Brain Evolution and Development: Allometry of the Brain and Arealization of the Cortex

Brain Evolution and Development: Allometry of the Brain and Arealization of the Cortex

Diarmuid J. Cahalane and Barbara L. Finlay

13.1 Introduction

The first images and descriptions of brains have an unusual leverage on how we pose further questions about the brain. The principal images seen in introductions to neuroscience are sagittal views of the human brain, outlining the large divisions from forebrain to hindbrain derived from the “primitive swellings” of the neural tube, and the lateral view showing some version of Brodmann’s original subdivisions of the cortex with its numbered areas and lobes (Figure 13.1, bottom). Such first impressions underlie the concepts of brain structure and function that we form as individual researchers and the conversations we have as a research community.

Schematic illustrations of the cortical areas (left), white matter connections of the cortex (right), and divisions of the developing brain as demonstrated by early gene expression (bottom). — **Figure 13.1** **Left**: Cortical areas as originally delineated by Brodmann (1913). **Right**: White matter connections of the cortex. Gigandet et al., 2008. **Bottom**: Divisions of the developing brain as demonstrated by early gene expression. Martinez‐Ferre & Martínez, 2012.

The first studies of brain evolution enumerated these divisions and described their allometry. Studies of brain function attempt to map adaptive behavior, the mechanisms of perception and action, onto the same regions and areas delimited in those early images. However, the existence of an easily accessible set of morphological subdivisions of the brain is no guarantee at all that those subdivisions have identically corresponding developmental mechanisms on which evolution may act independently (cf. Chapter 12, this volume). How different might our current understanding of the brain be if our first view of it had been its wiring diagram (Figure 13.1, right), or the gradients of gene expression in the developing brain and the initial segmentation those gradients imply (Figure 13.1, bottom)?

The subdivisions drawn up by the first descriptive neuroanatomists were exalted by physical anthropologists studying brain evolution. The anthropologists applied the prima facie reasonable idea of “proper mass” (Jerison, 1973; Stephan, Baron, & Frahm, 1988) which says that the brain of an individual species should devote more of its volume to those sensory and motor capacities most important to it. Insofar as brain segments embody these particular capacities, therefore, to select an animal on the basis of some particular capacity is to select on the volume of its associated brain segment—this is the basis of the “mosaic” view of brain evolution (Barton & Harvey, 2000; Striedter, 2005). This discussion is not to suggest that the divisions and strategies employed by the first brain researchers were poorly chosen or necessarily misguided, given the information which was at hand. The purpose of our brief, critical retrospective is to disabuse us of any unworthy preconceptions which those early visions of the brain have instilled in our current thinking.

In the present chapter we present an alternative to the traditional, area‐based view of brain evolution and development. We will first review in general what is known about the overall scaling of the brain and its parts, at size scales ranging from gross anatomical divisions down to identified cell classes, and the accompanying alterations of development. We will then focus on the evolution of the isocortex as an example, contrasting its emerging combinatorial sources of organization with the area‐based view. We will consider various sources of organization in the cortex—columns, layers, areas, and larger‐than‐area regions— and directionalities in information flow and neuromodulation. We will conclude with a general developmental model of the cortical sheet that can encompass within‐cortex, between‐individual and cross‐species differences in neural architecture.

13.2 Basic Vertebrate Brain Allometry

13.2.1 “Allometry of What?” or “What Should We Measure?”

The first studies of changing brain organization across vertebrates looked at the weights or volumes of brain subdivisions because that was the only comparison possible given the available technology (Jerison, 1973; Northcutt, 1981; Stephan, Frahm, & Baron, 1981). Computer‐assisted stereology improved measurements of the total number of neurons and their arrangement in brain structures (Coggeshall, 1992) and, even though this method remains time‐consuming, substantial data sets have been generated (e.g., Giannaris & Rosendene, 2012; Pakkenberg, 1993). More recently, neuron and glial number have been counted in relatively large volumes of tissue using flow cytometry techniques (Herculano‐Houzel & Lent, 2005). The repeated assertion of flow cytometry studies is that counting neurons (sometimes along with other cell types) provides a more useful unit of measurement of the brain than does comparing volumes (Chapter 2, see also Herculano‐Houzel, Collins, & Wong, 2007; Herculano‐Houzel, Mota, & Lent, 2006). However, different measures highlight different features and those features could be either the subject of selective pressure or the consequences of developmental mechanisms through which evolution must act.

Single neurons are not “the” definitive measure of information transmission or circuit complexity in the brain. Only in those cases where the transmission of information is solely by action potentials over axons could the number of neurons be a faithful measure of bandwidth or complexity. In structures featuring local, graded computation involving non‐spiking communication (such as found in the retina, the cortex and the cerebellum), axodendritic assemblies, rather than whole neurons, often serve as computational units. For example, in the retina, a single A17 amacrine cell provides reciprocal feedback inhibition to hundreds of bipolar cells in isolated parallel circuits (Grimes, Zhang, Graydon, Kachar, & Diamond, 2010). Quite apart from variations in neuronal density, brain structures exhibit very diverse morphologies in their architectures for information processing and assembly. Consider, for example, the three‐dimensional stellate arborizations of the neurons of the neocortex, the parallel coursing axons intercepting the close‐packed two‐dimensional dendrites of the cerebellum’s Purkinje cells, or the patch‐and‐matrix arrangement of the striatum, where patches with particular processing features are interposed in the general matrix of neurons of the caudate and putamen (Brown et al., 2002). All of said structures have different doses of transmission neurons, interneurons and modulating neurons, as well as differences in resting metabolic activity. In addition, the volume of axonal and dendritic connections required to scale up each structure for additional neurons will scale differently for each cytoarchitectural plan. Such diversity makes it plain that using neuron numbers as the single measure of difference between complex structures paints a very incomplete picture.

What if energy consumption, rather than a structure’s volume or neuron count, is the implicit variable of concern? Then it is the action potential at the synapse that is by far the greatest consumer of resources (Laughlin, Steveninck, & Anderson, 1998; Logothetis, 2003) and the most relevant piece of information would be the total number of synapses, or perhaps, the volume of neuropil minus the volume of cell bodies. Thus we see that cell number, volume, cellular constituents and energetic requirements all interact to influence both the cost and the functionality of any given structure (cf. Chapter 7 of this volume). Then, it is hardly surprising that no single measure suffices to characterize a region’s function or is the obvious subject of selection.

Brain volume, on the whole, does scale as a fairly regular exponent of neuron number (e.g., Zhang & Sejnowski, 2000). However, it will dissociate from neuron number due to a number of factors. The two principal sources of variation are, first, the structure‐by‐structure differences in neuron assemblies already described above, and second, different adaptations to the “save wire” problem that can become acute at large brain sizes (Cherniak, 1995; Cherniak, Mokhtarzada, Rodriguez‐Esteban, & Changizi, 2004). The latter relates to the potential for wild proliferation of the volume of connectivity compared to neuron number if the network architecture of a small brain cannot scale up gracefully. Murre and Sturdy (1995) have calculated that, if a human cortex retained the same per cent connectivity seen in a mouse cortex, that cortex could be approximated as a cube 21 meters on a side, comprised principally of axons with merely a superficial dusting of neurons. Employing “small world” connectivity, whereby a tiny fraction of the network’s links being nonlocal ensures a low number of intermediaries between any two neurons, solves this problem (Watts & Strogatz, 1998). However, the parameters generating small world connectivity in differently sized brain networks may well be the subject of selection, e.g. selection on the distribution of axon lengths.

Overall, brain volume, neuron number, and the ratio of neurons versus glia all scale dramatically across vertebrate brains. As we have seen, however, each single feature is a patently inadequate measure of the computational diversity and capacity we are attempting to assess, and it is not clear that any one of these measures ought to be the favored target of selection. The best and only solution is to keep in mind the advantages and drawbacks of each measure, and the computational, energetic, and developmental constraints that bear on changing the size or amount of each kind of neural architecture.

13.2.2 Brain Scaling, Macro Scale

Looking at the overall patterns of brain change, we can infer evolution’s trajectory from the variation seen in present‐day sharks and rays to current mammals including primates. This corresponds to an evolutionary period of about 450 million years and the pattern of brain change is astonishingly stable: no new brain divisions appear; the same structures always become disproportionately large when the brain gets large (forebrain and cerebellum); and while the brainstem, mesencephalon, forebrain (telencephalon) and cerebellum strongly covary in volume, the olfactory bulb and associated structures (olfactory cortices and hippocampus) vary more independently (Figure 13.2; Yopak et al., 2010).

Image described by caption. — **Figure 13.2** **Top:** Mammals demonstrate a “grade shift” in the relative volume of the forebrain (telencephalon, labelled “Tel”) and cerebellum (“Cer”) at all brain sizes, but, in both mammals and nonmammals, those two structures enlarge disproportionately in the largest brains, a fact whose magnitude is often hard to appreciate in log scale representations as in the top panel. Note the data series for the respective brain parts have been offset by the quantities in parentheses for easy of viewing.

Yopak 2010. Reproduced with permission of National Academy of Sciences.

**Bottom left:** Large and small shark brains provide an example of the forebrain’s disproportionate enlargement. Digital Fish Library.

**Bottom right:** An alternative representation of the relative scaling of brain parts. The vertical axis is in regular (non‐log) units to better demonstrate the isocortex’s volume scaling.

Finlay 1995. Reproduced with permission of American Association for the Advancement of Science.

These general features of brain scaling are now well known and we have given two kinds of accounts of the causes of brain scaling. For the first, given the radical diversity of the functions of homologous parts of the forebrain, for example comparing between chickadees and chimpanzees, a functional reason for the preferential enlargement of these regions must lie at a computational level common to these species and arrived upon early in the vertebrate lineage. We have suggested that the insights into what kinds of computational architectures are both robust and evolvable can be found in the computational and robotics literature (Charvet, Striedter, & Finlay, 2011; Finlay, Hinz, & Darlington, 2011). Regarding the second kind of account, a developmental mechanism capable of producing this repeated change is the conserved segmental organization of the vertebrate brain: the most lateral locations in the developing neural plate, namely the cerebellum and forebrain pallium, produce their neurons for the longest time and so have the potential for disproportionate enlargement (Finlay, Hersman, & Darlington, 1998).

13.2.3 Individual Variability

The developmental account we have given of how extending embryonic development can directly produce the disproportionate enlargment of the cortex and cerebellum should work at large and small scales, that is, over the phylogenetic scale and over the range of variability of individuals. We have investigated this prediction in three separate data sets: the reports of the volumes of multiple brain parts in feral and domestic minks and pigs (Finlay et al., 2011); in laboratory mice of identified genetic strains (Finlay et al., 2011); and in several publicly available databases of MRI scans in normal humans (Charvet, Darlington, & Finlay, 2013). In all cases, the patterns of phylogenetic variability were reproduced at the level of individual variation. That is despite the range of variation at the individual level being multiple orders of magnitude reduced from that at the phylogenetic level. In all cases, brain parts scale together with high regularity, with the exception of the olfactory bulb and limbic system (Reep, Finlay, & Darlington, 2007). Here again, the cortex and cerebellum become disproportionately large in the largest brains. In Figure 13.3 we see the scaling and the relative variability of 6 neural structures in N = 90 humans plotted against the size of the medulla in each.

13.2.4 Scaling of Cortical Areas

To complete our picture of general trends in phylogenetic variability, individual variability, and developmental mechanisms that link the two, we will consider two aspects of the regularities of scaling of cortex which impact within‐cortex structures. First, we will consider the size‐scaling of cortical areas. Second, we will examine how the number of cortical areas relates to the total volume of different cortices.

Analyzing cortical areas presents some problems which do not arise in dealing with gross brain divisions. The direct homologizing of a “cortical area” from one species to the next can be impossible. For example, in the cortical areas containing the multiple representations of sensory surfaces, or levels of the motor hierarchy, questions of whether there is a homologous V4 in both the macaque monkey and a human, or a homologue of “Broca’s area” in a rat cannot be answered because of the ambiguities in the connectional architecture in smaller and larger brains. There are two kinds of exceptions to this. The first is that primary sensory and motor areas are unambiguously identifiable from one species to the next (Kaas, 2011; Krubitzer & Seelke, 2012)—they lie in the same relative positions and have the same types of input and output. The second exception is if cortical areas are massed, into large divisions like “frontal” or “parietal,” with respect to the primary sensory and motor areas. Then, the large regions massed will be homologous by exclusion. That these two kinds of within‐cortex division have regular scaling has been known for some time. Jerison (1973) described the regular, hyperallometric scaling of the frontal cortex with respect to the rest of the brain. The visual cortex also scales somewhat hyperallometrically (Frahm, Stephan, & Baron, 1984). In Figure 13.4, taken from more current work, these two kinds of regularities in cortical scaling can be seen.

In the top panel of the figure, the relative increase in the size of V1 is shown versus primary auditory, somatosensory, and motor cortex, as measured in a sample of diurnal and nocturnal animals (Kaskan et al., 2005). The slope of the allometric equation of V1 (1.086) was significantly higher than the other three (0.697, 0.66 and 0.84 respectively). In the bottom panel of Figure 13.4, the predictability of the size of human frontal cortex compared to great apes is plotted (Semendeferi, Lu, Schenker, & Damasio, 2002).

A second feature of cortical scaling is the change in the number of cortical areas in cortices of different volumes. Using the complete maps of cortical areas in 20 mammalian species (where a cortical area is described as a full thalamocortical topographic map in any modality and having distinct patterns of outputs and inputs compared to its neighbors) the relationship in Figure 13.5 can be seen (Finlay, & Brodsky, 2006; Finlay, Cheung, & Darlington, 2005).

Cortical areas increase very rapidly in number up until a cortex size of about 200 mm² in total area; thereafter the rate of increase is slowed (see Figure 13.5). We have suggested that the faster increase in the number of areas in small brains is produced by the scale‐dependent segregating mechanisms of Hebbian sorting, very similar to the mechanisms which in large brains produce features like ocular dominance columns. However, quite why there should be a discontinuity in scaling at about the cortical area of a galago as yet demands a precise mechanistic explanation.

13.2.5 Summary of Allometry, Focusing on the Cortex

We have described how, in the vertebrate lineage, the pallium (cortex or its forebrain homologue) is preferentially enlarged whenever brains become large. We have also shown that the same pattern of preferential enlargement is recapitulated in the individual variability of several mammalian species, including humans. The consistent pattern of brain change at these levels suggests a similar developmental mechanism should account for both. Within the cortex, areas that can be identified consistently across species enlarge at predictable allometric rates: the number of overall cortical areas increases with overall cortical volume very rapidly in small to medium‐sized brains and then more slowly in the largest. Considering the brain as a whole, and extrapolating from our work with the cortex, the manner and rate in which each brain part scales with overall brain size will depend on the relative size of its progenitor pool and the relative duration over which it proliferates during neurogenesis. The volume produced by neuron cell bodies is further adjusted by the type of interconnectivity each neuronal type maintains.