Evolutionary specialization in mammalian cortical structure


Robert A. Barton, Evolutionary Anthropology Research Group, Department of Anthropology, Durham University, Durham DH1 3HN, UK.
Tel.: +44 191 334 6171; fax: +44 191 334 6101; e-mail: r.a.barton@durham.ac.uk


Changes in neocortex size were a prominent feature of mammalian brain evolution, but the implications for cortical structure, and consequently for the functional significance of such changes in overall cortical size, are poorly understood. A basic question is whether functionally differentiated cortical areas evolved independently of one another (adaptive specialization) or were allometrically constrained to co-vary tightly with the size of the whole. Here, I provide comparative evidence for adaptive specialization of cortical structure. First, the sizes of individual areas differ significantly between taxa after controlling for overall cortical size. Second, an analysis of separate visual cortical areas reveals that these exhibit statistically correlated evolution, independent of variation in nonvisual areas. Third, visual cortex size exhibits correlated evolution with peripheral visual adaptations (eye morphology and optic nerve size) and with photic niche. Thus, the evolution of mammalian cortical structure was closely associated with specialization for different sensory niches.


The vertebrate brain is a heterogenous organ comprised of distinct functional systems that mediate different behaviours and cognitive attributes. Comparative studies reveal a mosaic pattern in the evolution of these systems, in which size change in one system is partially independent of size change in other systems (Rilling & Insel, 1998; Barton & Harvey, 2000; Marino et al., 2000; Kaas & Collins, 2001; Whiting & Barton, 2003; Iwaniuk et al., 2004; Schoenemann et al., 2005; Striedter, 2005), and in which relative size change in specific systems correlates with functionally associated aspects of behaviour and ecology (Eisenberg & Wilson, 1978; Krebs, 1990; Barton et al., 1995; Barton, 1996, 1998; Szekely et al., 1996; Frahm et al., 1997; Safi & Dechmann, 2005).

The extent to which such mosaic size change characterizes fine-grained functional divisions within the mammalian neocortex has been questioned, however. The neocortex is considered to be a hallmark of mammalian brain evolution. It is unique to mammals, highly complex in structure and function, and disproportionately expanded in large-brained species (Jerison, 1973; Kaas, 1995; Barton & Harvey, 2000). Anecdotal evidence suggests that species differences in ecologically related specializations of peripheral sensory organs are associated with differences in cortical structure (Allman & McGuinness, 1988; Cooper et al., 1993; Catania, 2000), but quantitative comparative evidence on this question is lacking. In the only quantitative comparative study of the evolution of mammalian cortical regions, Kaskan et al. (2005) found that, despite marked differences in the density of retinal rod cells in nocturnal and diurnal mammals, regression plots of visual cortex size on the total size of the neocortex revealed no evidence of ecologically related variation in visual cortex size. They concluded that, in contrast to the extensive adaptive specialization of peripheral structures such as the eye, ‘the areal divisions of the cerebral cortex are considerably more conservative’ (Kaskan et al., 2005, p. 91). If true, this conclusion has considerable importance for understanding the nature of the neocortex as an information processing device. In particular, it would tend to support the view that it ‘is not a hodgepodge of specialized circuits, each chosen by (different) evolutionary pressures’ (Quartz & Szenowski, 1997).

Rejection of a role for natural selection in determining the fine structure of the neocortex may be premature, however. First, plotting the size of a cortical area against total cortical size is a conservative test of the adaptive specialization hypothesis, because the area itself contributes to variation in total cortical size. In primates, visual areas comprise over 50% of the neocortex (Drury et al., 1996), so plotting visual cortex size against total cortical size creates a serious problem of part–whole correlation. A more appropriate test would involve plotting the size of one functional area against that of a different area. Second, no statistical test of variation in the size of specific functional areas has yet been performed. Similarly, no quantitative test for correlated evolution of cortical areas and corresponding peripheral sensory structures has been conducted. Finally, the question of whether ecological factors explain variation in the size of specific cortical areas needs to take into account confounding effects of a variety of different aspects of ecological niches, either by controlling statistically for such differences in large samples of species or, as here, by restricting analysis to a taxon that is relatively ecologically homogenous, except with regard to a specific variable of interest. Primates, for example, vary in activity timing, but none fly, swim, burrow or echolocate, increasing the chance that correlated variation of visual cortex size and activity timing would be detected.

Here I analyse comparative data on 30 species of mammals to address these issues, and also analyse a larger though taxonomically and ecologically narrower data set on visual cortex, peripheral visual structures and ecological niche in primates. First, I test for taxonomic differences in the size of cortical regions relative to each other. If it is true that each area co-varies tightly with overall size (i.e. has not evolved independently of the whole in response to specific selection pressures), phylogenetic affiliation should not explain variance over and above that explained by overall size. Thereafter, I focus on the evolution of the visual system. The adaptive specialization hypothesis predicts correlated evolution among separate visual cortical areas, after controlling for variation in other functional regions. Next I test for correlated evolution between peripheral visual structures and the size of visual cortex, an explicit test between the ‘peripheral variability and central constancy’ hypothesis of Kaskan et al. (2005), which predicts no such correlation, and the adaptive specialization hypothesis, which does predict such correlations. Finally, I test whether activity period is associated with variation in relative visual cortex size in primates. More detailed explanations of these predictions and how they are to be tested are provided below.

Materials and methods

Data and predictions

Data on cortical region sizes in 30 mammalian species were kindly provided by P. Kaskan (pers. comm.); see Kaskan et al. (2005) for details of data acquisition and methods. Size of total extra-striate visual cortex was calculated from these data by subtracting the size of area VI from total visual cortex size. Data on visual and other brain region volumes in primates are from Stephan et al. (1981). Area VI was the only cortical area measured in the latter data set. Data on peripheral visual structures are as follows. For primates, Stephan et al. (1983) provide data on the size of the optic nerve (cross-sectional area), optic tract (mid-sagittal sectional area) and eye (half surface area). Optic foramen size, ganglion cell density and the ‘summation index’ (number of ganglion cells per 100 photoreceptor cells) are from Kay & Kirk (2000). Relative cornea size data are from Kirk (2004). Eye sizes (axial length) for a wider sample of mammals are from Howland et al. (2004). Each of these variables is likely to be related to visual abilities and information processing in central brain areas as follows. The size of the eyes influences the focal length, size of the retinal image and acuity (Walls, 1942; Kay & Kirk, 2000; Kiltie, 2000; Howland et al., 2004). Indeed, eye size correlates positively with behaviourally measured acuity amongst both birds and mammals (Kiltie, 2000). However, a confounding factor is that eye size also influences light-gathering capacity, and, in primates, is greater in nocturnal than in diurnal species (Kay & Kirk, 2000). Increased eye size in nocturnal primates does not result in greater acuity than in diurnal species with smaller eyes, because retinal adaptations for sensitivity in nocturnal primates inherently conflict with acuity (Kay & Kirk, 2000). Diurnal mammals have greater visual acuity than do nocturnal mammals (Kay & Kirk, 2000; Kiltie, 2000). Hence, eye size is determined by two different selection pressures, one on sensitivity and one on acuity. Because of these confounding effects of eye size on both acuity and sensitivity, the predicted correlation between eye size and visual cortex size should be stronger in comparisons between primate species with the same activity period, or when the effects of activity period are controlled for. The number of ganglion cells and the size of the optic nerve, optic tract and optic foramen are related to the amount of information transmitted from retina to brain. Retinal ganglion cells integrate signals from the photoreceptors and transmit these integrated signals to the brain. The optic nerves and tract are largely composed of the axons projecting from ganglion cells to the brain. The size of the optic foramen or canal (the perforation in the orbit through which the optic nerve passes) provides an independent, osteological measure of optic nerve size for a larger sample of species, and relative optic foramen size ‘is highly correlated with the degree of retinal summation and inferred visual acuity’ (Kay & Kirk, 2000). Similarly, the summation index is negatively related to receptive field size and photosensitivity, and hence positively to acuity, and it is higher in diurnal than in nocturnal taxa (Kay & Kirk, 2000). Relative cornea size, on the other hand, is reduced in diurnal species, and is negatively associated with acuity (Kirk, 2004). Hence, relative cornea size is predicted to correlate negatively with visual cortex size. Some of these variables may be related allometrically to body size, so estimates of body size (from Barton, 1998) were included in analyses. Data on primate activity periods were taken from the literature (Barton, 1998; Kay & Kirk, 2000; Kirk, 2006a) and scored in two different ways. In the first, species were classified dichotomously, as either nocturnal or diurnal/partly diurnal, the latter category including ‘cathemeral’ species. Second, species were ranked on a ‘diurnality’ scale of 0–2 (nocturnal = 0, cathemeral = 1, diurnal = 2).


All size data were log-transformed. Taxonomic effects were assessed by analyses of covariance (controlling either for total neocortex size or the size of other cortical areas) on the size of the three unimodal sensory areas (visual, auditory and somatosensory) in four taxonomic groups (haplorhine primates, strepsirhine primates, insectivores and marsupials) for which there were adequate data. Where possible, I controlled for the size of other cortical regions rather than for total neocortex size. For comparability with previous work (Kaskan et al., 2005), however, I also carried out additional analyses controlling for overall neocortex size, with the proviso that tests for area-specific size variation will be conservative, because of the part–whole correlation (see Introduction): significant results using this procedure will thus be indicative of particularly strong patterns of covariation between visual structures. In the analysis of visual area VI in primates, measurements of nonvisual cortical areas were not available. Here, again, I carried out two types of analysis, one, a conservative analysis, controlling for total neocortex size, and the other controlling for nonvisual hindbrain structures (cerebellum, medulla and pons).

I tested for correlated evolution among nervous system components using the method of independent contrasts (Felsenstein, 1985; Harvey & Pagel, 1991) as implemented in the computer programme caic (Purvis & Rambaut, 1995). Because caic standardizes the evolutionary change estimates, or contrasts, according to the amount of time since evolutionary divergence of the taxa, the output values should have homogenous variances and be suitable for analysis by standard linear methods. However, before proceeding to analysis, I checked that the contrast values met this and other assumptions (Purvis & Rambaut, 1995). For comparability, the phylogenetic tree used by Kaskan et al. (2005) was also used here for the data set of 30 mammals. For the primate analyses, a phylogenetic tree, including branch lengths, was taken from Purvis (1995). Independent contrasts were then analysed using multiple regression through the origin, to test for correlated evolution among functionally related structures whilst controlling for variation in other structures.

Analyses of the effects of activity period were carried out in two ways, according to which activity period scale was used. First, variation in relative visual cortex size in relation to dichotomously classified activity period was evaluated using the BRUNCH algorithm in caic (Purvis & Rambaut, 1995). For these analyses, relative visual cortex size was calculated as the residuals from the independent contrast-based regression of visual cortex size on size of other structures. Second, visual cortex size contrasts calculated using the CRUNCH algorithm for continuous variables were regressed on contrasts in diurnality and contrasts in other structures (neocortex or hindbrain). The occurrence of some diurnality contrasts with values of zero (i.e. those between lineages not differing in degree of diurnality) may violate statistical assumptions, so the robustness of the results was checked by repeating analyses after removal of such contrasts.


Taxonomic differences in relative cortical area size

Taxonomic differences in the relative size of cortical areas are most apparent when one area is plotted against another (Fig. 1). Reflecting the emphases of the different taxa on vision vs. touch, both primate taxa have large visual relative to somato-sensory cortices, and, on average, the predominantly diurnal haplorhines have relatively larger visual cortices than the nocturnal strepsirhines. Even using the conservative procedure of controlling for total cortical size (see Materials and methods), analyses of covariance reveal significant differences among taxa in all three cortical areas tested (Table 1), and post hoc tests reveal that all pairwise comparisons between taxa were significant at P < 0.01–P < 0.0001.

Figure 1.

 Taxonomic differences in visual cortex relative to somatosensory cortex surface areas. Open circles = haplorhine primates, closed circles = nocturnal strepsirhine primates, diamonds = marsupials, crosses = insectivores.

Table 1.   Taxonomic effects on the relative size of sensory cortical areas in mammals.
 d.f.Sum of squaresMean squareF-valueP-value
  1. Results of analyses of covariance are shown, in which effects of taxonomic group are found after controlling for total neocortex size. Note that the apparent allometric effects of total cortex size are inflated by the nonindependence between this variable and the dependent variables – see Materials and Methods).

Visual cortices
 Total cortex17.187.18170.6< 0.0001
Auditory cortices
 Total cortex16.386.38229.7< 0.0001
Somatosensory cortices
 Total cortex17.097.09299.9< 0.0001
 Taxon31.000.3314.0< 0.0001

Correlated evolution of separate visual cortical areas

A multiple regression analysis of independent contrasts in the size of V1 on other cortical areas reveals a significant positive correlation between the visual areas (V1 and V2), even with the effects of variation in the sizes of auditory cortex and somato-sensory cortex taken into account (t3,12 = 3.64, P = 0.003). In contrast, the correlations between V1 and nonvisual areas were, as predicted by the adaptive specialization hypothesis, nonsignificant (auditory cortex, t3,12 = −2.06, P = 0.07; somatosensory cortex, t3,12 = 0.45, P = 0.66). In a similar analysis, V1 was regressed on total extra-striate visual cortex, auditory cortex and somato-sensory cortex, and again, a significant relationship was found between the visual areas (t3,23 = 3.0, P = 0.006), but not between V1 and nonvisual areas (auditory cortex, t3,23 = −0.42, P = 0.68; somatosensory cortex, t3,23 = 0.99, P = 0.33). Hence, separate visual cortical regions exhibit positively correlated evolution independently of variation in other cortical regions. These patterns of correlated evolution are presented graphically in Fig. 2.

Figure 2.

 Correlations within and between sensory modalities. Relative contrasts are the residuals from regressions of independent contrasts in the size of each sensory area on contrasts in the size of other sensory areas. In (a) the y-axis variable is the residuals of contrasts in the surface area of visual area VI regressed on contrasts in the surface area of somatosensory and auditory cortices, whereas the x-axis is similar residuals for visual area VII. Hence, visual areas VI and VII exhibit positively correlated evolution after controlling for variation in auditory and somatosensory cortices. In (b), the same type of analysis has been performed, but the correlation is between VI and the surface area of all other cortical visual (extra-striate) areas. In (c) and (d), the relationships between visual area VI and auditory cortices, controlling for variation in somatosensory cortices and either visual area VII (c) or all other extra-striate visual areas (d), is presented. The correlations between visual areas, but not those between visual and auditory areas, are significant (see text for multiple regression statistics).

Covariation between visual cortex and peripheral nervous structures

Amongst mammals in the data set of 30 species, there is, despite the small sample size, a significant positive relationship between contrasts in eye size and contrasts in V1 size, controlling for body size and the size of nonvisual cortex (t3,8 = 3.02, P = 0.039). For total visual cortex, there is a nonsignificant trend in the predicted direction (t3,8 = 2.31, P = 0.07), which becomes significant with stepwise variable selection (t2,8 = 2.52, P = 0.046).

Results of multiple regression analyses of visual area VI in primates are presented in Table 2. These also reveal correlated evolution between area VI and peripheral visual structures, including the predicted negative correlation with relative cornea size (see Materials and Methods). As predicted, the relationships are all weaker, although with only one exception still significant, when controlling for total neocortex size. Also, as predicted, the relationships between eye size and V1 size are stronger when the effects of activity period are removed (controlling for body mass, hindbrain and activity period, t4,27 = 3.95, P = 0.0006; controlling for body mass, neocortex size and activity period, t4,27 = 3.29, P = 0.003).

Table 2.   Correlated evolution of visual cortex and peripheral visual anatomy.
Controlling for body mass along withRegression of visual cortex size on
Optic nerveOptic tractOptic foramenEyeCorneaGanglion cellsSI
  1. In each analysis, visual cortex volume contrasts were regressed on volume contrasts in a peripheral visual structure, body mass contrasts and either hindbrain or total neocortex volume contrasts. See Materials and Methods for definitions of variables. SI = summation index (Kay & Kirk, 2000. t-values and associated significance levels and degrees of freedom are shown. *P < 0.05, **P < 0.01, ***P < 0.001).

Hindbrain size2.82**4.66***2.85**2.64*−2.48*2.87*2.23*
d.f. = 3, 29d.f. = 3, 18d.f. = 3, 20d.f. = 3, 24d.f. = 3, 21d.f. = 3, 8d.f. = 3, 11
P = 0.009P = 0.0002P = 0.011P = 0.014P = 0.028P = 0.021P = 0.047
Total neocortex size2.24*3.43**2.03*2.29*−2.33*2.41*0.25
d.f. = 3, 29d.f. = 3, 18d.f. = 3, 20d.f. = 3, 24d.f. = 3, 21d.f. = 3, 8d.f. = 3, 11
P = 0.033P = 0.003P = 0.05P = 0.030P = 0.030P = 0.042P = 0.81

Visual cortex and activity period in primates

There is some overlap between nocturnal and diurnal primates in relative visual cortex size (Fig. 3), but this is partial, not complete. Amongst the predominantly diurnal haplorhine primates, the two nocturnal species (tarsier and owl monkey) have relatively small visual cortices, whereas amongst the predominantly nocturnal strepsirhine primates, the diurnal lemurids and indriids have relatively large visual cortices (Fig. 3). An independent contrasts analysis on relative V1 size confirms that visual cortices are relatively enlarged in diurnal lineages. First, treating activity period as a categorical variable, relative visual cortex size is larger in diurnal than in nocturnal lineages (t3 = 2.91, P = 0.03, one tailed). Second, treating activity period as a continuous variable, a multiple regression analysis reveals a significant relationship between contrasts in V1 size and contrasts in activity period, controlling for contrasts in body mass and either hindbrain size (t3,38 = 2.91, P = 0.006) or total neocortex size (t3,38 = 2.64, P = 0.013). These relationships strengthen after removing activity period contrasts with values of zero (controlling for hindbrain size, t3,14 = 4.06, P = 0.002; controlling for neocortex size, t3,14 = 4.23, P = 0.001). Hence, even using the conservative method of controlling for total neocortex size, activity period and relative visual cortex size are correlated.

Figure 3.

 Relative visual cortex size in diurnal and nocturnal primates. Open symbols = diurnal species, filled symbols = nocturnal species, circles = haplorhine primates (monkeys, apes and tarsier), squares = strepsirhine primates (lemurs and lorisids). Regression lines are drawn for the diurnal haplorhines and nocturnal strepsirhines, indicating a clear grade shift. Note that the species that differ from the common activity pattern for their respective taxonomic groups (two nocturnal haplorhines and four diurnal strepsirhines) are intermediate. Vertical residuals have been drawn to emphasize that each of these species deviates as predicted from the regression line for their phylogenetic close relatives. Numbered species referred to in the text are: 1 = tarsier (Tarsius spectrum), 2 = slender loris (Loris tardigradus), 3 = slow loris (Nycticebus coucang) and 4 = owl monkey (Aotus trivirgatus).


Each of the predictions of the adaptive specialization hypothesis for the evolution of neocortical structure was confirmed. First, there is significant variation in the size of functional areas after removing the effects of variation in other areas, or in overall neocortical size. Second, separate visual areas evolved together independently of variation in other functional areas. Third, cortical visual areas show positively correlated evolution with the size of visual inputs to the brain and aspects of eye morphology indexing acuity. This is the first demonstration of a quantitative evolutionary association between corresponding central and peripheral sensory structures. Finally, visual cortex size correlates with photic niche. Hence, cortical structure appears to be fine tuned to the demands of different lifestyles and sensory specializations.

Comparisons between species that have similar-sized cortical areas in one sensory domain illustrates the magnitude of variation in other cortical areas over and above that because of allometry. Hedgehogs (Erinaceus europaeus), for example, have slightly larger somatosensory cortices than do tamarins (Saguinus fuscicollis), yet visual cortices in the tamarin are nine times larger than in the hedgehog (Table 3). In relative terms, therefore, the tamarin is clearly a visual specialist compared with the hedgehog. Even more dramatically, squirrel monkeys (Saimiri sciureus) and platypuses (Ornithorhyncus anatinus) have similar somtaosensory cortex sizes, but visual cortices are 127 times larger in the squirrel monkey (Table 3). Such variation is far beyond the two- to threefold difference in brain structure size observed within species (Finlay & Darlington, 1995; Kaskan et al., 2005).

Table 3.   Variation in absolute size (mm2) of visual cortex among species with similar somato-sensory cortex sizes.
 Cortical area
Monotreme1 vs. primate2
Insectivore1 vs. primate2

These large and statistically significant interspecific differences are most parsimoniously explained as products of natural selection on specific information processing capacities. Whereas activity period has been identified as one factor associated with visual system evolution in primates, other behavioural factors are also likely to be influential (Barton et al., 1995; Barton, 1998). Two nocturnal strepsirhines, the lorises Nycticebus coucang and Loris tardigradus, have visual cortices that are close in relative size to those of diurnal primates (Fig. 3). Interestingly, lorises have unusually convergent orbits compared with other nocturnal strepsirhines (Ross, 1995), and binocular convergence correlates positively with the evolutionary expansion of visual cortex in primates (Barton, 2004). A high degree of convergence in lorises is thought to be functionally associated with visually guided predation (Ross, 1995). Hence, activity period is only one of the ecological factors implicated in visual system evolution. Determining the ecological correlates of cortical variation among mammals more widely will require comparative data sets large enough to permit separation of the effects of a variety of factors, including activity period, diet and foraging technique, sociality and substrate use (e.g. fossorial habits). It would also be useful to measure a variety of peripheral sensory structures to test quantitatively for correlated evolution between cortical structure and sensory specialization (Catania, 2000).

Early suggestions linking visual specialization and brain evolution in primates (Elliot Smith, 1927; Jerison, 1973; Cartmill, 1974) have now received strong support from phylogenetic comparative studies. The evolution of large brain size is associated with relative expansion of cortical and subcortical visual structures (Barton, 1998), with increased relative optic nerve size (Kirk, 2006b), and with relatively convergent orbits (Barton, 2004). The evolutionary change in the visual system associated with overall brain and neocortex size is exhibited by parvocellular, but not magnocellular pathways (Barton, 1997, 1998, 2004), implying natural selection on relatively fine-grained, acute vision, rather than on low-acuity spatial vision and movement detection (Allman & McGuinness, 1988; Livingstone & Hubel, 1988). The present study adds to these findings by demonstrating correlated evolution between relative visual cortex size and aspects of ocular and peripheral nerve morphology that support visual acuity. The extent to which size variation in multisensory cortical areas, such as inferotemporal and frontal regions, is associated with visual projections from unimodal visual cortices is currently unknown, but potentially has important implications for understanding the adaptive significance of neocortical evolution in primates. Similarly, correlations with noncortical brain regions, such as the cerebellum, may provide more detailed indications as to the specific cortical circuits that underlie variation in overall neocortex size (Whiting & Barton, 2003). Amongst mammals more generally, a variety of sensory specializations is likely to impact brain size and structure (Eisenberg & Wilson, 1978; Catania, 2000). Nevertheless, the finding here of a correlation between visual cortex size and eye size in the taxonomically wider sample of mammals implies that the role of visual specialization may not be restricted to primates (see also Kirk, 2006a, b). Relative eye size and brain size also correlate in birds, and both eye and brain size correlate with activity pattern (Garamszegi et al., 2002). Hence, similar evolutionary effects appear to have occurred in disparate lineages.

An important question concerns the developmental mechanisms by which species differences in cortical structure arise during ontogeny. One possibility is that such differences in cortical structure are produced by differences in neural input to the developing cortex from specialized peripheral sensory structures. For example, the number of retinal ganglion cells might directly influence the growth of subcortical and cortical brain structures. Indeed, there is experimental evidence that perturbations of sensory inputs cause reorganization at the cortical level (Rakic et al., 1991; Quartz & Szenowski, 1997). However, other evidence implies more direct genetic control of cortical structure. Regional differentiation of the cortex occurs early in ontogeny, prior to and/or independently of thalamic innervation (Kennedy & Dehay, 1997; Rakic & Kornack, 2001), mutations are known which target specific regions of the sheet of neuronal progenitor cells known as the ‘proliferative zone’, also prior to innervation by the periphery (Rakic, 2004), and abolition of peripheral input does not prevent the development of region-specific cytoarchitecture (Rakic & Kornack, 2001). Whatever the precise ontogenetic mechanisms underlying species differences in cortical structure, these evidently arose, at least in part, through natural selection on specific sensory systems rather than simply through allometrically coordinated growth of individual areas.


I thank Peter Kaskan for generously sharing the data used in Kaskan et al. (2005) and Charlie Nunn, Alex Bentley and Isabella Capellini for helpful comments on drafts of the manuscript.