A major focus of comparative neuroanatomy has been on whether the mammalian brain evolves in a concerted or a mosaic fashion. Workers have examined variation in the volume of different brain regions across taxa to test the degree to which selection is constrained by the timing of events in neural development. Whether a conserved neurogenetic program in the mammalian brain constrains the distribution of different cell types, however, has not yet been investigated. Here we tested for evidence of evolutionary constraints on the densities of different cell types in the primary visual cortex (V1) and the hippocampus in 37 primate and 21 carnivore species. Cellular densities in V1 and the hippocampus scale isometrically with respect to one another in carnivores, as predicted by the concerted evolution hypothesis. In primates, however, cellular distributions in the hippocampus and primary visual cortex show no correlations, which supports the hypothesis of mosaic brain evolution. We therefore provide evidence for the presence of constraints controlling the adult densities of different cell types in disparate regions of the mammalian brain, but also for specializations along the primate lineage. We propose that adaptations to modularity at the cellular level may carry a deep phylogenetic signal.

The mammalian brain is composed of structurally distinct cell groups, which are configured into topographical maps underlying sensorimotor and cognitive functions (Kaas 1982; Passingham et al. 2002; Krubitzer 2009). While it is clear that some species display remarkable behavioral specializations, and that certain brain areas are devoted to mediating quite specific behaviors, the degree to which one region can evolve independently of functionally unrelated regions is poorly understood. It has been suggested that the size of different brain regions evolves in concert due to constraints of neural developmental timing (Finlay and Darlington 1995). In contrast, it has been proposed that developmental constraints are not sufficient to overpower the ability of regions to evolve independently (Barton and Harvey 2000; de Winter and Oxnard 2001).

Comparative studies of connectivity and circuitry in the mammalian brain confirm many of the predictions of the concerted evolution hypothesis. Structural components in the trans-cerebellar loops, for example, have been observed to covary in size across species (Voogd 2003). Similarly, reduction in the amount of retinal afferents has been shown to cause corresponding reductions in the lateral geniculate nucleus and visual cortex (e.g., Rakic et al. 1991; Cooper et al. 1993; Dehay et al. 1996). Epigenetic population matching, wherein competition for some trophic factor produced by a target determines the number of projection neurons that survive the period of programmed cell death (Katz and Lasek 1978; Linden 1994; Yeo and Gautier 2004), may, in part, explain these phenomena. However, patterns predicted by epigenetic population matching are not observed universally—different species tend to elaborate pathways from a common source differently (Northcutt and Wulliman 1988)—and, without developmental data, it is impossible to say that the population matching is epigenetically controlled (see Bunker and Nishi 2002). It may be that epigenetic cascades operate successfully in linear circuits, but not in reticulate circuits, which is why an examination of the available evidence suggests that the structure of region sizes in the mammalian brain is neither completely constrained by developmental timing nor completely free to evolve independently.

Volumetric size, however, is a poor estimate of the cellular composition of brain tissue (Azevedo et al. 2009). Increasing evidence for phyletic variation in the cellular organization of homologous regions of mammalian brains (e.g., Preuss and Coleman 2002; Hammock and Young 2005; Hutsler et al. 2005; Sherwood and Hof 2007) has demonstrated that interspecific variation in factors underlying brain size variation (e.g., cellular density, degree of dendritic arborization, and cell soma size) may also reflect evolutionary adaptations within lineages in conjunction with morphological or volumetric changes. Comparing cellular properties in disparate brain regions across taxa provides a new perspective to explore the extent to which developmental constraints act on the evolving mammalian brain.

Although the number of neurons in the cortex is approximately determined by the number of progenitor cells (Fish et al. 2008; Noctor et al. 2008), the duration of cell-division cycles (Lange et al. 2009), and the number of cell cycles during neurogenesis (Kornack and Rakic 1998), glia are generated only after neuronal migration terminates (Voigt 1989). The most abundant type of glial cells are astrocytes, which are distributed homogeneously in the cortical gray matter (Bushong et al. 2003; Nedergaard et al. 2003) and support neurons and the neuronal environment by producing trophic agents (Hatten et al. 1986; Müller et al. 1993; Araque et al. 1999; Barres and Smith 2001; Hidalgo et al. 2001; Allen and Barres 2005). Oligodendrocytes, glial cells that synthesize myelin, a lipid-rich membrane that ensheaths axons and increases the conduction velocity of electrical impulses, begin their differentiation after neurons have been surrounded by astrocytes and formed functional synapses (Baumann and Pham-Dinh 2001). The implication that astrocytes may regulate the generation of new neurons (Song et al. 2002; Horner and Palmer 2003; Nedergaard et al. 2003), influence the development and synaptogenesis of those neurons (Pfrieger and Barres 1997; Kang et al. 1998; Haydon 2001; Ullian et al. 2001), monitor neurometabolic interactions at the synaptic cleft (Laming et al. 2000; Hertz et al. 2001), and generally be required for dense synaptic networks to achieve advanced degrees of local modulation and control, as well as the need for oligodendrocytes to bypass axonal size constraints in increasingly large brains (Wen and Chklovskii 2005), suggests that there might be an evolutionary role for a relative increase in glial cells.

Our aim was to test whether the density of neurons and glial cells covaries across different regions of the brain in carnivores and primates. The regions examined here—the primary visual cortex and subfields of the hippocampal formation—are not directly interconnected with one another and therefore may be free to evolve independently. Furthermore, these regions are well documented and can be reliably delineated in carnivores and primates. From evidence that the mammalian brain is loosely modularized (see Krubitzer 2007), such that one region is rarely isolated for specialization at the expense of others, but that the design of modularization itself can be selected, it is likely that the degree to which certain brain regions must evolve in concert and can evolve independently will carry a phylogenetic signal. In the current study, we compared neuronal and glial cell densities in the primary visual cortex (V1) and subfields of the hippocampus proper (CA1–3) in 37 primate species and 21 carnivore species. Our results provide evidence for concerted evolution of neuronal and glial cell densities in disparate regions of the carnivore brain, but also for specialization in the proportions of these different cell types along the primate lineage.

Materials And Methods


Samples of the left hemisphere of nonpathological postmortem brains representing 37 primate species and 21 carnivore species were used (Table 1). Eleven other eutherian species were also sampled (Table S1). All samples were from adult brains, except for Trachypithecus francoisi and Pithecia pithecia, which were from juveniles with brain sizes comparable to species-typical adult averages. Specimens from all collections were immersion fixed with either 10% formalin or 4% paraformaldehyde. Some brains were embedded in paraffin prior to sectioning. Brain sections were Nissl stained in the context of this research or unrelated experiments. The original research reported herein was performed under guidelines established by the Animals Scientific Procedures Act (ASPA). For the specimens sampled in the context of other investigations, it was impossible to control for artifacts related to discrepancies in fixation length and postmortem delay. The recorded brain weights in our sample, nonetheless, do not show significant deviations from species-typical average fresh weights. Comparable error in density estimates from tissue shrinkage and histological processing artifacts along both axes should be expected to affect the elevation of regressions, but not scaling exponents or residuals (see Sherwood et al. 2006). Because of concerns about the possible effect of unknown degrees of shrinkage artifact on variables, analyses were confined to only those that involved regressing density variables collected from the same specimens on themselves, containing equivalent effects of histological artifact in both x and y axes of the data derived from that individual.

Table 1.  List of species by taxonomic classification1.
Taxonomic groupSubgroupSpeciesTaxonomic groupSubgroupSpecies
  1. 1The number of individuals sampled for each species is listed. Where no number is listed, only one individual was sampled.

PrimatesStrepsirrhiniGalago senegalensisCarnivoraCaniformiaMustela nigripes
  Nycticebus coucang  Neovison neovison
  Lemur catta  Mephitis mephitis
  Eulemur mongoz  Taxidea taxus
  Microcebus murinus  Procyon cancrivorus
  Cheirogaleus medius  Procyon lotor
 TarsiidaeTarsius bancanus  Nasua nasua
  Tarsius syrichta  Bassaricyon gabbii
 PlatyrrhiniCallithrix geoffroyi  Potos flavus
  Leontopithecus rosalia  Ailurus fulgens
  Saguinus oedipus  Zalophus californianus
  Cebus capucinus  Callorhinus ursinus
  Saimiri sciureus  Phoca vitulina
  Aotus trivirgatus  Ursus maritimus
 PitheciidaeCallicebus moloch  Canis lupus familiaris
  Pithecia pithecia  Canis latrans
 CercopithecinaeAlouatta caraya  Vulpes vulpes
  Alouatta palliata FeliformiaPanthera pardus
  Ateles ater  Felis catus
  Macaca fascicularis (2)  Puma concolor
  Macaca mulatta (2)  Crocuta crocuta
  Macaca maura (5)  Cynictis penicillata
  Cercocebus torquatus   
  Mandrillus sphinx   
  Papio anubis (2)   
  Cercopithecus mitis   
  Cercopithecus nictitans   
  Erythrocebus patas (2)   
  Colobus angolensis   
  Trachypithecus francoisi   
 HominoideaPongo pygmaeus (2)   
  Pan paniscus   
  Pan troglodytes (5)   
  Homo sapiens (6)   
  Gorilla gorilla (2)   
  Hylobates muelleri   
  Symphalangus syndactylus   


We estimated the density of neurons and both astrocytes and oligodendrocytes (herein referred to as glia) in the primary visual cortex (V1) and hippocampal subfields (CA1–3). Demarcation of the hippocampus is explained in Figures 1 and S1L, and the demarcation of V1 has been explained previously (Lewitus et al. 2012). Excitatory and inhibitory neurons were not differentiated. No quantitative data were obtained on cell morphology, as it would have been outside the scope of our hypothesis, however, ease of identification of different cell types was not observed to change systematically across taxa. Cell counting was performed under bright-field microscopy using StereoInvestigator software (MBF Bioscience, Williston, VT). The thickness of sections cut from the microtome ranged from 25 to 100 μm. Mounted section thickness was measured at the first and final counting site for each section using the microcator with a 63× objective and used to calculate volume estimates for cellular densities by dividing the total estimated cell population by the mounted section thickness. For each individual, a random starting section was selected. Serial sections spaced at 300–400 μm were selected for analysis for each cell type. Boundaries of layers II–VI in V1 were outlined using a 10× objective, and a virtual 30 × 30 μm lattice of counting frames was randomly positioned on each slide to cover the sampled area with approximately 30 frames per section. Counting was performed under Koehler illumination using a 100× (NA 1.25, oil) objective—a 63× objective (NA 1.4, air) was used with one human individual and one chimpanzee individual, as the slides were too thick to allow for the working distance of higher power objectives. Because mounted section thickness varied, the disector thicknesses used also varied. A minimum 4-μm guard zone, defined as the space between the boundary of the tissue section and the part of the section used for counting, was set on either side of each section. Pilot tests were performed for each individual to determine the optimal size of the counting frame (approximately two cells per counting frame). The resulting coefficient of error (CE) was below 0.08 ± 0.01 for all analyses (Gundersen and Jensen 1987; Gundersen et al. 1999; Slomianka and West 2005). Cellular density was calculated as the sum of neurons counted with the disectors divided by the product of the sum of the disectors examined and the volume of the disector (Howard and Reed 1998). Volumetric estimates of the granular and molecular layer of the hippocampus and the granular layer of the cerebellum are presented as supplemental information (Table S2).

Figure 1.

Glia and neuron densities were counted in the primary visual cortex (A, B) and hippocampal subfields (C, D) using design-based stereology. (A) The mammalian V1 was demarcated (arrows) on the basis of its topological location and distinct appearance in materials stained for Nissl substance (Allman and McGuinness 1988; Hof and Morrison 1995; DeFelipe et al. 1999; Rosa and Krubitzer 1999; Rosa et al. 2005). The region of V1 sampled was restricted to layers II–VI due to tissue preservation in layer I. (C) Pyramidal cell regions of the hippocampus proper (cornu ammonis, CA) were demarcated at one end (CA3) by an abrupt change in the organization of neuronal cell bodies in the hilus (Rosene and Van Hoesen 1977; Amaral and Insausti 1990; Keuker et al. 2003) and at the other end (CA1) by the point at which the superficial cells of the hippocampus proper ceased to be contiguous (West et al. 1991; Keuker et al. 2003). Neurons in V1 (B) and the hippocampus (D) were distinguished from non-neuronal cells by the presence of dark, coarsely stained Nissl substance in the cytoplasm, a large nucleus, a distinct nucleolus, ovoid shape, and lightly stained proximal segments of dendritic processes. Glia were expected to lack a conspicuous nucleolus and contain less endoplasmic reticulum than neurons. Photo: A, Callicebus moloch; B, Homo sapiens; C, Sorex araneus; D, Saguinus oedipus.

Figure 2.

Log-log regression plots of glia–neuron ratio in V1 and CA1–3 in carnivores (top) and primates (bottom). A significant scaling relationship, represented by the dotted line (y= 0.923x+ 0.061, R2= 0.814, P < 0.001), was found in carnivores only. The bar graphs show group means for glia–neuron ratio in V1 (right) and CA1–3 (top), which are significantly different in carnivores and primates (χ2= 7.02, P= 0.030).

Table 2.  Stereologic estimates of cellular densities (cells per mm3) in (A) nonprimate mammals and (B) primates.
 Primary visual cortexHippocampus
 Glia–neuronNeuronalGlial cellGlia–neuronNeuronalGlial cell
 Mustela nigripes0.78275,423213,7960.63213,796135,060
 Neovison neovison0.54229,087120,2260.50269,153134,559
 Mephitis mephitis0.22169,824 38,0190.17263,027 43,758
 Taxidea taxus1.07 77,625 83,1760.52194,984102,329
 Procyon cancrivorus0.66144,544 93,3250.42186,209 79,004
 Procyon lotor0.78104,713 83,1761.35 79,308107,152
 Nasua nasua1.17109,648125,8930.59 95,333 56,234
 Bassaricyon gabbii1.05123,027128,8250.63208,930131,826
 Potos flavus0.76186,209141,2540.91147,911134,896
 Ailurus fulgens1.07154,882165,9591.12 95,499106,989
 Zalophus californianus1.86 30,903 57,5440.93 75,858 70,795
 Callorhinus ursinus1.7 63,096104,7131.82 60,256109,648
 Ursus maritimus2.19 44,668 95,4993.24 34,674112,202
 Canis lupus familiaris0.76204,174158,4890.35288,403102,304
 Canis latrans0.34 72,444 25,1190.46 74,131 33,884
 Vulpes vulpes0.95 81,283 77,6250.58138,038 79,433
 Panthera pardus0.68 75,858 51,2860.85 51,286 43,652
 Felis catus0.2114,815 22,9090.22 79,433 17,783
 Puma concolor1.07 69,183 74,1310.95 87,096 83,176
 Crocuta crocuta1.26 63,096 79,4330.98 64,565 63,096
 Cynictis penicillata0.87141,254123,0270.68115,025 78,101
 Callithrix geoffroyi0.29338,844 97,7240.76 85,114 64,565
 Leontopithecus rosalia0.24301,995 72,4440.93 56,234 52,481
 Saguinus oedipus0.3328,541102,3290.68114,815 77,625
 Cebus capucinus0.17245,471 41,6870.66114,900 75,858
 Saimiri sciureus0.25478,630117,4900.65113,200 74,131
 Aotus trivirgatus0.12410,950 59,9300.40100,009 39,811
 Callicebus moloch0.27467,735125,8930.71102,329 72,444
 Pithecia pithecia0.58169,824 97,7240.38 45,162 17,378
 Alouatta caraya0.22194,984 42,6583.24 36,308117,490
 Alouatta palliata0.28176,349 49,1680.81 81,283 66,069
 Ateles ater0.3218,776 67,6080.69116,505 79,006
 Macaca fascicularis0.14331,131 47,8631.78134,896234,423
 Macaca mulatta0.27422,149113,7831.58 67,608107,152
 Macaca maura0.18361,470 64,343   
 Cercocebus torquatus0.39426,580165,9591.02 60,256 61,660
 Mandrillus sphinx0.52263,027138,0380.74 87,099 64,569
 Papio anubis0.51275,423141,2540.47194,984 89,125
 Cercopithecus mitis0.16245,471 39,8110.78 89,125 69,202
 Cercopithecus nictitans0.26302,604 78,1680.74 97,146 71,950
 Erythrocebus patas0.37416,869154,8821.36 78,740107,345
 Colobus angolensis0.23223,872 51,2861.48 75,858109,648
 Trachypithecus francoisi0.33446,684147,9110.72112,202 79,433
 Pongo pygmaeus0.96151,356144,544
 Pan paniscus0.62218,776134,8963.09 44,668138,038
 Pan troglodytes0.59208,930123,0270.66 80,261 53,703
 Homo sapiens0.72234,423169,8241.35 32,359 43,652
 Gorilla gorilla0.95144,544138,038
 Hylobates muelleri0.41229,087 93,3251.66 57,544 93,325
 Symphalangus syndactylus0.42239,883101,1031.66 74,131123,027
 Tarsius bancanus0.26234,510 60,2560.48204,174 97,724
 Tarsius syrichta0.25200,103 50,4250.47331,131154,882
 Lemur catta0.85 70,795 60,2560.79 87,096 69,183
 Eulemur mongoz0.59234,423138,0380.61117,490 71,328
 Microcebus murinus0.59190,546112,2021.95111,302218,776
 Cheirogaleus medius0.59186,209109,6482.00107,152213,796
 Galago senegalensis0.45338,844151,3560.63245,471151,356
 Nycticebus coucang0.49109,648 53,7030.26158,489 41,687


Neuronal and glial cell densities of V1 and CA1–3 were plotted as functions of one another in carnivores and primates (Table 1). Scaling exponents were determined by standard major axis (SMA) line fitting based on log-transformed data. Independent contrasts were calculated using the PDAP:PDTREE module of Mesquite (Maddison and Maddison 2011) from a pruned mammalian phylogeny with the original branch lengths (Bininda-Emonds et al. 2007). Stepwise Akaike's information criterion (AIC) was used to determine the relative strengths of variables in predicting glia–neuron ratio in V1 and CA1–3 (Yamashita et al. 2007); Pearson product–moment correlations were used to determine linear dependence between cellular variables; and recursive trees and additional multiple regression metrics were used to determine the relative contributions of each variable to overall variation (Supporting information). Data within taxonomic groups were tested for homogeneity of variance with Bartlett's test and for normality with the Shapiro-Wilk's W test. Differences in distributions between taxonomic groups were tested using a two-sample Kolmogorov–Smirnov goodness-of-fit test. Kruskal–Wallis sum rank and multiple comparison tests were used to determine sample mean differences between groups. Statistical significance for all analyses was set at 0.05 (two tailed). All analyses were performed in R (version 2.13) with our own code and the package SMATR (Warton et al. 2006).


Cellular variables in V1 and CA1–3 showed significantly different scaling patterns in primates and carnivores (Table 2A, B). In carnivores, glia–neuron ratio, neuronal density, and glial cell density in V1 were shown to scale isometrically with glia–neuron ratio, neuronal density, and glial cell density in CA1–3, respectively, for species mean data and independent contrasts (Figs. 2 and 3; Table 3). In primates, none of the variables in V1 scaled significantly with any of the variables in the hippocampus for species mean data or independent contrasts (Table 3). Pearson product–moment correlations (PMCC) showed glia–neuron ratio (R2= 0.814, P < 0.001), neuronal density (R2= 0.751, P < 0.001), and glial cell density (R2= 0.862, P < 0.001) in V1 and CA1–3 to have strong linear dependences in carnivores. Furthermore, stepwise AIC multiple regressions showed glia–neuron ratio in CA1–3 to be the greatest predictor of glia–neuron ratio in V1 (t-value = 4.477, P= 0.001) and glial cell density in CA1–3 to be the greatest predictor of glial cell density in V1 in carnivores (t-value = 7.429, P < 0.001). Relative importance metrics and recursive tree models strongly supported these results (Figs. 4, S1A–S1K). In primates, PMCC showed no significant correlations between V1 and CA1–3 and no relative importance metric showed variables in V1 and CA1–3 to significantly predict or contribute to variance in one another.

Figure 3.

Log-log regression plots for neuronal density and glial cell density in CA1–3 and V1 for carnivores (A, B) and primates (C, D). The dotted line represents SMA regressions fitted to carnivore species mean data (R2 > 0.75, P < 0.001). All SMA exponents for species mean data and independent contrasts are presented in Table 3. See Figure 2 for legend.

Table 3.  Slope estimates for scaling relationships based on cell densities (cells per mm3).
TaxaDependent variableIndependent variableSMASpecies mean dataSMAIndependent contrasts
R2Lower 95% CIUpper 95% CIPR2Lower 95% CIUpper 95% CIP
CarnivoresGlia–neuron ratio in V1Glia–neuron ratio in CA1–30.9230.8140.6951.2260.000 0.9190.422 0.781 1.1210.003
(n=21)Neuron density in V1Neuron density in CA1–30.9220.7510.6711.2680.000 0.9970.292 0.808 1.2460.025
 Glia density in V1Glia density in CA1–31.110.8620.8711.4100.000 1.1830.685 0.994 1.4670.000
 Glia–neuron ratio in V1Brain mass (g)0.4190.5540.2840.6190.009 0.3000.483 0.308 0.6990.031
 Glia–neuron ratio in CA1–3Brain mass (g)0.4710.5490.3150.7050.012 0.2720.532 0.419 0.780.013
PrimatesGlia–neuron ratio in V1Glia–neuron ratio in CA1–30.9230.0190.6951.2260.565−0.650.071−0.528−0.8400.701
(n=37)Neuron density in V1Neuron density in CA1–30.9220.0150.6711.2680.605 0.6300.154 0.499 0.8050.361
 Glia density in V1Glia density in CA1–31.1080.0420.8711.4090.386 0.7280.080 0.537 0.9060.670
 Glia–neuron ratio in V1Brain mass (g)0.3410.2560.2440.4770.138−0.10.018−0.510 0.3620.469
 Glia–neuron ratio in CA1–3Brain mass (g)0.3990.2720.2820.5630.129 0.1190.051−0.277 0.4700.208
Figure 4.

Relative importance metrics (top) and recursive trees (bottom) for determining glia–neuron ratio in carnivores (left) and primates (right). In carnivores, the variables collectively explained 82.91% of the observed variance, with the contribution of glia–neuron ratio in CA1–3 shown to be significantly greater than that of any other variable for all metrics. The recursive tree shows glia–neuron ratio in CA1–3 to be the foremost and greatest contributor to variance in glia–neuron ratio in V1, stratum moleculare volume to be a significant contributor in species with a low glia–neuron ratio in CA1–3 (<0.51), and cerebellar granular layer volume to be a significant contributor in species with a large glia–neuron ratio in CA1–3 (>0.51). In primates, the variables collectively explained 28.79% of the observed variance. No variable is shown to contribute significantly more to variance than any other variable for all metrics. All variables in the recursive tree are log transformed and the branch lengths are representative of the deviance explained by each variable. Abbreviations: BdM = body mass (kg), BrM = brain mass (g), CA.GNI = glia–neuron ratio in CA1–3, CrbGc = volumetric estimate of the granule cell layer of the cerebellum (μm3), EQ = encephalization quotient, GstLth = gestation length (days), StrGr = volumetric estimate of the stratum granulosum of the dentate gyrus (μm3), StrMol = volumetric estimate of the stratum moleculare of the dentate gyrus (μm3), V1.GNI = glia–neuron ratio in V1.

We further tested glia–neuron ratio as a function of brain mass and showed significantly different correlations in carnivores and primates (Fig. 5; Table 3). In carnivores, glia–neuron ratio in both V1 and CA1–3 showed weak but significant (R2 < 0.590, P < 0.05) correlations with brain mass. No significant correlations were found between brain mass and glia–neuron ratio in either V1 or CA1–3 in primates.

Figure 5.

Carnivores show significant linear scaling relationships between brain mass and glia–neuron ratio in both CA1–3 (A; y= 0.469x−1.035, R2= 0.583, P= 0.007) and V1 (C; y= 0.434x−0.899, R2= 0.531, P= 0.013). Scaling relationships in primates for both CA1–3 (B) and V1 (D) do not reach significance for species mean data or independent contrasts (see Table 3). See Figure 2 for legend.

Group means of glia–neuron ratios in V1 showed homogeneity of variance (K2= 0.403, P= 0.818), as well as normality in primates (W= 0.974, P= 0.612) and carnivores (W= 0.923, P= 0.098). Group means of glia–neuron ratio in CA1–3 showed homogeneity of variance (K2= 3.558, P= 0.169), normality in primates (W= 0.960, P= 0.201) and carnivores (W= 0.979, P= 0.925), and normality of distribution between carnivores and primates (D= 0.315, P= 0.165).


Most quantitative comparative studies of the mammalian brain have focused on the evolutionary relationships among different brain-region volumes (Jerison 1973; Gould 1975; Stephan et al. 1981; Finlay and Darlington 1995; Barton and Harvey 2000; Clark et al. 2001; Lefebvre et al. 2004; Yopak et al. 2010). However, no studies have yet considered the coordinated evolution of neuronal and glial cell distributions in regions of the neocortex and allocortex. As recent evidence has confirmed that volume and the total number of neurons in a given brain region show phylogenetically variable relationships to one another (Herculano-Houzel et al. 2006, 2007; Sarko et al. 2009; Herculano-Houzel 2010), investigating species diversity at the cellular level can help to identify evolutionary physiological constraints acting on the mammalian brain. Our data revealed significant relationships between neuron and glia in the primary visual cortex (V1) and hippocampal subfields (CA1–3) in carnivores that are not present in primates. Specifically, primates showed no significant scaling relationship between neurons and glia in V1 and CA1–3 and a remarkably different relationship between glia–neuron ratios in V1 and CA1–3 compared to all other eutherian species (Fig. 6; Table S3). We propose that the pattern observed in carnivores is indicative of constraints acting on evolutionary processes affecting mammalian brain development, and that the alteration of that pattern observed in primates represents a removal or relaxation of certain constraints. It is possible, for example, that evolutionary adaptations in the visual cortex in primates have influenced certain neurogenetic mechanisms, such as apoptosis (see Lietzau et al. 2009), and thus affected late-stage cell proliferation in other regions. Our data show that variation in the cellular organization of two diverse brain regions may be constrained along a mammalian lineage, but may also be relaxed or specialized along another lineage. As an induction of evolutionary change, the removal or relaxation of constraints may be a condition for adaptation.

Figure 6.

Glia–neuron ratios in V1 and CA1–3 for nonprimate and primate (dashed box) species show consistently higher values in V1 (mean = 1.05 ± 0.36) than in CA1–3 (mean = 0.78 ± 0.22) in nonprimates, but consistently lower values in V1 (mean = 0.41 ± 0.17) than in CA1–3 (mean = 1.07 ± 0.24) in primates. Mean values are significantly different in nonprimates and primates for both V1 (T= 6.733, P < 0.001) and CA1–3 (T=−2.377, P= 0.021) when all species are sampled (see Table S3).

In addition to interspecific differences in the number of cortical areas, with a proliferation of cortical areas generally following an increase in brain size (Krubitzer and Huffman 2000), interspecific differences in cortical cytoarchitecture have been shown to exist (Hof et al. 2000). Adaptations in cellular organization, which often represent isolated functional or behavioral variations and may be more easily interpreted than differences in cortical size across taxa (Sherwood et al. 2003, 2009; Hof et al. 2005; Raghanti et al. 2008; de Sousa et al. 2009, 2010), suggest that cellular reorganization may be one pathway to the independent specialization of brain regions. Our data agree with existing evidence showing the primate visual cortex to be, among mammals, especially derived (Preuss et al. 1999; Preuss and Coleman 2002) and that interspecific diversity in the cytoarchitecture of the visual cortex has arisen independently of brain size evolution (de Sousa et al. 2009, 2010). However, because the adult forms of cortical areas are a result of developmental processes that associate diverse cortical regions, there is a fundamental difficulty in selecting one region without affecting all other developmentally associated regions. Although there is evidence for isolated cortical adaptation between, for example, nocturnal and diurnal rodents (Campi and Krubitzer 2010), concerted morphological evolution of cortical regions appears to be the prevailing trend (Finlay et al. 2001). But it is still unclear whether the same principle of concerted evolution applies to neurotransmitters, receptors, the expression of neuromodulators, and cell structure and organization. Is an increase in glia to neurons in V1 likely to be accompanied a priori by a similar increase in the hippocampus? Is the principle of concerted evolution relevant at the cellular level? The observation that glia–neuron ratios in the neocortex and allocortex—or even brain size and body size—show tight statistical correlations within a particular mammalian order may simply mean that the only selection pressure acting to stabilize the relationship is a constraint on a developmental process. Additionally, the especially derived V1 in primates makes it difficult to conclude that all brain regions are more free to evolve independently in primates than in other mammals. In fact, homologous corticogenesis in carnivores and primates indicates that development of the cortex in mammals is influenced by similar constraints (Reillo et al. 2010; Kelava et al. 2012) and it is more likely, therefore, that selection on the proliferation of glia, which occurs subsequently to the proliferation of neurons, is responsible for the variation observed in this study. Comparative work on gliogenesis in mammals would be needed to test such a hypothesis.

We have provided evidence that diverse regions of the brain along a mammalian lineage are not de facto capable of evolving independently at the cellular level, with the implication that regions may only evolve independently following deep phylogenetic adaptations to conserved developmental processes.

Associate Editor: C. Farmer


This work was supported in part by a grant from the J. S. McDonnell Foundation (22002078 to PRH and CCS; 220020293 to CCS), National Science Foundation (BCS-0827531 to CCS), National Institutes of Health (NS42867 to CCS), Brain Research Trust (EL), and University of London Central Research Fund (EL). EL would also like to thank Archibald Fobbs for help with the neuroanatomical collection at the NMHM (Washington, DC), C. Stimpson for assistance in preparation of many of the brains that were included in this study, and Evan Charles for discussion.