Human frontal lobes are not relatively large

Abstract

One of the most pervasive assumptions about human brain evolution is that it involved relative enlargement of the frontal lobes. We show that this assumption is without foundation. Analysis of five independent data sets using correctly scaled measures and phylogenetic methods reveals that the size of human frontal lobes, and of specific frontal regions, is as expected relative to the size of other brain structures. Recent claims for relative enlargement of human frontal white matter volume, and for relative enlargement shared by all great apes, seem to be mistaken. Furthermore, using a recently developed method for detecting shifts in evolutionary rates, we find that the rate of change in relative frontal cortex volume along the phylogenetic branch leading to humans was unremarkable and that other branches showed significantly faster rates of change. Although absolute and proportional frontal region size increased rapidly in humans, this change was tightly correlated with corresponding size increases in other areas and whole brain size, and with decreases in frontal neuron densities. The search for the neural basis of human cognitive uniqueness should therefore focus less on the frontal lobes in isolation and more on distributed neural networks.

Although it is widely assumed that human brain evolution involved relative expansion of the frontal lobes, no clear picture has emerged from comparative studies. Early claims (1, 2) have been both cited as evidence for relative expansion (3, 4) and questioned owing to small sample sizes and uncertainty about anatomical boundaries defined using older methods (5, 6).

Newer and larger data sets have not yet resolved these uncertainties but have been used as support for the claim of disproportionate expansion of frontal lobe structures and specific tissue types (4, 6–10), for the lack of thereof (5, 10, 11), or have given ambiguous results depending on factors such as which species are included in the analysis (12). In addition, there is variability among these studies as to whether and which other ape species are reported to exhibit similar relative expansion as humans (5, 6, 9). Finally, among those studies reporting larger than expected values for humans and/or other apes, there is little consistency in terms of which measures show differences, with claims based variously on whole frontal cortex volume (5), volume of prefrontal area 10 (7), prefrontal vs. nonprefrontal white matter volume (6), and left (but not right) prefrontal white matter volume relative to prefrontal gray matter volume (9).

Resolving this confused picture is of considerable importance to understanding the neural basis of human cognitive evolution. As Rilling (4) states, “In the field of comparative primate neuroanatomy, perhaps no question has engendered more interest than whether the human frontal lobes are larger than expected for a primate of our brain size.” Yet the strongest justifiable statement that could be made in a recent review was, “It can be tentatively concluded that it is questionable that the size of the prefrontal cortex can account for the human executive functions” (13).

One reason for ambiguity in the literature is the use of different types of measure that give different results. Claims about frontal lobe evolution in humans are sometimes based on unscaled measures (such as absolute size or percentage of brain volume comprising frontal areas or plots of raw—rather than logged—volumes) (3–5, 7). These measures suggest that human frontal cortices are larger as a proportion of total brain size than in other species. However, proportional size differences conflate selective enlargement with allometric scaling. Volumes of different brain structures change at different rates as brain and body size evolve (14, 15), and with different implications for variables more directly related to information processing than volume, such as numbers of neurons and synapses (16). For example, the volume of the whole neocortex increases with brain size more rapidly than does the volume of the cerebellum (15), but neuron density declines more rapidly in the neocortex, so that despite the volumetric ballooning of the neocortex as overall brain size increases, the proportion of the brain’s neurons in each of the two structures remains approximately constant (16). Volumetric ratios or proportions therefore progressively overestimate the contribution of the neocortex to neuron numbers as brain size increases.

The ballooning of cortical volume as brain size increases seems to be due largely to allometric constraints on connectivity, because cortical white matter volume increases substantially faster than gray matter volume (12, 15, 17, 18), and explains the positive allometry of neocortex size (15). This allometry is especially marked for frontal regions (8, 9, 11, 19). Although allometric scaling among brain structures is an interesting issue in its own right and probably reflects the way that functional properties such as connectivity and neural processing speed are conserved across body and brain sizes (17, 18), it would be illogical to conclude from the observation that human values are closely predicted from a general scaling relationship for other species that humans are specialized for frontal lobe functions in particular (as opposed to functions mediated by more distributed networks connecting different brain regions).

In addition to this theoretical point, unscaled measures of frontal cortex size give problematic empirical results when the comparative net is cast widely. For example, frontal gray matter volume is smaller as a proportion of total cortical volume in humans than in several nonhuman prosimian and anthropoid primates and two mustelid carnivores, and values for several prosimians exceed those for chimpanzees (Pan troglodytes), spider monkeys (Ateles), and baboons (Papio hamadryas) (calculated from table 2 in ref. 11). In terms of absolute size, human frontal cortex exceeds that in other measured species; but sea lions (Zalophus californicus)—which have not been noted for intelligence—exceed several anthropoids, including baboons (P. hamadryas) and gibbons (Hylobates lar), whereas llamas (Llama glama) exceed macaque monkeys (Macaca mulatta) (11). Hence, unless one is willing to take seriously the hypothesis that lemurs have more of the qualities bestowed by frontal cortices than do humans, or that llamas possess more than monkeys, it must be concluded that testing the hypothesis that any species is specialized for frontal cortex functions, as opposed to functions mediated by more extended networks, requires scaling to be taken into account. Here we test the hypothesis that human frontal cortex and prefrontal cortex (PFC) size is larger than expected from scaling against other brain structures, an empirical issue that is still contested (4, 6, 8–10).

There has also been variability in whether and how comparative analyses take account of phylogeny. Some studies have treated species values (or even multiple intraspecific values) as independent data points (3, 5–8, 10). It is now well established that species values cannot be assumed to be independent, because species share similarities due to common descent as well as to independent evolution. Such phylogenetic nonindependence can substantially bias estimates of scaling parameters and the measures of relative size based on them (20, 21). This problem affects both the prediction of human from nonhuman values and claims of different scaling relationships in different taxa (8, 10, 11). Two notable exceptions are a study of frontal cortex scaling in mammals using the method of phylogenetically independent contrasts (11), and recent studies by Smaers and colleagues using a more general phylogenetic method than independent contrasts, phylogenetic generalized least squares (PGLS) (e.g., refs. 9 and 19). Here we apply PGLS to available comparative data sets to determine whether there is any clear and consistent pattern of human values for frontal cortex size falling beyond the expected range for a primate of our neocortex or brain size.

In summary, the literature on the evolution of human frontal lobe size presents a confused picture in which different studies using different measurements and different analytical methods make starkly diverging claims. We seek to clarify this confused picture by bringing together all suitable data sets and analyzing them within a common, phylogenetic framework. Although these analyses inevitably involve replication of individual studies, this is necessary to determine the overall pattern of results when a common analytical framework is applied to them all, and to test alternative explanations for the patterns previously reported. We also present analyses to determine whether there were statistically significant shifts in the rate of size change along the phylogenetic branch leading to humans. The issue of which of the various anatomical demarcation criteria used in previous studies are most appropriate is important and is addressed in Discussion; however, our main concern is to determine which of the claims based on these data are empirically supportable and whether any consistent pattern across data sets emerges.

Results

Phylogenetically determined (PGLS) coefficients, r² values, and estimates of the phylogenetic signal in the data (λ values) for the relationship between frontal and nonfrontal structures in nonhuman species are presented in Table S1. In Fig. 1 the phylogenetic regression lines for nonhuman species are mapped back onto species values, with humans and 95% confidence intervals included. In all four data sets in Fig. 1, the size of frontal cortices in humans is within the 95% confidence limits for the size predicted from the size of other brain regions. This applies to both the whole human frontal cortex (Fig. 1 A and B) and to PFC (Fig. 1 C and D).

Fig. 1.

PGLS analysis of four comparative data sets on frontal cortex size relative to size of other brain structures. In each case the slope of the regression line was determined using PGLS models (Methods) on the nonhuman species. In each data set, humans (white circles) fall within the 95% confidence limits (dotted lines), indicating that they conform to the allometric relationships for nonhuman species. Great apes are shown as gray circles. Data from (A) Bush and Allman (11), (B) Semendeferi et al. (5), (C) Schoeneman et al. (6), and (D) Semendeferi et al. (7).

Two studies suggest that it is specifically prefrontal white matter that expanded disproportionately during human evolution (6, 9). In the first, a nonphylogenetic regression of prefrontal white matter volume indicated human values greater than predicted from nonprefrontal white matter volume (figure 4c in ref. 6). Phylogenetic reanalysis using PGLS initially seems to provide some support for this conclusion (although note that the mean human value is actually just within the 95% confidence limits; Fig. 2A). However, the trend seems to be due not to larger than expected volume of prefrontal white volume but to smaller than expected values of nonprefrontal white matter volume in this particular data set (Fig. 2 B and C). First, the human value for prefrontal white vs. gray matter is well within the 95% confidence intervals (Fig. 2B; see also ref. 10). Second, human nonprefrontal white matter volume relative to nonprefrontal gray volume is smaller than expected (Fig. 2C), explaining the trend in figure 4c in ref. 6 (data replicated here in Fig. 2A). Although it may seem tempting to seek an explanation for this smaller than expected nonprefrontal white matter volume, this should await confirmation in independent data sets, especially because this negative deviation is (like the positive deviation for frontal white on nonfrontal white matter) just within 95% confidence intervals. In the second study (9), most attention was paid to scaling trends, but analysis of PGLS residuals also indicated that humans deviated from allometry in one specific respect: prefrontal white matter volumes were observed to be greater than expected relative to prefrontal gray matter volume in the left (but not right) hemisphere (p. 71). This result was based on a phylogenetic analysis (using similar methods as here). Once again this analysis does not distinguish between larger than expected PFC white matter and smaller than expected PFC gray matter. Most importantly, however, both ref. 9 and our own analysis (Fig. 2D) indicate that there is no significant deviation of the human mean value for left prefrontal white matter volume relative to nonfrontal brain volume. Moreover, the different patterns of size difference in human prefrontal white matter volume in these two studies (6, 9) suggests that claims of deviation in a single species based on analyses of single data sets should be treated with great caution.

Fig. 2.

PGLS analysis of frontal white matter volume. Data from (A–C) Schoeneman et al. (6) and (D) Smaers et al. (9). Human values are plotted relative to the PGLS regressions for nonhuman primates, with 95% confidence intervals (dotted lines). In line with Schoeneman et al. (6), the human value for prefrontal white matter volume is greater than predicted by white matter volume in the rest of the brain. B and C show that this is because white matter volume in the rest of the brain is smaller than expected. (D) Data from ref. 9, indicating that left hemisphere white matter prefrontal volume is not large relative to the rest of the brain.

Two studies have claimed that apes, including humans, differ significantly from non-ape species (5, 9). In the first study, this claim was based on the fact that the percentage of brain volume made up by frontal cortex is larger in great apes than in lesser apes and monkeys (although not differentiating between humans and other great apes) (5). As explained above, percentages conflate scaling and relative size, and we observe no trend for great apes to deviate from phylogenetically determined scaling relationships (Figs. 1 and 2). In the second study, white matter volume relative to gray matter volume in the left (but not right) PFC was found to be significantly greater in apes than in monkeys (9). In contrast to the findings of ref. 5, the pattern applied to all apes, including lesser as well as great apes (9). However, using the same data, a t test on the residuals of the PGLS regression of log-transformed white on gray matter volume in the left PFC is not significant (t_2,16 = 0.76, P = 0.45). Although the mean ape residual is higher than that for monkeys, the two sets of residuals overlap, and the largest residual is for a New World Monkey (Pithecia monachus, residual = 0.23, mean ape residual = 0.04). These recent claims of distinct trends in apes (including humans) are therefore neither consistent between studies nor individually reliable.

For completeness, we also analyzed a data set available on limbic frontal cortex [Brodmann area (BA) 13], although this area has “few ‘higher-order’ functional attributes ascribed to it,” and sample size is small (22). The analysis again shows no evidence of relative enlargement in humans (Fig. S1).

Finally, we apply a recently developed Bayesian model of trait evolution that allows rates of change to vary in individual branches or entire monophyletic subclades of a phylogenetic tree (23) to two datasets in which sample sizes were sufficiently large. Over the course of billions of iterations this model allows us to estimate the mean rate of evolution along each branch of a phylogenetic tree (Methods). Fig. 3 shows the distribution of mean rates (x-fold increase in rate) for each branch of the phylogeny derived from applying the varying rates model to the first data set. As expected on the basis of overall brain size, the lineage leading to humans has the highest rate of change in log-transformed absolute prefrontal gray or white matter volume (Fig. 3 A and B, respectively). This is not the case, however, when we examine relative prefrontal white and gray matter volumes (residuals from the phylogenetic regression with nonprefrontal white and gray matter, respectively): here the rate along the human lineage is unremarkable (Fig. 3 C and D). A similar pattern is found in the second dataset (Fig. 4). Fig. 3 C and D suggest that the branches associated with the great apes are at the fast end of the distribution for evolutionary change in relative frontal region volume (shaded dark gray areas in Figs. 3 and 4). However, these fast rates are associated with an increase in variance rather than a directional increase in relative size (reflecting the fact that PGLS residuals are both positive and negative; Fig. 1).

Fig. 3.

Rates of evolutionary change in PFC size. The distributions of mean rates for each branch of the phylogeny are shown, derived from applying the varying rates model to data on prefrontal cortical volume (9). (A) Log-transformed prefrontal white matter; (B) log-transformed prefrontal gray matter; (C) relative prefrontal white matter; (D) relative prefrontal white matter. The branches associated with the great ape clade are shaded dark gray, and the rate associated with the branch leading to humans is in white. (Insets) Phylogenetic trees of the species examined, where the branches have been scaled according to the rate of evolution. The starting phylogenetic tree was ultrametric, as it was scaled to time, so any deviation from this is owing to rate variation. Branchwise rates are available from the authors upon request. Reading from top to bottom, the species at the tips are as follows: Cebus albifrons, Pithecia monachus, Alouatta seniculus, Lagothrix lagotricha, Ateles geoffroyi, Hylobates lar, Pongo pygmaeus, Gorilla gorilla, Homo sapiens, Pan troglodytes, Pan paniscus, Nasalis larvatus, Procolobus badius, P. hamadryas, Lophocebus albigena, Miopithecus talapoin, Erythrocebus patas, Cercopithecus mitis, and Cercopithecus ascanius.

Fig. 4.

Rates of evolutionary change in frontal cortex size. The distribution of mean rates for each branch of the phylogeny derived from applying the varying rates model to the Bush and Allman (11) dataset. (A) Log-transformed volume of frontal gray matter; (B) relative volume of frontal gray matter. Reading from top to bottom, the species at the tips are as follows: Otolemur crassicaudatus, Galago senegalensis, Nycticebus coucang, Perodicticus potto, Lemur catta, Eulemur mongoz, Daubentonia madagascariensis, Propithecus verreauxi, Cheirogaleus medius, Microcebus murinus, Tarsius syrichta, Saimiri sciureus, Aotus trivirgatus, Callicebus moloch, Ateles geoffroyi, Alouatta palliata, Hylobates lar, Homo sapiens, Pan troglodytes, Semnopithecus entellus, Macaca mulatta, P. hamadryas, Mandrillus sphinx, Cercocebus torquatus, Cercopithecus nictitans. Other details as for Fig. 3.

Discussion

The consistency of our results across independent data sets supports the view (e.g., refs. 10–12, 24, and 25) that human frontal cortex, and regions and tissue subtypes within it, are no larger than expected for a nonhuman primate of our overall cortex or brain size. Furthermore, our results do not support more specific claims of relative enlargement in human prefrontal white matter (6); or in frontal regions of the whole ape, or great ape clade (5, 9). Our analysis of evolutionary rates also shows that frontal cortex did not evolve especially fast relative to other brain regions after the lineages leading to humans and chimpanzees diverged. Although it may be problematic to draw firm inferences from any one study, particularly where sample sizes are limited and/or boundary delineation is contested (6, 8, 10), the overall pattern of results obtained here from a variety of data sets is clear. Thus, in attempting to define what makes humans cognitively distinct from other species, comparative data provide no clear justification for a continuing focus on functions mediated by frontal cortices in isolation.

Comparative measures of total frontal cortex volume include functionally heterogenous motor, premotor, and prefrontal cortices, each of which in turn also includes areas with distinctive connections and functional properties. In principle, therefore, the demarcation criteria used in particular studies could be critically important in determining whether nonallometric expansion occurred during human evolution. In particular, experimental and clinical evidence associates PFC with cognitive control (e.g., ref. 26). Even within comparative studies of anterior frontal regions, the absence of a clear morphological boundary (8, 10) means that different demarcation criteria have been used to measure proxies for PFC, including criteria based on both anatomical landmarks (7, 8) and on cumulative sampling along the anterior–posterior axis (9). These different methods will affect the functional correlates of the measurements used. Despite this heterogeneity of the comparative data, however, our results are remarkably uniform; irrespective of differences in methods and demarcation criteria used, humans do not deviate from allometric expectations.

An additional complication in interpreting allometric patterns, and deviations from them, is a recent report of sex-by-hemisphere interactions in the scaling of PFC (19). In this study, some indication that human values deviated from the sex- and hemisphere-specific allometric trends was observed, but “results (were) not fully conclusive” (19, p. 210). Such sex-by-hemisphere interactions and their possible implications for human brain evolution may be an interesting avenue for further inquiry.

It has been argued that, irrespective of whether human values deviate from interspecific allometry, the fact that the steep (>1) slope of the allometric regression itself, which leads to large-bodied and large-brained species having a greater absolute and proportional volume of frontal cortices, is sufficient justification for concluding that human frontal regions are disproportionately large (4, 24, 27). This argument is difficult to sustain on both theoretical and empirical grounds. First, it is known that the size of individual brain structures, including cortical regions, can and do deviate substantially from allometry, and that such deviations correlate with species-specific ecological niche and behavior (15, 28). So far no argument has been presented to explain why frontal cortices should be treated as a special case in this respect. Second, as noted in the Introduction, absolute or proportional volumes create anomalous species rankings in at least some comparative data. Third, scaling relationships between volume and variables more directly related to information processing, such as neuron numbers, vary between brain structures (16, 29), so that the volumetric proportions of brain structures may not reflect their relative computational capacity. Although human area 10 occupies 1.2% of brain volume in humans, it contributes only 0.29% of its neurons (computed from refs. 7 and 16). It is therefore clearly unsafe to derive functional inferences directly from volumetric proportions or other improperly scaled measures.

Our results do not demonstrate that specialization of frontal lobe regions played no part in human cognitive specialization. They leave open three possibilities. First, it is possible that a hitherto unmeasured frontal region or microanatomical feature will provide evidence of nonallometric evolutionary change. However, relative expansion of any specific regions or tissue type within the frontal lobes should in principle impact on their total size, yet we do not observe this. Theoretically, increased relative size in specific regions could be masked by corresponding reduction in other regions. However, no clear evidence currently exists for this idea, nor any well-articulated theory as to why such contraction would have been likely. A range of possible alternative adaptive specializations in human frontal cortex other than relative size has also been considered, including neurotransmitter innervation (30), neuron density, dendritic structure, and number and width of minicolumns (31). For example, gray matter density is lower and width of minicolumns greater in human BA10 than in other ape species (31). However, these data are consistent with known allometric effects. Cortical neuron density correlates negatively with brain size across mammals (16, 18, 32), and reflecting this general relationship gray matter density is lower and minicolumn spacing greater in human compared to nonhuman primary somatosensory and primary visual cortices, as well as in BA10 (31). Although the differences are most marked for BA10 (31), the explanation is still likely to be allometric scaling, because frontal regions expand faster with brain size than do nonfrontal regions (Table S1) (9, 11, 19). Indeed, using the data in reference 31, PGLS analysis reveals, even in this small comparative sample of six species, a significant positive correlation between brain size and horizontal spacing distance (t_2,6 = 4.39, P = 0.01) and a negative correlation between brain size and gray matter density (t_2,6 = −3.90, P = 0.02) in BA10. This means that the large difference between humans and nonhuman species in BA10-to-brain volume proportion does not equate to a similar difference in BA10-to-brain proportion in numbers of neurons, as noted above. The idea that there are nonallometric differences in number of synaptic connections (31) so far lacks direct evidence. In general, therefore, allometric scaling, as determined by appropriate phylogenetic analysis, needs to be given greater consideration before inferring adaptive specialization from comparative data.

The second possibility is that a nonfrontal region (or regions) expanded disproportionately during human evolution. Some comparative evidence points to relative expansion of temporal lobe structures, including temporal cortices and amygdala, after accounting for differences in brain size (25, 33–36), but broader comparative studies are needed to verify this intriguing suggestion. A difficulty is that, if there were other extensive areas that had expanded disproportionately (relative to the rest of the brain) and other things being equal, frontal regions should appear relatively small in humans, which they do not.

The third possibility invokes an alternative route to cognitive specialization; coordinated expansion of functionally and anatomically connected areas, potentially including both cortical and noncortical regions (15, 37). Neocortex, cerebellum, and intermediate nuclei, for example, show closely correlated evolution in terms of both volume and neuron numbers, after controlling for variability in the size or neuron numbers of other brain regions (15, 32, 37, 38), suggesting that selective expansion of cortico-cerebellar systems was a general feature of primate brain evolution. Frontal and more posterior cortical regions both exhibit this pattern, though they have different correlations with specific regions of the cerebellum and basal ganglia (38, 39). Thus, the evolution of frontal regions such as PFC may be best understood in terms of their participation in more distributed networks. Experimental evidence now implicates such distributed networks in uniquely human cognitive capacities. Language, for example, involves distributed networks within and beyond the cortex (40) including the cerebellum (32). We suggest that natural selection selectively enlarged such distributed networks and that these—rather than more localized size change of frontal cortical regions—are likely to form the basis of human cognitive specialization.

Methods

Volumetric data were collated from five recent comparative studies of total (5–7, 9, 11). A sixth data set, on limbic frontal cortex (22), was included at the suggestion of an anonymous reviewer, although this region is not usually associated specifically with human cognitive specializations (22). Results of analysis of data from a seventh study (12) are not included here because, although subdividing frontal cortex into three regions, it included a narrower range of species than in an earlier study by the same team (5), and incorporating these results does not change any of our conclusions. Two older studies (1, 2) were not included because these have been questioned owing to small sample sizes, use of surface area measurements rather than volumes, and uncertainty about the accuracy of morphometric tools (5). For example, ref. 2 includes data for only four nonhuman species, collected by several laboratories over a period between the 1930s and 1960s. However, these data, their limitations, and implications for our conclusions (or lack thereof) are considered in more detail in SI Discussion, with supporting information in Fig. S2 and Table S2. PGLS (41–43) implemented by the R-package CAICR (R v.2.11.1; The R Foundation for Statistical Computing, www.r-project.org/foundation/) was used to determine the allometric coefficients relating frontal areas to size of the rest of the brain among nonhuman species. In PGLS analysis, regression parameters are found by maximum likelihood (ML) and “weighted” by the variance–covariance matrix that represents the phylogenetic relationships among the species. In each regression the phylogenetic signal is estimated as the value of λ of the residuals, varying between 0 (where the data have no phylogenetic structure) and 1 [where the best fit to the data is provided by a “Brownian Motion” model of trait evolution (43), with variation at the tips proportional to the duration of common evolution (41, 42)]. The estimated ML value of λ is incorporated as a parameter in the model, thus controlling for phylogenetic dependence in the data. The phylogeny (including branch lengths) for the species in our dataset was extracted from a published supertree of mammals (44, 45), pruned to remove species not in the data stets, and is displayed in Fig. S3. The coefficients (Table S1) were then mapped back onto the species data to determine the extent and consistency of any deviation of the mean human values from the allometric expectation.

Unless correlations are high (r² > 0.9), line-fitting methods can influence coefficient estimates. In particular, least-squares regression tends to give lower slopes than does the reduced major axis (20). However, all adjusted r² values in the present study were greater than 0.9. Furthermore, increasing the slope coefficients would make it less rather than more likely that human values showed a positive residual. Hence the conclusions are robust to use of alternative line-fitting methods.

Our variable rates model applies a PGLS model of trait evolution implemented in a Bayesian framework to analyze the evolution of the continuously varying traits on a phylogenetic tree (23). This model estimates the posterior densities of the instantaneous variance of change along the branches of a phylogeny (the “rate” parameter). The homogeneous GLS model is modified using two scaling mechanisms that are designed to enhance the likelihood of detecting both individual and generalized rate shifts throughout the tree while protecting against adding random noise to the model. One scaling mechanisms modifies the instantaneous rate along a single branch, whereas the other modifies the rate in a whole monophyletic group or clade. Over billions of iterations this model records for each branch in the tree the probability that its rate has been changed and what its mean rate is (23). Further details are provided in SI Methods.