Mammalian Brains Are Made of These: A Dataset of the Numbers and Densities of Neuronal and Nonneuronal Cells in the Brain of Glires, Primates, Scandentia, Eulipotyphlans, Afrotherians and Artiodactyls, and Their Relationship with Body Mass

The availability of datasets on mammalian brains that make comparative studies possible has been instrumental for the advancement of evolutionary neuroscience. Most notable have been the datasets on the volumes of brain structures in 51 species of bats, 48 primates and 28 ‘insectivores' (currently recognized as a combination of afrotherians and eulipotyphlans) published by Heinz Stephan's group [Stephan et al., 1981a, b], on cortical surfaces and volumes for 44 mammalian species compiled by Hofman [1985, 1988], and on neuronal and glial cell densities for 11 species studied initially by Tower and Elliott [1952] and Tower [1954], and later extended to another 42 species by Haug [1987].

Although restricted in their scope to mostly structure volumes and to cell densities in the cerebral cortex, those datasets were, for a few decades, the major references for studies on brain evolution that established the basic notions that there is both concerted [Finlay and Darlington, 1995] and mosaic [Barton and Harvey, 2000] scaling across brain structure volumes in evolution, that larger brains were composed of more and larger neurons, resulting in smaller neuronal densities and increasing glia/neuron ratios in a uniform manner across species [Tower and Elliot, 1952; Haug, 1987; Stolzenburg et al., 1989; Marino, 2006], and that larger brains have relatively larger cerebral cortices but a cerebellum of constant relative size [Stephan et al., 1981a, b; Clark et al., 2001], with presumably larger relative numbers of neurons in the cerebral cortex over the rest of the brain.

Since 2005, with the development of the isotropic fractionator, a new, nonstereological method to determine the numbers of neuronal and nonneuronal cells that compose brain structures [Herculano-Houzel and Lent, 2005] that gives results comparable to those obtained with careful stereological analysis [Herculano-Houzel et al., 2015], we have been able to expand our understanding of brain evolution by examining the scaling relationships between the mass of brain structures and the number of cells that compose them. Through the analysis of 42 species of primates (including the human) [Herculano-Houzel et al., 2007; Azevedo et al., 2009; Gabi et al., 2010; Ribeiro et al., 2014], glires [Herculano-Houzel et al., 2006, 2011; Ribeiro et al., 2014], eulipotyphlans [Sarko et al., 2009], scandentians [Herculano-Houzel et al., 2007], afrotherians [Herculano-Houzel et al., 2014a; Neves et al., 2014] and artiodactyls [Kazu et al., 2014], we have been able to challenge a number of the initial notions regarding mammalian brain evolution. Specifically, we could show that while there is indeed a shared, single relationship between numbers of nonneuronal cells and the mass of brain structures across species, with relatively unchanging nonneuronal densities, neuronal densities do not vary uniformly across all species and brain structures [reviewed in Herculano-Houzel, 2011a, 2014; Herculano-Houzel et al., 2014b], that glia/neuron ratios vary with average neuronal cell size, not brain structure mass, across different brain structures and mammalian species [Mota and Herculano-Houzel, 2014], that the relationship between the number of brain neurons and body mass differs across mammalian orders [Herculano-Houzel, 2011b; Herculano-Houzel et al., 2014b], and that relatively larger cerebral cortices do not hold relatively more of all brain neurons [Herculano-Houzel, 2010; Herculano-Houzel et al., 2014b]. We could also show that the apparent uniform scaling of the energetic requirement of the brain with brain mass across species [Karbowski, 2007] is actually a spurious mathematical consequence of the apparent scaling of neuronal density across the brains included in that analysis, which conflated primates and nonprimates, then already known to have different relationships between brain mass and neuronal density [Herculano-Houzel et al., 2006, 2007]. Rather, the energetic requirement of the brain scales linearly with the number of neurons in the brain, and uniformly across rodents and primates, despite the different neuronal scaling rules that apply to these orders [Herculano-Houzel, 2011c].

The analysis of our new dataset on numbers of neurons and nonneuronal cells that compose mammalian brains allowed us to propose a new synthesis of the mechanisms of brain evolution [Herculano-Houzel et al., 2014b]. Briefly, we propose that the evolution of mammalian brains of a wide range of masses has been the result of both concerted and mosaic changes in the distribution of neurons across brain structures and in the relationship between number of neurons and average neuronal cell size (including the cell body and all arbors). In most mammalian groups, the addition of neurons to individual brain structures has been accompanied by predictable increases in the average size of neurons in each structure (as inferred from changes in neuronal cell densities), which allowed us to infer the ancestral neuronal scaling rules for each structure. From those ancestral scaling rules, we inferred that the primate cerebral cortex and cerebellum, the eulipotyphlan cerebellum, and the artiodactyl rest of brain (RoB) diverged with changes in the predicted mechanism that ties the number of neurons to the average size of the neurons generated. The distribution of neurons to the cerebral cortex and cerebellum, two structures generated by different progenitor cell populations, has varied little from what we infer to have been the ancestral mammalian rule of about 4 neurons in the cerebellum to every neuron in the cerebral cortex. At the same time, the allocation of neurons to the ensemble of these two structures has departed greatly from the inferred ancestral ratio of 2 neurons in the cerebral cortex (and 8 in the cerebellum) for every neuron in the RoB to much larger and variable ratios in primates and artiodactyls (while still maintaining the ratio between numbers of neurons in the cerebellum and cerebral cortex) [Herculano-Houzel et al., 2014b].

In the spirit of making this new body of data available for researchers with complementary interests and expertise to ours who will be able to advance the understanding of brain evolution in a much wider sense, here we provide the full dataset that we have generated on the mass and numbers of neuronal and nonneuronal cells that compose the brain as a whole and subdivided in its four major structures (cerebral cortex, cerebellum, olfactory bulb and RoB). All data have been thoroughly checked for consistency regarding the brain structures included, because of inconsistencies in a few of the original studies [Herculano-Houzel et al., 2006; Sarko et al., 2009], guaranteeing that comparisons across species are valid (for example, that numbers for ‘cerebral cortex' always include the hippocampus, and that numbers for ‘RoB' and ‘whole brain' always exclude the olfactory bulb). We also report new observations on the scaling of neuronal density with body mass that shed light on the different factors that may control cell size across brain structures.

Our full dataset consists of 42 mammalian species across 5 orders (Glires, Primata, Scandentia, Eulipotyphla and Artiodactyla) and the superorder Afrotheria. For two of these species (the orangutan and gorilla), data were available only for the cerebellum, and although these allow the inference of numbers of neurons in the whole brain, and in the cerebral cortex in particular [Herculano-Houzel and Kaas, 2011], we have limited the data presented here to the cerebellum alone. The phylogenetic relationships amongst the species, compiled according to Price et al. [2005], Purvis [1995], Blanga-Kanfi et al. [2009], Douady et al. [2002], Shinohara et al. [2003] and Murphy et al. [2001], are illustrated in figure 1. A total of 86 brains (or hemispheres) were analyzed, and all data are provided in tables 1, 2, 3, 4, 5, 6. All data provided are averages ± standard deviation across individuals where more than one individual of each species was available, or data obtained for single individuals. All data are reported for the two sides of the brain together, even when the original data were collected from a single hemisphere, in which case results were multiplied by 2.

Values are reported here for the cerebral cortex (defined as all structures lateral to the olfactory tract), which includes the hippocampus and subcortical white matter, the cerebellum, which includes the cerebellar cortex, subcortical white matter and deep cerebellar nuclei, olfactory bulbs, where available, and RoB. The RoB amounts to the ensemble of brainstem, diencephalon and striatum. Because the olfactory bulbs are not always available for analysis, we chose to report values for ‘whole brain' as the sum of cerebral cortex, cerebellum and RoB, excluding the olfactory bulbs.

All analyses were made across average values so as not to confound intraspecific and interspecific allometric relationships [Armstrong, 1990]. All analyses were performed with JMP 9.0 (SAS). Although we report the best currently known phylogenetic relationships across the species in the dataset (fig. 1), we do not correct the reported allometric relationships for phylogenetic relatedness across the species included. As shown before, accounting for phylogenetic relatedness hardly changes the exponent of these strong allometric relationships [Gabi et al., 2010]. Most importantly, however, we wish to address directly the mathematical relationships across some of the most basic variables related to how mammalian brains are built, and we do not wish these to be affected by assumptions of phylogenetic relationships that have been known to change upon reexamination, such as those for ‘insectivores' (now assigned to the distant clades Afrotheria and Eulipotyphla).

The mass of all brain structures reported refers to paraformaldehyde (PFA)-fixed brains postfixed for at least 2 weeks. The brains of glires, primates, scandentians and eulipotyphlans were stored in 4% PFA until processed; the brains of all afrotherians and artiodactyls were stored in an antifreeze solution after fixation and cryoprotection in 30% sucrose [Herculano-Houzel, 2012]. While the mass may vary slightly from the fresh mass depending on the time of postfixation, shrinkage and other alterations in tissue mass due to the substitution of water with the glycerol-based antifreeze are minor concerns in studies of allometric relationships, where data typically span 3 or more orders of magnitude, although future users of this dataset must keep in mind that they are likely sources of extraneous, nonbiological variation in tissue mass. Most importantly, however, any alterations in tissue mass or volume due to fixation or storage in antifreeze have no effect on the estimates of numbers of cells reported here, since they were obtained with the isotropic fractionator [Herculano-Houzel and Lent, 2005], a nonstereological method.

As mentioned above, most of the data were obtained from single hemispheres and multiplied by 2 to refer to the entire structures or brain. This allowed one brain hemisphere to be kept for histological analysis, while the other was used for the quantitative analysis discussed here. In all cases, dissections started with a mid-sagittal section through the whole brain. From the available hemisphere, the olfactory bulb was dissected by a transverse cut at the olfactory tract immediately proximal to the bulb, which left the olfactory tract included in the RoB. The cerebellum was dissected next by cutting the cerebellar peduncles at the surface of the brainstem. The cerebral cortex in all animals was defined as all cortical regions lateral to the olfactory tract, including the hippocampus, amygdala and piriform cortex, and dissected from each hemisphere in small brains by peeling it away from the subcortical structures, as described earlier [Herculano-Houzel et al., 2006], or from a complete series of coronal sections after removing the brainstem by a transverse cut along the plane anterior to the superior colliculus and posterior to the hypothalamus. In this manner, the cerebral cortex includes the underlying white matter. All other brain structures (the ensemble of brainstem, diencephalon and striatum) were pooled and processed together as RoB.

Some authors have expressed concerns about the isotropic fractionator, the method whereby the numbers of cells reported here were obtained [e.g. Carlo and Stevens, 2013; Charvet et al., 2015]. Concerns about the validity of estimates obtained with the isotropic fractionator in comparison to stereology were dispelled when two groups established independently that the isotropic fractionator yields estimates of cell numbers that are comparable in value and variation to those obtained with stereology for matching [Miller et al., 2014] or neighboring [Bahney and von Bartheld, 2014] tissue. The data presented here can therefore be considered to be at least as reliable as data obtained with stereological methods. Most importantly, given the time and histological effort required for stereology, the determination of total numbers of neurons for structures that include widely different subregions such as those in the entire cerebral cortex, entire cerebellum or entire brainstem, would not have been possible without the isotropic fractionator [Herculano-Houzel et al., 2015].

It should be kept in mind that the numbers of neurons in the dataset correspond to the numbers of nuclei that express the universal neuronal nuclear marker NeuN [Mullen et al., 1992]. NeuN is known not to be expressed in some particular neuronal cell types such as Purkinje cells, mitral cells of the olfactory bulb, inferior olivary and dentate nucleus neurons [Mullen et al., 1992], neurons in the substantia nigra pars reticulata of the gerbil [Kumar and Buckmaster, 2007], and possibly others as yet unidentified. While this of course impacts the total number of cells identified as neurons, and unduly inflates the population identified as other cells (nonneurons), we expect this impact to be negligible, given that these specific neuronal subpopulations are very small compared to the structures that they integrate and which were analyzed here - the entire cerebral cortex, cerebellum or RoB.

It should also be kept in mind that, for most species, only one individual was available for study, and typically only one of the two brain halves was used for quantification with the isotropic fractionator. This means that this dataset does not address individual differences or scaling rules across individuals, which are known not to be an extension of allometric rules across species either in terms of brain × body mass [Armstrong, 1990] or in the relationship between brain structure mass and number of neurons [Herculano-Houzel et al., 2015]. Importantly, since only averages or single individual values for a species are reported in the dataset, their use in comparative studies will not confound intraspecific and interspecific variation. Moreover, although intraspecific variation can be as large as 50% in brain structure mass or number of neurons in the mouse [Herculano-Houzel et al., 2015], in the scope of comparative studies, which typically span several orders of magnitude, such variation is usually insignificant.

Although our dataset still excludes the very extremes of brain size in mammals, it ranges from very small shrews (Sorex fumeus, Blarina brevicauda) to the African elephant (Loxodonta africana), spanning body masses from 8 to 5,000,000 g and brain masses from 0.2 to over 4,000 g. Total numbers of neurons span from 36 million to 257 billion (that is, 36 × 10⁶ to 257 × 10⁹), and total numbers of other (nonneuronal) cells range from 23 million to 216 billion (table 5). Importantly, in all species, the majority of neurons (53-98%) are located in the cerebellum, leaving the cerebral cortex with typically 15-25% of all brain neurons, and the RoB with not more than 21% and often less than 10% of all brain neurons (table 6). This translates into a smaller range of 6-742 million neurons in the RoB (table 3), in contrast to 6 million to 16 billion neurons in the cerebral cortex (table 1), and 16 million to as many as 251 billion neurons in the cerebellum (table 2). In comparison to the cerebral cortex and cerebellum, the number of neurons in the RoB is thus remarkably small: no species has over 1 billion neurons in the RoB, even in the primate and artiodactyl brains with several billion neurons in the cerebral cortex and cerebellum.

As described previously [Azevedo et al., 2009; Herculano-Houzel, 2009, 2012], the availability of data on the cellular composition of the cerebral cortex of humans and various other primates allowed us to establish that the human cerebral cortex is not an outlier in its cellular composition, when compared to other primate brains. The human cerebral cortex, in particular, is not an outlier in the number of neurons for its mass. As shown in figure 2, when either all species (including the human and mouse lemur; fig. 2a) or only the center species in the distribution (excluding the two extremes, human and mouse lemur; fig. 2b) are used to calculate the relationship between cortical mass (including white matter) and number of cortical neurons, the human data point is well within the 95% confidence interval. The human cerebral cortex is only outside the confidence interval when the mouse lemur is included in the comparison (fig. 2c), but in turn the mouse lemur is the outlier in the relationship that excludes it but includes the human cerebral cortex (fig. 2d). The discordance reflects the influence of extreme data points in the calculation of fitted functions, but importantly neither mouse lemur nor human are outliers in comparison to the relationships that either include or exclude both. Instead, it is another species - of the genus Saimiri - that systematically sits outside the confidence intervals because of its atypically high neuronal density and absolute number of neurons in the cerebral cortex. Still, because of its relatively central position in the distribution of primate species, the inclusion or exclusion of Saimiri does not markedly affect the scaling rules that apply to primates. It is those species that have either very small or very large brains that possibly have a much larger impact on scaling relationships.

One such clear outlier in the allometric scaling rules that we have described previously is the naked mole-rat, which has only about half the number of neurons expected in a rodent cerebral cortex and cerebellum of its size, possibly due to regressive events such as reduced eyes, lateral geniculate nucleus and visual cortex [Catania and Remple, 2002, Xiao et al., 2006] caused by its strictly fossorial lifestyle [Jarvis and Sherman, 2002]. As shown in figure 3, calculating the neuronal scaling rules that apply to the rodent cortex with the exclusion of the two smallest species, mouse and naked mole-rat, places the latter, but not the former, outside the 95% confidence interval (fig. 3a), and adding the mouse to the scaling relationship changes it little, while still excluding the naked mole-rat (fig. 3b). The naked mole-rat should therefore be included with caution in comparative studies of rodents.

Another outlier in our dataset is the giraffe, probably because the individual in our dataset was still a juvenile, and therefore while its numbers of neurons had probably already reached adult levels, its brain mass was still below the average reported for the species, thus presumably skewing scaling relationships for numbers of cells and densities calculated with the inclusion of the giraffe [Kazu et al., 2014]. In agreement with the possibility that adult numbers of neurons had already been reached while brain structure mass was still growing, the giraffe matches the scaling rules across numbers of neurons in the cerebral cortex and cerebellum (fig. 3c).

Finally, we have reported that while the elephant cerebral cortex fits the neuronal scaling rules that apply to afrotherians and other nonprimates, its cerebellum is an obvious outlier, with over twice the number of neurons expected for an afrotherian cerebellum of its mass and 10 times the number of neurons that would be expected for the number of neurons in the elephant cerebral cortex, holding an extraordinary 98% of all brain neurons [Herculano-Houzel et al., 2014] (fig. 3c). Thus, we recommend not including the naked mole-rat, the giraffe and the elephant in comparative analyses, except for the purpose of examining these species directly.

Our dataset on the cellular composition of mammalian brain structures has made possible a number of discoveries on the scaling rules that apply to the construction and evolution of mammalian brains, many of which have been the subject of previous reviews [Herculano-Houzel, 2011, 2012; Herculano-Houzel et al., 2014b]. Amongst the most notable is the finding that distinct neuronal scaling rules apply to the primate cerebral cortex in comparison to all other mammalian species in the dataset. Nonprimate cortices scale with decreasing neuronal densities as the number of neurons increases, which suggests that the increases in neurogenesis across species that necessarily underlie increased numbers of neurons in evolution are coupled to an increasing average size of neurons (which we define as including all of their arbors, besides the cell body). Primates have diverged away from the common ancestor with other lineages with an uncoupling between increased numbers of neurons and changed average neuronal cell size (fig. 4a) [Herculano-Houzel et al., 2014b]. As a result, primate cortices contain many more neurons than nonprimate cortices of a similar mass. The magnitude of the discrepancy can be observed in table 1, where the different species of all six orders and superorders have been listed in ascending order of cortical mass. Perusing table 1 makes clear the numerical advantage that primates have in comparison to other groups in terms of numbers of neurons in the cerebral cortex, even when the human cerebral cortex is compared to the much larger African elephant cortex.

We found that different neuronal scaling rules apply to the cerebellum of primates and eulipotyphlans in comparison to the ensemble of afrotherians, glires and artiodactyls, with neuronal densities that decrease with increasing numbers of neurons in the latter but not in the former (fig. 4b) [Herculano-Houzel et al., 2014b]. Again, perusing table 2 shows the larger number of neurons found in eulipotyphlan cerebella compared to even larger cerebella of glires and afrotherians. The much larger number of neurons in primate cerebella than in even larger artiodactyl cerebella is also documented in table 2.

In contrast, we reported recently that the neuronal scaling rules for the RoB are shared by primates, glires, afrotherians and eulipotyphlans, but not by artiodactyls [Herculano-Houzel et al., 2014b]. These latter animals have far fewer neurons in their RoB than nonartiodactyls in the dataset with an even smaller RoB (table 3). The difference translates into far smaller neuronal densities in the artiodactyl RoB than expected for its number of neurons or RoB mass, compared to the scaling rules that apply to the RoB of other species (fig. 4c). However, it will be argued here that artiodactyls are not outliers in their neuronal scaling rules for the RoB; rather, once other relationships are taken into consideration, as shown below, once again it is primates who have deviated away from the scaling rule that applies to other mammalian clades.

Although artiodactyls share a similar range of brain masses with primates, the former are typically much larger animals than primates of similar brain mass or number of neurons. Since the RoB includes a number of structures that are directly connected to targets or sensory sources in the body, we examined the possibility that the very low neuronal densities found in the artiodactyl RoB, which indicate very large average neuronal sizes [Mota and Herculano-Houzel, 2014], are related to the large body mass of these animals, in comparison to all other mammals in the dataset.

We found that neuronal densities in the artiodactyl RoB are indeed much better aligned across all species in the dataset as a function of body mass (fig. 5c), to the point that they can be well described by a single power function, with lower neuronal densities (and thus larger average neuronal mass) in animals with larger body mass. In contrast, although there is also an overall trend for lower neuronal densities in the cerebral cortex and cerebellum of larger animals, fitting a single power law to the entire dataset here excludes the primate cerebral cortex (fig. 5a). Similarly, the power law that fits the cerebellum of glires, afrotherians and artiodactyls excludes not only the cerebellum of primates and eulipotyphlans, but also the elephant (fig. 5b). Thus, while neurons in the RoB seem to increase uniformly in average mass with increasing body mass across all mammalian orders analyzed, neurons in the cerebral cortex and cerebellum vary significantly across mammalian orders in how average neuronal cell mass scales with increasing body mass. This is consistent with the existence of different neuronal scaling rules that govern how average neuronal cell size in the cerebral cortex in primates and in the cerebellum of primates and eulipotyphlans scale with numbers of neurons compared to other species, as we have suggested [Herculano-Houzel et al., 2014b].

If it remains the case that the scaling rules that link average neuronal cell size to numbers of neurons in the RoB have diverged in artiodactyls, as shown in figure 4c, then one possibility is that the driving force behind this divergence was a shift in the body × brain relationship in the species of this clade. However, as seen in figure 6, artiodactyls are a much closer fit to the scaling relationship between body mass and number of RoB neurons (as also found for the cerebral cortex and cerebellum) that applies to nonprimate species, while primates clearly have their own body × brain relationship. If artiodactyls shared with all mammals the relationship between neuronal density in the RoB and body mass (fig. 5c) but showed a faster decrease in neuronal density for the number of RoB neurons compared to other species (fig. 4c), as we had initially presumed [Herculano-Houzel et al., 2014b], then the number of neurons in the artiodactyl RoB should scale faster with body mass than in other species - but it does not (fig. 6c). In contrast, if artiodactyls shared with other nonprimate mammals both the scaling of neuronal density in the RoB and body mass (fig. 5c) and the scaling of neuronal density with the number of RoB neurons, and primates were instead the outliers as shown in figure 7, then artiodactyls would be expected to share with nonprimates the scaling of number of RoB neurons with body mass, as is indeed the case (fig. 6c). It thus appears more likely that the scaling rules that apply to the RoB have diverged not in artiodactyls, but rather in primates, as they did in the cerebral cortex and cerebellum, as indicated in figure 7.

While the neuronal scaling rules that apply to the RoB might thus have diverged not in artiodactyls, but in primates, it remains that for all species in the dataset, including primates, neuronal densities in the RoB decrease with increasing body mass, indicating that average neuronal mass in the RoB increases together with increasing body mass. Of all brain neurons, it is those situated in the RoB that are most directly related to the body, as many neurons in these structures, from the medulla to the diencephalon, are directly connected to structures in the body through sensory or motor nerves. Those neurons that are directly connected to bodily structures must have their fibers increase, at least in length, within the RoB (as in the body) as the body grows and those targets become more distant. Indeed, the exponent of the single power law that relates neuronal density in the RoB to body mass, -0.301 ± 0.019 (r² = 0.873, p < 0.0001), is not significantly different from 1/3 - the exponent that relates body length to body volume. It thus appears that all mammalian species in the dataset have neurons that become larger (longer) within the brain as body mass increases, with no distinction across orders. We suggest that it is this physical constraint that makes neurons in the RoB become larger (longer) with increasing body mass across all clades.

Importantly, and in contrast to the hypothesis that larger bodies require more neurons to operate them [Jerison, 1973], it is only the neuronal density in the RoB (and thus average neuronal cell mass) that varies uniformly with increasing body mass: as shown in figure 6c, primates are clear outliers, such that there is no single scaling rule that relates numbers of neurons in the RoB to body mass across all mammalian species in the dataset. Interestingly, although clear relationships exist between brain mass and the number of neurons in the cerebral cortex (fig. 6a), cerebellum (fig. 6b) or RoB (fig. 6c), primates are in all three cases subject to a different scaling rule, with more neurons for a given body mass compared to other mammalian clades. The clade specificity indicates that, while larger bodies have neurons in the RoB that are on average larger in proportion to the linear dimension of the body, the number of brain neurons is not dictated simply by body mass, either in the RoB or elsewhere.

As mentioned above, the main focus of our work has been the investigation of the scaling relationships that apply to mammalian brains and what they teach about the evolutionary origins of brain diversity in mammals. We expect the dataset that we have generated to be useful to researchers interested in many other aspects of diversity: how it is related to lifestyle, habitat, diet; how it evolved within particular clades; how it is constrained by physical aspects of brain morphology and function. As our research on brain diversity continues to grow, we will continue to expand our dataset on the cellular composition of different brain structures across mammalian species and clades and make it available to the scientific community. In the near future, we will be able to add chiropterans, carnivores, marsupials and cetaceans to the dataset, as well as a subdivision of nonneuronal ‘other' cells into the underlying cell types (endothelium, astrocytes, oligodendrocytes and microglial cells).

We thank all the colleagues that were involved in collecting the data reported in this review. This work was supported by grants from CNPq, FAPERJ, MCT/INCT and The James S. McDonnell Foundation (S.H.H.), The National Research Foundation of South Africa (P.R.M.), and The National Science Foundation grant 0844743 (K.C.).