Evolution of the neocortex: Perspective from developmental biology

Experimental tests of the radial unit hypothesis

The radial unit model has been put to the test by the manipulation of cell production in the VZ by either diminishing the rate of programmed cell death (apoptosis), or by increasing the rate of neural stem cell proliferation. Mice deficient in Caspase 3 and 9 proteases, which are essential for the normal process of programmed cell death (apoptosis), an essential process during development^35,36, have a larger number of neural stem (founder) cells and radial glial cells (RG). These cells, in turn, produce more than the normal number of radial columns and, hence, have a larger cortical surface, which then begins to form incipient convolutions (FIG. 4AB and REF. 37).

Similarly, an even more dramatic increase in cortical surface occurs with an experimentally induced increase in neuronal production. For example, the β-catenin-mediated signaling pathway acts on the decision of the neuronal precursor cells to exit the cell cycle³⁸ and hence, influences the number of postmitotic neurons. Thus, transgenic mice expressing a stabilized form of β-catenin transgene in neural precursors develop enlarged brains with an increased number of precursor (founder) cells in the VZ that lead to a larger number of radial columns and the formation of elaborate cerebral convolutions in the mouse, that normally has a smooth (lissencephalic) hemispheric surface (FIG. 4B, C and REFS. ^{38, 39}). Despite severe additional brain abnormalities in these mice, many of the overproduced neurons migrate radially, and settle into apparently normal laminar and radial positions. Thus, it is both impressive and instructive that the expansion of the precursor pool can generate a much larger surface of the neocortex of relative uniform thickness, rather than just lumps of the ectopic neurons. In contrast, over-expression of cyclin-dependent kinase (cdk) inhibitor p27, which acts on neuronal progenitors at later developmental stages, decreases the production of neurons destined for the more superficial layers within the radial columns without increasing its surface⁴⁰.

Mice that are deficient for caspase 9 or carry mutations in β-catenins or cdk are not viable, and thus, it is unlikely that mutation of these particular genes were involved in neocortical evolution. However, these experiments can serve as an example of the Evo-Devo approach to theorize how a single gene mutation, that may have occurred in our common, long extinct ancestors, could increase the number of proliferative founder cells in the VZ, which triggers a cascade of events that lead to an increase in the number of radial units and expanded cortical surface without an increase in its thickness. It also shows that these types of genetic and cellular mechanisms, by which the neocortex may have expanded during evolution, are research goals that can be achieved using methods of modern developmental neurobiology. It is indeed remarkable that by a single gene mutation, a lissencephalic (smooth) neocortex typical of the rodent brain can be transformed into a gyrencephalic (convoluted) cortex that is characteristic of the primate cerebral hemispheres.

Development of convolutions

While the cerebral cortex in rodents is smooth, as the cortical surface expands during evolution in many species with a larger cerebrum, it began to form convolutions, which are particularly prominent in primates (FIG. 1). It has been recognized as common sense that the increase in cortical surface has to have lead to its folding. However, formation of convolutions involves rather complicated logistical problems: one of them being that the pattern of convolutions is not random, but highly reproducible within each species with relatively small individual variations. To this account, several hypotheses for a role of the long cortico-cortical axonal tracts in this process have been proposed^41–43. Indeed, the formation of the convolutions in the human cerebrum is associated with the formation of the voluminous intermediate zone that eventually transforms into subcortical white matter. This zone is, in primates, crisscrossed by the large number of distinct axonal fascicles forming cortico-cortical connections that contribute to the shape of the convolution. It has been hypothesized that the tension created by these fascicles is responsible for the stereotyped shape and orientation of the convolutions in gyrencephalic brains⁴³.

Evolutionary adaptation of radial glial cells

The transient population of radial glial (RG) cells in the developing cerebral cortex serves both as neural stem cells as well as guides for migration of their offspring and adjacent neuronal progenitors^44–46. However, formation of convolutions requires longer, curvilinear migratory pathways and more durable RG shafts, as they continue to serve as guides for neuronal migration from the VZ to the distant cerebral cortex at later fetal stages. For example, in the primate forebrain, many RG cells transiently stop dividing while their fibers span the entire cerebral wall, retaining their endfeet at the ventricular and pial surfaces⁴⁷. A single RG fiber in primates may serve as a guide for cohorts of up to 30 simultaneously migrating neurons, including some of the interneurons that at late stages of corticogenesis, originate from the expanded subventricular (SVZ) zone^48,49 and from the larger pool of progenitors with short radial processes^46,50,51. It was suggested that greater stability and longevity of the RG scaffolding, which consists in part of the elongated fibers of non-dividing RG cells, may be an essential evolutionary adaptation to enable proper allocation of neurons to the expanded and convoluted cerebral cortex^32,49. Indeed, the RG in both human and non-human primates differentiate precociously and, for example, express GFAP and Vimentin at the onset of corticogenesis, which is much earlier than in rodents, where these proteins are expressed only after birth^{48, 50–55}. Thus, even basic cell types in the embryonic forebrain, such as the RG, have diversified during evolution to accommodate the needs for cortical expansion and the development of cerebral convolutions.

Duplication through morphogenetic factors

The evolution of the cerebral cortex involves both an increase in the number of new cytoarchitectonic areas, as well as duplication of sensory representations⁶³. It is generally accepted that the regionalization of the neocortex is initiated by patterning centers that secrete signaling molecules such as the fibroblast growth factors (FGFs), wingless (Wnts) and bone morphogenetic protein (BMP), which regulate the position and size of cortical areas (Reviewed in REFS. ^28,64,65). For example, the mouse forebrain has an anterior patterning center at the commissural plate that secretes Fgf8 and Fgf17. Fukuchi-Shimogori and Grove⁶⁴ have shown that perturbations of the site and expression level Fgf8 can profoundly change the somatosensory map of the mouse cerebral cortex: overexpression of Fgf8 level in the embryonic forebrain induced by electroporation results in formation of an additional barrel field, with a mirror image representation of mouse whiskers⁶⁴. The mechanism of this remarkable duplication of the well defined cortical area is unknown, but it is instructive that Fgf8 expression is coordinated by distinct sets of transcription factors, namely the zinc-finger and sonic hedgehog signaling systems that are involved in neuronal production and specification^65–69. Similar duplication of the sensory areas with a reverse representation of the periphery has occurred during evolution; and, these examples show how it could occur rapidly by a small genetic change⁷⁰.

Decisions in the proliferative centers

Numerous recent studies demonstrated that the subdivisions and area enlargements of the cortical areas in embryonic mice could be traced back to the proliferative VZ and SVZ zones (Reviewed in REFS. ^28,58,65). For example, the expression of Fgf8 in the anterior parts of the telencephalon suppresses the posteriorly expressed chick ovalbumin upstream transcription factor I (COUP-TFI) and homeobox (EMX2) in the telencephalic neuroepithelial cells^{65, 71–74}, decreasing their anterior but increasing their posterior expression. The development of frontal and sensory subdivisions is dependent on the graded action of Fgf8⁷⁴ and Fgf17^75,76 and provides an elegant example of how one area, such as the frontal cortex, can expand differentially and independently of the growth rate of other areas (FIG. 4C and REFS. ^76,77). Furthermore, the size of an area can be regulated by the change in Fgf17 expression at early embryonic stages independent of the synaptic input from the periphery via the thalamus (FIG. 4 and REF. 75). The fact that it is now possible to enlarge and/or duplicate distinct cytoarchitectonic areas in the neocortex by genetic manipulation opens an unprecedented opportunity to explore how these maps develop in each individual, as well as how they may have been introduced during neocortical evolution. As a next step, it is important to search for the additional genes and morphoregulatory molecules that may be involved in cortical specification, and to develop rodent, and possibly non-human primate models, of cortical dysgenesis that mimic specific genetic or acquired cortical disorders in human⁷⁸. Thus, the creation of transgenic mice, in which the number of neurons and radial units is increased, serves as an example of how a larger-than-normal number of founder cells can create a cortex with an increased surface area and number of neurons that eventually form new connections.

In the spirit of Evo-Devo reasoning, the above examples show that it is possible to experimentally duplicate a cortical area in embryos with a proper topographical representation by changing the expression of various morphoregulatory factors. Again, as in the case of cortical expansion, we do not know whether genes that produce these particular molecules have been involved in areal enlargement and/or duplication during evolution. However, these studies illustrate how a single mutation could have a sudden and profound effect during evolution on the pattern of cortical parcellation and gives us some insights into how it could have occurred at the genetic and cellular level⁷⁰.

More neurons more connections

Obviously, an increase in the number of neurons, radial columns and cytoarchitectonic areas is not sufficient to explain the multitude of functional advances made in the human cerebral cortex, which must also involve the elaboration of neuronal connections (e.g., REFS. ^2,62,79). There is a large amount of comparative anatomical data showing that, in spite of enormous variety and inter-individual variability, the basic pattern of neuronal connection is species-specific and highly reproducible. It is generally agreed that functional usefulness of the evolutionarily new cortical connection generated initially by random mutation must be validated through the process of natural selection and, if proven advantageous, may contribute to the survival and propagation of the species. However, since neurons are generated before they form axonal connections, an increase in neuronal number, diversification of their types and attainment of their positions is the first steps in cortical evolution. As reviewed below, the new classes of neurons form new pattern of connections. In this complex process, innate neuronal activity is essential for the validation and refinement of synaptic connections that leads to their selective stabilization and/or elimination via competitive interactions^80–86. This is a large research field that deserve separate review article and cannot be considered here in sufficient and deserving detail. However, within the subject of cortical evolution, it is important to mention that there should be made a clear distinction between the formation of stereotypic, species-specific connections, that are established by surface mediated attractions and innate activity-dependent refinements, from the experience-dependent synaptic plasticity that shapes connections in cooperation with the outside environment in each individual. These later, experience-dependent changes, are not inheritable, and thus, cannot contribute to evolutionary advances, which requires gene mutation in germinal cells.

The Subpial Granular Layer (SGL)

This prominent cellular layer situated below the pial surface of the fetal human neocortex was described by O. Ranke more than a century ago, but was examined in a more systematic way by Brun⁸⁷ and Gadisseux et al.⁸⁸. It is considered to be another transient embryonic layer, but so far, it has not been observed in rodents¹⁷. Although very prominent and voluminous in the developing neocortex of human and non-human primates²³, because of its absence or negligent size in mice, it has not been sufficiently explored. There are only a handful of studies on its developmental history and function of the SGL in human and non-human primates. It appears around the 11th gestational week in the human Marginal Zone (MZ), peaks in size during mid-gestation and disappears by the time of birth^17,33,89. Analysis of proliferative activity in the developing cat and primate cerebrum indicate that the SGL produces neurons that may descend to the underlying cortical plate 87,23. Based on the uneven size and high incorporation of tritiated thymidine in the SGL overlaying the developing cortex in monkey embryos, it has been hypothesized that it contributes to a wealth of interneurons in certain cortical areas, such as the primary visual cortex, which has an extraordinary large layer IV²³. Thus, better characterization of the subpopulations of interneurons that originate in the SGL may help in elucidating their function as well as their possible involvement in cortical disorders that do not occur or can not be mimicked in mice.

Migration from the Ganglionic Eminence to the Thalamus

Another human-specific migratory pathway formed during midgestation by neurons streaming from the Ganglionic Eminence (GE) of the ventral telencephalon to the dorsal thalamus of the diencephalon, called the corpus gangliothalamicum (CGT), was discovered four decades ago⁹⁰. This voluminous, easily identifiable stream of migrating bipolar neurons in human could not be found in any other species that have been examined^{33, 91}. More recently, cell labeling in organotypic slice cultures of human embryonic brain tissue enabled tracing the migration of these cells directly from the GE to the dorsal thalamic association nuclei⁹². These migratory cells, which rely on homotypic-neurophilic guidance and express Dlx1/2 homeodomain-containing proteins that guide their tangential migration, eventually form GABAergic neurons in the dorsal thalamus, in particular, the pulvinar and mediodorsal nuclei, that are anatomically related to the association neocortex involved in higher cognitive functions, including symbolic reasoning or language⁹². Thus, it appears that the human thalamus, in order to accommodate the increased axonal input from the expanding association neocortex, recruits an additional complement of neurons from the nearby, still mitotically active GE in the ventral telencephalon after the diencephalic proliferative VZ becomes exhausted⁹². This large, easily identifiable migratory stream in the human fetus can not be analyzed experimentally, as it is not present in any other species so far examined, including rodents, carnivores and New and Old World primates.

Two origins of interneurons - a working hypothesis

It was Ramon y Cajal⁹³ who suggested that extraordinary human mental abilities are closely related to the increase in the number and diversity of cortical interneurons that he discovered using the Golgi silver impregnation method. However, since that time, many subclasses of primate-specific interneurons have been characterized^94,95. In spite of their larger number and higher diversity in primates, our knowledge of the origin and development of the interneurons has been derived almost exclusively from experiments in rodents where we can use more advanced methods. Thus, tracing cell lineages by the use of retroviral vectors in mice showed that interneurons originate from different progenitors than the projection neurons⁹⁶. In addition, the concept that the great majority, if not all, of the interneurons originate from the GE in the ventral pallium and then migrate tangentially to the dorsal telencephalon, comes from ingenious experiments performed in mouse (FIG. 6A,A′ and REFS. ^14,97–100).

We know little about origin of GABAergic interneurons in other species. Experimental studies in primates, particularly human, are by their nature limited or impossible; and, the suggestion that the enlarged number of interneurons in human originate from the voluminous subventricular zone can be traced to classical histological observation³³. However, more recently, the few studies that have succeeded to explore the origin of interneurons in these species with more advanced methods, uniformly suggest that cortical GABAergic neurons seem to originate from both the GE as well as the VZ/SVZ in these species (Fig. 6B and REFS. ^101–106). This stands in contrast to the exclusively GE origin in rodents. Importantly, the Letinic et al., fate-mapping experiments in human mid-fetal slice cultures suggest that the percentage of cortical GABAergic neurons arising from the dorsal pallium VZ/SVZ is higher than those arising from ventral progenitors¹⁰¹. Furthermore, it has been shown that proliferative (BrdU+) cells in slices of the dorsal human proliferative VZ/SVZ during mid gestation express ventral transcription factors (Dlx1,2 and Nkx2.1)¹⁰⁵ and that bipolar morphology and orientation of Dlx+ cells also indicate local origin and radial migration⁸⁹. These studies collectively support the idea that in primates, there are two types of interneurons with respect to their origin: those that are born in the GE of the ventral telencephalon such as in rodents; and, in addition, those that are born in the VZ/SVZ of the dorsal telencephalon (FIG. 6B). Both GE- and VZ/SVZ-derived interneurons use GABA as a neurotransmitter, but their morphology as well as their molecular characteristics indicate that their function may be different and related to the human-specific higher cortical functions.

Recently, Fertuzinhos et al.¹⁰⁶ took advantage of the severe ventral midline defects present in the human congenital disorder named holoprosencephaly (HPE). In HPE, the development of the ventral forebrain structures, including the medial GE, is incomplete. Examination of the distribution of various classes of cortical GABAergic interneurons showed that the neocortex and hippocampus of HPE specimens with severe ventral forebrain hypoplasia lack specific subtypes of cortical GABAergic neurons that in mice are known to derive from the GE, while other subtypes seem to be little or not affected by ventral midline defects, suggesting that they could possibly originate in the preserved dorsal pallium VZ/SVZ. These findings, based on the detailed analysis of expression of various cell-specific markers in the fetal and infant brains of HPE specimens with mild and severe ventral forebrain defects, as well as in age-matched controls, revealed that NOS1/NPY/SST- and parvalbumin-positive cells are consistently absent in both fetal and infant stages while calretinin-positive GABAergic cell numbers seem to be little or not affected. Furthermore, the absence of NOS1/NPY/SST- and PVALB-positive cells correlated with the dramatic depletion of TITF1-positive medial GE progenitors, which in mice are the main source of these subtypes of interneurons. Although this type of study on human genetic malformations cannot be conclusive, the results suggest that NOS1/NPY/SST- and PVALB-positive cells originate in the GE and migrate tangentially into the cortex, while the calretinin-positive cells could have arisen from the dorsal neuroepithelium of the lateral ventricular wall and migrated to the cortex radially together with glutamatergic projection neurons. The study cannot eliminate the possibility that the calretinin-positive cells could have arisen from less affected GE progenitors. It is noteworthy that the neurons expressing NOS1/NPY/SST are most numerous in the derivatives of the MZ and subplate zone (SPZ) in monkeys¹⁰⁷. In contrast, the calretinin expressing interneurons populate cortical layers II-VI and include the double bouquet cells that are absent from the rodent cortex (108,109, 110; reviewed in REF. 95).

The finding of evolutionarily novel origins that produce different subtypes of interneurons in human and non-human primates does not negate the significance of the tangential migration of cortical GABAergic cells from the GE that were discovered in the mouse. In fact, studies in human^101,102,106 and cynemogoulus monkey¹⁰³ indicate that the GE produces a significant portion of interneurons that migrate to the neocortex via the tangential route. However, human and non-human primates possess a greater diversity of inhibitory interneurons that have to be recognized and better understood to uncover a possible role in human psychiatric disorders⁹⁵ as the VZ/SVZ generated interneurons may be involved in a variety of disorders of higher brain function¹¹¹.

Predecessor Neurons

The most recently discovered new type of neuron in human, the Predecessor cell, has so far not been observed in any other species³⁴. These large, bipolar cells are the earliest generated neurons in the human forebrain that emerge under the pial surface of the ventro-lateral cerebral wall at the end of the 1^st gestational month, before completion of the neural tube closure and before onset of neurogenesis in the VZ/SVZ of the dorsal telencephalon. They are TU20 and Tbr1 positive, have long horizontal processes and form an extensive network over the forebrain primordium³⁴. We have speculated that predecessors may be a transient population, involved in determining the number of functional radial units in the cerebral cortex, but experimental proof is missing. As a next step, it is essential to examine where they exist in non-human primates in order to determine their lineage, pattern of gene expression, function and eventual fate.

The selected examples of human-specific developmental features and cellular events described above are by no means exclusive. Such evolutionary novelties are likely the result of gene mutations and/or the introduction of new genes and/or the addition of the developmental mechanisms and cell-cell interactions that act during the progenitor’s exit from the mitotic cycle. They generate a different outcome depending on new evolutionary contexts. Recent studies indicate that even subtypes of the pyramidal cells destined for layer V, that are the main efferent system connecting the cortex with subcortical structures, are determined at the time of their last division in the VZ^112–114. These experimental studies in mice indicate that evolutionary novelties that have emerged during evolution, including introduction of new neuronal types, may start with a mutation of the neuronal stem cells in the proliferative zones.

Region specific differential gene expression

A recent study from Johnson et al.¹²⁵ has provided a major step forward, examining 13 brain regions from both the left and right sides of the mid-fetal human brain. Using the most advanced generation of microarrays, these researchers profiled over one million known and predicted exons in nine distinct neocortical areas, including four subdivisions of the prefrontal cortex, as well as the temporal and parietal association areas of the perisylvian cortex (Figure 7A), thus enabling researchers to look for the first time on a genome-wide scale at the genetic and molecular mechanisms that distinguish these areas from each other and from the primary sensory cortical areas during development. The data revealed a huge number of transcriptional differences, including genes both differentially expressed and alternatively spliced. For example, the authors found that while the greatest distinction within the neocortex is between lobes – most notably between the prefrontal and non-frontal areas – there are also expression patterns common to the perisylvian areas (VLPFC, prospective Broca’s area; PAS, prospective Geschwind’s area; and TAS, prospective Wernicke’s area), which, although distributed across three lobes, are functionally linked by their involvement in speech and language. The functions of these genes commonly enriched in perisylvian areas might therefore hold some clues to the evolution of structure and connectivity that allowed for the development of language in our species. In addition, Johnson et al. have noted that the prospective Broca’s area (and its right hemisphere homolog) is, on the basis of gene expression patterns, more similar to the neighboring motor cortex than to the PFC¹²⁵. Interestingly, these authors did not identify any statistically significant left-right transcriptional asymmetry at the population level, concluding that such asymmetries most likely occur earlier during fetal gestation as previously suggested¹²⁶; but, also noting that more sensitive technologies, as well as greater sample sizes, may yet uncover such asymmetry at the late mid-fetal stage.

The application of weighted gene co-expression network analysis to microarray-derived expression data has previously provided unique insights into the genetic mechanisms of human brain evolution and disease¹²⁷. Applying this analysis to their data from the developing human brain led to identification of networks of genes linked both to brain regions, including the prefrontal cortex and to biological processes such as alternative splicing¹²⁵. Genes with the highest degree of interaction within such networks, called “hubs”, may represent critical transcription or other regulatory factors, and are prime candidates for functional characterization.

Several recent studies indicate that tissue- and cell type-specific enhancers play critical roles in neuronal specification and development¹²⁸; and, other work has identified subsets of these enhancers that, through recent evolutionary modification, may have led to the emergence of novel gene expression patterns and neocortical areal phenotypes¹²⁹. By generating such spatially comprehensive expression data in the fetal neocortex, Johnson et al. have opened the door to the systematic investigation of the relationship between such recently evolved enhancers and transcriptional specificity. They identified specific examples of genes differentially expressed in the developing human neocortex associated with putative recently-evolved enhancers, providing a short list of candidate genes and regulatory elements for functional characterization. They further found that the number of such associations was statistically disproportionate relative to the genome-wide frequency of such putative enhancers, providing further evidence that regulatory evolution may have played an important role in the evolution of the human cortex.

The recent data in human¹²⁴ have uncovered an order of magnitude of greater transcriptional differences between neocortical areas than has been obtained in comparable studies in rodent^128,129. For example, the gene CNTNAP2, previously studied for its role in autism and specific language impairment^132–135 is specifically and highly enriched in the orbital prefrontal cortex, an area involved in regulation of social behavior in humans and which has no comparable analog in rodents¹²⁴. The mouse homolog Cntnap2 has not been found to be expressed in any areal pattern or gradient in the mouse brain at any stage of development (REF. ¹³⁶, and M.B. Johnson and N. Sestan, personal communication). Thus, this gene exemplifies the importance of studying the human cortex, wherein structures and gene expression patterns not found in rodents give rise to abilities, such as language, but also disorders, such as autism, for which there is also no currently accepted mouse model ¹³⁷. Importantly, the type of data promises to provide new insights into the complex transcriptional and molecular underpinnings of cortical evolution and development of human specific neuronal and synaptic organization.

Future analyses contrasting these fetal human brain gene networks with comparable data from mouse (as has been done for human and chimpanzee adult brain¹³⁸) should aid in identifying those genes whose importance in cortical development has increased in the course of human brain evolution. Although the importance of spatial and temporal specificity of gene expression during development and evolution has long been recognized¹³⁸), it is only recently that the molecular mechanisms of this specificity have begun to be investigated due to availability of the complete genome sequences of human, mouse and other mammalian species.