is independent of patterns of neural activity (Rakic et al.,1986; Granger et al.,1995).

Synaptic pruning may be partially the result of the elimination of overabundant axons. However, there is such a dramatic decrease in the number of synapses per unit of neuropil during development that it is likely that synapse elimination involves unique processes of its own.  Furthermore, axons and of synapses are eliminated through different time courses (Figure 5). In primates, where these phenomena have been best studied, axon elimination is completed in the first three months after birth, while the process of synapse elimination is relatively slow. Indeed, the adult number of synapses in the primate cerebral cortex is not achieved until near adolescence (Rakic et al.,1986). The fact that in primates synapse elimination and, therefore, refinement of the cortical circuitry, is not complete until puberty may be important with regard to age on onset of schizophrenia and other psychiatric disorders of peripubertal onset.

The processes of elimination of axonal branches and synapses are part of a phase of anatomical rearrangements that result in the adult pattern of connectivity. During this phase, both spontaneous and synaptically evoked activity shape these processes.

14. During critical periods, exuberant connections are pruned off as a consequence of activity-related processes. Although activity-dependent changes can occur through life, they are characteristically most pronounced during specific critical periods. These are periods in which the synaptic circuitry of a given brain region becomes stabilized in a mature conformation to establish a given function.  The best known example is the maturation of ocular dominance columns and other physical and functional characteristics of the synaptic circuitry within the visual cortical system. The primary visual cortex, like the rest of the cerebral cortex, is organized in vertical assemblies of neurons called cortical columns that are the functional units of cortical information processing (Mountcastle, 1957).  Cells in the layer IV of the visual cortex initially receive synapses from thalamic neurons from both the right or the left eye. However, eventually these inputs segregate such that separate cortical neurons or columns receive input from only one eye. To determine the role of visual activity in the formation of eye-dominance columns, cats were deprived of visual input from one eye. If this visual deprivation is during adulthood, there is no effect on the columnar organization of the visual cortex. However, if animals are monocularly deprived during the first two to three weeks after birth (the critical period), the cortical area occupied by columns for the functional eye enlarged at the expense of the deprived eye, that eventually becomes virtually unable to drive the activity of the cortex (Wiesel and Hubel, 1963; LeVay et al.,1990).  This physical and functional "disconnection" of an inactive input only occurs during the critical period, as eye suture after the first weeks of life no longer affect cortical representation; however, the effects are permanent for the life of that individual. One mechanism responsible for these effects is synaptic plasticity. A general rule is that connections whose activity is temporally correlated are strengthened, whereas connections that display non-correlated activity tend to be weakened. These phenomena generally depends on the activity of excitatory receptors (Antonini and Stryker, 1993b; Antonini and Stryker, 1993a).

Critical periods may vary depending on the area of the brain and the activity involved. Furthermore, it is likely that each cell's developmental history constrains these processes. In certain areas, such as the hippocampus, a degree of plasticity persists through life, and the cellular and synaptic circuitry will always be to some extent responsive to changing pattern of activity (Kempermann et al.,1997; Collingridge and Singer, 1990). This capacity reflects the important role that the hippocampus plays in learning and memory.

Despite the evidence that patterns of neural activity influence the organization of neuronal circuitry, the mechanisms involved remain elusive. Neuronal activity both stabilizes existing synapses and drives the formation of new ones, thereby influencing at least three processes: the pruning of synaptic connections from inappropriate areas, and the selective survival and sprouting of branches and collaterals, accompanied by the local addition of synapses, within appropriate areas (Jessel and Kandel, 1993; Shatz, 1990; Katz and Shatz, 1996).  Characteristically, in the cortex and in the optic tectum, axonal pruning involves a reduction in terminal arbor of each axon, and a decrease in the amount of overlap between terminals. These processes increase the refinement and precision of maps in these laminated brain areas.  An interesting example has been provided by the transplantation of an ectopic eye in frogs, which causes additional retinal fibers from the ectopic eye to innervate the tectum. The resulting phenotype is alternating bands of normal and third eye fibers in the tectum (Cline and Constantine-Paton, 1990; Constantine-Paton et al.,1990). This pattern is similar to the segregation of retinal ganglion cell axons into different, eye-specific laminae in the lateral geniculate nucleus and to the ocular dominance columns in the mammalian visual cortex.  Plasticity in the retino-tectal system of the three-eye frogs, in the retino-geniculate system and in the geniculate-cortical system of mammals is achieved by competitive processes driving the elimination of axon collaterals and synaptic pruning as well as by the active growth of collaterals within appropriate areas (Kandel and Jessel, 1991; Antonini and Stryker, 1993b; Antonini and Stryker, 1993a).  In all these cases, axon segregation and synaptic plasticity is  abolished by blocking neuronal activity in the input neurons; specifically, the NMDA-type glutamate receptors appear to be involved (Hahm et al.,1991; Kleinschmidt et al.,1987; Collingridge and Singer, 1990; Constantine-Paton et al.,1990). In addition to excitatory receptors, the process of local elimination/addition of collaterals and synapses may involve the competition for locally released growth factors, a similar mechanism to that operating in the regulation of neuronal number (Thoenen, 1995; Lo, 1995; Inoue and Sanes, 1997). For example, intracortical or intraventricular delivery of neurotrophins prevents the shift in ocular dominance after monocular deprivation (Maffei et al.,1992; Cabelli et al.,1995; Shatz, 1997) and the shrinkage of LGN neuron cell bodies receiving input from the deprived eye (Riddle et al.,1995).

Synaptic connections are remodelled through life, and these changes in synaptic structure and strength are thought to underlie learning. Long term potentiation (LTP) and long term depression (LTD), for example, consist of increases or decreases in synaptic strength in specific areas of the adult CNS and depend on patterns of activity.  Interestingly, NMDA receptors and growth factors are also involved in LTP and LTD. Thus, similar or identical mechanisms are used during the development of synaptic connections and the remodeling of these connections during learning (Kandel and Jessel, 1991; Jessel and Kandel, 1993).

15. The impact of gonadal steroids and the development of sexually dimorphic areas.   Gonadal steroids act on the developing nervous system to create a variety of sex differences in neural organization. Sexually dimorphic behaviors in invertebrate and vertebrate species have been linked to structural differences in the CNS (Allen and Gorski, 1991; Goy and McEwen, 1980). Although some of these effects are likely to be hormone-independent, gonadal steroids (estrogens and androgens) acting during the course of CNS development can influence the number, size, and connectivity of neurons in a variety of brain regions (Arnold and Gorski, 1984; Balan et al.,1996; Pilgrim and Hutchinson, 1994).  For example, the increased size of the anteroventral periventricular nucleus of the hypothalamus in male rats appears to depend on the action of testicular hormones during the neonatal period, although the actual structural difference between the sexes is not obvious until puberty (DAVIS et al.,1996).

Intriguingly, for many areas of the brain, the action of testosterone depends on its conversion to estradiol by aromatase. The emergence of sexually dimorphic regions may depend on the creation of gender-specific networks of estrogen forming neurons.  For example, investigators have measured aromatase activity in two strains of mice selectively bred for behavioral aggression.  The animals bred to have short attack latency showed a different developmental pattern of aromatase activity in both the amygdala and the hypothalamus  (Hutchinson et al.,1995).

Traditionally, the investigators have focused on the role of steroid receptors acting via the nucleus and the binding of the steroid-receptor complex to specific DNA regions to alter the transcription of specific genes  (Evans, 1988). More recent studies have indicated that estrogen may also act through effects on the signaling pathway of nerve growth factors to induce changes in dendritic arborization and synapse formation.  For example, ovariectomized females rats lose dendritic spines in specific hippocampal regions.  When treated with estrogen, these animals show a 30% increase in NMDA receptors in the same hippocampal regions (Gazzaley et al.,1996; Wooley et al.,1997).

16. Summary: Alteration of neuronal development and vulnerability to psychopathology.   The relevance of developmental neurobiology to psychopathology and psychopharmacology are no longer remote. The past decade has seen unprecedented advances in our understanding of the mechanisms involved in the morphogenesis and activity-mediated sculpting of brain circuitry. The reciprocal interplay of conserved genetic programs and the ever changing macro- and micro-environment are a recurrent theme. These events set the stage for individual differences and range of phenotypic diversity seen within our human species (Bartley et al.,1997). A deeper understanding of these mechanisms should lead the way to improved treatments and preventive interventions.

Many psychopathological states, such as schizophrenia, autism, or Tourette's syndrome, are fundamentally developmental disorders that likely involve allelic variants that confer vulnerability - particularly in the context of disorder-specific environmental risk factors  (Bailey et al.,1996; Ciccheti and Cohen, 1995; Leckman et al.,1997; Weinberger, 1996).   A developmental perspective is of value not only in considering childhood onset disorders but will likely prove to be broadly useful.  For example, the ability of estrogens to maintain a rich dendritic arborization in regions of the hippocampus may herald an effective means to maintain mental function (Simpkins et al.,1997) and prevent the toxic effects of substances such as ß-amyloid (Tang et al.,1996).

published 2000