Journal of Neurocytology 31, 289298 (2002)

Scaling laws in the mammalian neocortex: Does form provide clues to function?
K I M B E R LY H . H A R R I S O N 1 , PAT R I C K R . H O F 2 a n d S A M U E L S . - H . WA N G 1,
Department of Molecular Biology, Princeton University, Princeton, NJ 08544; 2 Fishberg Research Center for Neurobiology, Kastor Neurobiology of Aging Laboratories, and Advanced Imaging Program, Mount Sinai School of Medicine, New York, NY 10029, USA sswang@princeton.edu Received October 22, 2002; accepted January 22, 2003
1

Abstract
Although descriptions of form have been a mainstay of comparative neuroanatomy, less well explored is the use of quantitative approaches, especially at the cellular level. In the neocortex, many gross and cellular anatomical measures show striking regularities over a wide range of brain sizes. Here we review our recent efforts to accurately characterize these scaling trends and explain them in functional terms. We focus on the expansion of white matter volume with increasing brain size and the formation of surface folds, in addition to principles of processing speed and energetics that may explain these phenomena. We also consider exceptional cases of neocortical morphology as a means of testing putative functional principles and developmental mechanisms. We illustrate this point by describing several morphological specializations at the cellular level that may constitute functional adaptations. Taken together, these approaches illustrate the benefits of a synthesis between comparative neuroanatomy and biophysics.

Introduction More than one hundred years of experimental investigation have revealed a massive, almost bewildering, amount of variation in the form (Nauta & Feirtag, 1986; Ramon y Cajal, 1990; Niewenhuys et al., 1998), cellular and ultrastructural composition (Peters & Palay, 1991; Cowan et al., 2001), and physiology (Shepherd, 1990; Johnston & Wu, 1995; Cowan et al., 2001) of neural tissue. This variation has attracted much attention from anatomists, developmental biologists, and evolutionary biologists. Gross components of the central nervous system have been investigated quantitatively (Stephan et al., 1981; Jerison, 1987; Finlay et al., 2001), and comparative descriptions comprise a rich literature (Johnson et al., 1994; Niewenhuys et al., 1998). However, the use of quantitative approaches at the cellular level is much less well explored. Comparisons of this type are noteworthy because many parameters of neocortical anatomy show striking regularities across many species. Could these variations be the manifestation of invariant design principles (Kaas, 2000)? Here we use cellular architecture as a basis for deducing possible principles of neocortical function. Our approach
To

is quantitative, providing a means of recasting anatomical description in terms of biophysical principles such as speed, energetics, and mechanical forces. General form of the neocortex The neocortex has features common to all mammals (Mountcastle, 1998). While the neocortex can occupy anywhere from 25 to 80 percent of the brain (Jerison, 1991; Clark et al., 2001) and range in mass from about 14 mg in shrews (Stephan et al., 1981) to over 9 kg in whales (Rice, 1967; Kamiya & Yamasaki, 1974), the following design elements are always observed. On the outside is a sheet of gray matter 0.4 to 4 mm thick (Hofman, 1988), consisting mainly of neuronal cell bodies (chiefly pyramidal neurons) and synaptic terminations arranged in visible cell and fiber layers. Interior to the gray matter is a core of white matter composed of myelinated and unmyelinated axons. Each neuron makes synaptic connections with thousands of other neurons, mostly elsewhere within the neocortex, thus creating a structure with strong internal

whom correspondence should be addressed.

C

03004864

2003 Kluwer Academic Publishers

290 connectivity. In fact, pathways into and out of the neocortex comprise only a small fraction of white matter volume (Braitenberg & Schuz, 1998). These features indicate that much neocortical processing takes place over white matter projections. This architecture is reminiscent of large, human-designed computational devices (connection machines) (Mead & Conway, 1980; Hillis, 1986). As such devices increase in size, they encounter the limit that the links between processing elements occupy more space than the processors themselves. Scaling-up of such interconnections can cause wiring to dominate the size of a working system (Cherniak, 1994; Feynman, 1996). Since neural tissue is energetically expensive (Aiello & Wheeler, 1995), limits on white matter volume may thus impose a strong constraint on neocortical design. An additional principle in designing computing systems is maximization of processing speed. In axons, this can be achieved by adding an insulating myelin sheath and by increasing the axons diameter (Jack et al., 1975), thus decreasing the time required for impulse transmission through white matter. However, these adaptations impose several costs: first, such axons cost more metabolic energy to build, maintain, and operate; and second, larger axons increase the total white matter volume, thereby increasing path lengths (Chklovskii & Stevens, 2000) and negating in part the benefits of increased speed. With these principles in mind, we hypothesize that processing via these connections is under selection constraints to consume the minimum necessary amount of space and energy while maintaining satisfactory processing speed (Braitenberg, 1998; Chklovskii & Stevens, 2000). If the strongly interacting factors of speed, volume, and energetic costs are constraints on brain developmental mechanisms, they may dictate an optimal neocortical architecture which results from trade-offs among these limitations. Quantitative parameters Insight into the nature of size and speed trade-offs may come from examination of detailed anatomical parameters. A number of such parameters may be easily defined at both gross and cellular/ultrastructural levels. Gross quantitative parameters include the volumes of the white matter and gray matter, the external (exposed) and total (folded) surface areas, and the thickness of the cortical sheet. Cellular and ultrastructural parameters include the density per unit volume of neuronal cell bodies, synapses, and glia; the size of neuronal components such as soma, axon and dendrite; the number of synapses per neuron; and arborization and form parameters of individual neurons. An idea of which of these parameters show unusual scaling can be obtained simply by examining large brains. In large brains, the most conspicuous variations

H A R R I S O N , H O F A N D WA N G occur in three quantities: the number and variety of functional areas, the folded surface area, and the ratio of white matter to gray matter volume. These quantities increase with brain size (Changizi, 2001). Most apparent is the increasingly folded appearance of the neocortical surface in larger animals (Fig. 1A). For example, in the human neocortex, total folded surface area exceeds exposed surface area by a factor of three. Accompanying this increased folding is an increase in the thickness of the gray matter sheet. Furthermore, the white matter increasingly dominates neocortex, comprising less than 10% of the neocortex in shrews and galagos but over 40% in man and whales (Frahm et al., 1982; Hofman, 1991; Allman & Hasenstaub, 1999). Quantitative reconciliation The causes of these scaling trends are in general not yet known; in many cases, even the basic phenomena are not well described. An important first step towards characterization is to identify how many actual degrees of freedom are likely to exist in the system. These are reduced by the fact that neocortical parameters have a number of mutual dependencies (Fig. 1B). For instance, synapse density per unit volume, number of synapses per neuron, and neuron density are three distinct quantities, but since they can be defined in terms of one another, only two true degrees of freedom exist. Such arguments can be used to express macroscopic measures in terms of constituent microscopic components (Zhang & Sejnowski, 2000; Changizi, 2001); thus, such interdependencies reduce the true number of separate phenomena to be explained. Analysis of these dependencies is often made more convenient by the empirical observation that many anatomical quantities have allometric scaling properties (Huxley, 1932; Thompson, 1942; Hofman, 1989). Allometric scaling trends approximately follow power laws of the form Y = kG , where Y is one measured parameter, such as volume, area, or cell density; k is an intercept parameter of the fit; G represents a macroscopic reference parameter, usually gray matter or whole brain volume; and is the power-law slope. When quantities are related to one another, this power-law scaling can be used to reduce degrees of freedom as follows. If A, B , and C all follow powerlaw relationships A = G a , B = G b , and C = G c , and A = B C , then a = b + c . In the case of synapse density per unit volume, A = the density of synapses in gray matter, B = the density of neurons in gray matter, and C = the number of synapses per neuron. In this example, existing evidence suggests that the synapse density is invariant (b = 0); evidence comes from comparisons of synapse density made by investigators using uniform methods. Comparing different areas within a single species, overall synapse density spans a range of no more than 1.2-fold (Rakic et al., 1986; Schuz &

Scaling laws in the neocortex

291

Fig. 1. Morphological relationships and scaling trends. A. Total surface area (open symbols) and external surface area (filled symbols) versus gray matter volume (from Hofman, 1985). Total surface area includes the folded surface area due to gyrification. B. Interdependencies among anatomical parameters of neocortex. Each colored region indicates a set of interrelated parameters. C. White matter volume versus gray matter volume (from Zhang & Sejnowski, 2000). D. Neuron density versus brain weight (from Tower, 1954 and refitted). E. Electron micrographs of corpus callosum indicating scaling trends in axon caliber, density, and myelination (Shultz & Wang, 2001).

Palm, 1989). Comparing species, synapse density spans a range of no more than 1.5-fold (Cragg, 1967; OKusky & Colonnier, 1982; Beaulieu & Colonnier, 1985; Schuz & Demianenko, 1995; especially see discussion in Schuz & Demianenko, 1995). These data, from a small number of taxa so far, indicate that synapse density per unit volume is indeed approximately constant. Because synapses in gray matter are principally corticocortical, this constancy implies that on average the number of synapses received by each neuron is inversely proportional to the density of neurons. Direct testing of this deduction has commenced with measurements of the number of synapses per neuron in three primates (Elston et al., 2001), and of neuron density in nine mammals, including one primate (Tower, 1954; Cragg, 1967). This principle would reduce the number of synapses per neuron and neuron density to a single scaling relationship linked by the principle that synapse density is constant. Constant synapse density could arise as a result of selection pressure to bring synapse volume to a working minimum in the face of functional, energetic, and structural constraints (OKusky & Colonnier, 1982; Rakic et al., 1986; Allen & Stevens, 1994).

Preliminary evidence suggests that another quantity may be approximately constant: the number of neurons per cortical column or module (Rakic, 1995). The functional organization of gray matter into radially oriented columns has been established for many sensory areas (Mountcastle, 1998). It has been argued that the rest of neocortex is also organized into functional modules, and a characteristic lateral scale has been measured in a variety of species (Krubitzer, 1995). Analysis of these measurements indicates that the diameter of a module increases with brain size with a power law factor of = 0.13 (Changizi, 2003). The thickness of the cortical sheet also increases with = 0.08 for all but the smallest mammals (Hofman, 1988). Together, these scaling trends imply that the volume of a module increases with brain size ( = 0.34). However, this volume increase is matched by an approximately equal decrease in neuron density ( = 0.30 to 0.32) (Tower, 1954; Cragg, 1967). This provides indirect empirical support for the proposal that all mammals have a comparable number of neurons per column or module (Rakic, 1995). It is important to note that the foregoing calculation is not a restatement of previous erroneous claims (Rockel

292 et al., 1980). In that previous work, the neuron density per unit surface area was said to be constant, with the implication that the number of neurons per column is constant. The last statement may in fact be valid, but the particular argument made in that work contains two major errors. First, to count neurons vertical strips of gray matter were used that were so narrow as to cause errors of overestimation of neuron density (see DeFelipe et al., 2002). Second, columns were assumed to be of constant diameter. As mentioned, the characteristic diameter of columns in fact appears to increase with brain size. However, by a fortuitous accident, the two errors were of similar magnitude, and thus in the end the number of neurons per column may still be relatively unvarying. Architectonic and interspecies differences will lead to some degree of variation, so needless to say our prediction must be tested directly by comparisons in cases of well-delineated columns. Finally, the neocortex has been suggested to have a globally conserved parameter, network diameter (Changizi, 2001). Network diameter is defined as the mean number of steps used to get from one point (i.e., node) in the network to any other. The network diameter has been shown to depend only weakly on the precise pattern of network connectivity, so long as a minimal number of long-distance connections exist (Watts & Strogatz, 1998). In such a small-world system, the network diameter is approximately (log N)/ (log k ), where the network contains N nodes that make an average of k connections each. The estimated scaling in the total number of neurons and the number of synapses per neuron leads to an estimated network diameter between 2 and 3 (Changizi, 2003; S. S.-H. Wang, unpubl. results). This small number suggests that all neocortical neurons are separated from one another by at most a few synaptic steps. This architecture might be well suited for generation of patterned activity and allow associations to be formed at a global level. Analysis of relationships among other measurements leads to the identification of a number of sets of interrelated neocortical parameters (Fig. 1B). They include (a) gray matter volume, gray matter thickness, and surface area; (b) gray matter volume, axon length, axon caliber, and neuron density; (c) white matter volume, axon length, axon caliber, and neuron density; and (d) gray matter volume, neuron density, synapse density in gray matter, and number of synapses per pyramidal neuron. Of these parameters, only the macroscopic ones (gray matter volume, gray matter thickness, neocortical surface area, and white matter volume) are well measured. In contrast, ultrastructural measurements are less common, involve a narrower range of species, and are often compromised by poor stereological technique. Reconciliation of macroscopic neocortical structure in terms of components is thus currently limited by the availability of accurate microscopic measurements.

H A R R I S O N , H O F A N D WA N G Principles underlying neocortical scaling Can these scaling relationships be explained in biological terms? Explanations can take two forms: functional significance, that is, why they exist and in what sense they optimize nervous system function; and mechanism, that is, how they have arisen through developmental processes. Functional significance can be investigated by examination of cases shaped by natural selection. We have addressed this both within and among species by examining white matter ultrastructure and region-specific cellular specializations of the gray matter. Questions of mechanism may be best addressed by perturbation of normal processes by surgical or genetic means.

Functional implications of scaling in cellular components Axons are the principal component of white matter and as such are the principal determinants of its volume. This suggests that the runaway in white matter volume relative to gray matter, W = (G 1.23 ) (Fig. 1C) (Zhang & Sejnowski, 2000), may be accounted for by disproportionate increases in the total number of neurons giving rise to axons, their average length, and/or the average axon cross-sectional area. Because the neuron density actually decreases rather steeply with brain size (Fig. 1D) (Tower, 1954), calculations suggest that the only way to account for white matter runaway is for axon cross-sectional area to increase with brain diameter (Changizi, 2001). In recent measurements on axons of the corpus callosum, we have confirmed this prediction. In preliminary results (Shultz & Wang, 2001; M.W. Wagers, S. S.-H.W. & M.J. Burish, submitted for publication) (Fig. 1E), callosal axons appear to be closely packed, and axonal density is consistent with a decreasing power law with increasing brain size. Average axon cross-sectional area steadily increases with increasing brain size. Our measurements thus complete a geometric set of relationships that together account for the runaway in white matter volume with increasing brain size. Why would axons scale up with increasing brain size? One possibility is that scaling of axon size may optimize conduction of action potentials over long distances. Conduction time is a limiting factor in cortical processing, and propagation of action potentials through the white matter may be a major component of overall neocortical processing time (Swadlow, 2000). Cross-brain conduction times, which can exceed 100 ms (Swadlow & Waxman, 1976), may dominate processing time since other neural events take much less time, such as action potentials (1 ms) (Jack et al., 1975; Johnston & Wu, 1995) and synaptic transmission delays (0.3 ms) (Sabatini & Regehr, 1999). However, cross-brain delays are variable because brains vary in diameter and

Scaling laws in the neocortex because conduction properties of axons depend on their diameter and type. Axon conduction velocity is faster in wider axons due to a relative reduction in axial vs. leak resistance (Jack et al., 1975). Another major speedenhancing innovation in vertebrate nervous systems is the wrapping of axons in a fatty myelin sheath (Ritchie, 1995). Myelination increases the conduction velocity of action potentials by decreasing membrane capacitance and increasing leak resistance (Jack et al., 1975). Wider brains may require faster impulse conduction to maintain low cross-brain conduction times. This suggests that myelination would be more prevalent in largebrained mammals. We have confirmed this experimentally (Fig. 1E), and our results indicate that myelination occurs principally for axons wider than 0.60.8 m (see also Waxman & Swadlow, 1976). Another notable scaling phenomenon in white matter axons is a subpopulation of large, myelinated fibers that scale dramatically with brain size (Fig. 1E). For example, the widest axons are approximately 2 m in diameter in mice (Schuz & Preissl, 1996) and 8 m in macaque (LaMantia & Rakic, 1990), which is proportional to the increase in brain diameter in these species from 2 cm to 8 cm. Since in this type of fiber conduction velocity is proportional to axon diameter (Ritchie, 1995), the largest axons may therefore hold cross-brain conduction time relatively constant across species. However, the benefits of building larger axons come with several metabolic costs. Wider axons require more metabolic energy to construct and maintain, as does the addition of myelin. These adaptations also increase mean conduction distance by increasing the total white matter volume. Moreover, axonal capacitance must be discharged and recharged during an action potential. Thus wider axons tend to consume more energy (though myelination drastically reduces the energetic cost of an action potential). These trade-offs may place white matter under selection pressure to balance the benefits of fast conduction velocity against the costs of building and operating wider axons. Neocortical variations Thus far we have discussed general scaling trends. Another approach is to look for naturally occurring exceptions to the trends. For instance, the neocortical surface area to brain volume observes a power law relationship that accurately predicts surface area in most species within 15 percent (Hofman, 1985). Deviations from this relationship can be categorized relative to the overall trend as either less folded than expected (lissencephalic) or more folded than expected (gyrencephalic). Animals such as the manatee, beaver, and platypus (Fig. 2A) are lissencephalic, while the spiny anteater is gyrencephalic. A variation on this theme is cetaceans, which have folds that are more numerous than expected but also are decreased in diameter and

293 depth (polymicrogyria). In some cases the thickness of the gray matter sheet is also aberrant: in manatees the gray matter is unusually thick (approx. 4 mm; Reep & OShea, 1990), while in cetaceans the cortical sheet is unusually thin (Fig. 2A). If white matter is composed of closely packed axons, surface area is expected to be proportional to both the number and the mean cross-sectional area of white matter-projecting axons (Changizi, 2001). This model makes quantitative predictions about how these cellular parameters are related to macroscopic folding. One possibility is that lissencephaly may be caused by a shortage in the number of neurons or white matterprojecting axons. Alternatively, axons in lissencephalic animals may be unusually narrow relative to the general scaling trend. In some cases, such as the beaver, natural lissencephaly is accompanied by the presence of an exceptionally thick gray matter layer compared with similarly sized mammalian brains. It could be that the optimization principles operating in other mammals do not apply to lissencephalic animals, or that some biophysical or other functional adaptation leads to both the appearance of surface smoothness and thicker gray matter. In either case the coincidence of several unusual macroscopic parameters warrants further scrutiny on a microscopic level. Unusual cellular specializations Further insight may be gained from examining deviations from normal scaling in the case of individual cell types. Such comparisons are of interest, although baseline anatomical and physiological measurements are not yet abundant, with the possible exceptions of primates (anatomy) and rodents (physiology). Highquality comparative measurements allow the application of the biophysical scaling principles described previously. For instance, assuming that the dependence of conduction velocity (and length constant) on fiber diameter is similar across species, the existence of exceptionally large neurons with wide axons may reflect specialized functional physiological properties in those neurons. A number of cellular specializations can be recognized in the mammalian cerebral cortex using simple cytoarchitectural or chemoarchitectural analysis to define specific neuronal subclasses. Such classification criteria may then be used to infer functional correlates as well as phylogenetic traits (Hof et al., 1999, 2000a). The possibility of such an association is suggested by the fact that anatomically classified areas often correspond to functionally defined areas (Brodmann, 1905; Mountcastle, 1998). For instance, using calcium-binding proteins as markers, comparable patterns of expression can be found in members of the same clade. Whales and closely related artiodactyls

294

H A R R I S O N , H O F A N D WA N G

Fig. 2. Natural and induced exceptions to scaling trends. A. Left, cow brain, middle, manatee brain, demonstrating its lissencephalic nature relative to the cow. Right, dolphin brain (one hemisphere only) with its polymicrogyria and unusually thin sheet of gray matter. Black scale bars = 1 cm, white scale bars = 5 cm. Images courtesy of the Brain Museum, http://www.brainmuseum.org. B. Large Betz cells (arrows) in layer V of Pongo pygmaeus primary motor cortex. Scale bar = 250 m. C. Spindle cells in human cingulate cortex (arrows). Scale bar = 75 m. D. Left, wild-type mouse brain, right, gyrencephalic -catenin overexpressing mouse brain (Chenn & Walsh, 2002).

share patterns of neocortical expression of calbindin and calretinin, while other groups such as archontans have distinct cytoarchitecture and distribution of parvalbumin-expressing interneurons (Hof et al., 1999, 2000a). A functional rationale for cellular scaling relationships should ultimately be able to explain differences not only among species, but also among different cell types. During brain evolution the neocortex has undergone expansion, addition of areas, and the elaboration of novel neuronal subtypes leading to great diversity in sensorimotor and cognitive specializations (Krubitzer, 1995; Nimchinsky et al., 1999; Allman, 1999). In particular, primates possess three unusually large neuron types of the neocortex: Betz, Meynert, and spindle cells. These cell types are exceptionally large in big-brained primates, and may give rise to axons that are disproportionately large and have faster conduction velocity. This would be consistent with the possibility that

certain neocortical functions require high transmission speed. Betz cells of primary motor cortex (Brodmann area 4) and Meynert cells of primary visual cortex (Brodmann area 17) have attracted particular attention due to their large cellular volume, unique dendritic arborization patterns, distinctive connections, and thick myelinated axons (Fig. 2B) (Chan-Palay et al., 1974; Scheibel & Scheibel, 1978; Hof & Morrison, 1995; Hof et al., 2000b; Rivara et al., 2003). Although large neurons have been described in other large-brained mammals, the exceptionally large size and unusual architecture of these cells in primates (Le Gros Clark, 1942; Kaas, 2000) suggests the possibility that these neuronal subtypes constitute cellular substrates for specialized sensorimotor capacities such as nimble digital movement and vision (Le Gros Clark, 1959; Martin, 1990). Betz cells of primary motor cortex and Meynert cells of primary visual cortex are of particular interest for their potential role

Scaling laws in the neocortex in specialized adaptations of primates. Betz cells are involved in setting muscle tone prior to fine motor output and Meynert cells participate in the processing of visual motion (Heffner & Masterton, 1983; Fries et al., 1985; Rivara et al., 2003; Sherwood et al., 2003). Stereologic analyses show that Betz cells become disproportionally enlarged relative to other pyramidal neurons in larger brains, whereas Meynert cells are also larger than other pyramidal cells but in fixed proportion. Phylogenetic variance in the volumetric scaling of these neuronal subtypes may be related to species-specific adaptation. For example, Betz and Meynert cells are markedly enlarged in terrestrial primates such as the patas monkey (Sherwood et al., in press). Another remarkable cellular specialization is the spindle cell, a particular type of projection neuron that is characterized by a vertical, fusiform morphology and very large size (Nimchinsky et al., 1995; Nimchinsky et al., 1999) (Fig. 2C). These particular neurons may represent a well-defined projection, similar to the Meynert cells or the Betz cells. We recently observed that these spindle cells are prevalent in a restricted sector of the anterior cingulate cortex and in the agranular insular cortex. Among the mammals only certain primates, pongids and hominids, have these cells (Nimchinsky et al., 1999). In the human anterior cingulate cortex, spindle cells occur most often in small clusters, located exclusively in layer Vb, and are conspicuous due to the relatively low cellular density in this layer. Spindle cells account for a low percentage of the neuronal population and are larger in chimpanzees and humans than in gorillas and orangutans. Unlike the volume of other neurons, their volume is strongly correlated with encephalization (Nimchinsky et al., 1999). These observations reveal possible adaptive changes and functional modifications in the anterior cingulate cortex, a region that plays a major role in the regulation of many aspects of autonomic function and cognitive processes. The exclusive presence of these unusual cells in humans and their closest relatives suggests that the anterior cingulate cortex may have been a target of strong adaptive pressure during the last 1520 million years of primate evolution. Developmental perturbations In addition to these natural exceptions, other useful tools for investigating neocortical form are induced malformations. Such exceptions are informative because they usually generate suboptimal neocortical layouts by perturbing developmental mechanisms, thereby illuminating the link between development and evolutionary optimization. Large-scale manipulations of developing neocortex were first accomplished by fetal surgical ablations (Goldman-Rakic & Rakic, 1984). These pioneering experiments, done in macaque monkeys, provided key

295 findings on the formation of gyral patterns. First, gyrus formation persisted after removal of parts of the developing cortex, even though the remaining cortex did not entirely fill the cranium. This rules out the possibility that pressure from the cranium is needed as a force in fold formation. Since cell-cell adhesion forces confer a degree of rigidity to tissue, additional forces intrinsic to developing tissue are necessary to account for the buckling of the cortical sheet. Second, local removal of tissue led to abnormalities in gyri and sulci not only at the site of surgery, but also at distal locations with strong corticocortical connectivity to the ablated region. This result suggests that the formation of folds at stereotyped locations is strongly influenced by the presence of white matter axon projections. It has recently been proposed that major axon tracts, which grow towards their targets by epigenetically regulated mechanisms, guide the morphogenesis of neural tissue by generating force (Van Essen, 1997). Axons, which are filled with incompressible fluid and contain microtubules, exert force both laterally and axially (Bray, 1979). The tension exerted by long-distance axons might therefore be able to draw different parts of the cortical sheet together to form folds (Van Essen, 1997). This model suggests several predicted outcomes: (a) folds form to bring strongly interconnected regions closer together; (b) the white matter attains a configuration that is close-packed and minimizes surface area; (c) the total length of axon paths is globally minimized; and (d) disruption of white matter axons should alter folding patterns in regions to which the axons project. Experimental evidence is consistent with several of these predictions. Analysis of connectivity between cortical regions indicates strongly connected regions tend to face one another within gyri (Van Essen, 1997). A role for long-distance connections is also suggested by surgical manipulations that affect primary visual cortex, which in macaques is normally smooth (GoldmanRakic & Rakic, 1984). Surprisingly, removal of frontal lobes leads to the formation of convolutions in visual cortex. This result is consistent with the interpretation that when long-distance projections are removed, local connections generate much of the remaining force. A more extreme version of this manipulation is global chemical ablation of projection neurons of layer III, a major source of long-distance corticocortical axons; this manipulation leads to polymicrogyria (Volpe, 1981; reviewed in Goldman-Rakic & Rakic, 1984); in mice, ibotenate injections during development, which ablate deep cortical neurons, have analogous effects (Marret et al., 1995). Eye enucleation, which leads to reduction in the size of the lateral geniculate nucleus, induces folds in primary visual cortex (Rakic, 1988). In this case a key event may be reduction in the amount of input from the lateral geniculate nucleus, and a concomitant reduction in force or synaptic volume due to the axons of that pathway.

296 These ideas may be probed further by use of developmental defects that disrupt surface folding and white matter morphogenesis (Lammens, 2000; Schwartzkroin & Walsh, 2000; Feng & Walsh, 2001). These malformations include small folds (microgyria), excessively large folds (pachygyria), and lack of folds (lissencephaly). One well-studied developmental defect is type I classical lissencephaly, also referred to as Miller-Dieker syndrome. Miller-Dieker syndrome is characterized by mental retardation and severe epilepsy; and is caused by mutations of the gene LIS1, which in mice encodes a protein involved in dynein-dependent neuronal migration and axon growth (Smith et al., 2000). LIS1+/ mutant mice and Miller-Dieker patients reportedly have reduced neuron number (Lammens, 2000; Schwartzkroin & Walsh, 2000) which, in the context of the tension-generation model, points to a possible structural basis for lissencephaly: fewer axons. It will be valuable to know if mechanisms perturbed in LIS1 mutants are the same as those that lead to the natural smoothness of brains in animals such as the manatee. Conversely, manipulations have been discovered that lead to the ectopic formation of additional folds, that is, gyrencephaly. The first mouse model for gyrencephaly was a caspase-9 knockout (Kuida et al., 1998). The absence of caspase-9 disrupts apoptosis during development, resulting in an enlarged brain that protrudes through the skull, and is fatal. As the brain enlarges, folds are often observed, suggesting that failure to eliminate excess neurons can force greater expansion of the cortical sheet. A striking example of genetically induced folding comes from a recently produced transgenic mouse in which -catenin is overexpressed (Chenn & Walsh, 2002). This mutation results in increased production of neurons and marked cortical folding (Fig. 2D) and is perinatally fatal. These mutations suggest that the number of neurons can force the cortical sheet to occupy more area, perhaps because of the space demanded by axons as they cross the boundary between gray and white matter. Conclusions In recent years, variations in the gross form and cellular composition of the mammalian neocortex have been used by many investigators as landmarks or purely descriptive features. However, these anatomical features are also likely to reflect functional design principles and developmental mechanisms that shape architecture and form. These features have been optimized through millions of years of mammalian evolution. We have described efforts to recast these issues in terms of principles of speed and energetics. It is our hope that this approach can make some aspects of comparative neuroanatomy more theoretically driven.

H A R R I S O N , H O F A N D WA N G Acknowledgments We thank Mark Changizi and Damon Clark for discussions; Joseph Erwin, Chet Sherwood, Jennifer Shultz, Mark Burish, and Matt Wagers for experimental collaboration; and Anjen Chenn and the Brain Museum http://www.brainmuseum.org (with the support of the National Science Foundation Division of Integrative Biology and Neuroscience) for permission to use brain images.

References
AIELLO, L. C. & WHEELER, P. (1995) The expensive-tissue

hypothesis: The brain and digestive system in primate evolution. Current Anthropology 36, 199221. ALLEN, C. & STEVENS, C. F. (1994) An evaluation of causes for unreliability of synaptic transmission. Proceedings of the National Academy of Sciences USA 91, 1038010383. ALLMAN, J. & HASENSTAUB, A. (1999) Brains, maturation times, and parenting. Neurobiology of Aging 20, 447454. ALLMAN, J. M. (1999) Evolving Brains. New York: W.H. Freeman. BEAULIEU, C. & COLONNIER, M. (1985) A laminar analysis of the number of round-asymmetrical and flatsymmetrical synapses on spines, dendritic trunks, and cell bodies in area 17 of the cat. Journal of Comparative Neurololy 231, 180189. BEAULIEU, C. & COLONNIER, M. (1989) Number and size of neurons and synapses in the motor cortex of cats raised in different environmental complexities. Journal of Comparative Neurology 289, 178181. BRAITENBERG, V. (1998) Selection, the impersonal engineer. Artificial Life 4, 309310. Z, A. (1998) Cortex: Statistics BRAITENBERG, V. & SCH U and Geometry of Neuronal Connectivity. 2nd ed. Berlin: Springer-Verlag. BRAY, D. (1979) Mechanical tension produced by nerve-cells in tissue-culture. Journal of Cell Science 37, 391410. BRODMANN, K. (1905) Beitrage zur histologischen Lokalisation der Grosshirnrinde: Die Rindenfeldern der niederen Affen. Journal of Psychology and Neurology 4, 177226. CHANGIZI, M. A. (2001) Principles underlying mammalian neocortical scaling. Biological Cybernetics 84, 207215. CHANGIZI, M. A. (2003) The Brain from 25,000 Feet: High Level Explorations of Brain Complexity, Perception, Induction and Vagueness. Dordrecht: Kluwer Academic.
CHAN-PALAY, V., PALAY, S. L. & BILLINGSGAGLIARDI, S. M. (1974) Meynert cells in the primate

visual cortex. Journal of Neurocytology 3, 631658.

CHENN, A. & WALSH, C. A. (2002) Regulation of cerebral

cortical size by control of cell cycle exit in neural precursors. Science 297, 365369. CHERNIAK, C. (1994) Component placement optimization in the brain. Journal of Neuroscience 14, 24182427. CHKLOVSKII, D. B. & STEVENS, C. F. (2000) Wiring optimization in the brain. Advances in Neural Information Processing Systems 12, 103107.

Scaling laws in the neocortex

CLARK, D. A., MITRA, P. P. & WANG, S. S.-H. (2001)

297
and aged macaque monkeys. Journal of Comparative Neurology 420, 113126. HOFMAN, M. A. (1985) Size and shape of the cerebral cortex in mammals. I. The cortical surface. Brain, Behavior and Evolution 27, 2840. HOFMAN, M. A. (1988) Size and shape of the cerebral cortex in mammals. II. The cortical volume. Brain, Behavior and Evolution 32, 1726. HOFMAN, M. A. (1989) On the evolution and geometry of the brain in mammals. Progress in Neurobiology 32, 137158. HOFMAN, M. A. (1991) The fractal geometry of convoluted brains. Journal fur Hirnforschung 32, 103111. HUXLEY, J. (1932) Problems of Relative Growth. London: Methuen. JACK, J. J. B., NOBLE, D. & TSIEN, R. W. (1975) Electric Current Flow in Excitable Cells. Oxford: Clarendon Press. JERISON, H. (1987) Brain size. In Encyclopedia of Neuroscience (edited by ADELMAN, G. ) pp. 168170. Boston: Birkhaeuser. JERISON, H. J. (1991) Brain Size and the Evolution of Mind. New York: American Museum of Natural History.
JOHNSON, J. I., KIRSCH, J. A. W., REEP, R. L. & SWITZER III, R. C. (1994) Phylogeny through brain

Scalable architecture in mammalian brains. Nature 411, 189193. COWAN, W. M., S UDHOF, T. C. & STEVENS, C. F. (2001) Synapses. Baltimore: Johns Hopkins University Press. CRAGG, B. G. (1967) The density of synapses and neurones in the motor and visual areas of the cerebral cortex. Journal of Anatomy 101, 639654.
D E FELIPE, J., ALONSO-NANCLARES, L. & ARELLANO, J. I. (2002) Microstructure of the neo-

cortex: Comparative aspects. Journal of Neurocytology 31(35), 299316.

ELSTON, G. N., BENAVIDES-PICCIONE, R. & DEFELIPE , J. (2001) The pyramidal cell in cogni-

tion: A comparative study in human and monkey. Journal of Neuroscience 21, RC163. FENG, Y. & WALSH, C. A. (2001) Protein-protein interactions, cytoskeletal regulation and neuronal migration. Nature Reviews in Neuroscience 2, 408416. FEYNMAN, R. P. (1996) Feynman Lectures on Computation. Reading, MA: Perseus Books.
FINLAY, B. L., DARLINGTON, R. B. & NICASTRO, N.

(2001) Developmental structure in brain evolution. Behavior and Brain Sciences 24, 263278. FRAHM, H. D., STEPHAN, H. & STEPHAN, M. (1982) Comparison of brain structure volumes in insectivora and primates. I. Neocortex. Journal fur Hirnforschung 23, 375389. FRIES, W., KEIZER, K. & KUYPERS, H. G. (1985) Large layer VI cells in macaque striate cortex (Meynert cells) project to both superior colliculus and prestriate visual area V5. Experimental Brain Research 58, 613616. GOLDMAN-RAKIC, P. S. & RAKIC, P. (1984) Experimental modification of gyral patterns. In Cerebral Dominance: The Biological Foundation (edited by GESCHWIND, N. & GALABURDA, A. M. ) pp. 179192. Cambridge, MA: Harvard University Press. HEFFNER, R. S. & MASTERTON, R. B. (1983) The role of the corticospinal tract in the evolution of human digital dexterity. Brain, Behavior and Evolution 23, 165183. HILLIS, W. (1986) The Connection Machine. Cambridge, MA: MIT Press.
HOF, P. R., GLEZER, I., CONDE, F., FLAGG, R. A., RUBIN, M. B., NIMCHINSKY, E. A. & VOGT WEISENHORN, D. M. (1999) Cellular distribution of

traits: More characters for the analysis of mammalian evolution. Brain, Behavior and Evolution 43, 319347. JOHNSTON, D. & WU, S. M.-S. (1995) Foundations of Cellular Neurophysiology. Cambridge, MA: MIT Press. KAAS, J. H. (2000) Why is brain size so important: Design problems and solutions as neocortex gets bigger or smaller. Brain and Mind 1, 723. KAMIYA, Y. & YAMASAKI, F. (1974) Organ weights in Pontoporia Blainvillei and Platanista Gangetica. Sci. Rep. Whales Res. Inst. 26, 265270. KRUBITZER, L. (1995) The organization of neocortex in mammals: Are species differences really so different? Trends in Neuroscience 18, 408417.
KUIDA, K., HAYDAR, T. F., KUAN, C.-Y., GU, Y., TAYA, C., KARASUYAMA, H., SU, M. S., RAKIC, P. & FLAVELL, R. A. (1998) Reduced apoptosis and

the calcium-binding proteins parvalbumin, calbindin, and calretinin in the neocortex of mammals: Phylogenetic and developmental patterns. Journal of Chemical Neuroanatomy 16, 77116.
HOF, P. R., GLEZER, I., NIMCHINSKY, E. A. & ERWIN, J. M. (2000a) Neurochemical and cellular specializations

in the mammalian neocortex reflect phylogenetic relationships: Evidence from primates, cetaceans, and artiodactyls. Brain, Behavior and Evolution 55, 300310. HOF, P. R. & MORRISON, J. H. (1995) Neurofilament protein defines regional patterns of cortical organization in the macaque monkey visual system: A quantitative immunohistochemical analysis. Journal of Comparative Neurology 352, 161186.
HOF, P. R., NIMCHINSKY, E. A., YOUNG, W. G. & MORRISON, J. H. (2000b) Numbers of Meynert and

cytochrome c-mediated caspase activation in mice lacking caspase 9. Cell 94, 325337. LAMANTIA, A.-S. & RAKIC, P. (1990) Cytological and quantitative characteristics of four cerebral commissures in the rhesus monkey. Journal of Comparative Neurology 291, 520537. LAMMENS, M. (2000) Neuronal migration disorders in man. European Journal of Morphology 38, 327333. LE GROS CLARK, W. E. (1942) The cells of Meynert in the visual cortex of the monkey. Journal of Anatomy 74, 369 376. LE GROS CLARK, W. E. (1959) The Antecedents of Man. Edinburgh: Edinburgh University Press.
MARRET, S., MUKENDI, R., GADISSEUX, J. F., GRESSENS, P. & EVRARD, P. (1995) Effect of ibotenate

layer IVb cells in area V1: A stereologic analysis in young

on brain development: An excitotoxic mouse model of microgyria and posthypoxic-like lesions. Journal of Neuropathology and Experimental Neurology 54, 358370. MARTIN, R. D. (1990) Primate Origins and Evolution. Princeton: Princeton University Press. MEAD, C. & CONWAY, L. (1980) Introduction to VLSI Systems. Reading, MA: Addison-Wesley.

298
MOUNTCASTLE, V. (1998) Perceptual Neuroscience: The

H A R R I S O N , H O F A N D WA N G
dendritic spines in hedgehog and monkey. Journal fur Hirnforschung 36, 113122. SCH UZ, A. & PALM, G. (1989) Density of neurons and synapses in the cerebral cortex of the mouse. Journal of Comparative Neurology 286, 442455. SCH UZ, A. & PREISSL, H. (1996) Basic connectivity of the cerebral cortex and some considerations on the corpus callosum. Neuroscience and Biobehavioral Reviews 20, 567570. SCHWARTZKROIN, P. A. & WALSH, C. A. (2000) Cortical malformations and epilepsy. Mental Retardation and Developmental Disabilities Research Reviews 6, 268280. SHEPHERD, G. M. (1990) The Synaptic Organization of the Brain. 3rd ed. New York: Oxford University Press.
SHERWOOD, C. C., LEE, P. W. H., RIVARA, C.-B., HOLLOWAY, R. L., GILISSEN, E. P. E., SIMMONS, R. M. T., HAKEEM, A., ALLMAN, J. M., ERWIN, J. M. & HOF, P. R. (2003) Evolution of specialized pyra-

Cerebral Cortex. Cambridge, MA: Harvard University Press. NAUTA, W. J. H. & FEIRTAG, M. (1986) Fundamental Neuroanatomy. New York: Freeman.
NIEWENHUYS, R., TEN DONKELAAR, H. & NICHOLSON, C. (1998) The Central Nervous Sys-

tem of Vertebrates. Berlin: Springer.

NIMCHINSKY, E. A., GILISSEN, E., ALLMAN, J. M., PERL, D. P., ERWIN, J. M. & HOF, P. R. (1999) A

neuronal morphologic type unique to humans and great apes. Proceedings of the National Academy of Sciences USA 96, 52685273.
NIMCHINSKY, E. A., VOGT, B. A., MORRISON, J. H. & HOF, P. R. (1995) Spindle neurons of the human ante-

rior cingulate cortex. Journal of Comparative Neurology 355, 2737. O KUSKY, J. & COLONNIER, M. (1982) A laminar analysis of the number of neurons, glia, and synapses in the visual cortex (area 17) of adult macaque monkeys. Journal of Comparative Neurology 210, 278290. PETERS, A. & PALAY, S. L. (1991) The Fine Structure of the Nervous System: Neurons and Their Supporting Cells. 3rd edn. New York: Oxford University Press. RAKIC, P. (1988) Defects of neuronal migration and the pathogenesis of cortical malformations. Progress in Brain Research 73, 1537. RAKIC, P. (1995) A small step for the cell, a giant leap for mankind: A hypothesis of neocortical expansion during evolution. Trends in Neuroscience 18, 383388.
RAKIC, P., BOURGEOIS, J.-P., ECKENHOFF, M. F., ZECEVIC, N. & GOLDMAN-RAKIC, P. (1986) Con-

midal neurons in the primate motor and visual cortex. Brain, Behavior and Evolution 61, 2844. SHULTZ, J. R. & WANG, S. S.-H. (2001) How the neocortex got its folds: Ultrastructural parameters underlying macroscopic features. Society for Neuroscience Abstracts.
SMITH, D. S., NIETHAMMER, M., AYALA, R., ZHOU, Y., GAMBELLO, M. J., WYNSHAW-BORIS, A. & TSAI, L. H. (2000) Regulation of cytoplasmic dynein be-

current overproduction of synapses in diverse regions of the primate cerebral cortex. Science 232, 232235. Y CAJAL, S. (1990) New Ideas on the Structure of the RAM ON Nervous System in Man and Vertebrates. Cambridge, MA: MIT Press. REEP, R. L. & O SHEA, T. J. (1990) Regional brain morphometry and lissencephaly in the Sirenia. Brain, Behavior and Evolution 35, 185194. RICE, D. W. (1967) Cetaceans. In Recent Mammals of the WorldA Synopsis of Families (edited by ANDERSON, S. & JONES, J. K. ) pp. 291324. New York: Ronald Press. RITCHIE, J. M. (1995) Physiology of axons. In The Axon: Structure, Function and Pathophysiology (edited by WAXMAN, S. G., KOCSIS, J. D. & STYS, P. K. ) pp. 6896. New York: Oxford University Press.
RIVARA, C.-B., SHERWOOD, C. C., BOURAS, C. & HOF, P. R. (2003) Stereologic characterization and spatial dis-

tribution patterns of Betz cells in human primary motor cortex. Anatomical Record 270A, 137151. SABATINI, B. L. & REGEHR, W. G. (1999) Timing of synaptic transmission. Annual Review of Physiology 61, 521542. SCHEIBEL, M. E. & SCHEIBEL, A. B. (1978) The dendritic structure of the human Betz cell. In Architectonics of the Cerebral Cortex (edited by BRAZIER, M. A. B. & PETS, H. ) pp. 4357. New York: Raven Press. SCH UZ, A. & DEMIANENKO, G. P. (1995) Constancy and variability in cortical structure. A study on synapses and

haviour and microtubule organization by mammalian LIS1. Nature Cell Biology 2, 767775. STEPHAN, H., FRAHM, H. & BARON, G. (1981) New and revised data on volumes of brain structures in insectivores and primates. Folia Primatologica 35, 129. SWADLOW, H. A. (2000) Information flow along neocortical axons. In Time and the Brain. Conceptual Advances in Brain Research (edited by MILLER, R. ) pp. 131155. Amsterdam: Harwood Academic Publishers. SWADLOW, H. A. & WAXMAN, S. G. (1976) Variations in conduction velocity and excitability following single and multiple impulses of visual callosal axons in the rabbit. Experimental Neurology 53, 128150. THOMPSON, D. W. (1942) On Growth and Form. Cambridge: The University Press. TOWER, D. B. (1954) Structural and functional organization of mammalian cerebral cortex: The correlation of neurone density with brain size. Journal of Comparative Neurology 101, 952. VAN ESSEN, D. C. (1997) A tension-based theory of morphogenesis and compact wiring in the central nervous system. Nature 385, 313318. VOLPE, J. J. (1981) Neurology of the Newborn. Philadelphia: Saunders. WATTS, D. J. & STROGATZ, S. H. (1998) Collective dynamics of small-world networks. Nature 393, 440 442. WAXMAN, S. G. & SWADLOW, H. A. (1976) Ultrastructure of visual callosal axons in the rabbit. Experimental Neurology 53, 115127. ZHANG, K. & SEJNOWSKI, T. J. (2000) A universal scaling law between gray matter and white matter of cerebral cortex. Proceedings of the National Academy of Sciences USA 97, 56215626.