Finlay, B., L., Darlington, R., B. & Nicastro, N. (2001) Developmental structure in brain evolution. Behavioral and Brain Sciences 24 (2): XXX-XXX.

This is the unedited final draft of a BBS target article that has been accepted for publication (Copyright 2000: Cambridge University Press) and is currently being circulated for Open Peer Commentary.

This preprint is for inspection only, to help prospective commentators decide whether or not they wish to prepare a formal commentary.

Please do not prepare a commentary unless you have received a formal invitation indicating that it has been possible to include you in the final list of invited commentators.

For information on becoming a commentator on this or other BBS target articles, write to bbs@soton.ac.uk

For information about subscribing or purchasing offprints of the published version, with commentaries and author's response, write to:

journals_subscriptions@cup.org (North America)
journals_subscriptions@cup.cam.ac.uk (All other countries).

Developmental structure in brain evolution.

Corresponding author:

Barbara Finlay is Professor in the Departments of Psychology, and Neurobiology and Behavior at Cornell University. She holds the William R. Kenan Chair of Psychology. She has worked in developmental neuroscience since her doctorate in 1976, and has framed that developmental work in an evolutionary context dating from a series of lectures by Glen Northcutt in 1979.

Richard Darlington is Professor of Psychology at Cornell University,where he has been on the faculty since 1963. He is a Fellow of AAAS. He is author or coauthor of 3 textbooks and numerous papers on statistical methods, especially regression and linear models.His active involvement in allometry and neurogenesis stems only from 1994.

Nicholas Nicastro is a doctoral candidate in Psychology at Cornell University, and is author of the upcoming book Deep in the World: Of Nature, Nurture, and Neither.

Abstract

How does evolution grow bigger brains? It has been widely assumed that the growth of individual structures and functional systems in response to niche-specific cognitive challenges is the most plausible mechanism for brain expansion in mammals. Comparison of multiple regressions on allometric data for 131 mammalian species, however, suggests that for 9 of 11 brain structures taxonomic and body size factors are less important than covariance of these major structures with each other. Which structures grow biggest is largely predicted by a conserved order of neurogenesis that can be derived from the basic axial structure of the developing brain. This conserved order of neurogenesis predicts the relative scaling not only of gross brain regions like the isocortex or mesencephalon, but also the level of detail of individual thalamic nuclei. Special selection of particular areas for specific functions does occur, but it is a minor factor compared to the large-scale covariance of the whole brain. The idea that enlarged isocortex could be a"spandrel," a byproduct of structural constraints later adapted for various behavior, contrasts with approaches that look to selection of particular brain regions for cognitively advanced behaviors, as commonly assumed in the case of hominid brain evolution.

Keywords

Allometry, brain size, cortex, development, heterochrony, hominid evolution, limbic system, neurogenesis.
 

1. Introduction

When we speak of brain evolution, what exactly do we imagine to be evolving? This is not a trick question. Natural selection, after all, acts on particular systems and capacities based on differential survival of whole organisms. If some change in brain structure is selected for, how can change be implemented? Such questions arguably have more to do with architectural constraints born of the phylogenetic history of brains than they do with some putative optimal engineering (with "optimal" defined functionally, energetically, or any way the engineer chooses). Based on a legacy of prior change, some patterns of adaptation are more likely to be hit upon by the evolving organism than others. It is our contention here that developmental processes are a primary locus of architectural constraints on brain evolution.

There would seem to be two broad models for how brains change. On the one hand, their parts might be taken to be fundamentally discriminable in function and independently variable. Brain evolution in that case would be a matter of growing a bigger auditory processing system, resource mapper, olfactory system, etc., with the rest of the system left mostly unchanged. Alternatively, the size of the entire brain might be taken to vary in response to selection on any of its constituent parts. In the latter model, developmentally inspired architectural constraints make part/whole size dissociations inherently less workable responses to selection. (These views are necessarily simplified for the purposes of this introduction.)

Considering just the sensory and motor periphery, the case for special selection looks strong. Sensory systems can vary wildly --consider the ears of the echolocating bats, the eye-shine of nocturnal animals, the near telephoto vision of raptors, or the special chemical communication systems of many rodents. On the motor and sensorimotor side, specializations are no less impressive: prehensile tails and trunks, the precision grip of social-grooming primates, the palpating noses of moles. Features of the sensory and motor periphery may diverge quite strikingly in relative size and conformation, in components from the morphological to the biochemical level, and in function.

To what extent is such idiosyncratic organization also a property of the nervous system that organizes the information provided and action afforded by the sensory periphery? If one looks at mature isocortex, one finds overrepresentation of the fovea in animals with high acuity vision in the striate cortex, overrepresentation of the "acoustic fovea" of bats in their auditory cortex, and disproportionate topographic layout of specializations like vibrissae and fingers in any somatosensory cortex. In these cases, however, there is strong evidence from developmental or adult manipulations that the sensory or motor periphery may impose its form on a "generic" nervous system (Van der Loos & Welker 1985; Gilbert et al. 1996; Florence et al. 1997).

Peripheral-to central isomorphisms of the above kind are not the only sort of specialization we could consider, however. Complicated behaviors might have complex and idiosyncratic internal circuitry, from elaborated specialized capacities to single percepts or actions. We have such examples as "the song system" as a coordinated perceptuomotor system; we speak of animals as being "more visual" or "more olfactory." Complex abilities like foraging and food-storing have been shown to relate to hippocampus size and (by implication) the type of cognitive map the hippocampus can support (Jacobs & Spencer, 1994). Precocial ungulates without experience (and humans with some experience) recognize a certain pattern of visual stimulation as a cliff and inhibit motion (Gibson & Walk, 1960). Human infants recognize a different pattern as a face, orient to it and reproduce its expressions (Meltzoff, 1966). In looking more closely at the structure of brain evolution, we may hope to understand how the general and the specialized can cohabit in a single brain.

The predominant quantitative anatomical and allometric techniques used by evolutionary neurobiologists are differentiative. For example, in most allometric studies, the question typically asked is what part of the brain is largest in an animal with a special behavioral capacity, "controlling for" general enlargement in brain size. In this article, we wish to turn attention back to the coordination of brain change. We want to look at the relationship of brain evolution to specialized and distributed circuits, and to get a better methodological idea of what might be involved in "controlling for" baseline changes in brain sizes. Does the brain have selectable, covarying "units" from the level of single circuits, structures, functional systems or anatomical divisions? What is the range of independent variation observed at each one of these levels of analysis?

We will first review some published work on the structure of relative changes in size of gross brain parts, and the close relationship of a highly conserved schedule of neurogenesis and other neurodevelopmental events to this change in brain size. We include consideration of the statistical and methodological issues involved in determining the amount and type of variance accounted for in allometric and developmental data. We will present some new data showing how well the developmental constraint hypothesis works in predicting size changes in the di- and telencephalon, as well as instances where it does not work. Finally, we will make an the argument that structural change must often precede functional allocation in the developing brain, and explore some implications this has for essential structure-function relationships in cognitive neuroscience and elsewhere.

Even a complete analysis of the adult brain, using the full array of current techniques in neuroscience, will leave unexamined central questions about the essential relationship between structure and function. The study of development promises unique insights into the nature of functional architecture. Likewise, patterns of comparative brain evolution show structure/function links in a different light than that cast by any one species. The problem we concern ourselves with here, then, is establishing the precise developmental substrate on which brain evolution selects. Do the brain and its information-gathering organs divide themselves up in evolution into components, modules or circuits that can be the independent objects of special selection, either for sensory and motor performance, or for whole coordinated chunks of motivated behavior? Or does selection attack along a broader front, working change by adjusting the parameters of a "standard" developmental program?

The analysis we have done is drawn entirely from information on relative mammalian brain sizes and neurodevelopmental events. There is reason to view mammals as a special case among vertebrates, in that the vast majority of their neurogenesis is confined to very early development and not extended over the lifespan, as it is in fish, amphibians, birds and reptiles. On the other hand, issues in space allocation in brains transcend mammals, and the general issue of the segmental structure of the brain and its relationship to neurogenesis should relate to vertebrates generally. In the following discussion, we will take examples and raise issues somewhat more broadly than for mammals alone, though we make no assumptions the detail of the patterns we see in mammals will apply directly to non-mammalian vertebrates.
 

2. The structure of variation in mammalian brain size.

In the early 1980’s, Heinz Stephan et al. published their comprehensive volumetric data set for eleven precisely delimited divisions of the brain and for more discrete nuclei and zones for a large sample of insectivores, prosimians, simians and bats . (Stephan et al. 1981; Frahm et al. 1982; Stephan et al. 1982; Baron et al. 1983; Frahm et al., 1984a; Frahm et al. 1984b; Baron et al. 1988; Baron et al. 1987; Stephan et al. 1987; Stephan & Frahm et al. 1988; Baron et al. 1990). The strength of this data set is the number of species analyzed, with its information on niche and classes of behavioral specialization, and its comprehensive brain coverage. It has limitations: most of the brain divisions measured subsume multiple functional systems, and the fact that measures are of volumes, which include neurons, their processes, and supporting elements of all kinds, raises secondary problems of how these related elements scale with each other and with brain and body size.

Even so, much has been learned about how brains vary from this valuable source, and it has become a playing field for two opponent approaches to comparative brain structure. One class of analysis has sought to differentiate the size of particular subregions from the overall coordinated enlargement of the brain and map those onto behavioral or niche variables. A second class of analysis has sought to demonstrate internal covariation and structure in the evolution of brain components. Through all of these analyses runs the question of the proper way to handle variance in analyses of data that has an intrinsic structure of relatedness, as all mammalian species (and, indeed, all vertebrates) have with each other.

2.1 "Correcting" for body and brain size: niche-specific variation

Jerison's (1973, 1991) analysis of the allometric relationship between body size and brain size established many of the core assumptions and analytical methods subsequently used to investigate structural brain evolution. Brain size increases with body size at a characteristic exponential rate. The reason for this well-characterized relationship (Martin, 1982) has always remained essentially unexplained, though there have been many intriguing attempts. The neural machinery for controlling muscles and for innervating the sensory surface might reasonably increase at some function of body size. However, in a medium-sized brain like a cat's, the representation of the body surface and the primary motor representation occupies less than a tenth of the surface extent of isocortex, and it is not at all clear (depending upon your model of the brain) why specialized sense organs like eyes and ears should also increase so regularly with body size (and they do - for example Hughes 1977).

Why aren't the basic mechanisms of action, memory, communication and cognition scale-independent? Compare the North American ruby-throated hummingbird, with a brain size of less than a gram, with a baleen whale with a brain in excess of 5000 grams. Both show a marvelous variety of behaviors. Both sing (the hummingbird adds a courtship dance), defend territories and mates, raise young and migrate seasonally for long distances. The hummingbird also builds nests and solves some interesting pattern-recognition problems in finding flowers. Uncertainty over the full range of cetacean capabilities notwithstanding, there is really no justifiable metric of behavioral complexity that would account for most of the excess poundage of the whale brain.

Correcting for body weight via the "encephalization quotient" does offer some explanatory power. The assumption is essentially that some constant ratio of brain to body size is required for a "basic" behavioral repertoire, and that additional brain may be selected for more specialized or elaborate behaviors or demanding niches. In fact, those animals with high EQ's do show a wider range of behavioral complexity. Carnivores have higher EQ's than their prey; among prosimians and primates, frugivores beat out folivores, and careful parents outrank the careless. In general, the bottom-feeders of each vertebrate radiation stake out the lowest edge of the EQ range (Jerison 1973; Eisenberg, 1981; Stephan & Frahm 1988; Gittleman 1994; 1995).

Stephan and other researchers employing his data set subsequently turned to a finer grained structural analysis to see if the allometric data would support a closer mapping of behavioral capacities to specialization of brain structures. Do animals with impressive motor skills have larger-than-expected cerebellums, do nocturnal animals have larger olfactory bulbs, etc. (Stephan & Frahm 1988)? We might also hope that the reverse analysis will illuminate unsuspected structure-function relationships -- maybe all carnivorous animals have larger-than-expected entorhinal cortices, for example, implicating that structure in an unsuspected function.

On the first pass, this type of analysis proved disappointing. (Though researchers might have been more impressed with the correlational structure of the data that they uncovered - for example, in the way Hofman (1989) noted how precisely isocortex volume could be predicted from simple total brain volume). What are quite obvious are the strong positive correlations of all the individual structure volumes with brain volumes, shown in its least-processed forms in Figures 1A and B (Pirlot & Jolicoeur 1982; Jolicoeur et al. 1984; Stephan & Frahm 1988). This relationship persists even if the overall effect of brain size is removed in any number of ways, from simple ratio to statistical residuals (Finlay & Darlington 1995). What is more clearly revealed is what Stephan termed the "progression index" - that each structure has a characteristic rate of change in size with increase in brain size, with the isocortex the steepest, and basal forebrain and medulla the flattest (Figure 1C and D). This holds for even "recent" evolutionary events -- brain size regresses overall consequent to domestication across mammalian orders, with the structures with the highest progression indices regressing the most (Kruska, 1988). With this strong correlational structure in the data, accounting for around 95% of the variance in this data set (for primates and insectivores (Sacher 1970); including bats Finlay & Darlington 1995), it is not surprising that it is difficult to link variation in any individual structure to niche or behavior-specific variation. For example, contrary to one of the most obvious predictions one might make, researchers found that the cerebellum was relatively larger in the slower-flying, (larger-brained) fruit eating bats than in the acrobatic (smaller-brained) insect eating bats (Stephan et al. 1974).

Figure 1. Scaling of brain components on total brain volume

Figure 1A. Scaling of the volumes of brain components against brain size for the collection of bats, insectivores, prosimians and simians measured by Stephan and colleagues, reprinted (with permission) from Finlay and Darlington (1995). The regression lines for each structure are stepped by the constant indicated in the margin, to separate them for better visualization. All axes on this and other volume/volume regressions are log/log.

Figure 1B. Scaling of the olfactory bulb on brain size for the same set of mammals depicted in (A). Note that unlike the other brain parts, the both the slope and intercepts differ substantially for each taxonomic group.

Figure 1C. Regression lines for the volumes of neocortex, diencephalon, and medulla against total brain volume, this time adjusted with new constants (indicated to the right of the structure name) so as to place each intercept at the origin so that slopes may be more directly compared.

Figure 1D. Replot of graph (B) adjusted to the origin to show more clearly that the increase in olfactory bulb volume for greater brain volume is lower for simians and prosimians compared to insectivores and bats.

Successes in the attempt to link more specific brain structures to behavior and niche have been found mostly with respect to the olfactory bulb and associated limbic structures. The variation in the olfactory bulb with respect to total brain size (Figure 1B,D) is correlated with the volume of a number of other features of the limbic system. The olfactory bulb is smaller overall in simians at any brain size than in prosimians, and both simians and prosimians show a flatter slope of increase of olfactory bulb size with brain size than do insectivores and bats. These lineage descriptions map onto the nocturnal and diurnal niches inhabited by these radiations in a fairly direct way (Barton et al. 1995). Aquatic carnivores such as otters, cetaceans (Oelschlaer & Kemp 1998) and semi-aquatic insectivores (Bauchot & Stephan 1968; Gittleman, 1991) all show reduced olfactory bulb size, so lability in olfactory bulb size is not restricted to the primate lineage.

There are three major grounds for arguing that Figure 1A exaggerates the appearance of high correlation among the sizes of various brain structures. First, the correlations that are visually inferred from Figure 1A are part-whole correlations, since a brain part is plotted against the whole brain. Part-whole correlations are well-known to be exaggerated, often substantially. Second, much of the association visible in Figure 1A may be produced by the correlation of each brain part's size with body size. Third, correlations across species can be exaggerated by treating species as independent units, ignoring the fact that a single evolutionary split, such as the appearance of primates, can affect many species. All these concerns are addressed in Section 3.

2.2 Factor analytic approaches

The importance of the olfactory bulb and associated limbic structures has been demonstrated most clearly in various multivariate approaches to the Stephan and related data sets. Using factor analysis, several investigators (Sacher, 1970; Gould 1975; Holloway, 1979; Pirlot, 1987; Finlay & Darlington, 1995) have found two primary factors, the first associated with brain size, and the second an olfactory bulb factor loaded not only on the olfactory bulb but on a number of other limbic structures. Similar covariation can be seen in domesticated mammals (Kruska, 1988). In stepwise discriminant analysis, these two measures could be used with close to perfect accuracy to discriminate simians, prosimians and insectivores (Gould 1975). As we will describe in more detail later, a "limbic factor" also accounts for substantial variation in the structural development of the brain (Kaskan et al., in press; Clancy et al. 1999). A third factor, accounting for about an order of magnitude less of the variance, and associated variously with body size, the medulla and the cerebellum, has also been described (Sacher, 1970; Fox and Wilczynski 1986; Pirlot, 1987; Finlay et al.1998). The fundamental two factor structure of allometric brain growth, with the isocortex the most highly-loaded component, also tends to associate greater-than-expected isocortex size with various behavioral capacities in primates, such as social group size (Barton, 1996; Dunbar and Bever 1998) or tactical deception (Whiten and Byrne 1988).

2.3 Moving closer to functional systems

A fair criticism of all of this work is that the units of brain studied are so large and intrinsically multifunctional that it is unsurprising that few direct behavior-to-brain part links have been made. For that reason, we and a number of other investigators have turned our analysis to explicitly-defined functional systems that cut across the parts of brain segmentation used in the overall analysis of Stephan's data. Like other investigators, using the more detailed analyses of Stephan and associates (Stephan et al, 1981; 1982; 1987; Baron et al. 1983; 1988; 1987; 1990; Frahm et al. 1982; 1984a; 1984b), we have been unable to capture any more of the correlational structure of the data set by defining, for example, all visual system structures, or all motor structures, or all auditory structures and seeing if a separate fit for those designated structures can improve the variance accounted for (with the sole and striking exception of the limbic factor). The same is true for the component structure of timing of brain development which we will discuss subsequently (Finlay & Darlington 1995). Alternatively put, neither comparative structure of adult nor developing brains reveal a covarying unit, distributed across structures, that is the "visual system" or the "motor system." In an elegant analysis of the relationship of "dexterity" (hooves to hands), Nudo and Masterton (1990) found that the amount of isocortex that was the origin of the corticospinal tract was in fact positively correlated with dexterity, but that that amount in turn was associated with total isocortex size, which accounted for all of the correlation. The result of a recent detailed analysis of the subcortical auditory system (Glendenning & Masterton 1998) was striking for the conservatism it showed in system organization across species, and for the lack of support for overall increase in the relative size of the auditory system in species one might guess to be "more auditory."

2.4 Variation in single structures

Considering the dimension of size alone (we will return later to organizational changes in brains), increase in the size of individual structures has often been linked to special behavioral capacities. The size of hippocampus has been linked to range size in mammals (Jacobs & Spencer 1994) and food caching in birds (Sherry et al. 1992). The particular social system of anthropoid apes has been postulated to be associated with relative enlargement of anterior thalamic nuclei (Armstrong et al. 1987). Within the somatosensory representation of the isocortex, the increased size allotted to specialized sensory organs (whether vibrissae, trunks, tongues or hands) has long been the most notable example of nervous system divergence, in the context of a general mammalian plan for isocortical somatosensory organization (Krubitzer, 1995). The sizes of components of the song system in birds have been linked to such variables as repertoire size, though this story presents interesting complexities (Nottebohm & Pandazis 1981; DeVoogd et al. 1993; Airey 1999). Sex differences (which figure in both of the examples above) are far and away the most fertile area for locating size differences in nuclei associated with a range of behaviors from direct motor control to parental care (Sengelaub 1989). A virtual industry thrives in attempting to link isocortex size, cell density, fissurization patterns and corpus callosum size with sex differences in cognitive abilities in humans, with ambiguous results (for example, Witelson et al. 1995; reviewed in Bishop & Wahlsten 1997).

One study of special note underscores the importance of knowing the absolute size of a niche/brain effect, which is often not underlined in such studies. Barton (1996) showed consistent covariation in primates between neocortex size and social group size with several other factors controlled. However, the largest difference in neocortex size that Barton observed that could be attributed to social group size was below 3%, although the range in neocortex size in his data set (from Microcebus murinus to Gorilla gorilla) was over 400:1. 3% falls in the unaccounted variance we described previously (Finlay & Darlington 1995). For our purposes, the size of the effect is very relevant for understanding what mechanism might produce the observed difference, whether it be an increase in the number of neurons or an increase in dendritic arborization. To show that the effect on size is small is not to discount the effect, but to try to understand its context-for example, experience-related effects on cortex volume of rats, including social enrichment, are in the 5-10% range (Rosezweig, 1972).

The coverage of structures and functions in the above-listed sample is far from systematic. Overall, though, it is quite clear that while individual structures may show behaviorally-linked variation, coordinated structural variation in spatially distributed functional systems is rarely greater than the natural coordination of the whole brain. We will further discuss how to integrate these various kind of variation into a general picture of the development structure of brain change below, after consideration of some central issues in statistical analysis of this type of data.
 

3. Documenting the case for coordination of brain-part sizes

Figure 1A illustrates our claim that except for the olfactory bulb, major brain structures grow or shrink together in evolution. The analyses in this section are designed to make that point while controlling for three factors that admittedly affect Figure 1A: part-whole correlations, correlation with body size, and the use of species rather than evolutionary lineages as the units of analysis.

In this discussion we refer to the same data set used by Finlay and Darlington (1995). This is a collection of measures by Heinz Stephan and colleagues of body size and the sizes of 11 brain structures for 131 mammalian species: 40 insectivores, 43 bats, 21 prosimians, and 27 simians including humans. Whenever we refer here to four "taxonomic groups," we mean those four groups. Of course these groups represent only a small fraction of all mammalian orders, so any conclusions we draw are subject to later modification.

In our first analysis we tried to find the simplest or most parsimonious way to accurately predict the sizes of brain structures from other variables. We used three sets of "other variables": other brain structures, body size, and taxonomy. All predictions were made by simple or multiple regression. All size variables (of the 11 brain structures and body) were used in natural log form. Taxonomy was represented by dummy variables distinguishing among the four taxonomic groups mentioned above.

Since we never predict the size of any structure from the size of the whole brain, we avoid the problem of part-whole correlations. If variation were produced primarily by a structure’s association with body size, then body size would contribute. (Later we deal with the criticism that body size tends to contribute less to regressions because of intraspecific variation and measurement error.)

If variation in some structure were produced primarily by the evolutionary splits among our four taxonomic groups, then the taxonomic variables would contribute heavily to the regression. It might be objected that these groupings represent only the grossest taxonomic distinctions among these 131 species, but we respond that this is actually an advantage for "taxonomy" in our analyses, since we measure the contribution of each set of variables in relation to the number of variables in the set. Thus "taxonomy" has the best chance of making a good showing in our analyses if we represent it by just the few most important taxonomic variables, as we have opted to do. If variables measuring differences among orders, or between simians and prosimians, cannot make an impressive showing in our analysis, it seems clear that more subtle taxonomic variables, measuring differences among families or genuses, will be even less important.

Thus except for the problem of intraspecific variation in body size, this analysis addresses all three of the issues raised in the final paragraph of Section 2.1 and the opening paragraph of this section. We conclude from this analysis that by far the most useful predictors of structure sizes are the sizes of other brain structures. We don't deny that taxonomy and body size add some predictive power, but we argue they are minor factors in comparison to the sizes of other brain structures.

For each regression we computed the familiar MSE or mean squared error. Perhaps confusingly, MSE is not actually the mean of the squared errors. Rather it is defined as

The subtraction in the denominator essentially corrects for the fact that the sample sum of squared errors will always decline as new predictors are added to a regression model, even if those predictors are useless in the population. Thus it is reasonable to compare the MSE values of complex models (models with many predictors) to those of simpler models.

Rather than work with MSE itself, we find it more reasonable to work with the standard error of estimate, denoted SEE, and defined as SEE = ÖMSE. (For readers with older browsers, the equation is SEE=SQRT(MSE).) SEE has the advantage that it is measured in the same units as the dependent variable. When this variable is the natural log of a structure size, then an SEE value of, say, 0.1 means that the structure size is predicted with a typical error of about 10% of the true size.

"Parsimony" is measured by the number of parameters fitted in a regression. This includes one regression slope for each variable in the regression, plus the additive constant.

For each of our 11 brain structures, we predict the structure in 7 different ways:

Figure 2.  Relative accuracy of 7 ways to predict sizes of 11 brain structures.

These 7 models are ranked here by complexity, with the most parsimonious first (Figure 2). Model 1 has just two parameters: the additive constant, and the slope for body size. Model 2 has 8: the two just mentioned, for each of 4 taxonomic groups. Model 3 has 11: 10 slopes plus the additive constant. Model 4 has 12. Model 5 has 15: the same 12 as model 4, plus 3 for variables distinguishing among taxonomic groups. Model 6 has 44: 11 for each of 4 groups. Model 7 has 48: 12 for each of 4 groups.

As mentioned earlier, the subtraction in the denominator of MSE makes it scientifically meaningful to compare the SEE values of complex models with those of simple models. If a more complex model has a smaller SEE value than a simpler model, that is not a statistical artifact but rather suggests that the more complex model is genuinely more accurate.

Even though it is mathematically possible for a more complex model to have a larger SEE value than a simpler model, that almost never happened in this data set, and any such effects were very small. Thus it is generally true that the very best predictions are made by the most complex models. Since the two most complex models differ from simpler models mainly in their inclusion of many interactions involving taxonomic terms, this suggests the existence of evolutionary adaptations specific to particular taxa.

We give two illustrations of this general point. The first concerns the relation between the olfactory bulb and the other structure most closely related to it as measured by correlation in size across species: the paleocortex. These two structures are very highly correlated in size in insectivores and in bats, and in a scatterplot the points for the two orders fall essentially on top of each other. But prosimians show both smaller olfactory bulbs than would be predicted from their paleocortex sizes, and a lower correlation between the two than is observed in bats or insectivores. Simians show both these tendencies to a much larger degree.

The second illustration concerns body size in bats. When one predicts neocortex size from body size in each of our four taxonomic groups, three of the regression slopes are nearly identical, but the slope for bats is about 1.5 times that of the other groups. This seems to suggest that every additional ounce of weight is more important for a bat than for other orders, or in terms of its relation to the neocortex, every 2% increase in body size is about equal in importance to a 3% increase in other orders. Our faith in this interpretation is increased by the fact that a similar effect is observed when body size is predicted from olfactory bulb size: 3 of the 4 within-group regression slopes are almost equal, but the bat slope differs widely, such that a variable of 1.5*log(body size) has almost the same regression slope relative to olfactory bulb size as log(body size) has in other orders.

These examples illustrate the ubiquity and complexity of evolutionary adaptation. We have little doubt that many similar interesting findings still lie hidden in this rich data set. However, we now turn to our original question: When the importance of each group of variables is measured by its contribution to regression fit per parameter added to a model, which groups of variables are most important?

The answer to this question appears in Figure 2. Here we show SEE when each of the 11 brain structures is predicted by each of the 7 models described above. We have plotted SEE on a logarithmic scale because as SEE gets smaller, each additional 10% drop in SEE shows up as a smaller and smaller change in SEE itself. But each such drop appears equally large when SEE is plotted on a log scale.

Figure 2 is complex and must be taken in doses. The overall picture is that prediction of the olfactory bulb and medulla each follow their own unique patterns, while the remaining 9 structures share a single pattern. Temporarily ignoring the olfactory bulb and medulla, as one moves from left to right (that is, from simple models to more complex ones) by far the largest drop in SEE values occurs between Models 2 and 3. This is associated with a rather small change in model complexity: from 8 to 11 parameters, or switching from a model using body size, taxonomic group, and their interaction, to a model that completely ignores body size and taxonomic group in favor of other brain structures. As the figure shows, further declines in SEE do occur after model 3, but the only noticeable declines are associated with very large increases in model complexity, as one moves from 11 parameters to 44 or 48 parameters. Thus for these 9 structures, model 3 seems to represent by far the best combination of accuracy and parsimony. As mentioned earlier, neither model 3 nor any other model gains any unfair advantage from the use of part-whole correlations, since each brain part is always predicted only from other brain parts.

The olfactory bulb follows a different pattern, in which the largest drops in SEE occur in moving from model 1 to 2, from 4 to 5, and from 5 to 6. All three of these moves involve just the addition of taxonomic variables, either as linear terms or as interaction terms. Thus taxonomic variables are clearly most important for the olfactory bulb.

The medulla also has its own pattern. For it, the decline in SEE is not concentrated in just a few parts of the graph; almost every increment in model complexity produces a decline in SEE. This suggests that other brain structures, body size, and taxonomy are all important determinants of medulla size.

3.1 Finding the two best predictors for each structure's size

These points can be illustrated in still another way. We created 10 dummy or indicator variables to represent every possible taxonomic division of our four taxonomic groups. Four of the 10 variables were coded 1 for one of the 4 groups and 0 for the other 3 groups. Each of the other 6 variables was coded 1 for a pair of groups--e.g., bats and simians--and 0 for the other two groups (insectivores and prosimians in this case). Thus the 10 variables together represented every possible division among our 4 groups, whether reasonable or not in an evolutionary sense.

Consider now the problem of predicting logged diencephalon size, across all 131 species, as accurately as possible by regression from just two variables, where the two can be chosen from any of the following: (a) logged body size, (b) the logged sizes of the other 10 brain structures, (c) the 10 dummy taxonomic variables just described. We predicted logged diencephalon size from every possible pair of these 21 variables, and found it to be predicted most accurately from the logged sizes of the mesencephalon and neocortex.

We performed a similar analysis for every one of the 11 brain structures, finding the best two variables for predicting it from the other 21 variables just described. Even though taxonomic variables constituted essentially half of the pool of variables, a taxonomic variable was selected just once: the paleocortex and simian variables formed the best pair of variables for predicting olfactory bulb size. Body size also appeared only once: body size and mesencephalon size formed the best pair of variables for predicting medulla size. For every brain structure except medulla and olfactory bulb, the best two-variable prediction was made from the sizes of other brain structures, even though those variables always formed just 10 of the 21 variables in the predictor pool.

3.2 More on body size

Any attempt to demonstrate such an effect must allow somehow for the well-known fact that the correlation of body size with other structures is reduced because body size is more subject to random individual variation than the sizes of brain structures (Harvey & Krebs 1990). But this argument cannot explain the pattern of correlations mentioned at the end of the previous section, since it cannot explain why body size would so consistently correlate highly with medulla size.

We also addressed the body-size theory, with its random variation corollary, through a two-stage least squares procedure. First we predicted body size, across the 131 species, from the sizes of the 11 brain structures. We took this estimated value as a measure of body size with random variation largely removed. This is clearly not a perfect measure of that parameter, but fortunately this measure is biased against a positive finding in our subsequent analysis. We then examined the partial correlations among the three telencephalic structures (neocortex, striatum, diencephalon) with this measure partialed out. Across all 131 species the lowest of these three partial correlations was .90. When the same partial correlations were computed separately in our four taxonomic groups, the lowest value in each group was: Insectivore .91, Bat .77, Prosimian .88, Simian .87. All these values have been rounded down to two digits. Thus there is clearly a high association among the sizes of these three structures that cannot be explained by their common association with body size, and which exists even within the four taxonomic groups we studied. Like our other analyses, this analysis avoids problems with part-whole correlations. And the claim that these associations are produced by a few evolutionary splits cannot explain why these relations appear so consistently in each of our four taxonomic groups.

3.3 A three-factor model

Given that other brain structures are the most useful predictors for all structures except the medulla and olfactory bulb, can we come up with a plausible model of brain structure based on just those variables? We repeat that one does lose some real predictive power by such simplification, but describing such a model nevertheless appears useful.

We propose a three-factor model of brain structure. Each of these three factors is anchored, and indeed defined, by a single structure or object. In order of importance for predicting the sizes of various brain components, the three factors are a telencephalic factor anchored by the neocortex, a limbic factor anchored by the olfactory bulb, and a somatic factor anchored by body size.

In our model, each factor also has one or two secondary structures. The telencephalic factor has the striatum and diencephalon as secondary structures, the olfactory factor has the paleocortex, and the somatic factor has the medulla. This leaves 5 structures (hippocampus, cerebellum, schizocortex, septum, and mesencephalon) not clearly identified with any one factor. Even the secondary members of each factor are influenced at least slightly by the other factors. The primary members, by definition, are not so influenced, since the factors are simply defined as those structures.

The utility of the three-factor model is illustrated by some simple statements. In each of our four major taxonomic groups, body size correlates more highly with medulla size than with any other structure size, and olfactory bulb size correlates more highly with paleocortex size than with any other structure size. Nevertheless, in each group, each of the three telencephalic structures correlates more highly with the other two telencephalic structures than it correlates with either body size or olfactory bulb size.
 

4. Striking conservation of the order of events in development and its significance for brain evolution

We now turn attention back to the first observations of Jerison and Stephan, which must rank among the most significant in neuroscience:

1. Increased general "encephalization" is associated with greater behavioral complexity.
2. Enlargement of brain parts is highly coordinated and predictable.
3. Different brain components enlarge with different slopes compared to brain size.

There is no doubt that this regular enlargement of the brain carries a cost to the organism, and must confer some benefit; the metabolic rate of brain tissue is nine times higher than the average mass-specific rate of the human body (Aiello & Wheeler 1995). A curious feature of primate evolution is that the basal metabolic rate does not increase as it should with increasing brain mass. To balance the energetic budget, another expensive tissue needed to be sacrificed, and that appears to have been intestinal length. Less elaborated intestines require that higher quality food be found, which in turn might require greater memory for foraging strategy, or coordinated predation. By whatever evolutionary route this enlargement occurred, it suggests that the amount of brain mass is important for terrestrial mammals, and there is every reason to think the brain should be configured optimally for use. (For very large marine mammals, interestingly, the amount of brain tissue produced obligatorily by its relationship to body size may become insignificant compared to body mass, and it may be possible for large cetaceans to carry more brain than they "need".)

The fundamental force driving change in brain size is the number of neurons, coupled with the amount of body the neurons must innervate, directly or indirectly. The somal diameters of some cells, particularly those with long-range connections, varies somewhat with brain size - for example, the pyramidal cells of the cortex of an elephants are a little more than twice as large as those of a mouse. The volume of their connections varies substantially more (Purves, 1988). The volume of connections and requirements for supporting glia and vasculature, when they have been studied, can be seen as a regular functions of total neuron number (Murre & Sturdy 1995). We have chosen to concentrate on the factors that control the number of neurons in phylogeny. The number of cells in a structure can increase either via change in the rate at which precursor cells are produced, or increase in the length of time over which they are produced.

We have investigated both of these possibilities, beginning with changes in duration, measured by determining the peak of "neuronal birthdays" in the structure or cell group under consideration. Early in the development of the nervous system, each precursor cell located in the ventricular zone divides and produces two daughter cells, each of which can further divide. These symmetric divisions produce precursors whose numbers increase at an exponential rate. The birthday of a neuron is said to occur when a precursor cell divides "asymmetrically" and the resulting cell migrates from its initial position in the ventricular zone of the neural tube to a distant position, where it differentiates into a neuron. The time from conception to the peak of neuronal birthdays in a structure is a measure of the duration of cytogenesis for that particular structure. The longer peak birthday is delayed, the more precursor cells can be produced, which will increase the size of the particular structure. Therefore, if a single structure in the brain were to gain more cells by this method, its peak of neurogenesis would be delayed.

We initially examined data on peak neurogenesis for 51 separate structures or cell groups in 7 laboratory animals for which neuronal birth dates were adequately known. Our strategy was to find the simplest possible linear model to capture the alterations of schedules of neurogenesis across species, which ranged in duration of neurogenesis from less than 20 days in the hamster to over a hundred in rhesus; residual variation unaccounted for by this model would be the potential source of variation in brain development that might produce "brain speciation". This model is Y = ln(PC days -7) = SP + ST (later work increasing the number of species and events altered the constant to 5.37; Darlington et al., 1999) . That is, we derived from our data a score for each species (SP) that conveniently represents its duration of neurogenesis, and a score for each structure (ST) that represents the characteristic order of generation of the structures across species. (We found we predicted log of days better than days; the subtraction of the constant essentially finds the true zero for the scale of early neural events, 5-7 days after conception.)

The correlation between observed and predicted Y-values is .988, indicating that the order of neurogenesis is very precisely conserved across the animals we initially examined (four rodents, the possum, cat and rhesus macaque). This result is of course quite consistent with the prior analyses of extreme conservation of the relative size of brain components, and also with a prior analysis limited to the visual system by Robinson and Dreher (1990). Yet the strength of the relationship is remarkably robust.

4.1 The limbic system

The two-component structure that appears in the analysis of structure sizes -- a whole brain factor versus a limbic factor expressed most strongly in simians and prosimians -- was reflected in detail in the pattern of neurogenesis. Note in figure 3A how the slope of increase in size for the two example structures of the hippocampus and paleocortex is higher in insectivores and lower in simians, while for isocortex, the pattern is reversed. In our original paper, we noted that the onset dates for neurogenesis of limbic system structures were systematically advanced in the rhesus compared to the rat (Finlay & Darlington 1995). More recently, using an extended data set that includes humans (Darlington et al., 1999; Clancy et al., in press), we found that we consistently overestimated the time of developmental events in the limbic system for primates (in this data set, the rhesus macaque and humans) and underestimated the time of events associated with generation and wiring up of the isocortex (Figure 3B). Modification of the model for primates produced a better fit (Clancy et al. 1999).

Figure 3. Scaling of cortical and limbic volumes with total brain volume in insectivores and simians compared to relative time of generation.

The fact that the one consistent deviation from whole-brain predictability was the limbic system, and that neurogenesis is altered in primates in the predicted manner to produce a smaller limbic system, supports our proposal of the way development can be altered to produce larger or smaller structures. In mammals, the distributed components of the limbic system can all be labeled with a marker protein associated with that system ("LAMP" -- Levitt 1984). This marker raises the possibility that a single molecular signal might modulate neurogenesis in a number of spatially separated cell groups. No such coordinating molecular marker has ever been found for any other functionally-defined system, like the visual, auditory or motor systems. Nor do these systems show any evidence of coordinated neurogenesis. The isocortex, unlike the limbic system, is not distributed spatially, and the problem of its coordination is less complicated. The terminal neurogenesis of its component neurons might be systematically delayed in primates by a known molecular mechanism, such as later expression of the symmetry-breaking Notch protein (Cepko et al. 1996).

4.2 Late equals large

The most important finding in the comparison of schedules of neurogenesis and brain size is the simple relationship between how structures increase in relation to brain size (the slopes of the curves in Figures 1A and C) and the order of neurogenesis. Structures whose neurons are born late get disproportionately large as absolute brain size increases. The reason for this can be appreciated in schematic figure 4. The three events A, B and C, when transformed from the schedule of the mouse (left) to the schedule of a monkey (center) undergo a nonlinear rearrangement. The neural precursors for the last structure C are in production for about 80 days in the monkey versus 18 in the mouse. Note that this assumes that the onset of precursor production is fairly synchronous across species, and that the rate of neurogenesis is similar in this set of eutherian mammals. Both of these assumptions have empirical support, as we will discuss more fully later.

Figure 4. Lengthening of the duration of neurogenesis predictably affects numbers of neurons in comparable structures. Reprinted from Finlay, Hersman and Darlington, 1998, Journal of Comparative Neurology.

Schematic representation of time of peak neurogenesis of three brain structures A,B,C for the mouse, monkey and a marsupial. The mouse and monkey move on the same curve of neuronal production, and the monkey's extended neurogenesis causes late-generated structures to be enlarged predictably. Marsupials move along this same function to produce brains of different sizes, but the pace of neuronal production is slowed overall.

4.3 Extension of the developmental model to more types of developmental events and more species.

Recently, in collaboration with Dunlop, (Darlington et al., 1999), we increased the number of eutherian (placental) species studied from 6 to 9 (including humans), the number of metatherian (marsupial) species from 1 to 6, and the number of developmental events on our scale from 51 to 94. Using data published by Robinson & Dreher (1990); Ashwell et al. (1996) and Dunlop et al. (1997), we expanded the list to include not only neurogenesis, but also other commonly measured events like cell death, axon extension, formation of commisssures, and segregation of overlapping projections. First, we found that there was no difference in the predictability of neurogenesis versus those events associated with forming and refining connections: both proceeded on the stereotyped plan we had already described for neurogenesis. Second, the rate of neuronal production was characteristically slower in marsupials, which require considerably longer periods of neurogenesis to produce comparably-sized brains than do placental mammals (Figure 4) The event scale derived from eutherian data also works for marsupials, though the latter group exhibits a different relation between scale values and time. The marsupial slowdown is particularly marked for later developmental events. We think it quite likely that slowed development will not only be characteristic of metatherian mammals, but also will be found for single species or related groups with slowed maturation among eutherians.
 

5. Deep structure in nervous system scaling: From lampreys to language

One conclusion of this work is that big isocortices may be spandrels-- byproducts of structural constraints for which some use is found later (Gould 1997). The reason that the isocortex gets unusually large, and the medulla stays unusually small, is that the former is simply produced later by the rules of brain development. None of the intrinsic virtues of isocortex, such as the oft-cited laminar structure or multimodality, caused its disproportionate enlargement.

Why is the order of neurogenesis is the way it is? Is it a chance vestige of the sequence of events in early mammalian divergence, or it is orchestrated by something deeper? To address these questions, we made use of a fairly recent recharacterization of the segmental structure of the forebrain: the prosomeric model (Finlay et al. 1998). Here the segmental relationships of the topologically tortuous forebrain (Rubenstein et al. 1994) are found from the the expression of homeobox genes and others. The prosomeric model demonstrates segmentation with respect to the initial neural tube, giving every mature structure a location on the anterior to posterior and dorsal to ventral (or basal to alar) axis of that tube. Everything that arises from a single location on the neural tube is a single point in an analysis of prosomere-based patterns of neurogenesis, which is somewhat different from the cell-group and nucleus-based analysis we have described so far.

For example, for any tangential location in the isocortex, events of interest extend from the first generation of the subplate to layers 2-3; for the retina they extend from the first retinal ganglion cells to rods. For some areas of the neural tube, most notably the nuclear regions of the neuraxis, we were less likely to know all the neural progeny of a single neural tube region. Still, we included what was available, albeit with the awareness that total duration of neurogenesis is likely to be underestimated due to missing data for parts of the neuraxis giving rise to diverse nuclei, as in the diencephalon. Figure 5 shows a schematic of the total duration of terminal neurogenesis plotted onto the anterior-posterior and basal-alar axes of the prosomeric model, using combined data from the rat and monkey normalized into a single time frame.

Figure 5 Duration of neurogenesis is predicted by position on the embryonic neural tube. Reprinted from Finlay, Hersmann and Darlington, 1998 Journal of Comparative Neurology

Termination of neurogenesis by location anterior to posterior and alar to basal in the neural tube organized according to the prosomeric model, combining data from the rat and the monkey. Structures plotted are: (posterior to anterior, basal to alar) Row 1, Cranial motor nuclei, cranial sensory nuclei, vestibular nuclei. Row 2, Inferior olivary nuclei. Row 3, Cochlear nuclei. Row 4, Pontine nuclei. Row 5, Locus coeruleus, deep cerebellar nuclei, purkinje cells, granule cells. Row 6, Red nucleus, substantia nigra, raphe complex, inferior colliculus, superior colliculus. Row 7, Lateral geniculate nucleus, medial geniculate nucleus. Row 8, Ventrolateral geniculate nucleus, reticular nuclei. Row 9, Amygdala, dentate gyrus, granule cells, CA-1-2. Row 10, Globus pallidus, caudoputamen, subplate and cortical layers 2-6. Row 11, Anteroventral, anteromedial , and anterodorsal nuclei, suprachiasmatic nucleus, ventroposterolateral and ventrobasal nuclei, retinal structures, magnocellular basal forebrain, preoptic nucleus, nucleus accumbens, subicular structures, septal nuclei, olfactory structures, entorhinal cortex.

A substantial proportion of the extreme conservation of timing we see in mammalian development (but not all!) may therefore be referred to the basic spatial organization of the two neuraxes. This organization certainly precedes mammals, and in part, it precedes vertebrates. Language locates itself in the human brain in the progeny of the part of the neuraxis that could have been predicted to become unusually large by its position in the first jawless vertebrates. So, the pattern of duration of neurogenesis we see derives in large part from a very fundamental feature of the intrinsic organization of the embryonic neural plate. We should emphasize that this is not a claim that every vertebrate has the same sequence of neurogenesis - there are in fact some interesting transpositions of this order in different vertebrate radiations, resulting in some quite divergent brain structures. Much, however, is conserved.
 

6. A closer look at the relationship between birthdays and structure size

The foregoing analysis has taken a broad view of the relationship between birthdate of a structure and its pattern of size change with brain size across species. Our general claim is that if we know the order of neurogenesis of any class of structures in one species, we should be able to predict the pattern of relative change in size of the nuclei for any other set of mammals, whether or not the species are closely related. This should be useful, since there is excellent and systematic knowledge about neurogenesis in the rat from the work of Bayer and Altman (1987). The rat data can be compared to extant allometric data for a variety of other species to see how our predictions hold up on a smaller scale of brain nuclei. Using "found" data to explore this hypothesis is instructive, though differences in nomenclature of neuroanatomical regions used by different investigators makes comparisons somewhat imprecise and precludes much statistical analysis. We will discuss two examples from the neurogenetic and allometric literature we could locate.

6.1 Volumetric reconformation of the amygdala in primates and insectivores compared to generation of the amygdala in the rat.

The amygdala is a complex structure with multiple subdivisions. Bayer (1980) has studied the genesis of subdivisions of the amygdala and gradients within those subdivisions in the rat in great detail, while Stephan, Frahm and Baron (1987) have studied changes in the relative volumes of corresponding subdivisions of the amygdala across insectivora and primates. The amygdala shows notable volumetric alteration across brains, and can be divided into two regions, the centromedial group versus the corticobasolateral region. The latter increases much more steeply in volume with brain size than the centromedial group - it comprises only 52.4% of the amygdala in insectivores versus 81.1% in Homo sapiens. The order of birthdates in the rat predicts this relative enlargement. A "weighted mean" for the birthdate of these regions was derived from the Bayer data by averaging over nuclei, with each nucleus weighted by its relative volume in the rat. (The same order of neurogenesis is conserved in the monkey, though the manner of data presentation does not permit the semi-quantitative analysis done here (Kordower et al. 1992)). For the rat, the mean peak day of neurogenesis for the lesser scaling centromedial group was 14.7, while the mean peak day for the corticobasolateral group was 16.1. This is a large difference, considering that the complete range of all amygdalar birthdays in the rat is embryonic day 13 to 20.

6.2 Changes in the relative volumes of thalamic nuclei in hominoids versus genesis of the same nuclei in the rat.

Armstrong has done an extensive quantitative study of the changes in the relative volumes of thalamic nuclei in hominoids. With a sample including two gibbons, one gorilla, one chimpanzee and three human brains, she looked at which nuclei increased in volume at greater, equal, and lesser rates than the volume of the remaining thalamus (Armstrong 1979a; 1979b; 1980; 1981; Armstrong et al. 1987). The goal of these studies was to map thalamic nucleus volume onto social niche. In this analysis, we ask if the rate of change in volume of the thalamic nuclei can be predicted simply by their relative order of neurogenesis in the rat (Altman & Bayer 1988a; 1988b). A plot of the peak day of neurogenesis in the rat versus the slope of increase of the volume of each nucleus with respect to thalamus size for the nuclei studied in both cases is shown in Figure 6. There is a strong positive relationship between the two, with the partial exception of the medial geniculate nucleus, which increases in size at a greater slope than predicted. Because the pulvinar and lateral posterior nuclei appear to derive from different embryonic origins in the rodent, ape and human, it was necessary to omit those structures from this analysis (Rakic & Sidman, 1961; Ogren & Rakic 1981).

Figure 6. Late equals large: relative enlargement of thalamic nuclei in anthropoid apes is predicted by the order of neurogenesis of the same nuclei in the rat.

Order of generation of thalamic nuclei was derived from the published work of Bayer and Altman. The relative increase in size for the same nuclei with respect to changing thalamic volume was determined from the published work of Armstrong, as cited in the text. MGb, medial geniculate body (primary auditory); LGb, lateral geniculate body (primary visual); VB, ventrobasal nucleus (primary somatosensory); VL, ventrolateral nucleus (motor); AD, anterodorsal nucleus (limbic);MD, mediodorsal (frontal).

6.3 A caveat about the scaling of layered structures like the retina and isocortex.

The relative numbers of cells in individual laminae of layered structures have proven to have a more complicated relationship to birthdate than "late makes large," though they fit the pattern generally. In these cases (Cepko et al. 1996), a single progenitor area in the ventricular zone with multipotent precursor cells gives rise to a region that changes in both its tangential extent and its depth as overall brain size changes. In addition, the rate of cell proliferation and the mechanism that assigns a cell to a "type", which can be synonymous with layer, may be functionally dissociated (Artavanis-Tsakonas et al. 1999; Cepko 1999). The changing geometry of a layered population is complex, though it can and is being modeled. A full discussion of the processes connecting cell proliferation to cell type is beyond the scope of this review.

The relationship of extended proliferation of an area of the neural tube to the variable size of the resulting structure is thus best applied to whole nuclei or brain divisions that derive from identifiable areas of the neural tube, including the divisions first analyzed in the Stephan data set, or the thalamic nuclei described above.
 

7. Functional considerations and the structure of brain growth: return to whales and hummingbirds

In the preceding sections, we have basically described four kinds of growth of the brain that can be statistically distinguished and are supported by both the developmental and allometric literature. The first is the coordination of brain size with body size: the brain enlarges predictably with body size producing no change in "EQ", and perhaps little or no increase in behavioral complexity, such as memory capacity, formal problem solving capacity, etc. (tigers vs. housecats). Second, the entire brain may grow while body size is held constant, increasing EQ. This is the type of growth that has been associated with generally enhanced behavioral and cognitive abilities (Jerison, 1973). In both of these two cases, the brain appears to grow in the coordinated, predictable, but nonlinear manner we have described with late generated structures increasing in size at a greater rate than early generated ones (though it will be interesting to see if these two patterns may be distinguished in any further ways). The first type of growth may produce size differences of many thousandfold. The smallest shrew brain in our data set is .0584 grams, while a baleen whale has a brain over 5000 grams--a ratio of about 100,000. The second type of growth generally appears to produce ranges of about 10-fold. This not only was the case for the insectivore/primate/bat data set we analyzed, but also seems to be generally true for all the vertebrate radiations (Northcutt, 1981). The third type of growth is that we have described for the limbic system, whose size may vary relative to the brain as a whole, even with brain and body size held constant.

The fourth type is the independent variation of individual brain parts, the "unaccounted-for variance" of our model. Including all sources of error, the variation of individual structure size at any particular brain size did not exceed two- to threefold. Deviations in the large collection of auditory nuclei observed across species by Glendenning and Masterton (1998) were of this nature - rather than wholesale enlargement or reduction of auditory nuclei, particular nuclei deviated from the rest in different species. Induced experiential effects on brain weight and cortical thickness fall generally in the range of 5-10% (Rosenzweig, 1971) and could account for some of the variation observed in this range. Independent variation of individual brain parts has often associated with a specific behavioral advantage, like foraging ability (Jacobs and Spence, 1994). The "limbic factor" presents a puzzle here. The limbic factor is most often associated with the loss of olfactory ability in principally diurnal primates, and not any direct behavioral advantage. Since the structures influenced by the limbic factor are decidedly not all olfactory (such as the hippocampus), it would be interesting to determine if there is any behavioral advantage conferred by other features of a large limbic system. Taking the example of foraging rodents above, one might predict that an insectivore might do better at spatial memory in foraging than a comparably-brain-sized small primate.

Our ideas bear some similarity to the views of Aboitiz 1996, though they also differ in certain ways. Aboitiz distinguishes between "passive" and "active" evolutionary brain growth. He suggests that passive growth is associated mainly or entirely with increase in body size, produces little or no increase in processing capacity, and follows a "conservative allometric rule." Active growth is in response to cognitive selection, may affect just one or a few parts of the brain, and increases computational power. He makes other distinctions between the two types of growth that are not germane to the present discussion.

We believe Aboitiz has made an important contribution, but our own views differ from his in at least three ways. First, we agree that a brain part, or the entire brain, may grow as a secondary consequence of some other growth, such as body growth. However, we hold that neural material so added may later be pressed into service--an important feature of a "spandrel"--even if it takes a very long time to do so. The term "adjunct" growth may capture this property more fully than "passive." Second, we hold that adjunct growth of brain parts is not always adjunct to body growth. Rather, the entire brain may grow in response to evolutionary demand for greater processing power, producing adjunct growth in many individual parts of the brain. Such growth may be associated with some increase in body size but not be adjunct or secondary to that increase. Rather, the causal relation may go the other way--a larger body may be needed to safely carry such a large brain, making body growth secondary to brain growth. Third, we are not as convinced as Aboitiz seems to be of the significance of growth in just one brain part or system (other than the limbic system, which seems to be a special case). There is such variation of individual parts, as we have described above. As we have said before, this is substantial in one sense, but is still trivial in comparison to the many-thousand-fold variation in structure sizes produced by other mechanisms.

We have given here a mechanistic account as to why the brain grows in a coordinated fashion, and why some parts of the brain increase in size at the expense of others. We have no definitive answer as to why this is an advantageous way to organize the brain, but we will offer a speculation.

Why do all sensory systems and brain parts scale so resolutely together? Why do whales and hummingbirds have so similar a range of capacities? Perhaps the answer lies in the distributed nature of sensory processing. If each sensory and cognitive system is a separate "module" that produces some sort of calculated output, like an "on" or "off" decision, its size does not matter, as long as it influences the circuit. However, in widely distributed systems where many sensory and computed systems are feeding input into a decision, a minority input, in terms of simple volume of connections, might fail to retain any voice. Therefore retinal neuron number must scale with the number of neurons in the skin, or the number of neurons in the cochlea with all the multiple sensory inputs to structures like the superior colliculus, to remain influential in making orienting decisions.

Conversely, functions do migrate within the brain with the changing size of brain parts -- the best example of this is the "corticalization" of many functions that are carried out in the midbrain in smaller-brained animals that become dominated by the isocortex in primates. We argue that the development of the brain acts to keep some kind of volumetric parity between sensory and motor components of its input (unless deletion of one is actually desirable, as in the olfactory/limbic system). In contrast, development is non-committal about the location in the brain where computations should occur, and lets those volumes vary widely with respect to each other. Better understanding of these dynamics will only be possible when we know much more about the properties of distributed neural systems.
 

8. Conclusions and implications: A legacy of evolvability

There are several ways in which the results of this work might seem counter-intuitive. The relative metabolic expense of neural tissue would seem to make adjustment of global growth contingencies an inefficient mechanism for increasing the size of particular subsystems. Insofar as the adult brain is functionally differentiated, it seems reasonable to expect selection pressures to act directly upon areas most relevant to adaptive behaviors. Indeed, one might expect the energetic demands of brain tissue preferentially to disentangle the developmental fates of these systems, insofar as organisms with brains that are just big enough, in just the right ways, might bear the advantage of lighter metabolic loads over the globally-boosted competition. Gould articulated the present consensus when he argued "The concept of 'mosaic evolution'...refuted the notion of harmonious development by affirming that individual organs could have independent phyletic histories, despite the evident correlation of parts within any organism. Correlations are no more immutable than species themselves" (Gould 1977).

While correlations may not be immutable, they appear to be surprisingly resilient with respect to brain size evolution. Indeed, the model presented here is entirely consistent with a number of important theoretical frameworks of current developmental biology. First, it underlines the relevance of ontogenetic processes to any deeper understanding of how organisms evolve; natural selection does not do its work on some equipotent substrate, but on a complex mechanism with a history of previous change that makes some adaptations more "workable" than others. As Gould also noted, heterochrony is a "pervasive phenomenon among evolutionary processes" precisely because it is such a productive mechanism for working disparate changes on the developing organism, from recapitulation (immature descendants resemble adult forms of their ancestors) on the one hand, all the way to paedomorphosis (mature animals resemble their immature ancestors) on the other. As we have seen, heterochrony in the evolution of brain size is manifested in adjustment of a relatively simple non-linear function -- or linear with respect to Y = ln(days - 5.37) -- that determines the timing of terminal neurogenesis. This adjustment has almost exactly predictable effects along the course of differential growth described by the prosomeric model. Though such non-linear functions are seldom the first models considered by researchers, they are commonly seen in natural processes and arguably under-utilized by students of development (Elman et al. 1996).

Another way of thinking about historical contingencies on evolutionary change is to acknowledge that not only physical and behavioral traits are under selection, but also the processes that produce the traits. Developmental mechanisms that are both robust and flexible are often in the best position to "solve" adaptational problems, with the consequence that evolution tends to conserve those mechanisms. This is the essence of von Baer's explanation for the similarity of embryonic forms treated by Haeckel's "biogenetic law." More recently, notably in Gerhart and Kirshner (1997), the concept has gone under the term evolvability, or the capacity of organisms to transcend ontogenetic constraints by conserving robust and flexible developmental mechanisms. Examples include the near-ubiquity of the pax-6 transcription factor implicated in development of eyes as structurally disparate as Drosophila and human, or the almost endless variations on a theme afforded by the versatility of actin-based cytoskeletons in the evolution of sperm cells. Indeed, evolutionary biologists are becoming increasingly aware that analyzing physical traits atomistically, as independent objects of selection, fails to account for the consequences of the developmental process. McCollum (1999), for instance, presents an account of how the facial morphology of robust australopithecines is affected by alteration of a single key component. In this case, dental morphology is held to have induced wide-ranging effects on the modeling of the entire cranium.

8.1 Cortical enlargement and multiple representations

One of the surprising revelations of recent neuroscience has been the apparent multiplicity and redundancy of sensory and motor representations in isocortex (Kaas, 1989). For example, instead of a single cortical map of the visual world, the visual processing system appears to have at least several. These range from the "basic" functions of Brodmann's Area 17 (V1) through the somewhat more integrative representations in Areas 18 and 19, toward the parallel processing facilities of the ventral ("what") and dorsal ("where") streams. Additional processing areas are also suspected, including extrastriate connections that may be responsible for subconscious perception or "blindsight," and maps for particular objects, as in suspected object-centered representations in Areas 7a and LIP (Ungerleider & Haxby 1994; Weiskrantz 1996; Olson & Gettner 1996). The persistence of the phantom limb phenomenon in young amputees and even subjects born without particular limbs has been used to argue for multiple and distributed body representations in iso- and allocortex, thalamus and the limbic system that survive local somatosensory reorganization (Melzack et al. 1997). Even emotional affect has been attributed to parallel streams for primary and secondary emotions involving the amygdala and ventromedial areas of the right frontal lobe (Damasio 1994; LeDoux 1995).

What are we to make of nature's penchant for engineering multiple representations based on slightly different processing needs? It is conceivable that each processing stream "grew" its own cortical domain based on the importance of its function to survival-- that there existed some early versions of the brain that were, say, good at "what" or "where" visual processing but not both, or perhaps not so good at either until selective pressure expanded the computational resources available to each. It is conceivable, but not plausible. Following our suggestion that structure leads function, it would be our contention that the form of these sensory, motor and cognitive systems are the result of competitive recruitment of processing resources from a super-abundant pool of cortical neurons made available more or less at the same time. This super-abundance may go some way toward explaining the redundancy of certain representations: in the developing brain, the limiting factor is not space, but the task of solving formidable processing challenges in a finite amount of time. The brain's solution - massively parallel processing - looks very much like the strategy adopted by computer engineers with a complex problem to master (grandmaster-level chess, realistic modeling of weather) and a multitude of cheap semiconductors to do it with.

8.2 Implications for hominid evolution

If we accept this account, certain deeply entrenched dispositions in most theorizing about human brain evolution may need to be revisited. At the risk of caricature, much speculation on this topic makes the explicit or implicit presumption that some behavioral challenge, such as finding food, using language, learning to manipulate social competitors, etc., led to functions that took up residence in the 400 gram brains of our australopithecine or early Homo ancestors. The imperative to perform these functions better drove the evolution of bigger and better facilities to serve them. The result was presumably a kind of "mosaic evolution" characterized by differential hypertrophies of the physical subsystems essential to humanness. Indeed, one could infer the behavioral repertoire by taking the relative measure of these hypertrophies: Neanderthal endocasts, for instance, supposedly showed smaller frontal lobes compared to modern human brains of the same size (but see Holloway 1985). This appraisal seemed to justify the long-standing judgement of Marcellin Boulle that Neanderthals were capable of only "vegetative or bestial" preoccupations, or more recent claims that these recent ancestors were hampered by "expedient" or "15-minute cultures" marked by poor planning depth (Hayden 1993; Stringer and Gamble 1993; Mithen 1996; Noble and Davidson 1996).

The advent of a "prix fixe" over the old "Chinese menu" model of brain size evolution alters the scenario. Instead of function dictating the evolution of structure, additional structure preceded enhanced function in hominid brains. To be sure, adaptation has subsequently "tailored" each subsystem to the processes that tend to take up residence in them. On balance, however, the current model posits a far greater role for exaptation of structure to function in the natural history of the brain.

Based on estimates of endocranial capacity in fossil specimens, the expansion of the hominid brain appears to have proceeded in stepwise fashion from a baseline close to African apes approximately 4 million years ago to the modern average around 1350 grams (Harvey and Krebs 1990). Ambiguities in assigning "typical" body sizes to ancestral forms, and Homo habilis in particular, makes brain/body size ratios fairly speculative. While some within-species variation has been observed over the long career of Homo erectus, the most dramatic jumps in mean endocranial capacity occurred first between the australopithecines and the later variants of H. habilis (500 to 750 grams) at about 2 million years, and between H. erectus and archaic H. sapiens sometime around 400,000 years (about 1000 to about 1250 grams).

Based on data presented in a recent review of endocranial capacity and estimated body size for a number of hominid species (Wood and Collard 1999), regression analysis of mean brain volume on body mass alone accounts for some 70% of the variance (p<0.001). By far the largest residual (more than 2.2 standard deviations) is associated with modern humans. If the data for moderns is removed from the set, adjusted R² rises to 91.2%. This suggests that the great majority of the brain size increase from australopithecines to Neanderthals is a straightforward function of body mass. Only with the appearance of anatomically modern humans did brain size become somewhat disproportionate.

The advent of lithic technologies around the time of the first growth spurt and of complex tool reduction techniques concurrent with the second have quite naturally been implicated with cognitive enhancements made possible by larger brains. Indeed, the assumption that cognitive function renovated hominid brain structure has become something of a fixture in the paleoanthropological literature (for instance, Deacon, 1997 and Mithen, 1996). Paradoxically, this notion has become yet another way to set humans apart from the rest of the animal kingdom: where evolution is something that happens to other organisms, in all the important senses humanity authors itself (cf., for instance, the tradition from Childe 1951 ["Man Makes Himself"] to Kingdon 1993 ["Self-Made Man"]).

The current study suggests that a different emphasis is in order. As all the vaunted capabilities of modern brains are fairly recent developments, their location in isocortex is a straightforward consequence of the fact that that area is the latest brain structure to undergo terminal neurogenesis, and therefore in the best position to provide the additional processing capacity for those functions. The exponential growth of isocortex relative to the rest of the brain is due to its prosomeric location, not to accelerating technological demands "remaking" the brain.

Most importantly there is no reason to presume selection pressures for cortically-based functions drove brain expansion at all. As we have argued, the brain grows as a covarying whole, increasing in size according to a fairly straightforward log function. It is just as likely, therefore, that pressures for enhanced archicortical, corticoid, or sub-cortical processing could have triggered the adjustment of global timing constraints that led, incidentally, to much bigger isocortices. Such demands on subcortical processing could have had little or nothing to do with the suite of cognitive traits (language, advanced tool-making) we prefer to think of as distinctly human. The advantage conferred by an enhanced motor control via the basal ganglia, for instance, or a more responsive amygdala in regulating affective function, or a bigger hippocampus for memory of fruiting trees or water, would have done just as well in driving hominid encephalization. The cognitive traits would have been fortuitous by-products, afforded by the "spandrel" of greater isocortical capacity.

This formulation has further implications for our expectations of finding behavioral correlates of bigger brains. One of the current puzzles of paleoanthropology, for instance, is the apparent gap between the appearance of anatomically-modern humans some 100,000 years ago, and the advent of unequivocally modern behavior. The latter is conservatively marked by new lithic blade industries in Europe around 45,000 years ago (and probably earlier in Africa), and more liberally by the over-water peopling of Australia, at around 60,000 years. If function leads structure in brain evolution, then why are modern-looking people with modern-size brains not acting modern for some 40,000 years (Klein, 1989)?

One answer might be that the archaeological evidence for advanced behavior at deeper time ranges has not been found yet. Alternately, and in accord with our argument here, it is possible that early modern brains reached near-current dimensions for reasons unrelated to modern cognitive functions. Only later, after "extra" cortical volume had become exapted to the physical correlates of modern behavior, would the full panoply of familiar functions have appeared. We would therefore expect to find evidence for exactly such an anatomical/behavioral gap.

The broad and tight correlations between mass increases all over the brain have implications for primates and every other mammalian order. In moving away from essentialism in our thinking about brain structure and function, we are alive to a wider range of causal scenarios, and perhaps to an understanding of brain evolution that goes deeper than a few millimeters of cortex.

This paper was prepared with support from NIH Grant R01 19245 and NSF/INT grant 96-04599 to B.Finlay, and an NIMH predoctoral research fellowship to N. Nicastro (T32 MN19389). We thank Marcy Kingsbury, and Michael Spivey for their helpful comments on the manuscript, and Jennifer Judson for the preparation of the figures.

Aboitiz, F. (1996). Does bigger mean better? evolutionary determinants of brain size and structure. Brain Behavior and Evolution 47:225-245. Aiello, L.C., & Wheeler, P. (1995). The expensive-tissue hypothesis: the brain and digestive system in human and primate evolution. Current Anthropology 36:199-221.

Airey, D. (1999) Song bird neuroanatomical correlates of song behavior: individual differences and evolution Doctoral thesis Cornell University

Altman, J., & Bayer, S.A. (1988a). Development of the rat thalamus. I. Mosaic organization of the thalamic neuroepithelium. Journal of Comparative Neurology 275:346-377.

Altman, J., & Bayer, S.A.. (1988b). Development of the rat thalamus. II. Time of origin and settling pattern of neurons derived from the anterior lobule of the thalamic neuroepithelium. Journal of Comparative Neurology 275:378-405.

Armstrong, A. (1979a.) A quantitative comparison of the hominoid thalamus. I. Specific sensory relay nuclei. American Journal of Physical Anthropology 51:365-382.

Armstrong, A. (1979b.) A quantitative comparison of the hominoid thalamus. II. Limbic nuclei anterior principalis and lateralis dorsalis. American Journal of Physical Anthropolog y 52:43-54.

Armstrong, A. (1980). A quantitative comparison of the hominoid thalamus. III. A motor substrate- the ventrolateral complex. American Journal of Physical Anthropology 52:405-419.

Armstrong, A. (1981). A quantitative comparison of the hominoid thalamus. IV. Posterior association nuclei -- the pulvinar and lateral posterior nucleus. American Journal of Physical Anthropolog y 55:369-383.

Armstrong, E., Clarke, M.R. & Hill, E.M. (1987). Relative size of the anterior thalamic nuclei differentiates anthropoids by social system. Brain Behavior and Evolution 30:263-271.

Artavanis-Tsakonas, S., Rand, M. D.& Lake, R. J. (1999). Notch signaling: Cell fate control and signal integration in development. Science 284:770-776.

Ashwell, K. W. S., Waite, P. M. E. & Marotte, L.. (1996). Ontogeny of the projection tracts and commissural fibres in the forebrain of the tammar wallaby (Macropus eugenii): timing in comparison with other mammals. Brain Behavior and Evolution 47:8-22.

Baron, G., Frahm, H.D. , Bhatnagar, K.P & Stephan, H. (1983). Comparison of brain structure volumes in insectivora and primates: Main olfactory bulb (MOB). Journal für Hirnforschung 24:551-558.

Baron, G., Frahm, H.D. & Stephan, H. (1988). Comparison of brain structure volumes in Insectivora and Primates VII. Vestibular Complex. Journal für Hirnforschung 29:509-523.

Baron, G., Stephan, H. & Frahm, H.D. (1987). Comparison of brain structure volumes in Insectivora and Primates VII. Paleocortical components. Journal für Hirnforschung. 28:463-477.

Baron, G., Stephan, H. & Frahm, H.D. 1990. Comparison of brain structure volumes in Insectivora and Primates VIII. Trigeminal complex. Journal für Hirnforschung 31:193-200.

Barton, R. A. (1996). Neocortex size and behavioural ecology in primates. Proceedings of the Royal Society of London (B) 263:173-177.

Barton, R.A., Purvis, A. & Harvey, P.H. (1995). Evolutionary radiation of visual and olfactory brain systems in primates, bats and insectivores. Philosophical Transactions of the Royal Society of London (B) 348:381-392.

Bauchot, R. & Stephan, H. (1968). Etude des modifications encephaliques observees chez les insectivores adaptes a la recherche de nourriture en lilieu aquatique. Mammalia 30:160-196.

Bayer, S.A. (1980.) Quantitative 3H-Thymidine radiographic analysis of neurogenesis in the rat amygdala. Journal of Comparative Neurology 194:845-876.

Bayer, S.A., & Altman, J. (1987). Directions in neurogenetic gradients and patterns of anatomical connections in the telencephalon. Progress in Neurobiology 29:57-106.

Bishop, K. M., & Wahlsten, D. (1997). Sex differences in the human corpus callosum: Myth or reality? Neuroscence and Biobehavioral Reviews21:581-601.

Byrne, R. W. (1995). Primate cognition: comparing problems and skills. American Journal of Primatology 37:127-141.

Cepko, C. L. (1999). The roles of intrinsic and extrinsic cues and bHLH genes in the determination of retinal cell fates. Current Opinion in Neurobiology 9:37-46.

Cepko, C. L., Austin, C. P., Yang, X. J. & Alexiades, M. (1996). Cell fate determination in the vertebrate retina.Proceedings of the National Academy of Sciences USA 93:589-595.

Childe, V.Gordon (1951) Man Makes Himself (Mentor, NY)

Clancy, B., Darlington, R.B. & Finlay, B.L. (1999). The course of human events: predicting the timing of primate neural development. Developmental Science.

Damasio, A. (1994) Descartes' Error (Avon, New York)

Darlington, R.B., Dunlop, S.A. & Finlay, B.L. (1999) Commentary: Neural development in metatherian and eutherian mammals:variation and constraint. Journal of Comparative Neurology 411: 359-368.

Deacon, T.W. (1997) The Symbolic Species: The co-evolution of language and the brain (W.W. Norton, New York)

DeVoogd, T., Krebs, J.R., Healy, S.D. & Purvis, A. (1993). Relations between song repertoire size and the volume of brain nuclei related to song: comparative evolutionary analyses among oscine birds. Proceedings of the Royal Society of London B 254:75-82.

Dunbar, R. I. M., & Bever, J. (1998). Neocortex size predicts group size in carnivores and some insectivores. Ethology 104:695-708.

Dunlop, S. A., Tee, I. B. G., Lund, R. D. & Beazley, L. D. (1997) Development of primary visual projections occurs entirely postnatally in the fat-tailed dunnart, a marsupial mouse, Sminthopsis crassicaudata. Journal of Comparative Neurology 384:26-40.

Eisenberg, J.F. ( 1981). The Mammalian Radiations: An Analysis of Trends in Evolution, Adaptation and Behavior. Chicago: The University of Chicago Press.

Elman, J.L. et al. (1996) Rethinking Innateness: A Connectionist Perspective on Development (MIT, Cambridge)

Finlay, B. L., & Darlington, R. B.. (1995). Linked regularities in the development and evolution of mammalian brains. Science 268:1578-1584.

Finlay, B. L., Hersman, M. N. & Darlington, R. B. 1998. Patterns of vertebrate neurogenesis and the paths of vertebrate evolution. Brain, Behavior and Evolution 52:232-242.

Florence, S. L., Jain, N. &Kaas, J. H. (1997). Plasticity of somatosensory cortex in primates. Seminars in Neuroscience 9:3-12.

Fox, J.H., & Wilczynski,, W. 1986. Allometry of major CNS divisions: towards a reevaluation of somatic brain-body scaling. Brain, Behavior and Evolution 28:157-169.

Frahm, H.D., Stephan, H. & Stephan, M.. (1982). Comparison of brain structure volumes in Insectivora and Primates I. Neocortex. Journal für Hirnforschung. 23:375-389.

Frahm, H.K., Stephan, H. & Baron, G.. (1984a). Comparisons of brain structure volumes in Insectivora and Primates: V. Area striata. Journal für Hirnforschung 25:537-557.

Frahm, H.K., Stephan, H. & Baron, G. (1984b). Comparisons of accessory olfactory bulb volues in the common tree shrew (Tupaia glis). Acta Anatomica 119:129-135.

Gerhart, J. & Kirschner, M. (1997) Cells, Embryos, and Evolution (Blackwell, Malden MA)

Gibson, E.J., & R.D. Walk. 1960. The "visual cliff". Scientific American 202:64-71.

Gilbert, C. D., Das, A. , Ito, M. , Kapadia, M. & Westheimer, G. (1996). Spatial integration and cortical dynamics. Proceedings of the National Academy of Sciences USA 93:615-622.

Gittleman, J.L. (1991). Carnivore olfactory bulb size: allometry, phylogeny and ecology. Journal of Zoology, Lond. 225:253-272.

Gittleman, J.L. (1994). Female brain size and parental care in carnivores. Proceedings of the National Academy of Sciences USA 91:5495-5497.

Gittleman, J.L. (1995). Carnivore brain size, behavioral ecology and phylogeny. Journal of Mammology 67:23-36.

Glendenning, K. K., & Masterton, R. B. (1998). Comparative morphometry of mammalian central auditory systems: Variation in nuclei and form of the ascending system. Brain Behav Evol 51:59-89.

Gould, S. J. (1975). Allometry in primates, with emphasis on scaling and the evolution of the brain. In Approaches to Primate Paleobiology, edited by Szalay. Basel: Karger.

Gould, S.J. (1977). Ontogeny and Phylogeny. Cambridge, Massachusetts: Harvard University Press.

Gould, S. J.( 1997). The exaptive excellence of spandrels as a term and prototype. Proceedings of the National Academy of Sciences USA 94:10750-10755.

Harvey, P.H., & Krebs, J.R. (1990). Comparing brains. Science 249:140-146.

Hayden, B. (1993) The cultural capacities of Neanderthals: a review and re-evaluation, Journal of Human Evolution 24:113-146

Hofman, M.A. (1989). On the evolution and geometry of the brain in mammals. Progress in Neurobiology 32:137-158.

Holloway, R. (1979). Brain size, allometry and reorganization. In Development and Evolution of Brain Size, edited by M. E. Hahn, C. Jensen and B. C. Dudek. New York: Academic Press, Inc.

Hughes, A. (1977). The topography of vision in mammals of contrasting life style: comparative optics and retinal organization. In Handbook of Sensory Physiology, edited by F. Crescitelli. Berlin: Springer-Verlag.

Jacobs, L.F., & Spencer, W.D. (1994). Natural space-use patterns and hippocampal size in kangaroo rats. Brain, Behavior and Evolution44:125-132.

Jerison, H.J. (1973). Evolution of the Brain and Intelligence. New York: Academic Press.

Jerison, H.J. (1991).  Brain size and the evolution of mind.  The 59th James Arthur Lecture on the evolution of the human brain.  New York, American Museum of Natural History.

Jolicoeur, P., Pirlot, P. , Baron, G. & Stephan, H. (1984). Brain structure and correlation patterns in insectivora, chiroptera and primates. Systematic Zoology 33:14-29.

Kaas, J.H. (1989) The evolution of complex sensory systems in mammals. In Principles of Sensory Coding and Processing, edited by S. B. Laughlin. Cambridge, UK: The Company of Biologists, LTD.

Kingdon, J. (1993) Self-Made Man: Human Evolution from Eden to Extinction? (John Wiley & Sons; NY)

Klein, R. (1989) The Human Career (University of Chicago Press, Chicago)

Kordower, J.H., Piecinski, P. & Rakic, P. (1992). Neurogenesis of the amygdaloid nuclear complex in the rhesus monkey. Developmental Brain Research 68:9-15.

Krubitzer, L. (1995) The organization of neocortex in mammals: are species differences really so different? Trends in Neuroscience 18:408-417.

Kruska, D. 1988. Mammalian domestication and its effect on brain structure and behavior. In Intelligence and Evolutionary Biology, edited by H. J. Jerison & I. Jerison. Berlin Heidelberg: Springer-Verlag.

LeDoux, J.E. (1995) Emotion: Clues from the Brain, Annual Review of Psychology 46:209-235

Levitt, P. (1984). A monoclonal antibody to limbic system neurons. Science 223:299-301.

Martin, R.D. (1982). Allometric approaches to the evolution of the primate nervous system. In Primate Brain Evolution: Methods and concepts, edited by E. Armstrong and D. Falk. New York: Plenum Press.

McCollum, M. A. (1999) The robust Australopithecine face: A morphogenetic perspective, Science 284:301-305

Meltzoff, A. N. (1996). The human infant as imitative generalist: a 20-year progress report on infant imitation with implications for comparative psychology. In Social Learning in Animals: The Roots of culture. New York: Academic Press.

Melzack, R. et al. (1997) Phantom limbs in people with congenital limb deficiency or amputation in early childhood, Brain 120:1603-1620

Mithen, S. (1996) The Prehistory of the Mind (Thames & Hudson, NY)

Murre, J. M. J., & Sturdy, D. P. F. (1995). The connectivity of the brain: multi-level quantitative analysis. Biological Cybernetics 73:529-545.

Noble, W. & I. Davidson (1996) Human Evolution, Language and Mind: A Psychological and Archaeological Inquiry (Cambridge University Press, Cambridge)

Northcutt, R.G. (1981) The evolution of the telencephalon in nonmammals. Annual Review of Neuroscience 4:301-350.

Nottebohm, F., Kasparian, C. & Pandazis, S. (1981). Brain space for a learned task. Brain Research 213:99-109.

Nudo, R.J., & Masterton, R.B. (1990). Descending pathways to the spinal cord IV: Some factors related to the amount of cortex devoted to the corticospinal tract. Journal of Comparative Neurology 296:584-597.

Oelschlager, H. H. A., & Kemp, B. (1998). Ontogenesis of the sperm whale brain. Journal of Comparative Neurology 399:210-228.

Olson, C.R. & Gettner, S.N. (1996) Brain representation of object-centered space, Current Opinion in Neurobiology, 6:165-170

Ogren, M.P., & Rakic, P. (1981). The prenatal development of the pulvinar in the monkey: 3h-Thymideine autoradiographic and morphometric analyses. Anatomy and Embryology 162:1-20.

Pallas, S.L. (1990). Cross-modal plasticity in sensory cortex: visual response properties in primary auditory cortex in ferrets with induced retinal projections to the medial geniculate nucleus. In The Neocortex: Ontogeny and Phylogeny, edited by B. L. Finlay, G. Innocenti & H. Scheich. New York: Plenum Press.

Pirlot, P. (1987). Contemporary brain morphology in ecological and ethological perspectives. Journal für Hirnforschung. 28:145-211.

Pirlot, P., & Jolicoeur, P. (1982). Correlation between major brain regions in Chiroptera. Brain, Behavior and Evolution 20:172-181.

Purves, D. (1988) Body and Brain: A Trophic Theory of Neural Connections Cambridge, Ma.: Harvard University Press.

Rakic, P., & Sidman, R.L. (1969). Telencephalic origin of pulvinar neurons in the fetal human brain. Zeitschrift für Anatomie und Entwicklungsgeschichte. 129: 53-82.

Robinson, S.R., & Dreher, B. (1990). The visual pathway of eutherian mammals and marsupials develop according to a common timetable. Brain, Behavior and Evolution 36:177-195.

Rosenzweig, M.R. (1972) Effect of environment on development of brain and behavior. In: The Biopsychology of Development edited by E. Tobach, L.R. Aronson and E. Shaw. New York: Academic Press

Rubenstein, J.L.R., Martinez, S. , Shimamura, K. &Puelles, L. (1994). The embryonic vertebrate forebrain: the prosomeric model. Science 266:578-579.

Sacher, G.A. (1970). Allometric and factorial analysis of brain structure in insectivores and primates. In The Primate Brain, edited by C. R. Noback & W. Montagna. New York: Appleton-Century-Crofts.

Sengelaub, D.R. (1989). Cell generation, migration, death and growth in neural systems mediating social behavior. In Advances in Comparative and Environmental Physiology, edited by J. Balthazart. Berlin: Springer-Verlag.

Sherry, D.F., Jacobs, L.F. & Gaulin, S.J. C. (1992). Spatial memory and adaptive specialization of the hippocampus. Trends in Neurosciences 15:298-302.

Stephan, H., Baron, G., & Frahm, H.D. (1982). Comparison of brain structure volumes in insectivora and primates. VII. Accessory olfactory bulb. Journal für Hirnforschung 23:571-595.

Stephan, H, Frahm, H. & Baron, G.. (1981). New and revised data on volumes of brain structures in insectivores and primates. Folia Primatologica 35:1-29.

Stephan, H., Frahm, H.D. & Baron, G. (1987). Comparison of brain structure volumes in Insectivora and primates. VII. Amygdaloid components. Journal für Hirnforschung 28:571-584.

Stephan, H., Baron. G., & Frahm, H.D. (1988). Comparative size of brain and brain components. In Comparative Primate Biology: Alan R. Liss.

Stephan, H., Pirlot, P. & Schneider., R. 1974. Volumetric analyses of pteropid brains. Acta Anatomica 87:161-192.

Stringer, C. & Gamble, C. (1993) In Search of the Neanderthals (Thames & Hudson, NY)

Sur, M., Pallas, S.L. & Roe, A.W. (1990). Cross-modal plasticity in cortical development: differentiation and specification of sensory neocortex. Trends in Neurosciences 13:227-233.

Ungerleider, L.G. & J.V. Haxby (1994) 'What' and 'where' in the human brain, Current Opinion in Neurobiology, 4:157-165

Van der Loos, H., & Welker, E.. (1985) Development and plasticity of somatosensory brain maps. In Development, Organization, and Processing in Somatosensory Pathways; Neurology and Neurobiology, edited by M. Rowe and J. W.D. Willis. New York: Alan R. Liss.

Weiskrantz, L. (1996) Blindsight revisited, Current Opinion in Neurobiology, 6:215-220

Williams, R. W., R. C. Strom, and D. Goldowitz. (1998). Natural variation in neuron number in mice is linked to a major quantitative trait locus on Chr 11. Journal of Neuroscience 18:138-146.

Whiten, A. and Byrne, R.W. Tactical deception in primates. The Behavioral and Brain Sciences, 11: 233-273

Witelson, S. F., Glezer, II, & Kigar, D. L. (1995). Women have greater density of neurons in posterior temporal cortex. Journal of Neuroscience 15:3418-3428.

Wood, B. & Collard, M. (1999) The Human Genus, Science 284:65-71