evolution of increased glia–neuron ratios in the human frontal...

Evolution of increased glia–neuron ratios in thehuman frontal cortexChet C. Sherwooda,b, Cheryl D. Stimpsona, Mary Ann Raghantic, Derek E. Wildmand,e,f, Monica Uddind,Lawrence I. Grossmand, Morris Goodmanb,d,g, John C. Redmondh, Christopher J. Bonari, Joseph M. Erwinj,and Patrick R. Hofk,l

aDepartment of Anthropology, The George Washington University, Washington, DC 20052; cDepartment of Anthropology and School of BiomedicalSciences, Kent State University, Kent, OH 44242; dCenter for Molecular Medicine and Genetics and Departments of gAnatomy and Cell Biology andfObstetrics and Gynecology, Wayne State University School of Medicine, Detroit, MI 48201; ePerinatology Research Branch, National Institute ofChild Health and Human Development, National Institutes of Health, Bethesda, MD 20892; hDivision of Neuroscience and Center for BehavioralNeuroscience, Yerkes National Primate Research Center, Emory University, Atlanta, GA 30329; iCleveland Metroparks Zoo, Cleveland, OH 44109;jFoundation for Comparative and Conservation Biology, Needmore, PA 17238; kDepartment of Neuroscience, Mount Sinai School of Medicine,New York, NY 10029; and lNew York Consortium in Evolutionary Primatology, New York, NY

Contributed by Morris Goodman, July 14, 2006

Evidence from comparative studies of gene expression and evo-lution suggest that human neocortical neurons may be character-ized by unusually high levels of energy metabolism. The currentstudy examined whether there is a disproportionate increase inglial cell density in the human frontal cortex in comparison withother anthropoid primate species (New World monkeys, Old Worldmonkeys, and hominoids) to support greater metabolic demands.Among 18 species of anthropoids, humans displayed the greatestdeparture from allometric scaling expectations for the density ofglia relative to neurons in layer II�III of dorsolateral prefrontalcortex (area 9L). However, the human glia–neuron ratio in thisprefrontal region did not differ significantly from allometric pre-dictions based on brain size. Further analyses of glia–neuron ratiosacross frontal areas 4, 9L, 32, and 44 in a sample of humans,chimpanzees, and macaque monkeys showed that regions in-volved in specialized human cognitive functions, such as ‘‘theory ofmind’’ (area 32) and language (area 44) have not evolved differ-entially higher requirements for metabolic support. Taken to-gether, these findings suggest that greater metabolic consumptionof human neocortical neurons relates to the energetic costs ofmaintaining expansive dendritic arbors and long-range projectingaxons in the context of an enlarged brain.

allometry � human evolution � prefrontal cortex � brain energymetabolism � language evolution

Humans are distinguished from other primates by a dramat-ically enlarged neocortex and the elaboration of cognitive

capacities that have culminated in the evolution of language,technological innovation, and complex social behavior. Expan-sion of the human brain entails high metabolic costs (1).Although the human brain comprises only �2% of body mass,it captures �20% of the body’s total glucose utilization (2). Atthe same time, because the metabolic rate per gram of neuraltissue generally decreases with larger brain size, the human brainis more energetically efficient than that in smaller-brainedprimate species (3). Despite this evidence for relatively lowermass-specific brain metabolism in humans, recent microarraystudies have shown that genes involved in neuronal signaling andenergy production are up-regulated in the human neocortexcompared with chimpanzees and other great apes (4, 5). Fur-thermore, evidence for positive selection in the human lineagefor genes that encode components of the mitochondrial electron-transport chain suggests that there has been evolutionary pres-sure for high rates of aerobic energy consumption in metabol-ically active cells, such as neurons (6). Taken collectively, thesefindings suggest that neuronal activity level and energy expen-diture per neuron have become enhanced in human evolution,even as mass-specific rates of brain metabolism declined. Thispattern is consistent with a model that predicts that, with

increases in brain size, a progressively smaller fraction of thetotal neuron population may be concurrently active, leading tomore modularized information coding (7, 8).

An indirect measure of metabolic support supplied to neuronscan be obtained by examining the ratio of glia to neurons. Glialcells, particularly astrocytes, play a crucial role in the flux ofenergy substrates to neurons by regulating the rate of glucoseuptake and phosphorylation in response to glutamate concen-trations in the synaptic cleft (9). The other main type of glial cell,oligodendrocytes, synthesize the myelin that ensheathes axons inthe brain to facilitate long-range propagation of action poten-tials. The proliferation of both these glial cell types is responsiveto trophic cues associated with neuronal activity (10–12). Thus,the local density of glia in the normal brain provides anindication of the metabolic demand of neighboring neurons.Indeed, because adult neocortical neuron numbers are attainedat around the time of birth in humans, the tremendous growthof the neocortex during postnatal development is due to elab-oration of dendritic arbors and a 4-fold increase in glial cellnumbers (13, 14). Several experimental studies in rodents,furthermore, have demonstrated that the glia–neuron ratioincreases with greater environmental enrichment and stimula-tion (15, 16).

Here, we investigate whether glial cell densities are relativelyincreased in human frontal cortex compared with other anthro-poid primates (i.e., New World monkeys, Old World monkeys,and hominoids). Because the human brain is approximatelythree times larger than that of great apes, the increased meta-bolic demand of human neocortical neurons might simply be theexpected allometric consequence of cell physiology associatedwith larger overall brain size. Alternatively, the energetic con-sumption of neurons in the human neocortex might exceedallometric predictions and represent a unique trait that hascontributed to the evolution of human intelligence. To examinethese possibilities, we tested the hypothesis that humans havedisproportionately more glial cells than would be expected basedon allometric scaling within a region of the dorsolateral pre-frontal cortex (area 9L) that is involved in ‘‘working memory.’’

In addition, because the neocortex is functionally and archi-tecturally heterogeneous, particular regions that subserve hu-man cognitive specializations might display differential energeticdemands relative to other primates. Arguably, the most distinc-

Conflict of interest statement: No conflicts declared.

Abbreviations: CI, confidence interval; LS, least-squares; PI, prediction interval; RMA,reduced major axis.

bTo whom correspondence may be addressed at: Department of Anthropology, TheGeorge Washington University, 2110 G Street NW, Washington, DC 20052. E-mail:[email protected] or [email protected].

© 2006 by The National Academy of Sciences of the USA

13606–13611 � PNAS � September 12, 2006 � vol. 103 � no. 37 www.pnas.org�cgi�doi�10.1073�pnas.0605843103

tive cognitive capacities expressed by humans are the ability touse symbolic language in a recursive manner (17) and to makeinferences regarding the mental states of others, i.e., ‘‘theory ofmind’’ (18). Area 44 in the inferior frontal cortex is consideredpart of Broca’s area for language production (19) and theanterior paracingulate cortex (area 32) has been shown to beselectively activated in tasks that use theory of mind (20, 21).Therefore, we hypothesized that humans would differ fromchimpanzees (Pan troglodytes) and macaque monkeys (Macacamaura) in displaying relatively higher glia–neuron ratios withinanterior paracingulate cortex (area 32) and Broca’s area (area44) versus primary motor cortex (area 4) and dorsolateralprefrontal cortex (area 9L).

ResultsThe results of stereologic estimates of cell densities are shown inTable 3, which is published as supporting information on thePNAS web site. The glia–neuron ratios derived from these cellcounts are presented in Table 1. We did not perform completeanalyses of neuron and glial density scaling directly against brainweight because of possible artifacts related to tissue shrinkage ofindividual specimens in this sample; however, plots and regres-sion exponents and a discussion of these relationships areavailable in Fig. 3, which is published as supporting informationon the PNAS web site.

Our first analyses concerned the allometric scaling of glial celldensity against neuron density in dorsolateral prefrontal cortex(area 9L) across anthropoid species (Fig. 1A). Glial cell densityscaled against neuron density with a significant positive allo-metric exponent, as revealed by analysis of both species meandata [reduced major axis (RMA) slope � 1.38, lower confidenceinterval (CI) � 0.88, upper CI � 2.15; r � 0.499, P � 0.035] andindependent contrasts (RMA slope � 1.25, r � 0.756, P � 0.001).Although human glial density was contained within the 95%prediction intervals (PIs) of the least-squares (LS) regressionline of nonhuman species data (observed log glial density � 5.19;predicted � 4.94; upper PI � 5.27, lower PI � 4.60), it was thegreatest positive outlier (studentized deleted residual � 1.85).Based on the nonhuman species mean LS regression, humansdisplayed a 46% greater density of glial cells per neuron thanexpected. Glial densities were predicted based on neuron densityfor each individual human in our sample (n � 6) by using thenonhuman LS regression function. As a group, the observed

human glial densities were significantly greater than their pairedpredictions [paired samples t test: t � 10.06, degrees of freedom(df) � 5, P � 0.001]. Next, we used independent contrasts datato generate a LS regression line to predict glial density based onneuron density for a hypothetical species attached to the branchleading to humans in the phylogenetic tree. From this prediction,glial density in humans fell within the 95% PIs (observed log glialdensity � 5.19; predicted � 5.02; upper PI � 5.40, lower PI �4.63) and represented 32% more glia than expected.

Across anthropoid primates, the RMA fit of area 9L glia–neuron ratio against brain weight (Fig. 1B) yielded a significantnegative allometric relationship based on both species meandata (RMA slope � 0.26, lower CI � 0.20, upper CI � 0.35,r � 0.840, P � 0.001) and independent contrasts (RMAslope � 0.31, r � 0.649, P � 0.004). When the nonhumanspecies data were used to generate a LS regression prediction,the human value was only 5% less than expected (observed logglia–neuron ratio � 0.22; predicted � 0.23; upper PI � 0.44,lower PI � 0.03). The paired comparisons between individualobserved human glia–neuron ratios and predictions based onthe nonhuman LS regression line did not reveal a significantdifference (paired-samples t test: t � �1.10, df � 5, P � 0.321).The independent contrasts prediction for glia–neuron ratio inhumans based on nonhuman data showed that humans haveonly a 5% greater glia–neuron ratio than expected when takingphylogeny into account (observed log glia–neuron ratio �0.22; predicted � 0.20; upper PI � 0.53, lower PI � 0.12).

Our next analyses investigated whether human glia–neuronratios in the frontal cortex display a pattern of regional variationthat differs from other primates. Individual data on glia–neuronratios in layer II�III of areas 4, 9L, 32, and 44 in humans,chimpanzees, and macaque monkeys (Table 2) were analyzed byusing repeated-measures ANOVA (Fig. 1C). The ANOVArevealed a significant main effect of species (F2,15 � 101.80, P �0.001) and cortical area (F3,45 � 22.46, P � 0.001) but nointeraction effect (F6,45 � 1.19, P � 0.331). Bonferroni post hoccomparisons showed that areas 4 and 32 differed significantly inglia–neuron ratio from one another in macaques and chimpan-zees (all comparisons, P � 0.05). In humans, glia–neuron ratiosin area 4 differed significantly from all other frontal areas (allcomparisons, P � 0.05). Post hoc tests of homologous corticalareas among species indicated that chimpanzees and macaquesdid not differ significantly in glia–neuron ratios from each other.In contrast, humans had significantly greater glia–neuron ratiosthan macaques for every cortical area (all comparisons, P � 0.05)and greater glia–neuron ratios than chimpanzees in area 4 (P �0.05) and area 32 (P � 0.05) but not area 9L (P � 0.104) or area44 (P � 1.000).

DiscussionThe human frontal cortex displays a higher ratio of glia toneurons than in other anthropoid primates. However, this rel-ative increase in glia conforms to allometric scaling expectations,when taking into consideration the dramatic enlargement of thehuman brain. We suggest that relatively greater numbers of gliain the human neocortex relate to the energetic costs of main-taining larger dendritic arbors and long-range projecting axonsin the context of a large brain.

Brain Size Determines Interspecific Differences in Glia–Neuron Ratio.Previous studies of glia–neuron ratios in the neocortex alsoconcluded that brain size is the main determinant of this variableacross mammals (22–24). However, these data are problematicfor interpreting human brain evolution in a number of ways.First, they included relatively few species drawn from verydiverse phylogenetic groups, and they represented a small num-ber of cortical areas. Second, because scaling relationships werecalculated based on contemporary species data, it was not clear

Table 1. Brain weights and glia–neuron ratios for layer II�III ofprefrontal area 9L (species mean)

Species n Brain weight, g Glia–neuron ratio

Homo sapiens 6 1,373.3 1.65Pan troglodytes 6 336.2 1.20Gorilla gorilla 2 509.2 1.21Pongo pygmaeus 2 342.7 0.98Hylobates muelleri 1 101.8 1.22Papio anubis 2 155.8 0.97Mandrillus sphinx 1 159.2 1.02Macaca maura 6 92.6 0.84Erythrocebus patas 2 102.3 1.09Cercopithecus kandti 1 71.6 1.15Colobus angolensis 1 74.4 1.20Trachypithecus francoisi 1 91.2 1.14Alouatta caraya 1 55.8 1.12Saimiri boliviensis 1 24.1 0.51Aotus trivirgatus 1 13.2 0.63Saguinus oedipus 1 10.0 0.46Leontopithecus rosalia 2 12.2 0.60Pithecia pithecia 1 30.0 0.64

Sherwood et al. PNAS � September 12, 2006 � vol. 103 � no. 37 � 13607

AN

THRO

POLO

GY

whether these associations would be maintained after correctingfor statistical nonindependence due to phylogenetic relatedness.In contrast, the current study was designed specifically to eval-uate human glial densities in reference to the allometric scalingpatterns in close phylogenetic relatives, other anthropoid pri-mates. It is noteworthy that our analyses of species data andindependent contrasts yielded similar allometric exponents:Independent contrast slopes were always contained within the95% CIs of slopes calculated from species data. This findingindicates that the scaling relationships observed in contempo-rary taxa represent changes that have occurred repeatedlythroughout multiple nodes in the phylogenetic tree and, hence,represent robust functional regularities in neocortical design.

Our results showed that area 9L of the human prefrontalcortex has higher glial density than expected for neuron densitybased on allometric scaling in nonhuman anthropoids, a findingthat is consistent when both species mean data (46% more gliaper neuron) and independent contrasts (32% more glia perneuron) are used. Additional analyses indicated that increaseddensities of glia relative to neurons also characterize other areasof the human frontal cortex compared with chimpanzees (areas4 and 32) and macaque monkeys (all cortical areas examined).What might explain this relative increase in the numbers of gliain the human frontal cortex? Although only a small proportionof the variance in glial density is explained by neuron density(r2 � 0.25, P � 0.035), the glia–neuron ratio is strongly associatedwith brain weight (r2 � 0.71, P � 0.001). Indeed, humanglia–neuron ratios did not significantly differ from nonhumananthropoid allometric predictions based on brain weight. Mam-mals with even larger absolute neocortex size than humans, suchas whales, are reported to have correspondingly higher glia–neuron ratios (23, 24). Taken together, these results suggest thatoverall brain size comprises an important factor that governs theratio of glia to neurons.

Metabolic Correlates of Increased Glia–Neuron Ratio. The increasedratio of glia to neurons in the human frontal cortex might beexplained by energy expenditures that are correlated with in-creasing neocortex size, which, in humans, accounts for 44% ofthe brain’s total energy consumption (8). The two highest costsassociated with neuronal activity come from excitatory postsyn-aptic potentials in the dendritic tree and propagation of actionpotentials along the axon (8). Thus, the energy required perneuron to sustain the Na��K� pumps that restore ion gradientsto generate electrical potentials is expected to increase inneurons with longer dendrites and axons. Comparative studies ofpyramidal neurons in granular prefrontal cortex have shown thathuman neurons display much larger dendritic arbors and morespines than those in primates with smaller brains (althoughhominoids have not yet been studied) (25). Axon diameter and

Log neuron density in area 9L

Log

glia

l den

sity

in a

rea

9L

4.7 4.8 4.9 5.0 5.1 5.24.6

4.7

4.8

4.9

5.0

5.1

5.2

5.3New World monkeysOld Word monkeys

Gorilla gorilla

Homo sapiens

Hylobates muelleriPongo pygmaeus

Pan troglodytes

Area 4 Area 9L Area 32 Area 44

0.4

0.8

1.2

1.6

2.0

2.4

Glia

-neu

ron

ratio

Macaca mauraPan troglodytesHomo sapiens

Log brain weight

Log

glia

-neu

ron

ratio

in a

rea

9L

0.8 1.2 1.6 2.0 2.4 2.8 3.2 3.6-0.4

-0.3

-0.2

-0.1

0.0

0.1

0.2

0.3

A

B

C

Fig. 1. Glial cell distributions in anthropoid primates. (A) The allometric scalingof species mean glial cell density (cells per cubed millimeter) against neurondensity (cellspercubedmillimeter) in layer II�III ofarea9L inanthropoidprimates.A solid line represents the LS regression line that is fit to the nonhuman anthro-poids based on species mean data (y � 0.68x � 1.54; r � 0.534; P � 0.027). Adashed line represents the LS regression line that is fit to independent contrastsmapped back into ‘‘tip’’ species space, calculated to predict a hypothetical speciesattached to the branch leading to humans by pruning humans from the tree andrerooting it. It should be noted that absolute values for cell densities may showsome error from fixation artifact; however, because each individual data point isaffected by such error equally in both x and y, the slope and residuals are reliable.(B) The allometric scaling of the glia–neuron ratio against brain weight in layerII�III of area 9L in anthropoid primates. A solid line represents the LS regressionline that is fit to the nonhuman species mean data (y � 0.23x � 0.47; r � 0.804;

Table 2. Glia–neuron ratios in layer II�III of different areas infrontal cortex

Species n Area 4 Area 9L Area 32 Area 44

Homo sapiens 6 2.19 (0.06) 1.65 (0.09) 1.63 (0.09) 1.55 (0.18)Pan troglodytes 6 1.54 (0.05) 1.20 (0.06) 0.91 (0.09) 1.24 (0.13)Macaca maura 6 1.29 (0.09) 0.84 (0.03) 0.68 (0.05) 0.95 (0.06)

Data are presented as mean (standard error).

P � 0.001). A dashed line represents the LS regression line that is fit toindependent contrasts mapped back into tip species space with humanspruned from the tree. (C) Glia–neuron ratio across frontal areas in humans,chimpanzees, and macaque monkeys (n � 6 for each species), showing meansand 95% CIs.

13608 � www.pnas.org�cgi�doi�10.1073�pnas.0605843103 Sherwood et al.

length also increase in larger brains (26), contributing to thehigher cost of neuronal activity. In this context, it is interestingthat glia densities have been shown to increase along differentsegmental levels of Clarke’s nucleus surrounding neurons thathave progressively longer axon projections in the spinocerebellartract (22).

Considering the relatively greater surface area of neuronalsoma, dendrites, and axons that accompany brain enlargement,it has been estimated that each human neocortical neuronconsumes 3.3 times more ATP to fire a single spike than in rats,and 2.6 times more energy to maintain resting potentials (8).Accordingly, our results suggest that relatively more glia mightproliferate to provide metabolic support to neurons that areincreasingly energetically expensive in larger neocortices, such asin humans. In fact, certain populations of neurons that providelong corticocortical association projections linking the prefron-tal cortex, temporal cortex, and parietal cortex in primates showmetabolically costly characteristics, such as highly myelinated,large axons and fast firing properties (27–30). In humans,neurons in the prefrontal cortex with specialization such as theseexhibit enhanced susceptibility to neurodegeneration in thecourse of disorders like Alzheimer’s disease (31–34). Theseneurons may also be preferentially vulnerable to alterations ofenergy metabolism that occur in many age-related degenerativebrain illnesses. It is, furthermore, conceivable that increasednumber of glia in humans may increase susceptibility to certaindiseases that involve these cell types, such as schizophrenia (35).

The Correlation Between Glia–Neuron Ratio and Brain Size Corre-sponds to Gene Evolution. Relatively higher glia–neuron ratios inthe human neocortex are most likely an epigenetic consequenceof the elevated metabolic costs of maintaining membrane po-tentials in neurons that are situated within an enlarged brain.Hence, whenever significant evolutionary changes in brain sizeoccur, there might be corresponding modifications to the pro-cesses underlying energy metabolism. Indeed, considerable ev-idence indicates that the biochemical mechanism of generatingenergy has been under evolutionary selection going back at leastto the anthropoid primate stem, presumably to fuel increases inbrain size. First, an evolutionary switch in the lactate dehydro-genase enzyme occurred among haplorhines from expression ofan isozyme that predominantly supports anaerobic metabolismto one that predominantly supports oxidative metabolism (36).Lactate generated in glia, particularly astrocytes, can serve as asubstrate for the oxidative metabolism of adjacent neurons (37).Second, the process of mitochondrial electron transport showsaccelerated evolution of multiple subunits of complex III, com-plex IV, and cytochrome c (6, 38), suggesting that the engine ofenergy generation has been a prominent target of Darwinianpositive selection in anthropoid primates.

Comparative microarray experiments have also revealed up-regulation of genes in the human neocortex related to energymetabolism (4, 5). For example, the gene encoding carbonicanhydrase 2, CA2, is up-regulated in the human cortex comparedwith chimpanzees and other primates (5). Carbonic anhydrasesare expressed in glial cells and are involved in modulating CO2and other metabolic end products to levels that may enablebuffering�secretion of lactate (37, 39). Given our current results,it is possible that part of the increase in transcript levels of CA2and other glia-expressed genes observed in the human cerebralcortex is due to relatively greater numbers of these cell types.

Regional Variation in the Glia–Neuron Ratio. Our analyses of glia–neuron ratios in functionally distinct areas of the frontal cortex(areas 4, 9L, 32, and 44) did not reveal a significant species �cortical area interaction effect in the ANOVA model, indicatingthat increased glia–neuron ratios in the human frontal cortexfollows a pattern of regional variation that resembles both

chimpanzees and macaque monkeys. This finding suggests thatregional differences in the energetics of frontal cortex have beenlargely conserved over �25 million years of catarrhine evolutionsince the last common ancestor shared by macaques and humans.These data provide further support for the idea that humancognitive and linguistic specializations have emerged by elabo-rating on higher-order executive functions of the prefrontalcortex, such as planning, working memory, attention, commu-nication, and social cognition, that evolved earlier in the primatelineage (40–42).

ConclusionIn summary, our results are consistent with the hypothesis thatinterspecific variation in the glia–neuron ratio in the neocortexof anthropoid primates is driven by the metabolic costs borneby single neurons in association with the energy required bylarger dendritic arbors and longer axons. In this regard, humanglia–neuron ratios in frontal cortex are well predicted byscaling expectations for brain size calculated from nonhumananthropoid primates. These findings might help to explain theselection for cellular mechanisms to enhance energy metab-olism that have occurred in the lineage leading to humans. Thisstudy demonstrates the potential of integrating data fromhistology to gain insight into the genetic changes underlyingbrain evolution.

Materials and MethodsSample and Histological Preparation. The left hemisphere frompostmortem brain samples representing 18 anthropoid primatespecies were used for these analyses (see Table 4, which ispublished as supporting information on the PNAS web site, fordetails regarding sample sizes, age, and sex characteristics). Themajority of specimens came from adults, with the exception oftwo juveniles that had a brain size comparable with their speciestypical adult average (Trachypithecus francoisi and Pithecia pithe-cia). None of the brain samples showed evidence of neuropatho-logical abnormality on routine examination. Many postmortembrain specimens used in this study were donated by zoological orresearch institutions after immersion fixation in 10% bufferedformalin. Some samples were obtained from animals that wereperfused transcardially with 4% paraformaldehyde in the con-text of unrelated experiments (Papio anubis, Macaca maura,Erythrocebus patas, Saimiri boliviensis, and Aotus trivirgatus).Human brain specimens from nongeriatric adults under 55 yearsof age were obtained from the Northwestern University Alzhei-mer’s Disease Center Brain Bank. These specimens were im-mersion-fixed in 10% buffered formalin. Brain weights wererecorded from each specimen and used in scaling analyses. Forall nonhuman specimens, the entire frontal lobe was sectionedin the coronal plane. Human brain samples were dissected fromthe regions of interest as 4-cm-thick slabs. Tissue blocks werecryoprotected by immersion with increasing concentrations ofsucrose solution up to 30%. Blocks were frozen on dry ice,sections were cut at 40 �m by using a sliding microtome, and a1:10 series of sections was stained for Nissl substance with asolution of 0.5% cresyl violet. Additional details on the materialsand experimental procedures can be found in Supporting Mate-rials and Methods, which is published as supporting informationon the PNAS web site.

Identification of Cortical Areas, Neurons, and Glia. Interspecificallometric scaling analyses used cell densities in layer II�III ofdorsolateral prefrontal area 9L (Fig. 4, which is published assupporting information on the PNAS web site). Although thecytoarchitecture of area 9L has not been described for everyspecies included in this study, reports from a diverse range ofanthropoids generally agree in the location and principal cyto-architectural features of area 9L: Callithrix jacchus (43), Saimiri


AN

THRO

POLO

GY

oerstedii (44), Macaca sp. (45–47), Papio hamadryas (48), Pantroglodytes (49), and Homo sapiens (45, 50). Therefore, we wereable to unambiguously identify this homologous cortical areaacross species upon histological observation, and our stereologiccounts of cell densities were confined to its borders.

We performed further analyses of glia–neuron ratios in hu-mans (n � 6), chimpanzees (n � 6), and macaque monkeys (n �6) across a larger number of frontal cortical areas, including area4 (in the region of hand representation), 9L, 32, and 44 (Fig. 5,which is published as supporting information on the PNAS website). These regions of interest were initially identified based ontopological location and confirmed by their characteristic cyto-architecture by using descriptions from previous parcellations inthese species (42, 45, 49, 51–54).

Based on Nissl-stained sections, neurons were distinguishedfrom glia and endothelial cells by the presence of dark, coarselystained Nissl substance in the cytoplasm, a large nucleus, adistinct nucleolus, and lightly stained proximal segments ofdendritic processes (Fig. 2). In contrast, glial cells lack a con-spicuous nucleolus and contain less endoplasmic reticulum.Similar to previous comparative studies of glia–neuron ratios(22, 23), we did not distinguish between astrocytes and oligo-dendrocytes, and microglia were excluded.

Stereologic Analyses. The optical disector method was used tocalculate densities of neurons and glial cells (55). Quantifica-tion of cellular densities within layer II�III was performed byusing a computerized stereology system consisting of a Zeiss(Oberkochen, Germany) Axioplan 2 photomicroscopeequipped with a Ludl (Hawthorne, NY) XY motorized stage,Heidenhain (Plymouth, MN) z-axis encoder, an Optronics(East Muskogee, OK) MicroFire color video camera, a Dell(Round Rock, TX) PC workstation, and StereoInvestigatorsoftware (MicroBrightField, Wiliston, VT). Within each cor-tical area, beginning at a random starting point, three sections

equidistantly spaced 400 �m apart were selected for analysisfor each cell type. After outlining the boundaries of layerII�III, a set of counting frames (30 � 30 �m) were placed ina systematic random fashion to cover the sampled area with�30 frames per section by using a fractionator samplingdesign. Disector analysis was performed under Koehler illu-mination using a 63� objective (Zeiss Plan-Apochromat, N.A.1.4). Each cell type was counted when its nucleolus wasencountered within the counting frame. The thickness ofdisectors was consistently set to 6 �m to allow for a minimum2-�m guard zone on either side of the section. On average, thecoefficient of error (56) of optical disector analyses was0.085 � 0.018 (mean � SD). Cellular densities (Nv) werederived from these stereologic counts and corrected for shrink-age from histological processing by the number-weighted meansection thickness as described in Sherwood et al. (57).

Data Analysis. Logarithmic (base 10)-transformed speciesmeans were used in allometric scaling analyses of glial cells inprefrontal area 9L. To determine the exponent of scalingrelationships, we used RMA line-fitting to bivariate databecause this method accounts for error in both independentand dependent variables (58). All RMA routines and tests werecalculated by using (S)MATR software version 1.0 (Falster D.S., Warton, D. I., and Wright, I. J.; www.bio.mq.edu.au�ecology�SMATR). Phylogenetic independent contrasts werealso calculated from the data to examine scaling relationshipswhile controlling for the effects of phylogenetic relatedness inthe data set (59). Standardized independent contrasts werecalculated by using the PDAP:PDTREE module (60) ofMesquite software version 1.06 (61) from log-transformeddata based on a phylogeny of primates in Goodman et al. (62).Branch lengths were transformed according to Pagel’s method(63), which assigns all branch lengths to 1, with the constraintthat tips are contemporaneous.

In addition to estimating scaling exponents, we also examinedwhether human glial cell densities represent significant devia-tions from allometric expectations based on other anthropoidprimates. We calculated LS prediction equations and PIs forhumans based on nonhumans using both species mean data andindependent contrasts. To generate phylogenetically informedpredictions based on independent contrasts, humans werepruned from the tree, the tree was rerooted so that the lastcommon ancestor of humans and chimpanzees was at the baseof the phylogeny, and then an independent contrasts regressionline and prediction intervals were computed and mapped backinto the original tip species space (60). After logarithmic de-transformation of predictions, the percentage difference be-tween observed and predicted values was calculated as the ratioof (observed � predicted)�observed.

Differences in glia–neuron ratios among macaques, chimpan-zees, and humans in different frontal cortical areas were analyzedby using a two-way ANOVA with repeated-measures design. Be-fore these analyses, data within each group were tested for nor-mality with the Shapiro–Wilk’s W test, and homogeneity of vari-ances was confirmed with Brown–Forsythe tests. Statisticalsignificance level for all tests was set at � � 0.05 (two-tailed).

We thank Drs. S. S.-H. Wang, J. K. Rilling, and M. Caceres for providinghelpful discussions and comments on an earlier version of this manu-script and E. Wahl, J. Hill, A. Zigo, A. Wolters, and K. Lever fortechnical assistance. This work was supported by National ScienceFoundation Grants BCS-0515484, BCS-0549117, and BCS-0550209; theWenner–Gren Foundation for Anthropological Research; and James S.McDonnell Foundation Grant 22002078. Brain material used in thisstudy was loaned by the Great Ape Aging Project, the Foundation forComparative and Conservation Biology, the Cleveland Metroparks Zoo,the New England Primate Research Center, and the NorthwesternUniversity Alzheimer’s Disease Center Brain Bank Grant P30 AG13854.

Fig. 2. Nissl staining of neurons and glia from different species in layer II�IIIof area 9L. Arrows indicate examples of neurons, and arrowheads indicateexamples of glia. Aotus trivirgatus (A), Saguinus oedipus (B), Alouatta caraya(C), Pithecia pithecia (D), Trachypithecus francoisi (E), Colobus angolensis (F),Cercopithecus kandti (G), Hylobates muelleri (H), and Pongo pygmaeus (I).(Scale bar, 50 �m.)

13610 � www.pnas.org�cgi�doi�10.1073�pnas.0605843103 Sherwood et al.

1. Aiello LC, Wheeler P (1995) Curr Anthropol 36:199–211.2. Sokoloff L (1960) in Handbook of Physiology, Section I, Neurophysiology, eds

Field J, Magoun HW, Hall VE (Am Physiol Soc, Washington, DC), Vol 3, pp1843–1864.

3. Mink JW, Blumenschine RJ, Adams DB (1981) Am J Physiol 241:R203–R212.4. Uddin M, Wildman DE, Liu G, Xu W, Johnson RM, Hof PR, Kapatos G,

Grossman LI, Goodman M (2004) Proc Natl Acad Sci USA 101:2957–2962.5. Caceres M, Lachuer J, Zapala MA, Redmond JC, Kudo L, Geschwind DH,

Lockhart DJ, Preuss TM, Barlow, C (2003) Proc Natl Acad Sci USA 100:13030–13035.

6. Grossman LI, Wildman DE, Schmidt TR, Goodman M (2004) Trends Genet20:578–585.

7. Attwell D, Laughlin SB (2001) J Cereb Blood Flow Metab 21:1133–1145.8. Lennie P (2003) Curr Biol 13:493–497.9. Tsacopoulos M, Magistretti PJ (1996) J Neurosci 16:877–885.

10. Hatten ME (1985) J Cell Biol 100:384–396.11. Kelic S, Levy S, Suarez C, Weinstein DE (2001) Mol Cell Neurosci 17:551–560.12. Barres BA, Raff MC (1999) J Cell Biol 147:1123–1128.13. Koenderink MJ, Uylings HB, Mrzljak L (1994) Brain Res 653:173–182.14. Larsen CC, Bonde Larsen K, Bogdanovic N, Laursen H, Graem N, Badsberg

Samuelsen G, Pakkenberg B (2006) Neuroscience 139:999–1003.15. Diamond MC, Krech D, Rosenzweig MR (1964) J Comp Neurol 123:111–120.16. Szeligo F, Leblond CP (1977) J Comp Neurol 172:247–263.17. Hauser MD, Chomsky N, Fitch WT (2002) Science 298:1569–1579.18. Tomasello M, Call J, Hare B (2003) Trends Cogn Sci 7:153–156.19. Bookheimer S (2002) Annu Rev Neurosci 25:151–88.20. Gallagher HL, Frith CD (2003) Trends Cogn Sci 7:77–83.21. Frith CD, Frith U (2006) Neuron 50:531–534.22. Friede RL, Van Houten WH (1962) Proc Natl Acad Sci USA 48:817–821.23. Hawkins A, Olszewski J (1957) Science 126:76–77.24. Tower DB, Young OM (1973) J Neurochem 20:269–278.25. Elston GN, Benavides-Piccione R, Elston A, Zietsch B, DeFelipe J, Manger P,

Casagrande V, Kaas JH (2006) Anat Rec 288:A26–A35.26. Harrison KH, Hof PR, Wang SS (2002) J Neurocytol 31:289–298.27. Hof PR, Nimchinsky EA, Morrison JH (1995) J Comp Neurol 362:109–133.28. Hof PR, Ungerleider LG, Webster MJ, Gattass R, Adams MM, Sailstad CA,

Morrison JH (1996) J Comp Neurol 376:112–127.29. Fries W, Keizer K, Kuypers HG (1985) Exp Brain Res 58:613–616.30. Raiguel SE, Lagae L, Gulyas B, Orban GA (1989) Brain Res 493:155–159.31. Hof PR, Morrison JH (2004) Trends Neurosci 27:607–613.32. Bussiere T, Giannakopoulos P, Bouras C, Perl DP, Morrison JH, Hof PR

(2003) J Comp Neurol 463:281–302.33. Hof PR, Cox K, Morrison JH (1990) J Comp Neurol 301:44–54.34. Hof PR, Morrison JH (1990) J Comp Neurol 301:55–64.

35. Davis KL, Stewart DG, Friedman JI, Buchsbaum M, Harvey PD, Hof PR,Buxbaum J, Haroutunian V (2003) Arch Gen Psychiatry 60:443–456.

36. Syner FN, Goodman M (1966) Nature 209:426–428.37. Ames A, III (2000) Brain Res Rev 34:42–68.38. Grossman LI, Schmidt TR, Wildman DE, Goodman M (2001) Mol Phylogenet

Evol 18:26–36.39. Deitmer JW (2002) J Neurochem 80:721–726.40. Holloway RL (1967) Gen Syst 12:3–19.41. Deacon TW (1997) The Symbolic Species: The Co-Evolution of Language and

the Brain (Norton, New York).42. Petrides M, Cadoret G, Mackey S (2005) Nature 435:1235–1238.43. Burman KJ, Palmer SM, Gamberini M, Rosa MG (2006) J Comp Neurol

495:149–172.44. Rosabal F (1967) J Comp Neurol 130:87–108.45. Petrides M, Pandya DN (1999) Eur J Neurosci 11:1011–1036.46. von Bonin G, Bailey P (1947) The Neocortex of Macaca mulatta (Univ of Illinois

Press, Urbana, IL).47. Walker AE (1940) J Comp Neurol 73:59–86.48. Watanabe-Sawaguchi K, Kubota K, Arikuni T (1991) J Comp Neurol

311:108–133.49. Bailey P, von Bonin G, McCulloch WS (1950) The Isocortex of the Chimpanzee

(Univ of Illinois Press, Urbana, IL).50. Rajkowska G, Goldman-Rakic PS (1995) Cereb Cortex 5:307–322.51. Sherwood CC, Holloway RL, Erwin JM, Schleicher A, Zilles K, Hof PR (2004)

Brain Behav Evol 63:61–81.52. Sherwood CC, Broadfield DC, Holloway RL, Gannon PJ, Hof PR (2003) Anat

Rec 271:A276–A285.53. Paxinos G, Huang XF, Toga AW (2000) The Rhesus Monkey Brain in Stereotaxic

Coordinates (Academic, San Diego).54. Vogt BA, Nimchinsky EA, Vogt LJ, Hof PR (1995) J Comp Neurol

359:490–506.55. Mouton PR (2002) Principles and Practices of Unbiased Stereology: An Intro-

duction for Bioscientists (The Johns Hopkins Univ Press, Baltimore).56. Schmitz C, Hof PR (2000) J Chem Neuroanat 20:93–114.57. Sherwood CC, Raghanti MA, Stimpson CD, Bonar CJ, de Sousa AA, Preuss

TM, Hof PR Brain Behav Evol, in press.58. Sokal RR, Rohlf FJ (1995) Biometry: The Principles and Practice of Statistics in

Biological Research (Freeman, San Francisco).59. Felsenstein J (1985) Am Nat 125:1–15.60. Garland T, Ives AR (2000) Am Nat 155:346–364.61. Maddison WP, Maddison DR (2005) Mesquite: A Modular System for Evolu-

tionary Analysis, Ver 1.06, http:��mesquiteproject.org.62. Goodman M, Grossman LI, Wildman DE (2005) Trends Genet 21:511–517.63. Pagel MD (1992) J Theor Biol 156:431–442.


AN

THRO

POLO

GY

Log brain weight

Log

neur

on d

ensi

ty in

are

a 9L

0.8 1.2 1.6 2.0 2.4 2.8 3.2 3.64.6

4.7

4.8

4.9

5.0

5.1

5.2

5.3

Log brain weight

Log

glia

l den

sity

in a

rea

9L

0.8 1.2 1.6 2.0 2.4 2.8 3.2 3.64.6

4.7

4.8

4.9

5.0

5.1

5.2

5.3

A

New World monkeysOld Word monkeys

Gorilla gorilla

Homo sapiens


Pan troglodytes

B

Figure 3

Saguinusoedipus

Pitheciapithecia

Pongopygmaeus

Hylobatesmuelleri

Cercopithecuskandti

Colobusangolensis

I

II/III

IV

V

VI

I

II/III

IV

V

VI

I

II/III

IV

V

VI

I

II/III

IV

V

VI

I

II/III

IV

V

VI

I

II/III

IV

V

VI

Figure 4

cc32cc

32

Human(Homo sapiens)

Chimpanzee(Pan troglodytes)

Macaque monkey(Macaca maura)

cc32

44

49L

44

9L 4 9L 4

44

A

B

Hum

an(H

omo

sapi

ens)

Chi

mpa

nzee

(Pan

trog

lody

tes)

Mac

aque

mon

key

(Mac

aca

mau

ra)

Area 4 Area 9L Area 32 Area 44I

II/III

V

VI

I

II/III

IVVVI

I

II/III

IV

V

VI

I

II/III

IV

V

VI

I

II/III

V

VI

I

II/III

V

VI

I

II/III

IV

V

VI

I

II/III

IV

V

VI

I

II/III

IV

V

VI

I

II/III

IV

V

VI

I

II/III

IVV

VI

I

II/III

IV

V

VI

Evolution of increased glia-neuron ratios in the human frontal cortex -- She... http://www.pnas.org/cgi/content/full/0605843103/DC1

1 of 7 9/11/06 12:26 PM

Institution: MT SINAI SCHOOL OF MEDICINE Sign In as Member / Individual

This Article

Abstract

Services

Alert me to new issues of the journal

Request Copyright Permission

Sherwood et al. 10.1073/pnas.0605843103.

Supporting InformationFiles in this Data Supplement:

Supporting Table 3Supporting Figure 3Supporting Table 4Supporting Materials and MethodsSupporting Figure 4Supporting Figure 5

Supporting Figure 3

Fig. 3. The allometric scaling of neuron and glial density against brain weight. There was not a significant correlation between log neuron density and log brain weight from species mean data (r = -0.189, P = 0.452) or independent contrasts (r = 0.203, P = 0.418). Glial cell density, however, showed a significant negative allometric scaling relationship with brain weight, as revealed by analysis of both species mean data [reduced major axis (RMA) slope = 0.29, lower confidence interval (CI) = 0.20, upper CI = 0.44; r = 0.616, P = 0.007] and independent contrasts (RMA slope = 0.48, r = 0.588, P = 0.010). Because these results differ in significant ways from previously reported data on neuron and glial density scaling in the neocortex of mammals (1-4), further discussion is merited. It is important to note how the methods in the current study differ from previous research. First, our results are based on design-based stereologic methods, whereas others used assumption-based counting methods that relied on various correction factors. Second, our data represent 18 species of anthropoid primates. In contrast, the analysis of Shariff (4) was based on five primate species; the analysis of Tower (2) was based on 11 species distributed across Rodentia, Lagomorpha, Carnivora, Primates, and Cetartiodactyla; the analysis of Haug (1) was based on 26 mammalian species for neuron density and 14 for glial density, with 14 primates included for neuron density and 5 primates for glial density. Third, some of the discrepancies in data may be explained by the fact that these other studies were based on cell density estimates taken from the entire cortical width, whereas our study included a more restricted sampling, confined to only layer II/III. In addition, our study obtained counts from area 9L, whereas the data reported in other studies were derived from different regions: Shariff (4) reported values from area 6, area 17, and "average"


2 of 7 9/11/06 12:26 PM

eulaminate cortex sampled from frontal, parietal, and temporal regions; Tower (2) reported values from "agranular (motor)" cortex for most mammals but the average of various cortical regions in whales and elephants; Haug (1) reported values from the "lateral convexity of the frontal cortex" in primates and from various other cortical areas in cetaceans, elephants, and humans. These older studies all generally agreed in finding an approximate -1/3 power scaling relationship between neuron density and brain weight based on LS regression of species mean data. In contrast, our neuron density data from area 9L did not show a significant correlation with brain weight. Furthermore, Haug (1) did not find a significant correlation between brain weight and glial density, which led him to conclude that glial densities are essentially invariant across mammals. In contrast, we did find a positive correlation between glial density and brain weight in area 9L of anthropoids. Although it is a possibility that shrinkage artifacts have contributed to this discrepancy in results, it is noteworthy that our recent study of neuron density scaling in areas V1 and V2, which used a sample of specimens that largely overlapped those used in the current study, found the expected -1/3 scaling exponent based on species mean data (5). Therefore, it is seems implausible that shrinkage artifact would negate the typical scaling association between neuron density and brain weight in area 9L but not in areas V1 and V2 from the same individuals. In our opinion, therefore, it is possible that neuron density scaling relationships may show differences among cortical areas. Nonetheless, the raw neuron density estimates we have obtained for humans in the current study appear to be unusually high. Indeed, the mean human neuron density for area 9L in our data set is higher than any other hominoid and within the range of New World and Old World monkey species. These human neuron densities, furthermore, are higher than any others found in the literature for frontal cortex. The human brains used in this study were originally prepared by the Northwestern University Alzheimer's Disease Center Brain Bank before they were received at our laboratory. Thus, we must admit that it is a strong possibility that institutional differences in specimen handling might have contributed to this apparent high degree of shrinkage. In any event, even after removing the suspect human data point, the scaling relationships seen in Fig. 3 are fundamentally unchanged. In summary, once the methodological differences between our study and previous studies are taken into account, it seems to be a possibility that neurons and glia display differential scaling depending on the cortical area, layers, and subset of taxa analyzed. This question will require examination in future studies. Nonetheless, because of uncertainty regarding the congruence of our data with those in the literature, the primary analyses presented in this article concern only measures of glia relative to neurons. Stereologic counts of these cell types were obtained concurrently during optical disector analysis, so any experimental error whatsoever would affect both variables equally. In this regard, it is notable that the scaling of glia-neuron ratio on brain weight that we observed in the current study (b = +0.26) is in good agreement with that reported in Tower and Young (3) of b = +0.23.

1. Haug, R. (1987) Am. J. Ana.t 180, 126-142.

2. Tower, D. B. (1954) J. Comp. Neurol. 101, 19-52.

3. Tower, D. B. & Young, O. M. (1973) J. Neurochem. 20, 269-278.

4. Shariff, G. A. (1953) J. Comp. Neurol. 98, 381-400.

5. Sherwood ,C. C., Raghanti, M. A., Stimpson, C. D., Bonar, C. J., de Sousa, A. A., Preuss, T. M. & Hof, P. R. Brain Behav. Evol., in press.

Supporting Figure 4

Fig. 4. Examples of the cytoarchitecture of area 9L in various anthropoid primates. The cytoarchitecture of area 9L displays similar characteristics in all species examined, including a cell-sparse lower layer III with large pyramidal neurons, a poorly developed layer IV, and a prominent layer V (1). (Scale bar, 500 mm.)

1. Petrides M, Pandya DN (1999) Eur J Neurosci 11:1011-1136.

Supporting Figure 5

Fig. 5. The location and cytoarchitecture of areas 4, 9L, 32, and 44 in macaque monkeys, chimpanzees, and humans.


3 of 7 9/11/06 12:26 PM

(A) The position of cortical areas is indicated on the surface of the cerebral cortex. (B) The cytoarchitecture of cortical areas is shown to demonstrate the fundamental similarities among the species examined. Primary motor cortex (area 4) is distinguished by the presence of large Betz cells in layer V, an incipient layer IV, and indistinct lamination overall (1). The cytoarchitecture of area 9L is described in Fig. 4. The anterior paracingulate cortex (area 32) is defined as the part of the paracingulate located anterior to the genu of the corpus callosum (2) and is identifiable by the absence of a distinct layer IV, diffuse layers III and V, and a particularly thick layer VI that merges into the underlying white matter (3). Broca's area homologue (area 44) is characterized as a dysgranular cortex with large pyramidal cell somata at the bottom of layer III (4, 5). Cytoarchitectural features were relied upon to identify the regions of interest because there may be interindividual variation in the topological location of cytoarchitectural areas. (Scale bar, 500 mm.)

1. Sherwood, C. C., Lee, P. W. H., Rivara, C.-B., Holloway, R. L., Gilissen, E. P. E., Simmons, R. M. T., Hakeem, A., Allman, J. M., Erwin, J. M., Hof, P. R. (2003) Brain Behav. Evol. 61, 28-44.

2. Gallagher, H. L. & Frith, C. D. (2003) Trends Cogn. Sci. 7, 77-83.

3. Öngür, D., Ferry, A. T. & Price, J. .L (2003) J. Comp. Neurol. 460, 425-449.

4. Sherwood, C. C., Broadfield, D. C., Holloway, R. L., Gannon, P. J. & Hof, P. R. (2003) Anat. Rec. 271,A276-A285.

5. Petrides, M., Cadoret, G. & Mackey, S. (2005) Nature 435, 1235-1238.

Table 3. Stereologic estimates of cell densities in layer II/III of prefrontal area 9L (species mean cells per mm3)

Species nNeuron density

Glial density

Homo sapiens 6 93,432 155,064

Pan troglodytes 6 73,755 89,091

Gorilla gorilla 2 81,417 99,220

Pongo pygmaeus 2 59,594 58,070

Hylobates muelleri 1 53,618 65,679

Papio anubis 2 91,806 83,803

Mandrillus sphinx 1 135,417 137,985

Macaca maura 6 144,649 122,726


4 of 7 9/11/06 12:26 PM

Erythrocebus patas 2 143,783 156,354

Cercopithecus kandti 1 89,538 102,725

Colobus angolensis 1 81,019 97,032

Trachypithecus francoisi 1 91,733 104,299

Alouatta caraya 1 93,326 104,707

Saimiri boliviensis 1 127,266 65,539

Aotus trivirgatus 1 74,074 46,974

Saguinus oedipus 1 111,679 51,263

Leontopithecus rosalia 2 105,365 62,976

Pithecia pithecia 1 71,549 45,894

Table 4. Sample used for neuron and glial density counts

Taxonomic group Species Sex Age, y Fixation

Hominoids Homo sapiens F 40 I

Homo sapiens F 43 I

Homo sapiens F 53 I

Homo sapiens M 35 I

Homo sapiens M 48 I

Homo sapiens M 54 I


5 of 7 9/11/06 12:26 PM

Pan troglodytes F 19 I



Pan troglodytes M 17 I



Gorilla gorilla M 13 I

Gorilla gorilla M 49 I

Pongo pygmaeus M 11 I

Pongo pygmaeus M 33 I

Hylobates muelleri M 19 I

Old World monkeys Papio anubis M 11 P

Papio anubis M 11 P

Mandrillus sphinx F 33 I

Macaca maura F 5 P

Macaca maura F 7 P

Macaca maura F 7 P

Macaca maura F 8 P

Macaca maura M 8 P

Macaca maura M 10 P


6 of 7 9/11/06 12:26 PM

Erythrocebus patas F 12 P

Erythrocebus patas M 12 P

Cercopithecus kandti M adult I

Colobus angolensis M 18 I

Trachypithecus francoisi M 1 I

New World monkeys Alouatta caraya M 21 I

Saimiri boliviensis F 12 P

Aotus trivirgatus M >18 P

Saguinus oedipus F 6 I

Leontopithecus rosalia F 11 I

Leontopithecus rosalia M 6 I

Pithecia pithecia F 1 I

M, male; F, female; I, immersion; P, perfusion.

Supporting Materials and Methods

Specimen Preparation. After initial fixation, all brains were transferred to 0.1 M PBS containing 0.1% sodium azide and stored at 4°C to prevent continued tissue shrinkage. Brain weights of nonhuman specimens were measured either immediately after perfusion or directly upon receipt from the donating institution. Human brain weights were measured at autopsy and were provided by the brain bank. Because many brain specimens were collected opportunistically, it was impossible to control for idiosyncratic artifacts related to differences in fixation length and postmortem delay. Nonetheless, the recorded brain weights in our sample do not show significant deviations from typical species average fresh weights (see discussion in reference 1). Regardless, our statistical analyses circumvented potential problems related to differential tissue shrinkage. In particular, if it is assumed that comparable error from tissue shrinkage is contained in both independent and dependent axes for each data point in bivariate plots, such error would affect the elevation of regressions but would not have a significant influence on scaling exponents and residuals. Additionally, because many analyses used the ratio of glia to neurons, equal shrinkage-based error


7 of 7 9/11/06 12:26 PM

contained in the numerator and denominator is factored out.

Data Analysis. We examined the sensitivity of independent contrasts to the method of branch length transformation by also setting all branch lengths to 1. This alternative method did not significantly alter the results. We followed the recommended diagnostic procedures of Garland et al. (2) to ensure that branch length transformation for analyses of independent contrasts met statistical requirements.

Rather than using reduced major axis line-fitting to generate human predictions from nonhuman data, we usedleast-squares (LS) regression because this model was designed to generate predictions for y when x is known and it produces residuals that are uncorrelated with the independent variable (3).

In generating LS prediction lines from independent contrasts data, we also calculated a "generic" prediction for aspecies of unspecified phylogenetic position attached to the root of the entire anthropoid tree. This prediction produced similar results to the phylogenetically informed regression that is "human-specific," and so details are not reported.

1. Sherwood, C. C., Raghanti, M. A., Stimpson, C. D., Bonar, C. J., de Sousa, A. A., Preuss, T. M. & Hof, P. R. Brain Behav. Evol., in press.

2. Garland, T., Harvey, P. H. & Ives, A. R. (1992) Syst. Biol. 41, 18-32.

3. Sokal, R. R. & Rohlf, F. J. (1995) Biometry: The Principles and Practice of Statistics in Biological Research(Freeman, San Francisco).

This Article

Abstract

Services

Alert me to new issues of the journal

Request Copyright Permission

Current Issue | Archives | Online Submission | Info for Authors | Editorial Board | About Subscribe | Advertise | Contact | Site Map

Copyright © 2006 by the National Academy of Sciences

Absolute brain size: Did we throw the baby outwith the bathwater?Lori Marino*Neuroscience and Behavioral Biology Program, 1462 Clifton Road, Suite 304, Emory University, Atlanta, GA 30322

The recent study by Sherwood etal. (1) in this issue of PNAS isa study in irony. It is thus sobecause the authors used some

of the most current and sophisticatedhistological and data-analytic tools to, inthe end, unearth . . . a dinosaur. Now theauthors did not do this in the literalsense. Rather, they laid bare an ‘‘ex-tinct’’ measure of brain anatomy: thatold fossil known as absolute brain size.Even as far back as the 1800s, Alex-ander Brandt (2) and, a bit later, OttoSnell (3) demonstrated the importanceof scaling the brain to body size, signal-ing the impending demise of absolutebrain size as a valid measure of braincapacity. The final death knell wasstruck in its most popular form byHarry Jerison, who reminded us onceagain that brain–body allometry plays amajor role in the size of any species’brain. He effectively argued that thebrain, like any other organ, scales withbody size, and the validity of the use ofbrain size as a measure of intelligenceor information processing capacity restsupon the size of the brain relative to thesize of the body. He introduced the en-cephalization quotient (EQ) as a propermeasure of relative brain size acrossspecies (4). EQ is a number that essen-tially quantifies how much larger orsmaller the average brain size of a givenspecies is relative to the expected brainsize based on body size. Likewise, theresidual of a brain weight–body weightregression across a sample of speciesprovides similar information. And, al-though different authors have differenttheoretical justifications for the parame-ters of the brain–body size relationshipacross species [see Deaner et al. (5) forreview], all measures of relative brainsize are based on the common assump-tion that it is only meaningful to con-sider brain size if body size, or somerelative measure, is taken into account.Therefore, nowadays, absolute brain sizeas a measure of cognitive capacity isconsidered obsolete.

However, in the rush to abandon ab-solute brain size in favor of measures ofrelative brain size, have we been alto-gether too hasty? Well, no, not entirely.There are valid reasons for taking intoaccount the allometric relationshipbetween brain and body size when at-tempting to infer how much of any

species’ brain is devoted to cognitiveprocesses. But could we have, at thevery least, missed something importantabout absolute brain size in the process?Did we throw the baby out with thebathwater? The Sherwood et al. (1)study suggests that we might have.

Recent microarray studies have shownthat neuronal signaling and energy pro-duction genes are up-regulated in thehuman neocortex compared with thegreat apes (6, 7). These findings suggestthat there was selection for higher ratesof energy expenditure in human neocor-tex compared with in other primates (8).One way to measure metabolic support

for neurons is by examining the ratio ofglia to neurons. The distribution densi-ties of glia provide an indication of themetabolic demand of neighboring neu-rons. Sherwood et al. (1) investigatedwhether glial cell densities are rela-tively higher in human frontal cortexcompared with other primates. Theyreasoned that significantly higher glia–neuron ratios in humans might exem-plify the kind of unique trait that wouldprovide a basis for human intelligence.However, the Sherwood et al. (1) studyis much more than a study in cell densi-ties, metabolism, and allometry. Whatmakes this article so compelling is thatit has profound implications for thequestion of neurobiological continuityacross species, particularly across hu-mans and our closest nonhuman rela-tives, the great apes. It addresses thegeneral question of whether humanbrains should best be thought of as largehominoid brains, or, alternatively, as asingularly endowed product of evolutionsomewhat apart from the rest of primatebrain evolution.

Sherwood et al. (1), did indeed findthat the human frontal cortex displaysa higher ratio of glia to neurons than inother primates. However, and impor-tantly, this relative difference is pre-

dicted by the allometric scaling inherentin the enlargement of the human brain.In other words, Sherwood et al. (1) showthat overall or absolute brain size con-stitutes a key factor in the ratio of gliato neurons. The authors suggest that thegreater numbers of glia in the humanneocortex may be due to the increasedenergetic costs of larger dendritic arborsand longer fiber projections within thecontext of the large human brain. Thebottom line is that Sherwood et al. (1)show that the human brain conformsto the general mammalian pattern ofhigher glia–neuron ratios with largerbrains.

Furthermore, Sherwood et al. (1) de-termined whether there were regionaldifferences in glia–neuron ratios acrosshumans and nonhuman primates in cor-tical regions associated with specific hu-man abilities, such as area 44, a key areafor language production, and area 32,which is active during theory-of-mindtasks in humans. Sherwood et al. (1)found no significant species differencesin these critical areas and suggest thatthe energetics of frontal cortex, even inthese regions, have been largely con-served over the past 25 million years ofprimate brain evolution. Their overallconclusion is striking: ‘‘. . . human cog-nitive and linguistic specializations haveemerged by elaborating on higher-orderexecutive functions of the prefrontalcortex . . . that evolved earlier in theprimate lineage’’ (1).

Sherwood et al. (1) provide supportfor the idea that the human brain ismore or less a large hominoid brain andcan be understood in that context. How-ever, these findings also engender manyquestions about the relationship be-tween brain size and cognition. If hu-man brains fall in line with generallyexpected patterns of relationship amongoverall size, neuron density, and glia–neuron ratios and yet possess a numberof striking cognitive features, howshould the absolute size of brains beinterpreted? What happens to brainswhen they enlarge over time? How can

The author declares no conflict of interest.

See companion article on page 13606.

*E-mail: [email protected].


Absolute brain sizeconstitutes a keyfactor in the ratioof glia to neurons.

www.pnas.org�cgi�doi�10.1073�pnas.0606337103 PNAS � September 12, 2006 � vol. 103 � no. 37 � 13563–13564

CO

MM

EN

TA

RY

these changes provide a substrate forchanges in cognitive abilities?

When More Is More than Just MoreWhen brains enlarge, they can do so ina couple of different ways. They mayadd more neurons or make existing neu-rons larger, or some combination ofboth. However, both of these changespose design problems that need to besolved because larger neurons havelonger axons and dendrite conductiontimes will be slower. When the numberof neurons increases, it becomes moredifficult for each individual neuron tomaintain connectivity. Therefore, brainscannot enlarge (either by adding neu-rons or by making them bigger) withoutthe organization of the brain changing.The main way most mammal brainshave enlarged is by increasing modular-ity to accommodate increases in thenumber of neurons. Increased modular-ity increases the number of processingareas. Indeed, there is substantial evi-dence that mammals with more neocor-tex possess more cortical areas andmore units of parcellation (e.g., col-umns, etc.) than smaller-brainedmammals (9).

So enlarged brains can support in-creases in cognitive ability because theymay be more highly segmented and dif-ferentiated, thereby possessing morecomplexity than their smaller counter-parts. This is a case of more being morethan just more. Enlarged brains not onlyhave more neurons, they also havegreater complexity, which is more thanjust more. Therefore, brain enlargement,and the increases in complexity thatnecessarily accompany it, may providethe substrate for enhanced cognitiveabilities or even the emergence of newcognitive abilities. In this context, abso-lute brain size regains its value as abrain metric because it becomes a proxyfor increased organizational complexity.So maybe it should be retrieved fromthe bathwater!

Implications for NeurobiologicalContinuityThe deep fundamental insight supportedby Sherwood et al. (1) is that the humanbrain is not unique or anomalous.Rather, the human brain is a product ofchanges in brain anatomy that are wellpredicted by scaling expectations for anynonhuman anthropoid primate. The

study by Sherwood et al. (1) is a particu-larly elegant example of a growing bodyof evidence for this conclusion. For in-stance, several studies have shown thatthe human frontal cortex occupies thesame proportion of total cortex in hu-mans as it does in great apes (10, 11).Therefore, humans are typical primateswith regard to the portion of their cor-tex devoted to frontal cortex. Similarly,the human brain possesses the degree ofcortical gyrification expected for a pri-mate of our brain size (12). These find-ings show that there are ways to obtainnew or enhanced cognitive abilities inhuman brains that are perfectly consis-tent with the way brain evolution oc-curred in other primates. [To be fair,there is also evidence that the humanbrain might depart from certain allomet-ric expectations (13).]

Regardless of which features of thehuman brain are typical of other pri-mate brains, the more general point isthat a consideration of absolute brainsize provides insight into which featuresof the human brain are predictable andwhich are not. Therefore, absolute brainsize has turned out to be an importantvariable in its own right for understand-ing mammalian brain evolution.

1. Sherwood CC, Stimpson CD, Raghanti MA, Wild-man DE, Uddin M, Grossman LI, Goodman M,Redmond JC, Bonar CJ, Erwin JM, Hof PR(2006) Proc Natl Acad Sci USA 103:13606-13611.

2. Brandt A (1867) Bull Soc Imp Nat Moscou 40:525-543.3. Snell O (1891) Sitz Ges Morph Physiol (Munchen)

7:90-94.4. Jerison HJ (1973) The Evolution of the Brain and

Intelligence (Academic, New York).5. Deaner RO, Nunn CL, van Schaik CP (2000)

Brain Behav Evol 55:44-52.6. Uddin M, Wildman DE, Liu G, Xu W, Johnson

RM, Hof PR, Kapatos G, Grossman LI, Good-man M (2004) Proc Natl Acad Sci USA 101:2957-2962.

7. Caceres M, Lachuer J, Zapala MA, Redmond JC,Kudo L, Geschwind DH, Lockhart DJ, Preuss T,Barlow C (2003) Proc Natl Acad Sci USA100:13030-13035.

8. Grossman LI, Wildman DE, Schmidt TR,

Goodman M (2004) Trends Genet 20:578-585.9. Kaas JH (1993) Perspect Dev Neurobiol 1:101-107.

10. Semendeferi K, Lu A, Damasio H (2002) NatNeurosci 5:272-276.

11. Bush EC, Allman JM (2004) Proc Natl Acad SciUSA 101:3962-3966.

12. Zilles K, Armstrong E, Moser KH, Schleicher A,Stephan H (1989) Brain Behav Evol 34:143-150.

13. Rilling JK, Insel TR (1999) J Hum Evol 37:191-223.

13564 � www.pnas.org�cgi�doi�10.1073�pnas.0606337103 Marino

cc32cc

32

Human(Homo sapiens)

Chimpanzee(Pan troglodytes)

Macaque monkey(Macaca maura)

cc32

44

49L

44

9L 4 9L 4

44

A

B

Hum

an(H

omo

sapi

ens)

Chi

mpa

nzee

(Pan

trog

lody

tes)

Mac

aque

mon

key

(Mac

aca

mau

ra)

Area 4 Area 9L Area 32 Area 44I

II/III

V

VI

I

II/III

IVVVI

I

II/III

IV

V

VI

I

II/III

IV

V

VI

I

II/III

V

VI

I

II/III

V

VI

I

II/III

IV

V

VI

I

II/III

IV

V

VI

I

II/III

IV

V

VI

I

II/III

IV

V

VI

I

II/III

IVV

VI

I

II/III

IV

V

VI

Saguinusoedipus

Pitheciapithecia

Pongopygmaeus

Hylobatesmuelleri

Cercopithecuskandti

Colobusangolensis

I

II/III

IV

V

VI

I

II/III

IV

V

VI

I

II/III

IV

V

VI

I

II/III

IV

V

VI

I

II/III

IV

V

VI

I

II/III

IV

V

VI

Figure 4

Log brain weight

Log

neur

on d

ensi

ty in

are

a 9L

0.8 1.2 1.6 2.0 2.4 2.8 3.2 3.64.6

4.7

4.8

4.9

5.0

5.1

5.2

5.3

Log brain weight

Log

glia

l den

sity

in a

rea

9L

0.8 1.2 1.6 2.0 2.4 2.8 3.2 3.64.6

4.7

4.8

4.9

5.0

5.1

5.2

5.3

A

New World monkeysOld Word monkeys

Gorilla gorilla

Homo sapiens


Pan troglodytes

B

Figure 3

Absolute brain size: Did we throw the baby outwith the bathwater?Lori Marino*Neuroscience and Behavioral Biology Program, 1462 Clifton Road, Suite 304, Emory University, Atlanta, GA 30322

The recent study by Sherwood etal. (1) in this issue of PNAS isa study in irony. It is thus sobecause the authors used some

of the most current and sophisticatedhistological and data-analytic tools to, inthe end, unearth . . . a dinosaur. Now theauthors did not do this in the literalsense. Rather, they laid bare an ‘‘ex-tinct’’ measure of brain anatomy: thatold fossil known as absolute brain size.Even as far back as the 1800s, Alex-ander Brandt (2) and, a bit later, OttoSnell (3) demonstrated the importanceof scaling the brain to body size, signal-ing the impending demise of absolutebrain size as a valid measure of braincapacity. The final death knell wasstruck in its most popular form byHarry Jerison, who reminded us onceagain that brain–body allometry plays amajor role in the size of any species’brain. He effectively argued that thebrain, like any other organ, scales withbody size, and the validity of the use ofbrain size as a measure of intelligenceor information processing capacity restsupon the size of the brain relative to thesize of the body. He introduced the en-cephalization quotient (EQ) as a propermeasure of relative brain size acrossspecies (4). EQ is a number that essen-tially quantifies how much larger orsmaller the average brain size of a givenspecies is relative to the expected brainsize based on body size. Likewise, theresidual of a brain weight–body weightregression across a sample of speciesprovides similar information. And, al-though different authors have differenttheoretical justifications for the parame-ters of the brain–body size relationshipacross species [see Deaner et al. (5) forreview], all measures of relative brainsize are based on the common assump-tion that it is only meaningful to con-sider brain size if body size, or somerelative measure, is taken into account.Therefore, nowadays, absolute brain sizeas a measure of cognitive capacity isconsidered obsolete.

However, in the rush to abandon ab-solute brain size in favor of measures ofrelative brain size, have we been alto-gether too hasty? Well, no, not entirely.There are valid reasons for taking intoaccount the allometric relationshipbetween brain and body size when at-tempting to infer how much of any

species’ brain is devoted to cognitiveprocesses. But could we have, at thevery least, missed something importantabout absolute brain size in the process?Did we throw the baby out with thebathwater? The Sherwood et al. (1)study suggests that we might have.

Recent microarray studies have shownthat neuronal signaling and energy pro-duction genes are up-regulated in thehuman neocortex compared with thegreat apes (6, 7). These findings suggestthat there was selection for higher ratesof energy expenditure in human neocor-tex compared with in other primates (8).One way to measure metabolic support

for neurons is by examining the ratio ofglia to neurons. The distribution densi-ties of glia provide an indication of themetabolic demand of neighboring neu-rons. Sherwood et al. (1) investigatedwhether glial cell densities are rela-tively higher in human frontal cortexcompared with other primates. Theyreasoned that significantly higher glia–neuron ratios in humans might exem-plify the kind of unique trait that wouldprovide a basis for human intelligence.However, the Sherwood et al. (1) studyis much more than a study in cell densi-ties, metabolism, and allometry. Whatmakes this article so compelling is thatit has profound implications for thequestion of neurobiological continuityacross species, particularly across hu-mans and our closest nonhuman rela-tives, the great apes. It addresses thegeneral question of whether humanbrains should best be thought of as largehominoid brains, or, alternatively, as asingularly endowed product of evolutionsomewhat apart from the rest of primatebrain evolution.

Sherwood et al. (1), did indeed findthat the human frontal cortex displaysa higher ratio of glia to neurons than inother primates. However, and impor-tantly, this relative difference is pre-

dicted by the allometric scaling inherentin the enlargement of the human brain.In other words, Sherwood et al. (1) showthat overall or absolute brain size con-stitutes a key factor in the ratio of gliato neurons. The authors suggest that thegreater numbers of glia in the humanneocortex may be due to the increasedenergetic costs of larger dendritic arborsand longer fiber projections within thecontext of the large human brain. Thebottom line is that Sherwood et al. (1)show that the human brain conformsto the general mammalian pattern ofhigher glia–neuron ratios with largerbrains.

Furthermore, Sherwood et al. (1) de-termined whether there were regionaldifferences in glia–neuron ratios acrosshumans and nonhuman primates in cor-tical regions associated with specific hu-man abilities, such as area 44, a key areafor language production, and area 32,which is active during theory-of-mindtasks in humans. Sherwood et al. (1)found no significant species differencesin these critical areas and suggest thatthe energetics of frontal cortex, even inthese regions, have been largely con-served over the past 25 million years ofprimate brain evolution. Their overallconclusion is striking: ‘‘. . . human cog-nitive and linguistic specializations haveemerged by elaborating on higher-orderexecutive functions of the prefrontalcortex . . . that evolved earlier in theprimate lineage’’ (1).

Sherwood et al. (1) provide supportfor the idea that the human brain ismore or less a large hominoid brain andcan be understood in that context. How-ever, these findings also engender manyquestions about the relationship be-tween brain size and cognition. If hu-man brains fall in line with generallyexpected patterns of relationship amongoverall size, neuron density, and glia–neuron ratios and yet possess a numberof striking cognitive features, howshould the absolute size of brains beinterpreted? What happens to brainswhen they enlarge over time? How can

The author declares no conflict of interest.

See companion article on page 13606.

*E-mail: [email protected].


Absolute brain sizeconstitutes a keyfactor in the ratioof glia to neurons.

www.pnas.org�cgi�doi�10.1073�pnas.0606337103 PNAS � September 12, 2006 � vol. 103 � no. 37 � 13563–13564

CO

MM

EN

TA

RY

these changes provide a substrate forchanges in cognitive abilities?

When More Is More than Just MoreWhen brains enlarge, they can do so ina couple of different ways. They mayadd more neurons or make existing neu-rons larger, or some combination ofboth. However, both of these changespose design problems that need to besolved because larger neurons havelonger axons and dendrite conductiontimes will be slower. When the numberof neurons increases, it becomes moredifficult for each individual neuron tomaintain connectivity. Therefore, brainscannot enlarge (either by adding neu-rons or by making them bigger) withoutthe organization of the brain changing.The main way most mammal brainshave enlarged is by increasing modular-ity to accommodate increases in thenumber of neurons. Increased modular-ity increases the number of processingareas. Indeed, there is substantial evi-dence that mammals with more neocor-tex possess more cortical areas andmore units of parcellation (e.g., col-umns, etc.) than smaller-brainedmammals (9).

So enlarged brains can support in-creases in cognitive ability because theymay be more highly segmented and dif-ferentiated, thereby possessing morecomplexity than their smaller counter-parts. This is a case of more being morethan just more. Enlarged brains not onlyhave more neurons, they also havegreater complexity, which is more thanjust more. Therefore, brain enlargement,and the increases in complexity thatnecessarily accompany it, may providethe substrate for enhanced cognitiveabilities or even the emergence of newcognitive abilities. In this context, abso-lute brain size regains its value as abrain metric because it becomes a proxyfor increased organizational complexity.So maybe it should be retrieved fromthe bathwater!

Implications for NeurobiologicalContinuityThe deep fundamental insight supportedby Sherwood et al. (1) is that the humanbrain is not unique or anomalous.Rather, the human brain is a product ofchanges in brain anatomy that are wellpredicted by scaling expectations for anynonhuman anthropoid primate. The

study by Sherwood et al. (1) is a particu-larly elegant example of a growing bodyof evidence for this conclusion. For in-stance, several studies have shown thatthe human frontal cortex occupies thesame proportion of total cortex in hu-mans as it does in great apes (10, 11).Therefore, humans are typical primateswith regard to the portion of their cor-tex devoted to frontal cortex. Similarly,the human brain possesses the degree ofcortical gyrification expected for a pri-mate of our brain size (12). These find-ings show that there are ways to obtainnew or enhanced cognitive abilities inhuman brains that are perfectly consis-tent with the way brain evolution oc-curred in other primates. [To be fair,there is also evidence that the humanbrain might depart from certain allomet-ric expectations (13).]

Regardless of which features of thehuman brain are typical of other pri-mate brains, the more general point isthat a consideration of absolute brainsize provides insight into which featuresof the human brain are predictable andwhich are not. Therefore, absolute brainsize has turned out to be an importantvariable in its own right for understand-ing mammalian brain evolution.

1. Sherwood CC, Stimpson CD, Raghanti MA, Wild-man DE, Uddin M, Grossman LI, Goodman M,Redmond JC, Bonar CJ, Erwin JM, Hof PR(2006) Proc Natl Acad Sci USA 103:13606-13611.

2. Brandt A (1867) Bull Soc Imp Nat Moscou 40:525-543.3. Snell O (1891) Sitz Ges Morph Physiol (Munchen)

7:90-94.4. Jerison HJ (1973) The Evolution of the Brain and

Intelligence (Academic, New York).5. Deaner RO, Nunn CL, van Schaik CP (2000)

Brain Behav Evol 55:44-52.6. Uddin M, Wildman DE, Liu G, Xu W, Johnson

RM, Hof PR, Kapatos G, Grossman LI, Good-man M (2004) Proc Natl Acad Sci USA 101:2957-2962.

7. Caceres M, Lachuer J, Zapala MA, Redmond JC,Kudo L, Geschwind DH, Lockhart DJ, Preuss T,Barlow C (2003) Proc Natl Acad Sci USA100:13030-13035.

8. Grossman LI, Wildman DE, Schmidt TR,

Goodman M (2004) Trends Genet 20:578-585.9. Kaas JH (1993) Perspect Dev Neurobiol 1:101-107.

10. Semendeferi K, Lu A, Damasio H (2002) NatNeurosci 5:272-276.

11. Bush EC, Allman JM (2004) Proc Natl Acad SciUSA 101:3962-3966.

12. Zilles K, Armstrong E, Moser KH, Schleicher A,Stephan H (1989) Brain Behav Evol 34:143-150.

13. Rilling JK, Insel TR (1999) J Hum Evol 37:191-223.

13564 � www.pnas.org�cgi�doi�10.1073�pnas.0606337103 Marino

evolution of increased glia–neuron ratios in the human frontal...

Documents