Across the 5 new primate species examined – Microcebus murinus, Callimico goeldii, M. fascicularis, M. radiata and Papio cynocephalus – body mass varies 133 times, from about 60 g in Microcebus to about 8,000 g in Papio and Macaca. Cortical mass varies between Microcebus and Papio, the smallest and largest species examined, by a factor of 132 times; cerebellar mass varies by a factor of 35 times; and the mass of the remaining structures varies 34 times. The numbers of neurons in these structures vary proportionately by 129 times in the Cx, 35 times in the Cb, and 25 times in the remaining areas (table ). The percentage of neurons in the brain and the 3 subdivisions analyzed does not vary significantly with the mass of the structures (Spearman correlation, all values of p > 0.8), and in all species but M. radiata, neurons comprise more than 50% of all brain cells (table ).
Cellular composition of the brain of a new set of primate species
Conformity to the Expected Cellular Composition
To examine whether the cellular composition of the brains of the present 5 primate species conforms to the rules identified previously for a different set of species, we first determined how the numbers of cells in each structure depart from the expected values obtained by applying to each species the scaling rules determined previously for primates [Herculano-Houzel et al., 2007
]. For comparison, we also calculated how much numbers of cells depart from the expected for the brain structures of each of the species in the previous study (see online suppl. table 1
), and from the expected cellular composition according to the scaling rules observed for rodents [Herculano-Houzel et al., 2006
]. The relationships examined were between the mass of each structure (whole brain, Cx, Cb, or remaining areas) and its numbers of neuronal and other cells. Percent deviations from the expected numbers of cells were calculated as [100 × (observed – expected)] for each relationship. The reciprocal relationships were not examined, as deviations tend to be symmetrical and would bias the results toward an erroneous average of zero deviation.
Figure shows that the cellular composition of each of the present 5 primate species, as well as of humans, departs from the expected for structure mass within the same range of variation found for the 6 species from which the cellular scaling rules for primate brains were obtained (average deviation from the expected: −0.04 ± 10.47%). The average deviation for each species is not correlated to brain size (Spearman correlation coefficient −0.173, p = 0.612). In contrast, the cellular composition of the primate brains analyzed here is systematically larger than the expected for rodent brains of similar size (average deviation from the expected 114.67 ± 79.74%; fig. ), and the larger the size of the primate brain, the larger the deviation from the expected for a generic rodent brain (Spearman correlation coefficient 0.918, p = 0.000). Although numbers of neuronal and other cells in the brain structures of M. fascicularis
tend to be smaller than the values expected for a primate brain, similar trends are also found for some species in the previous sample: M. mulatta
tend to have larger numbers of neuronal and other cells in their brain structures than expected, while Cebus
tends to have smaller numbers of neuronal and other cells than expected in its brain structures. Most importantly, numbers of cells in each structure typically depart from the expected by ±20%, as shown by the filled intervals between the 25 and 75% percentiles in figure . This finding indicates that the brains of the 5 present species, like those of humans [Azevedo et al., 2009
], conform to the cellular scaling rules described previously for 6 other primate species.
Fig. 2 The current primate species deviate from the expected in their cellular composition by as much as the primate species studied earlier. Y-axis, percent deviation from the neuronal and non-neuronal composition expected from cerebral cortex, cerebellum and (more ...)
Cellular Scaling Rules
We next determined the cellular scaling rules that apply to the brains of the 5 present primate species. As observed for the previous set of 6 species, the mass of all structures analyzed – Cx, Cb and RoB – is found to vary as a function of their respective numbers of neuronal and non-neuronal cells in ways that can be described equally well as power laws with exponents of approximately 1.0, and as linear functions (table ). The only exception is that the mass of the remaining structures in the new sample fails to vary significantly as a power function of its number of neurons, although this relationship can still be described as a linear function.
Cellular scaling rules for the current sample of primate brains
Updated Cellular Scaling Rules
As a final step in examining whether the same cellular scaling rules apply to the previous and present sets of primate species, we determined the cellular scaling rules that apply to the combined dataset comprising all 12 available primate species, including humans, and calculated how the scaling is affected by phylogenetic relatedness. We find that the updated cellular scaling rules that apply to the Cx and Cb are linear and remain so even when phylogenetic independent contrasts are used, as were the rules determined for the previous sample. Asshown in figure , the mass of these structures varies coordinately with their numbers of neuronal (fig. ) and non-neuronal cells (fig. ) in ways that can be described equally well by power laws with exponents of approximately 1.0 and by linear functions (table ). Additionally, power law exponents as well as the linear fits for the Cx and Cb are not significantly affected by the exclusion of humans from the extended dataset (table ).
Fig. 3 Scaling of brain structure mass in the combined dataset as a function of numbers of neurons and non-neuronal cells. Each point represents the average mass and number of neurons (left) or other cells (right) in the cerebral cortex (circles), cerebellum (more ...)
Cellular scaling functions for the expanded primate dataset
The cellular scaling rules that apply to RoB, however, are less clear. Although RoB mass in the expanded dataset is best explained (with the larger r2 value) as a linear function of its number of neurons, and whether independent contrasts are used or not (table ), the power slope that relates MRoB to NRoB (see table footnote for abbreviations) increases dramatically to about 1.4 (with or without humans) when phylogenetic relatedness is taken into account (table , right column). This suggests that rather than scaling in size linearly with its number of neurons as do the Cx and Cb, primate RoB may scale in size hypermetrically with its number of neurons, as does the rodent RoB.
In each structure, we find that neurons represent similar percentages of all cells across species (table ), which fail to vary with structure mass (fig. ; Spearman correlation, all values of p > 0.1), such that the O/N ratio (number of other cells/number of neurons) for each structure is relatively constant across species, not varying significantly with structure mass (Spearman correlation, all values of p > 0.1).
Fig. 4 The percentage of neurons in each structure does not vary significantly with structure mass in the combined dataset. Each point represents the average percentage of neurons among all cells found in the cerebral cortex (circles), cerebellum (squares) or (more ...)
Cx and RoB neuronal densities in the combined dataset covary respectively with structure mass (Spearman correlation, p = 0.045 and p = 0.017; fig. ), while Cb neuronal density does not (Spearman correlation, p = 0.417). However, as in the previous dataset, the power laws relating neuronal density and structure mass in the Cx and Cb fail to reach significance (p = 0.071 and 0.296, respectively), while neuronal density in the RoB decreases significantly with increasing structure mass raised to the power of −0.428 in the combined dataset (p = 0.005), although the power relationship corrected for phylogeny fails to reach significance (slope −0.318; p = 0.064). As seen before in the smaller dataset, non-neuronal densities in the various structures do not covary with structure mass (Spearman correlation, all values of p > 0.1), and have similar values across the structures (fig. ).
Fig. 5 Neuronal and non-neuronal densities do not co-vary with structure mass. Variation in density of neuronal cells (a) and other cells (b) in cerebral cortex, cerebellum and remaining areas plotted against the mass of each structure. Spearman correlation (more ...)
Relative Distribution of Mass and Cells
In the combined dataset, the relative mass of the Cx, expressed as the percentage of whole brain mass, increases significantly with increasing brain mass (Spearman correlation coefficient 0.911, p = 0.000), and can be expressed as a power function of brain mass with exponent 0.062 (p = 0.000), while the relative mass of the RoB decreases significantly with increasing brain mass raised to −0.233 (p = 0.001, Spearman correlation coefficient −0.818, p = 0.002) and the relative mass of the Cb fails to correlate with brain mass (Spearman correlation, p = 0.143; fig. ).
Fig. 6 Relative size of the cerebral cortex and cerebellum does not reflect relative number of neurons in these structures. Each point represents the average relative mass or average relative number of neurons, compared to the whole brain, of the cerebral cortex (more ...)
Larger cerebral cortices and cerebella do not hold relatively larger numbers of brain neurons (Spearman correlation, p = 0.537 and p = 0.750, respectively; fig. ). Similarly, relatively larger cerebral cortices and cerebella do not hold relatively larger numbers of brain neurons (Spearman correlation, p = 0.564 and 0.162, respectively; fig. ). In contrast, the relative number of neurons in the RoB does accompany both the absolute and the relative mass of this ensemble of structures: the relative number of RoB neurons decreases steeply with increasing absolute RoB mass, as a power function of MRoB with exponent −0.752 (p = 0.008; Spearman correlation coefficient −0.752, p = 0.008; fig. ) and varies linearly with relative RoB mass whether phylogenetic relationships are unaccounted for (Spearman correlation coefficient 0.715, p = 0.013; linear regression slope 0.222, r2 = 0.670, p = 0.002) or used for correction (linear regression slope 0.224, r2 = 0.604, p = 0.005; fig. ). Thus, only for the RoB is the relative distribution of neurons among brain structures reflected in the relative distribution of brain mass.
Brain × Body Scaling
In our previous study, we found that brain size increased as a linear function of body mass, or as a power function of this variable with an exponent of 1.017 across the 6 primate species then examined [Herculano-Houzel et al., 2007
]. Addition of the Saimiri
body mass, missing in the original report, decreases the exponent to 0.951 ± 0.219 (p = 0.000), and further inclusion of humans to that dataset changes the exponent to 1.012 ± 0.095 (p = 0.000). In contrast, for the present 5 species, we find that whole brain mass varies as a power function of body mass with a smaller exponent of 0.788 ± 0.127 (p = 0.025), although with a very large confidence interval that includes the previous exponent of 1.017 (table ). When all non-human primates examined are combined, we find brain mass to vary as a function of body mass raised to an exponent of 0.805 ± 0.094 (p = 0.000). Inclusion of humans in the combined dataset raises the exponent further to 0.903 ± 0.082 (p = 0.000; fig. ), which is little affected when phylogenetic relationships are accounted for (exponent 0.924, p = 0.000). Therefore, and in contrast to the relationship between brain mass and number of neurons, which is largely insensitive to the species compared, the power law that relates brain mass to body mass is highly sensitive to the particular species included in the comparison.
Fig. 7 Variation in brain mass and body mass across species. Each point represents the average brain and body mass of a primate species. Filled symbols = current dataset; unfilled circles = previous non-human dataset (Herculano-Houzel et al., 2007); cross = (more ...)
In line with the notion that body mass is more variable than brain mass for a given number of neurons in the brain of a species, we find that, in the combined dataset, brain mass is better correlated than body mass with the total number of neurons in the brain, Nbr (Spearman correlation coefficients, 1.000 and 0.900, respectively; p = 0.000). Using the power functions that relate body and brain mass to Nbr in the combined dataset (Mbr = 7.134 × 10−10 Nbr1.130 and Mbo = 3.088 × 10−8 Nbr1.144; slopes after correcting for phylogenetic relatedness in the dataset, 1.138 and 1.232, respectively; fig. ), a comparison of the normalized residuals of body mass and brain mass onto the total number of neurons in the brain shows larger residuals for body mass than brain mass (fig. ), with significantly greater absolute values of normalized residuals for body mass than for brain mass (mean body mass residual, 0.511 ± 0.156; mean brain mass residual, 0.127 ± 0.032; Mann-Whitney test, p = 0.033).
Fig. 8 Brain mass is better correlated than body mass with total number of brain neurons. Variation in brain mass (a) and body mass (b) are shown plotted against the total number of brain neurons in each species. Filled symbols = present species; unfilled symbols (more ...)
Fig. 9 Residuals of brain and body mass regressed onto total number of neurons in the brain. Each point represents the residual of the regression of brain mass (Mbr, filled symbols) or body mass (Mbody, unfilled symbols) onto the total number of neurons in the (more ...)