The first evidence for functional localization came from medical data; patients with lesions in specific regions of the brain did not degenerate fully – rather, quite circumscript psychological functions were altered by those lesions. Broca's analysis of the question rendered him historical fame in the controversy, and his name was lent to Broca's Area, which is linked to language functions. Broca's area comprises Brodmann's areas 44 and 45 of cerebral cortex; the neurocytologist Brodmann, thus, provided independent evidence for modularity by describing form differentiation in the brain.
The cerebral cortex is divided in cytoarchitectural regions that are usually labeled by Brodmann's descriptions. This striking feature of the cerebral cortex is found everywhere in the brain: this structure is not an homogeneous organ, but can be separated in more-or-less well-defined regions which present morphological, physiological and chemical profiles which are unique to them. In fact, what allows us to define structures in the brain is precisely those differences.
The fact that the brain presents a considerable modularity – that is, it is composed of more-or-less independent parts – is very interesting from the point of view of evolutionary biology. Modular organization favors evolvability (the ability to respond to a selective challenge) by allowing one module to change without interfering with the rest of the organism . Fisher  demonstrated that the probability of a random mutation being favorable was a steeply decreasing function of the number of traits it affected; thus, simultaneous random changes in many parts of a highly integrated structure are not likely to improve its function, as the chance improvement of one part will almost always be hampered by deleterious effects in many other parts. However, if parts are variationally independent, selection can act on them one at a time – which raises the probability that any one change in a module is an improvement.
Of course, the argument for higher evolvability in modular traits is based on a concept of modularity that is not the same concept we used so far. While we defined the modularity of the brain as a phenotypical one, Fisher's argument concentrates on generative modularity – that is, a given trait is considered to be modular in relation to another trait if both traits present different origins, developmental or genetic. Thus, if it can be shown that two traits are not under the pleiotropic control by the same generative mechanism, then they are modular.
Is there any relation between both concepts of modularity? According to Luis Puelles, yes – and this is due to developmental pathways in the brain. He postulated that embryonic modularity – that is, the specification of histogenetic fields by position-dependent expression of patterning genes – is transformed into functional modularity by translating early-generated positional information into an array of adhesive cues, which regulate the binding of functional neural structures distributed across the embryonic modules [3, 4].
The neuromeric theory of brain development and evolution
Puelles begins his theory by describing two types of modularity (which roughly correspond to the two types of modularity we defined) in the brain: embryonic modules and functional modules:
[embryonic modules] represent discrete neuroepithelial domains in the wall of the neural tube and form a mosaic-like pattern along both the longitudinal (anterior-posterior) axis and transverse (dorsal-ventral) axis of the neural tube [...]. These divisions can be called modular because they represent largely independent histogenetic units of neural tissue, in which neural cells proliferate, migrate and differentiate into characteristic neurons and glia. In this respect, they resemble embryonic modules in other parts of the body (ref. 4, p. 1100).
Thus, within embryonic modules, two sub-types can be discerned: in the longitudinal (anterior-posterior) dimension, they are the roof, alar, basal, and floor plates; in the transverse (dorsal-ventral) axis, they are called “neuromeres”. Those divisions are defined developmentally (embryologically): despite some limited exchange of signalling molecules and cells between the modules, most neurons remain in the embryonic fields where they are born . Check this figure (taken from ref. 4):
It represents a generalized map of the embryonic brain in vertebrates. The hindbrain (rhombencephalon) consists of seven rhombomeres (r1-r7) and four pseudorhombomeres (r8-r11); rostral to it, the isthmic region (is), the midbrain (mesencephalon) and the prosencephalon are found. The diencephalon consists mainly of prosomeres p1-p4; the anlage of the telencephalon (tel) and the eye are outgrowths of the secondary prosencephalon which consists of prosomeres p5-p6. The dashed line in the figure indicates the boundary between the alar plate (light gray) and the basal plate (green); the thick salmon line indicates the floor plate of the brain.
A lot of genes which are involved in transforming the early embryonic neural tube into the modular structure seen in the figure have been identified; they belong mostly to several families of transcription factors or gene regulatory proteins, many of which are known to regulate pattern formation in other parts of vertebrate and invertebrate embryos (cf. this post). The majority of those genes are expressed in restricted regions of the brain – Hox genes are expressed in the hindbrain and spinal cord , while members of the Dlx, Pax, Otx, Emx, Gbx, Wnt, Sox, Nkx, and Six families of genes are expressed in regions of the midbrain and forebrain . Of course, the domains in which those patterning factors are expressed overlap extensively; the patterns of expression is one of the most striking examples of epistasis, and resemble some sort of “code bar” for those embryonic modules.
The functional module, as defined by Puelles and his colleagues [3, 4], is represented by the neural circuits of the mature brain. They serve the purpose of particularizing information processing ; sensory modules, for example, process particular types of information (visual, auditory, somatic, and so on), being, in this sense, “modules” as defined by the cognitive scientist Jerry Fodor  (not all functional modules are modular sensu Fodor, though; see ref. 10 for a discussion). Each functional module is typically composed of parts that are derived from several embryonic divisions (figure also from reference 4):
Here, embryonic modules represent spatially separate and largely independent histogenetic fields. Each field gives rise to a coherent domain of gray matter (represented in the figure as different patterns) that is later characterized by a particular way of information processing (e.g., cortical circuits, thalamic relay, basal ganglia gating, etc.). Each domain contains several brain nuclei or regions that are connected to nuclei or regions in other domains by white matter tracts, forming the behavioral circuits of different functional modules (represented in different colors on the figure). According to Puelles, specialized functional modules can be optimally adapted either in the evolutionary history of a species or by plasticity and experience in and individual; by doing so, they can better carry out the type of information processing that is required under environmental pressure in each case.
How does the brain transforms from the initial patterning of the embryonic neural tube in the first figure to the mature form it presents in mature age? The initial pattern is translated into the expression pattern of genes that are involved in the various processes of morphogenesis and circuit formation (cell migration, sorting, and aggregation; axon elongation and fasciculation; axonal target recognition and synapse formation; and so on). This process is regulated by molecules that mediate cell-cell and cell-substratum adhesion (such as integrins and cadherins ), diffusible molecules that set up molecular gradients for cell and axon migration (such as netrins and slit ), and molecules that mediate attraction and repulsion between the neural cells and their axons and dendrites and also regulate neuronal migration (such as ephrins, Eph receptors, and neuropilin ). After this process ends, the brain is organized in gray matter domains that retain their embryonic topological relationships (this is the reason why topology is so important for homology). If a gray matter domain emerges ventral, dorsal, caudal, or rostral to another embryonic division, it will be found at that same topological position in the mature brain, despite the growth and deformation that sometimes takes place during brain development . Those gray matter domains can be either subcortical nuclei or the cortical divisions defined by Brodmann.
Ok, so Luis Puelles and his collaborators laid down a theory (along with strong evidence) for the developmental origin of functional modules in the brain. By itself, this is an incredible accomplishment: they are mainly attempting to link causally the embryonic modules to functional modules, by describing the developmental mechanisms that pattern the mature brain and its modules. What does this means, however, from the point of view of evolution?
The number, topological relation, and molecular characteristics of the neuromeres (embryonic modules) and their subsequent subdivisions is well conserved between all vertebrates; even though an extensive test of the proposal that this embryonic bauplan is homologous in vertebrates has not yet been made, data is beginning to accumulate in such a way that we can state with some confidence that it is. In contrast with the situation of embryonic modules, the mature brains of the different vertebrate species show morphological and functional differences. If one type of module is causally related to the other type, the main evolutionary question that is raised by the neuromeric model is “What is the cause of the diversity in mature brains?” The neuromeric model actually provides an answer to this question by postulating that evolution can act on the mechanisms that translate an embryonic module into a functional module. This means that the modular nature of the neuromeres should increase their evolvability, or, alternatively, the evolvability of their final fates as functional modules.
Modularity, evolvability, and canalization
Modularity enhances evolvability because it allows characters to evolve without interference. However, modularity may also hamper evolvability by reducing the number of genes that can affect the character, thereby also reducing its mutational target size . Recruiting more genes can also increase the evolvability of a character. Genes available for recruitment typically already have pleiotropic effects on other characters. Consider this example, given by Thomas Hansen:
Consider a character under directional selection. An allele that introduces a novel effect on this character may be picked up by selection and increase in frequency. This will lead to compensatory changes in the other characters affected by this gene, and eventually the new allele may go to fixation. If the new effect was acquired through the appearance of a new enhancer that expresses the gene on the character under directional selection, then almost all subsequent mutations of this gene will inherit this pleiotropic effect. Thus, through integration, the character has acquired a new source of mutational variability, which makes it more evolvable (ref. 14, p. 3).
In this example, the character is originally uncorrelated with the other traits that will be subsequently altered by the new allele. The introduction of pleiotropic effects increases the genetic variation of the traits. Since there is ample genetic variation in both traits that can compensate for the correlated changes, this increase in variability is virtually costless. As correlation increases, the genetic architecture becomes less and less able to compensate for the correlated changes, however. Eventually, the addition of further pleiotropic effects will decrease the evolvability, because the costs become too high. In the limit, as the traits become completely correlated, evolvability drops to zero. A compromise between total pleiotropy (correlation between traits 1.0) and total modularity (correlation between traits 0.0) must be reached.
Another shortcoming of total pleiotropy is that it tends to increase the strength of stabilizing selection acting on individual loci, thus reducing the amount of variability that is mantained at a locus under a balance between stabilizing selection and mutation ; this reduction further reduces the amount of variation available for response to directional selection when the environment changes. Whenever the environment is stable, stabilizing selection around optimal adaptive peaks is most common; however, when the environment is changing, traits tend to change the adaptive peak by means of random genetic drift followed by directional selection [16, 17].
Another factor that may raise evolvability is epistasis. If the epistatic interactions are random and non-directional, the effects of alleles to the response to selection will tend to cancel out; if there is a systematic directional pattern of gene interaction, a modified response to selection will emerge. Positive epistasis – where genes tend to reinforce each other's effects along the direction of selection – will accelerate the response, while negative epistasis – where genes tend to diminish each other's effects in the direction of selection – will reduce the response .
In a sense, the opposite of evolvability is canalization, defined as the degree to which a phenotypic character is “buffered” against variation in the underlying generative processes that construct it . The simple observation that, in artificial selection experiments, mutations with a major effect on a quantitative character not only change the mean value of the trait but sometimes also increase the variance compared to the wild type led Waddington, the proposer of canalization, to postulate that wild-type phenotypes are “buffered” against genetic variation. Thus, we can view any single character as more or less sensitive to genetic or environmental perturbations – with less sensitive characters being more canalized, and more sensitive characters being more evolvable. There is some evidence that the sensitivity of a trait to perturbation is correlated with its influence on fitness: the stronger the influence of the trait on fitness, the less sensitive it is to perturbations .
Canalization can be understood in terms of environmental or genetic influences – that is, a trait can be insensitive to mutations (genetic canalization) or to environmental factors (environmental canalization). Both are influenced by stabilizing selection, which favors genes that decrease environmental variance of quantitative characters . In systems under epistatic interaction, a fairly narrow window of epistatic effects allow the evolution of genetic canalization. Only to the extent that the magnitude of epistatic effects happens to fall within this window of opportunity, stabilizing selection will lead to canalization. When the strength of the stabilizing selection is too high, though, it eliminates genetic variation of the trait, which is critical for genetic canalization; thus, if genetic variation is maintained by mutation-selection balance, strong stabilizing selection can inhibit the evolution of genetic canalization . Then, at mild strengths of selection, genetic canalization is more probable; when the strength of selective pressures in stabilizing selection reaches a threshold, environmental canalization overcomes potentially deleterious pleiotropic effects of the canalized gene, and emerges as more probable.
The considerable conservation of the embryonic modules of the vertebrate brain suggestthat their formation is canalized – that is, there is little phenotypic plasticity in the formation of neuromeres; genes that alter the developmental trajectories that lead to the bauplan have little effect on the outcome (genetic canalization), and differences in concentration gradients (which produce dramatic changes in patterns elsewhere) have very little impact the bauplan . The functional modules, on the other hand, are highly evolvable; there is considerable evidence that the number, structure, and size of modules changed repeatedly in the evolution of vertebrate brains . The evolutionary quantitative genetics axioms analysed above predict that stabilizing selection would favor genetic canalization of any trait under small to mild selection strength, while stronger selective pressures would favor environmental canalization. It has been suggested that early environmental stages of a complex organism are under stronger stabilizing selection than are later stages . That models quite nicely what seems to happen in the developmental trajectories of the vertebrate brain: the early stages of development (that is, the stages leading to the formation of embryonic modules and the immediately posterior stages) are buffered against both genetic and environmental changes, while the later stages are not.
In this sense, the evolutionary stasis of the embryonic modules is an intrinsically stable state of the developmental pathway. Canalization, in this case, is an emergent property of the developmental system. Under strong stabilizing selection, both genetic and environmental forms of canalization evolve to a higher degree than under weaker stabilizing selection; in this sense, the genetic canalization observed in the neuromeric organization of embryonic brains is a side effect of selection of general developmental stability.
Canalization explains the stasis in embryonic modules, and the conditions for the high evolvability of functional modules were set. However, both systems are modular; the problem of why one would be canalized and the other highly evolvable remains.
Of course, the internal organization of the phylotypic stage (i.e., neuromeric organization of the embryo) tends to buffer it against mutation and environmental (epigenetic) effects. The phylotypic stage is so important and embedded in the organism's development that any modification is lethal; development just before this stage involves highly interdependent, nonmodular processes which are subject to mutational damage . In spite of the fact that embryonic modules are highly compartmentalized and subject to epistatic interactions – which can increase evolvability under the right conditions – they remain canalized because of the intrinsic properties of the developmental trajectories, strong stabilizing selection, or both. The embryonic modules are the stage for the wide diversity of later development that happens in different classes and orders of the vertebrate radiation; the development of each “add-on” structure which will result in the adult, mature brain is semi-autonomous once it is activated, but initially depends on signals from the embryonic modules for placement, orientation, scale and timing of the overall organizational pattern. Because the embryonic brain in the phylotypic stage is organized in neuromeres, even its high conservation through canalization is able to produce further versatility in the use of compartments and deconstraint in their formation. To account for the variability and evolvability of functional modules, we must postulate that some mechanism in the intermediary stages between the phylotypic stage and the last stages of development is responsible for “deconstraining” properties that allowed the further use of developmental trajectories in individuals of a given clade of organisms.
Size as a determinant of evolution of functional modules
Brain size is a very important variable in evolution . As brains increase in size, they increase in the number of neural centers in one or more brain regions, in the number of neuronal classes within neural centers, and in behavioral complexity . The same phenomenon is observed when one brain region increases in size independently. Thus, changes in size (either of brains or of regions) seem to be an important factor in the evolution of functional modules in the brain.
This is precisely what Redies and Puelles  hypothesized. According to this hypothesis, an increase in the size of a region may eventually generate enough space for another series of morphogenetic patterning processes to occur within that division:
This hypothesis is based on the assumption that pattern formation in the brain, as in other systems, is based on local self-activation and long-range inhibition of molecular signals with defined space and time constants. These self-organizing mechanisms may be autonomously activated in brain regions where, due to growth, the molecular gradients set up by pre-existing patterns have become exceedingly shallow or the molecular signals, which are secreted from increasingly distant signalling centers, have become very diluted [...]. These signalling centers then form new divisional boundaries and, in turn, induce further differentiation in the intervening areas. Growth of a brain region may thus be a trigger for more differentiation within that region (ref. 4, p. 1108).
For example, when FGF8, a molecule that is normally expressed at the rostral pole of the developing telencephalon, is artificially expressed at the caudal pole as well, an additional somatosensory area develops just caudal to the normal one . This new cortical area seems to be a mirror image of the “old” somatosensory map, just as most phylogenetically added brain regions seems to be mirror images of their older neighbors . Since the number of cortical areas in a given species correlates rather tightly with the total amount of neocortex available , this suggests that cortical expansion is causally related to area addition because the presence of those morphogens in areas in which they are ancestrally absent re-routes developmental trajectories .
If variability in functional modules is a consequence of the alteration of developmental trajectories by changes in brain (or structure) size, where do this leaves us? Changes in size are correlated with changes in complexity, but their genetic basis seems to be different: while functional modularity is created by the transformation of embryonic modules (which are patterned by expression of homeotic genes), brain size is controlled by genes such as microcephalin and ASPM (at least in the Homo lineage; see refs. 29 and 30). Unless we postulate a third generative mechanism for both size and complexity – which would entail perpetual regress and would not solve the problem at all! –, the high evolvability of functional modularity in brains must be explained not by epistasis or pleiotropy, but as a “spandrel” of brain size evolutionary changes. Of course, selective pressures probably turn the causal arrow in the other direction: brains do not increase in size as a result of selection for size; instead, it is the secondary consequences of enlargement – functional modularity – that are selected for. In this sense, increases in brain size form the context in which the “genetic architecture” (the number, identities, and variational properties of the genes that participate in the development of a character ) of modularity can be expressed: by increasing size, the potential for increased response to selection may unfold.
Evidence for modularity in brain evolution
We now understand how modularity (or, rather, intermediate pleiotropy) can, given the right opportunity (brain enlargement), respond quickly to directional selection. Changes in the number of brain modules appear when organisms leave their previous adaptive peaks because the environment changed; they must reach other peaks in the adaptive landscape by means of directional selection on proliferation/segregation/addition of brain areas , but this can only happen if brains increase either in their total size or in the size of a given structure. This would set the opportunity for changes in the relative size of some region. How often, though, regions increase in size independent of other regions? That is, how often mosaic evolution happens in the brain?
It is to be predicted, from the evolutionary consequences of the neuromeric model, that mosaic evolution should be quite common. Also, because most species excel at only some behaviors, and since brain regions tend to be causally linked to different behaviors, it is to be expected that functionally distinct cell groups should evolve independently from each other. Some evidence have been gathered, for example, that meadow voles (which are polygynous and have home range size as a competitive advantage in sexual selection ) present sexual dimorphism in the size of hippocampus, a trait that is not found in monogamous prairie voles . In songbirds, there is an association between song repertoire size and size of area HVC , a region that is primarily involved in song learning. However, most of those studies focused on a region of interest (hippocampus, HVC), mainly ignoring whether differences between species could also be observed in other structures.
Striedter  discussed this in chapter 5 of his Principles of Brain Evolution (reviewed here). He called the independent, non-correlated evolution of the size of a given structure mosaic, while the evolution of size in a structure that is correlated to the evolution of size of another structure is called concerted. He suggested that, while concerted evolution seems to happen a lot in the evolution of brains – suggesting evolutionary constraints  –, these constraints are sometimes “breached”, and mosaic evolution results. For example, Finlay and Darlington  proposed that some limbic system structures, such as the piriform cortex and the hippocampus, evolve in concert with the olfactory bulb; the size of the olfactory cortex correlates tightly with the size of olfactory bulb size in prosimians and insectivores. Nonetheless, if we include simians in the analysis, this correlation breaks down: simians have larger olfactory cortices than would be predicted by the size of their olfactory bulb : “This [...] suggests that some components of the limbic system became developmentally and/or functionally uncoupled from one another as simians evolved” (ref. 20, p. 151).
It appears, then, that mosaic evolution of individual brain regions has occurred in evolution, even though we do not know how frequent this phenomenon appears in comparison to concerted evolution. Barbara Finlay proposed that increases in the magnitude of 2- to 3-fold changes in size are still within the limits of concerted evolution ; in most species, the majority of differences fall within this range . This means that concerted evolution is a general principle in brain evolution “that holds most of the time but not always” [22, p. 158], as most other principles.
If concerted brain evolution is the general rule, why are there cases in which it does not hold? From what we have seen up till now, two explanations are possible. One is that concerted brain evolution is under stabilizing selection; when the environment begins to change and species must change from adaptive peaks, directional selection disrupts the tight correlation between region sizes, and those species which are able to decouple brain region size evolution occupy new niches. This hypothesis is favored by Striedter :
As the vast literature on key innovations and mass extinctions exemplifies [...] even rare events can be of profound significance in evolution. Indeed, I suspect that mosaic evolution is more likely than concerted evolution to cause major changes in brain function and, therefore, more likely to open up new ecological niches and possibilities for further change. If this is true, then mosaic evolution should be more common between classes than between orders, more common between orders than between families, and so forth. Because the frequency of mosaic evolution seems, indeed, to increase with taxonomic level [...], mosaic evolution was probably at least as important as concerted evolution when we consider vertebrate brain evolution overall [22, p. 158].
A second explanation is that increases in brain size disrupt the “developmental canalization” that is built by concerted evolution, creating the opportunity for mosaic evolution. Of course, both explanations are not mutually exclusive: if we follow the explanations delineated before for the causal chains that links brain size, development and functional modules, it is at least plausible that those species which occupied new niches did so by exploiting opportunities made by brain enlargement for mosaic evolution.
Barbara Finlay and collaborators [35, 36] have argued for developmental constraints in the guidance of correlated (concerted) evolution of brain region sizes. However, as Striedter  pointed, the existence of mosaic evolution does not imply the absence of constraints. Mosaic evolution probably happens against a background of at least some constraints – for example, the bauplan of neuromeres described by Puelles and colleagues [3, 4]. Finlay and Darlington  proposed that concerted evolution occurs when developmental schedules are lengthened (heterocrony); in the species they analysed, the order in which brain regions are “born” correlates with how rapidly they enlarge with increasing brain size: the later a region is born, the larger it becomes as overall brain size increases: “late equals large”!
Georg Striedter  proposed that any regions which are born earlier than expected in some species would end up being smaller than expected in adult brains. Also,
evolutionary changes in the initial size of a precursor region (that is, shifts in the expression boundaries of genes that give the region its identity) should lead to deviations from concerted evolution, since the model assumes that initial size remains constant [...]. [s]uch hypotheses are testable and merely contemplating them reveals that evolutionary changes in a single developmental parameter can yield mosaic evolution even as other developmental constraints remain in place [22, pp. 158-159].
The evolution of novel structures
The neuromeric model does not only predict that mosaic evolution in brain region size should be rather common; it also predicts that functional modularity itself evolves: as brains (or brain regions) increase in size, novel structures should appear. This is related to the more general problem of alometric growth in brain region size: as brain regions enlarge, one expects novel structures to appear:
as individual brain regions change in size, they tend to change in internal structure [...]. Specifically, brain regions tend to fractionate into more subdivisions, nuclei, or areas as they enlarge phylogenetically. This size-related proliferation of brain subdivisions may be due to the addition of some truly “new” brain areas or to the segregation of components that are “old”. Either way, complexity increases. That, in turn, is likely to allow different areas to specialize for different functions, leading to improved performance in at least some tasks [...]. Specifically, I suggest that regions subdivide as they enlarge because the distance over which developmentally important molecules can diffuse or interact is physically limited. [22, pp. 10-11]
How common is this process? If we compare higher taxonomic levels, evolutionary changes in complexity (i.e., number of subdivisions in the adult brain) occurred quite regularly: neuroanatomists have described considerably more cell groups in the forebrain of amniotes than in the forebrain of cyclostomes or amphibians, for example . Check this figure (modified from ref. 22):
A parsimony analysis demonstrates that forebrain complexity increased and decreased several times during the course of vertebrate brain evolution. Again, clade-wise analysis demonstrate that fractionation is a common process; there is little data on whether this happened in smaller taxonomic levels, but I suspect that these events are more common in classes than in orders, in orders than in families, and so on – just like mosaic evolution in brain size. If this is the case, then a similar mechanism can be acting on fractionation: when environmental changes pushes organisms away from the asymptotic state – that is, away from their ancestral adaptive optima – genetic drift, mutation  and gene conversion  takes the populations to “adaptive valleys”, from where they begin to “climb” (by directional selection) to a new adaptive peak. This can only happen, however, if a given region (or, sometimes, the whole brain) increases in size, which causes spatial distortions in diffusion processes of developmentally important molecules. Those distortions, in turn, will create “new” structures. It is predicted that the appearance of novel structures should be more commonly associated with concerted evolutionary changes in other regions, simply because mosaic evolution of the structure that gave rise to the new region necessitates developmental decoupling of alometric expectations – but this does not mean that this latter phenomenon cannot happen! If one allele results in decoupling and brain region enlargement, it can greatly enhance evolvability, and rapid response to selection will ensue.
 Wagner GP (1996). Homologues, natural kinds and the evolution of modularity. American Zoologist 36: 36-43.
 Fisher RA (1958). The Genetical Theory of Natural Selection. 2n edition. New York: Dover.
 Puelles L (2001). Brain segmentation and forebrain development in amniotes. Brain Research Bulletin 55: 695-710.
 Redies C, Puelles L (2001). Modularity in vertebrate brain development and evolution. BioEssays 23: 1100-1111. http://dx.doi.org/10.1002/bies.10014
 Smith-Fernandez A, Pieau C, Reperant J, Boncinelli E, Wassef M (1998). Expression of the Emx-1 and Dlx-1 homeobox genes define three molecular distinct domains in the telencephalon of mouse, chick, turtle and frog embryos: Implications for the evolution of telencephalic subdivisions in amniotes. Development 125: 2099-2111.
 Lumsden A, Krumlauf R (1996). Patterning the vertebrate neuraxis. Science 274: 1109-1115.
 Rubenstein JLR, Shimamura K, Martínez S, Puelles L (1998). Regionalization of the prosencephalic neural plate. Annual Review of Neuroscience 21: 445-477.
 Mountcastle VB (1979). An organizing principle for cerebral function: The unit module and the distributed system. In: Schmitt FO, Worden FG (editors), The Neurosciences Fourth Study Program, pp. 21-42. Cambridge: MIT Press.
 Fodor J (1983). The Modularity of Mind. Cambridge: MIT Press/Bradford Books.
 Buller D, Hardcastle VG (2000). Evolutionary psychology meets developmental neurobiology: Against promiscuous modularity. Brain and Mind 1: 307-325.
 Stoeckli ET, Landmesser LT (1998). Axon guidance at choice points. Current Opinion in Neurobiology 8: 73-79.
 Brose K, Tessier-Lavigne M (2000). Slit proteins: Key regulators of axon guidance, axonal branching, and cell migration. Current Opinion in Neurobiology 10: 95-102.
 O'Leary DD, Wilkinson DG (1999). Eph receptors and ephrins in neural development. Current Opinion in Neurobiology 9: 65-73.
 Hansen TF (2003). Is modularity necessary for evolvability? Remarks on the relationship between pleiotropy and evolvability. BioSystems 2189: 1-12.
 Lande R (1980). The genetic covariance between characters maintained by pleiotropic mutations. Genetics 94: 203-215.
 Kauffman SA (1993). The Origins of Order: Self-Organization and Selection in Evolution. Oxford: Oxford University Press.
 Wright S (1977). Evolution and the Genetics of Population. Volume II: Experimental Results and Evolutionary Deductions. Chicago: University of Chicago Press.
 Hansen TF, Wagner GP (2001). Modeling genetic architecture: A multilinear model of gene interaction. Theoretical Population Biology 59: 61-86.
 Waddington CH (1957). The Strategy of the Genes. New York: MacMillan.
 Stearns SC, Kawcki TJ (1994). Fitness sensitivity and the canalization of life history traits. Evolution 48: 1438-1450.
 Wagner GP, Booth G, Bagheri-Chaichian H (1997). A population genetic theory of canalization. Evolution 51: 329-347.
 Striedter GF (2005). Principles of Brain Evolution. Sunderland: Sinauer Press.
 Raff RA, Kaufman TC (1983). Embryos, Genes and Evolution. New York: MacMillan.
 Northcutt RG (2002). Understanding brain evolution. Integrative and Comparative Biology 42: 743-756.
 Fukuchi-Shimogori T, Grove EA (2001). Neocortex patterning by secreted signaling molecule FGF8. Science 294: 1071-1074.
 Rosa MPG (1999). Topographic organization of exastriate areas in the flying fox: Implications for the evolution of mammalian visual cortex. Journal of Comparative Neurology 411: 503-523.
 Chanzigi MA (2001). Principles underlying mammalian neocortical scaling. Biology and Cybernetics 84: 207-215.
 Striedter GF (1998). Stepping into the same river twice: Homologues as recurring attractors in epigenetic landscapes. Brain, Behaviour and Evolution 52: 218-231.
 Evans PD, Gilbert SL, Mekel-Bobrov N, Vallender EJ, Anderson JR, Vaez-Azizi LM, Tishkoff SA, Hudson RR, Lahn BT (2005). Microcephalin, a gene regulating brain size, continues to evolve adaptively in humans. Science 309: 1717-1720. http://dx.doi.org/10.1126/science.111372
 Mekel-Bobrov N (2005). Ongoing adaptive evolution of ASPM, a brain size determinant in Homo sapiens. Science 309: 1720-1722. http://dx.doi.org/10.1126/science.111681
 Lynch M, Walsh JB (1998). Genetics and Analysis of Quantitative Traits. Sunderland: Sinauer Press.
 Spritzer MD, Meikle DB, Solomon NG (2005). Female choice based on male spatial ability and aggressiveness among meadow voles. Animal Behaviour 69: 1121-1130.
 Jacobs LF, Gaulin SJC, Sherry DF, Hoffman GE (1990). Evolution of spatial cognition: Sex-specific patterns of spatial behavior predict hippocampal size. PNAS 87: 6349-6352.
 DeVoogd TJ, Krebs JR, Healy SD, Purvis A (1993). Relations between song repertoire size and the volume of brain nuclei related to song: Comparative evolutionary analyses amongst oscine birds. Proceedings of the Royal Society of London B 254: 75-82.
 Finlay BL, Darlington RB, Nicastro N (2001). Developmental structure in brain evolution. Behavioral and Brain Sciences 24: 263-308.
 Finlay BL, Darlington RB (1995). Linked regularities in the development and evolution of mammalian brains. Science 268: 1578-1584.
 Wicht H, Northcutt RG (1992). The forebrain of the Pacific hagfish: A cladistic reconstruction of the ancestral craniate forebrain. Brain, Behavior and Evolution 40: 25-64.
 Hansen TF, Carter AJR, Chui C-H (2000). Gene conversion may aid adaptive peak shifts. Journal of Theoretical Biology 207: 495-511. http://dx.doi.org/10.1006/jtbi.2000.2189