Abstract
Microglial cells play essential volume-related actions in the brain that contribute to the maturation and plasticity of neural circuits that ultimately shape behavior. Microglia can thus be expected to have similar cell sizes and even distribution both across brain structures and across species with different brain sizes. To test this hypothesis, we determined microglial cell densities (the inverse of cell size) using immunocytochemistry to Iba1 in samples of free cell nuclei prepared with the isotropic fractionator from brain structures of 33 mammalian species belonging to males and females of five different clades. We found that microglial cells constitute ∼7% of non-neuronal cells in different brain structures as well as in the whole brain of all mammalian species examined. Further, they vary little in cell density compared with neuronal cell densities within the cerebral cortex, across brain structures, across species within the same clade, and across mammalian clades. As a consequence, we find that one microglial cell services as few as one and as many as 100 neurons in different brain regions and species, depending on the local neuronal density. We thus conclude that the addition of microglial cells to mammalian brains is governed by mechanisms that constrain the size of these cells and have remained conserved over 200 million years of mammalian evolution. We discuss the probable consequences of such constrained size for brain function in health and disease.
SIGNIFICANCE STATEMENT Microglial cells are resident macrophages of the CNS, with key functions in recycling synapses and maintaining the local environment in health and disease. We find that microglial cells occur in similar densities in the brains of different species and in the different structures of each individual brain, which indicates that these cells maintain a similar average size in mammalian evolution, suggesting in turn that the volume monitored by each microglial cell remains constant across mammals. Because the density of neurons is highly variable across the same brain structures and species, our finding implies that microglia-dependent functional recovery may be particularly difficult in those brain structures and species with high neuronal densities and therefore fewer microglial cells per neuron.
Introduction
Microglial cells are mesoderm-derived macrophages that reside in the CNS; and together with neuroectoderm-derived astrocytes and oligodendrocytes, as well as endoderm-derived endothelial cells, they compose the non-neuronal population of brain cells. Microglia play key surveillance and housekeeping roles as sensors of their local environment, responding to injury, phagocytosing apoptotic cells, pathogens, and cellular debris, and releasing inflammatory modulators and cytokines (Wake et al., 2013; Hickman et al., 2018). During development, they engulf both seemingly healthy and apoptotic neural precursors and immature neurons, thus contributing to establishing the final number of neurons in brain structures (and consequently tissue volume) (Cunningham et al., 2013; Kabba et al., 2018). In adult neurogenesis, microglia support survival, proliferation, and migration of neuronal precursors from the subventricular zone to the olfactory bulbs through the rostral migratory system (Xavier et al., 2015); and in neural plasticity, they regulate synaptic formation, maturation, and elimination (Wake et al., 2009, 2013; Tremblay et al., 2010; Kettenmann et al., 2011; Paolicelli et al., 2011; Schafer et al., 2012; Parkhurst et al., 2013).
All of the above functions concern formation and maintenance of brain tissue volume (Nimmerjahn et al., 2005; Kettenmann et al., 2011). As macrophages that they are, individual microglia can be construed to be monitors of their local environment, each in charge of its respective volume in the parenchyma, a view that is compatible with their typically homogeneous distribution in all adult cerebral cortices examined to date (Nimmerjahn et al., 2005; Cunningham et al., 2013; but for noncortical structures, see Lawson et al., 1990; Mittelbronn et al., 2001). Given that their functions are volume-based, a question of fundamental importance for understanding healthy and diseased brain function is how the volume of brain tissue monitored by the average individual microglial cell compares across species.
We have previously found that, while neuronal densities vary by 1000-fold across brain structures in >60 mammalian species, densities of non-neuronal cells as a whole vary only by 10-fold in the same sample, and are thus comparatively restricted, which is compatible with volume-related functions of non-neuronal cells (Herculano-Houzel and Dos Santos, 2018). As a consequence, the lower the neuronal density (i.e., the larger the average size of neurons), the larger the ratio of non-neuronal cells per neuron found in a tissue (Herculano-Houzel, 2014; Mota and Herculano-Houzel, 2014).
Our finding of constant densities of non-neuronal cells across brain structures and species does not translate automatically into constant densities of each non-neuronal cell type (Herculano-Houzel and Dos Santos, 2018). In the few instances where microglia have been quantified, they constitute a very small minority of all brain cells, amounting to 5% of cells in the cerebral cortex (Cx) or corpus callosum and 12% of cells in the substantia nigra of the mouse brain (Lawson et al., 1990), from 0.3% of cells in the gray matter of the cerebellum (Cb) to 11% of cells in the medulla oblongata of the human brain (Mittelbronn et al., 2001), and to 5% of glial cells in the human cortical gray matter (Pelvig et al., 2008). These small percentages imply that, while the restricted densities of non-neuronal cells found in our previous studies may apply to the different non-neuronal cell types separately, they might also mask marked variations in microglial size and distribution (Herculano-Houzel and Dos Santos, 2018).
Here we undertake a systematic examination of microglial densities across a wide range of brain structures and species of mammals to determine the extent of their variation. Our analysis is directed to determine how microglial densities vary at four levels: (1) across different sites within a same structure (gray and white matter of the Cx); (2) across brain structures; (3) across species in the same clade; and (4) across mammalian clades.
Materials and Methods
We used 2069 samples of free nuclei prepared with the isotropic fractionator technique (Herculano-Houzel and Lent, 2005) from previous studies that determined total numbers of cells and the fraction of neurons in different brain structures from 33 mammalian species of five clades: marsupials, afrotherians, primates, carnivorans, and artiodactyls (Ribeiro et al., 2013; Herculano-Houzel et al., 2014; Kazu et al., 2014; Neves et al., 2014; Gabi et al., 2016; Dos Santos et al., 2017; Jardim-Messeder et al., 2017; compiled in Herculano-Houzel et al., 2015a). These clades encompass over 218 million years of evolution since their divergence from the last common ancestor shared exclusively by all of them (Murphy et al., 2001, 2004; Bininda-Emonds et al., 2007).
To determine microglial densities and infer total numbers of these cells in each brain structure, we subjected free nuclei in each frozen stored sample to immunocytochemistry against the universal microglial marker Iba1 (ionized calcium binding adapter molecule 1), which is structurally and functionally conserved among Metazoa (Imai et al., 1996; Imai and Kohsaka, 2002; Janušonis, 2018). Total numbers of cells in the structures examined as well as structure mass are available for each species in the original publications, as listed above. Applying the percentage of nuclei found to be Iba1+ to total numbers of cells/mg reported in the same structure gives us the microglial density in each structure analyzed.
Sample
The 33 species analyzed here consist of 10 species of marsupials (Dos Santos et al., 2017), 4 species of artiodactyls (Kazu et al., 2014), 8 species of primates (Ribeiro et al., 2013; Gabi et al., 2016), five species of afrotherians (Herculano-Houzel et al., 2014; Neves et al., 2014), and 6 species of carnivorans (Jardim-Messeder et al., 2017) for which we had samples of suspensions of free nuclei in storage (see below). This excluded all rodent and primate species of our initial studies. All were adults of either sex, except for the giraffe, which was a juvenile (Kazu et al., 2014), with no signs of extreme aging, neurologic diseases, or brain abnormalities. With the exception of a cat and two dog brains, which were not perfused, and the human brain, which was perfused several hours postmortem, almost all specimens were perfused shortly after death. The majority of freshly perfused specimens combined with the similarity of results obtained in the few species processed differently suggests that postmortem interval and any infiltration of the brain by blood macrophages are not significant issues for the present study. For 5 of the 33 species (Petrodromus tetradactylus, Elephantulus myurus, Procavia capensis, Canis lupus familiaris, and Mustela putorius furo), two specimens of each were available; thus, data reported for those species correspond to the average data of both specimens each, so as to avoid duplication and bias. For the other 28 species, we examined only one specimen of each. While a single specimen per condition would generally be deemed insufficient for biological comparisons across conditions, we consider that for a study of scaling of cell numbers and densities across multiple orders of magnitude, intraspecific variation is mathematically irrelevant, while remaining an interesting issue of its own, which we have addressed previously (Herculano-Houzel et al., 2015b).
For all species, only one hemisphere of the brain was analyzed, as reported previously. Each brain was separated into different structures of interest: Cb, Cx, and what we report as the rest of the brain (RoB), constituted by the remaining hindbrain, midbrain, diencephalon, striatum, and hypothalamus (Dos Santos et al., 2017; Watson et al., 2017). Only cerebral cortical structures were available from primates. For all species, the Cx was sectioned into coronal sections (1 mm thick for Didelphis aurita; 12.8 mm for Loxodonta africana; 2 mm for all other species), and gray and white matter dissected from each coronal section were processed separately for all species but the four smallest marsupials and afrotherians (Metachirus nudicaudatus, Marmosops incanus, Petrodomus tetradactylus, and Elephantulus myurus). The hippocampus was separated from the rest of the cortical gray matter and counted separately. Where a single Cx value is reported in a species, it corresponds to the ensemble of cortical gray matter, white matter, and hippocampus; whenever values apply exclusively to gray or white matter, that is so indicated. “Cortical sites” refer either to the various slabs taken from coronal sections in the original studies (L. africana, C. familiaris, F. catus, P. leo, U. arctos, D. goodfellowi, D. aurita, M. fuliginosus, M. parma, M. rufogriseus, S. harrisii, W. bicolor, and all primate cortices) (Ribeiro et al., 2013; Herculano-Houzel et al., 2014; Gabi et al., 2016; Dos Santos et al., 2017; Jardim-Messeder et al., 2017) or to each of the whole coronal sections, as in Neves et al. (2014) and Kazu et al. (2014).
Immunostaining
All samples were isotropic suspensions of free isolated nuclei that had been stored in an antifreeze solution (30% glycerol, 30% ethylene glycol, 30% distilled water and 10% 0.244 M phosphate buffer 2xPO4) (Manger et al., 2009) at −20°C (Ribeiro et al., 2013; Herculano-Houzel et al., 2014; Kazu et al., 2014; Neves et al., 2014; Gabi et al., 2016; Dos Santos et al., 2017; Jardim-Messeder et al., 2017). The proportion of microglial cell nuclei in each sample was determined by immunocytochemical detection of the antigen Iba1, specifically expressed in both the nucleus and cytoplasm of active and resting microglia (Imai et al., 1996; Imai and Kohsaka, 2002), an antibody commonly used to label microglia both in vivo and in vitro (Sousa et al., 2017). We first incubated the samples of free nuclei in 0.2 m boric acid, pH 9.0, for 2 h at 70°C for antigen retrieval. To maximize the chances of compatibility across species, we used a rabbit polyclonal primary antibody against Iba1 (Wako, 019-19 741) diluted 1:100 in a buffer made of 0.1 M PBS, pH 7.4, 3% NGS and 2% BSA, incubated for 2 d at 4°C. All samples were further incubated with AlexaFluor-488-conjugated goat anti-rabbit secondary antibody (Abcam, ab150077) diluted 1:500 in a similar buffer as above and incubated for 2 d at 4°C, and finally labeled in 0.1 M PBS, pH 7.4, with the fluorescent DNA marker DAPI (Invitrogen) at 2 mg/L for 30 min at room temperature. The fraction of DAPI+ cell nuclei that were Iba1+ was determined by unbiased systematic counting of contiguous squares in a modified Neubauer chamber, under a Carl Zeiss AxioImager A2 fluorescent microscope with a 40× objective (Carl Zeiss EC-Plan-NEOFLUAR 40×/0,75 420360-9900), until at least 500 DAPI+ nuclei had been scored. Depending on the percentage of Iba1+ nuclei, typically the latter were typically 10-30 of all nuclei scored for each brain structure analyzed.
To confirm that Iba1 is conserved enough that the same polyclonal antibody labels nuclei with anatomic characteristics of microglia in all the mammalian species analyzed, we conducted immunohistochemistry on 50 μm coronal tissue sections of the Cx of 2 species of each clade (primates, Northern greater galago and owl monkey; artiodactyls, springbok and pig; marsupials, big-eared opossum and Bennett's wallaby; carnivorans, ferret and cat; afrotherians, African elephant and elephant shrew; Fig. 1). We incubated the tissue sections in 0.2 m boric acid, pH 9.0, for 2 h at 70°C for antigen retrieval and applied to them the same 1:100 dilution of anti-Iba1 antibody used on the respective free nuclei in the presence of goat and bovine sera (5% NGS; 3% BSA in 0.1 M PB/0.25% Triton X-100) for 2 d at 4°C followed by incubation with AlexaFluor-488-conjugated goat anti-rabbit secondary antibody (Abcam, ab150077) diluted at 1:500 in an incubation buffer made of 0.1 M PB, pH 7.4, 5% NGS, and 3% BSA for 2 d at 4°C and ultimately in 0.1 M PB, pH 7.4, with DAPI at 2 mg/L for 1 h at room temperature. As shown in Figure 1, Iba1+ cells have the typical shape of microglial cells with extensive thin processes, and their distribution seems homogeneous across different cortical and brain regions, as has been previously reported for the mouse Cx, where minimal distances across neighboring microglial cells were fairly constant across cortical layers (Nimmerjahn et al., 2005). Two-dimensional photomontages were acquired on a Carl Zeiss AxioImager M2 fluorescent microscope under 5×, 20×, 40×, or 63× objective magnification using StereoInvestigator software (MicroBrightField). Colabeling with antibodies against Iba1 and other glial and neuronal markers showed no overlap in mouse brain sections (Herculano-Houzel and Dos Santos, 2018), confirming the specificity of this marker for microglial cells only, in agreement with a recent report (Valério-Gomes et al., 2018). Heretofore, Iba1+ cells and cell nuclei will be referred to as microglial cells (or I, for Iba1+ cells or nuclei).
Data analysis
All statistical analyses were performed in JMP Pro 14.0 (SAS Institute). Regressions to power functions were performed to find the best fit for each distribution by applying linear fits to automatically log-transformed data. All averages and exponents are reported ± SE and with the corresponding p value; 95% CI are not reported explicitly as they can be easily calculated as average or exponent ± 2 SE. To evaluate the statistical dependence between two variables, we do not assume normality in the data and use a nonparametric Spearman's rank correlation coefficient instead. We consider a cutoff p value of p < 0.01 throughout the manuscript to estimate significance of our statistical analyses. Because we focus directly on the question of whether microglial cell densities are universal across species, we chose to analyze the raw data directly without introducing any inferences about phylogenetic relationships about the species analyzed. All data are presented in the tables for future use in other studies where accounting for phylogenetic relationships may be of interest.
Data reported in Table 1 represent the sums of masses, total microglial and neuronal cells, and their densities in each structure analyzed in the present study. Total numbers of microglial cells in each structure were estimated by applying the percentages of Iba1+ nuclei in the samples to their respective recorded original total number of cell nuclei in the sample. Microglial cell density is the estimated total number of microglial cells obtained this way divided by the original sample mass, in cells/mg of tissue. Where a brain structure had been subdivided into multiple samples, we first calculated the absolute number of Iba1+ cells in each sample, then the total tally for the whole structure, and only then calculated the average microglial cell density for the whole structure as above. That is, the averages reported in Table 1 represent the real average when sampling the entire tissue, not an average of averages in multiple samples.
Results
Microglial cells are ∼6% of all cells in the mammalian Cx
We first focus on the percentage of microglia among all cells across the 1776 cerebral cortical samples of gray and white matter analyzed across 30 species (Table 2). There is an overall 33-fold variation in the percentage of cells that are microglia across cortical sites, from 0.46% in the red-handed tamarin gray matter to 15.37% in the ferret white matter, but these percentages span overlapping ranges across the 5 mammalian clades examined (Table 2; Fig. 2A). When all species and clades are considered together, microglia represent similar, small percentages of all cortical cells in gray and white matter (5.88 ± 0.10% and 5.67 ± 0.11%, respectively; Wilcoxon p = 0.2650, Table 2; Fig. 2B). Compared across clades, the average percentage of microglia among all cortical cells is lowest in primate gray and white matter, and highest in afrotherian gray matter and carnivoran white matter (Fig. 2C; “all clades” in Table 2). Within each clade, we find significant differences in the percentage of cells that are microglial between gray and white matter (Wilcoxon p values all <0.01), but percentages were not systematically higher in either one (Fig. 2C; “per clade” in Table 2). Within an individual cortex, 12 of the 26 species with multiple cortical sites available have significantly different percentages of cells that are microglia between gray and white matter (Wilcoxon p values <0.01; “per species” in Table 2). In these cases, percentages of microglial cells were higher in the gray matter samples for all species, except carnivorans.
We next determined whether the smaller and larger percentages of cells that are microglial in primate and afrotherian cortical gray matter, respectively, are related to the high neuronal densities found in primate versus the low neuronal densities found in the elephant cortex (Herculano-Houzel et al., 2014; Gabi et al., 2016). When all data are combined, we find a strong correlation between local neuronal density and the percentage of cells that are microglial in gray matter cortical sites across species and clades, as shown in Figure 2D (Spearman correlation for all clades together: ρ = −0.6836, p < 0.0001). This correlation indicates that microglial cells are an up to 10-fold higher percentage of all gray matter cells where neuronal densities are lower. However, across cortical sites within each clade, this correlation only applies for carnivorans and primates; and within each species, only within C. familiaris, P. leo, W. bicolor, P. anubis cynocephalus, O. garnettii, H. sapiens sapiens, and C. apella.
A slight but detectable correlation between the local non-neuronal cell density and the percentage of microglial cells among all cells in each white matter cortical site across species and clades is observed, as shown in Figure 2E (Spearman correlation for all clades together: ρ = −0.2625, p < 0.0001). However, across cortical sites within a clade, this correlation only applies for carnivorans; and within each species, only in P. leo and M. fascicularis. It thus appears more likely that any apparent correlation between local non-neuronal cell density and percentage of microglial cells in the gray or white matter of the Cx is related to clade-specific (evolutionary) differences, and not to biological scaling.
Microglial cells are a variable percentage of all cortical non-neuronal cells across cortical sites and clades
To avoid the confound of variable percentages of cortical cells that are neurons across mammalian species, depending on cortical mass and clade (Herculano-Houzel, 2017), we next analyzed the representation of microglial cells among only the non-neuronal cells in each sample. In this case, microglia are found to represent a slightly larger average percentage of non-neuronal cells in gray matter sites (7.56 ± 0.11%) than in white matter sites (5.86 ± 0.11%, Wilcoxon p < 0.0001; Fig. 2F,G; Table 3, “all clades”). This is the case in each clade separately, except in carnivorans (Table 3, “per clade”). Again, microglial cells are a higher percentage of non-neuronal cells in the gray matter in afrotherians (10.18 ± 0.20%) and lower in primates (5.44 ± 0.16%). Across white matter sites, a higher percentage of microglia is found in carnivorans (9.15 ± 0.24% of all non-neuronal cells), and the lowest percentage in primates (4.06 ± 0.11% of all non-neuronal cells; Fig. 2F,H). Within each clade, with the exception of carnivorans, microglial cells are a significantly higher percentage of non-neuronal cells in the gray matter than in the white matter (Fig. 2H; Table 3, “per clade”). Within each individual cortex, 12 of the 26 species with multiple data points available have significantly different percentages of microglia among other cells between gray and white matter (Wilcoxon p values <0.01; Table 3, “per species”) with percentages of microglial cells higher in the gray matter than in the white matter for all 12 species, except carnivorans.
When data are combined across all species and clades, we find a detectable correlation between local neuronal density and the percentage of microglial cells among non-neuronal cells in each gray matter cortical site across species and clades, as shown in Figure 2I (Spearman correlation for all clades together: ρ = −0.4805, p < 0.0001). However, this correlation only applies across cortical sites within carnivorans, and only applies within the species F. catus, P. leo, H. sapiens sapiens, and C. apella.
In the white matter, a slight but detectable correlation is observed between local non-neuronal cell density and the percentage of microglial cells among non-neuronal cells in each cortical site across species and clades, as shown in Figure 2I (Spearman correlation for all clades together: ρ = −0.2613, p < 0.0001). However, this correlation does not apply across cortical sites within each clade, and only applies within each species in P. leo, D. aurita, D. goodfellowi, and M. fascicularis.
It thus again appears more likely that any apparent correlation between local non-neuronal cell density and percentage of non-neuronal cells that are microglial in the gray or white matter of the Cx is related to clade-specific (evolutionary) differences, and not to biological scaling. Overall, we find that microglial cells constitute systematically <10% of all non-neuronal cells in both gray and white matter of the Cx across all species examined, with an exact percentage that is smaller in primates than in other clades, and only slightly and nonsystematically variable across cortical sites in most species, independently of local neuronal densities.
No systematic variation of microglial density within the Cx
Although microglial cells are a slightly higher percentage of non-neuronal cells in the gray matter than in the white matter (Fig. 2G), we find that the average absolute density of microglial cells across all species and clades examined is actually higher in the white matter (5016 ± 93 microglia/mg) than in the gray matter (3620 ± 63 microglia/mg, Wilcoxon p < 0.0001; Fig. 3A, middle; Table 4, “all clades”), due to the overall slightly higher densities of non-neuronal cells in the former than in the latter (Herculano-Houzel and Dos Santos, 2018). The vast majority of individual gray matter sites examined has microglial densities between 1000 and 10,000 microglia/mg, whereas neuronal densities at the same sites vary between 1000 and 100,000 neurons/mg (Fig. 3D). Average microglial cell densities in gray and white matter by clade are shown in Figure 3A (bottom). Across clades, the lowest average microglial cell densities are found in primates, both in gray (2890 ± 92 microglia/mg) and white matter (3965 ± 123 microglia/mg; Table 4, “per clade”), although the broad overall ranges of microglial cell densities encompass largely overlapping distributions across clades and cortical sites (Fig. 3A, top, bottom). Within each clade, there is not much variation in microglial densities across species (Table 4, “per species”). Within an individual cortex, microglial densities are significantly higher in white matter sites compared with those in gray matter in 12 of the species sampled, including the human cortex (Table 4).
Across all cortical gray and white matter sites analyzed (n = 980 and n = 796, respectively), we find that the mass of each cortical site varies as a power function of the number of microglial cells found in it that is significantly different from linearity whether gray and white matter sites are analyzed together (exponent, 0.905 ± 0.008, p < 0.0001) or separately (exponents 0.882 ± 0.010, p < 0.0001 and 0.945 ± 0.012, p < 0.0001 for gray and white matter, respectively; Fig. 3B,B′,B′′). These strong correlations still hold within individual cortices, with significant sublinear or linear relationships between mass and number of microglial cells, except for four species (M. mungo gray and white matter, and gray matter in S. scrofa, F. catus, and S. harrisii; Table 4). With exponents that include unity, a linear relationship cannot be excluded for gray and/or white matter in 21 of the 26 species analyzed, which suggests that microglial density does not really vary systematically within the Cx gray and white matter in these species. Indeed, across all cortical gray and white matter sites, the local density of microglial cells increases only weakly with their number across species and clades, with a significant but weak correlation across gray and white matter sites together (Spearman ρ = 0.2373, p < 0.0001) or separately (Spearman ρ = 0.2942, p < 0.0001 and Spearman ρ = 0.1614, p < 0.0001 for gray and white matter, respectively; Figure 3C,C′,C′′). Within individual cortices, weak correlations, most of the times nonsignificant, within both gray and white mater sites or at least one cortical region are observed in 23 of the 26 species (Table 4). Thus, microglial cells appear to be added in highly variable numbers but fairly similar densities to different cortical sites.
The slight but significant variation in microglial cell density across cortical sites could reflect local variations in neuronal density, which is known to be significant even within an individual Cx (Ribeiro et al., 2013; Herculano-Houzel et al., 2015b; Gabi et al., 2016). However, variations in local microglial densities in the cortical gray matter are only poorly correlated with local neuronal densities when data points from all sites are pooled together across species and clades (Spearman ρ = −0.1108, p = 0.0005; Fig. 3D), and the fit to a power function is not significant (p = 0.1165; Fig. 3D).
Within each clade, pooling cortical sites across species reveals in all clades, except Afrotheria, a significant positive relationship between local microglial density and neuronal density in the cortical gray matter (Fig. 3D; Table 4). However, most individual cortices do not present significant relationships between local microglial and neuronal densities (Table 4). Thus, any apparent systematic variation in microglial cell density seen here is most likely the result of pooling data across clades, and not of a systematic, biological relationship between local microglial and neuronal densities.
Similarly, local microglial densities in the cortical white matter sites are slightly correlated with local densities of non-neuronal cells when data points from all sites are pooled together across species and clades (Spearman ρ = 0.2172, p < 0.0001; Fig. 3E), with a significant fit (exponent, 0.596 ± 0.010, p < 0.0001) but with a very low r2 of 0.112. As for the gray matter, pooling cortical sites across species within each clade reveals a significant positive relationship between local microglial density and non-neuronal cell density for all clades (Fig. 3E; Table 4). However, only half of the species analyzed present significant relationships between local microglial and non-neuronal cell densities within an individual Cx (Fig. 3E; Table 4), again suggesting that any apparent relationship in the overall dataset is the result of clade-specific differences, with slightly lower densities in primate cortices.
Microglia/neuron ratio in the cortical gray matter is a universal function of neuronal density across sites, species, and clades
We find that there are typically fewer microglial cells than neurons in the cortical gray matter: that was the case in 973 of all 1776 sites analyzed in the 30 species. More microglial cells than neurons in the cortical gray matter as a whole were only found in 5 species, all of them the largest-brained nonprimate species of carnivorans, artiodactyls, or the African elephant. Even in those cases, there were at most 2-3 microglial cells per neuron in the cortical gray matter. Generally, we find that there are 1-100 neurons per microglial cell in the gray matter of most mammalian species (Table 1, I/N).
As reported previously (Herculano-Houzel, 2014; Herculano-Houzel and Dos Santos, 2018), we find that the number of non-neuronal cells per neuron varies inversely with neuronal density in our current dataset pooled across all cortical sites, species, and clades as a sublinear power function of local neuronal density (exponent –0.793 ± 0.009, p < 0.0001, r2 = 0.888; Fig. 4A). There is a similar trend of fewer microglial cells per neuron (i.e., a smaller I/N ratio) where neuronal densities are higher: the I/N ratio varies as a power function of neuronal density across all gray matter cortical sites and across species and clades with an exponent that is not significantly different from inverse linearity (−1.020 ± 0.016, p < 0.0001, r2 = 0.799; Fig. 4B). This correlation holds within each clade, except for Artiodactyla (Table 4). Within the cortex of each species, strong correlations, with significant sublinear or linear relationships between I/N and neuronal density, are found for all but six species, and are present in the only afrotherian species included, the African elephant (Table 4). In contrast with neuronal density, there is no single scaling relationship that applies between the I/N ratio and microglial density across all cortical sites, clades, and species (Fig. 4C; Table 4). These findings strongly indicate that variations in I/N across cortical sites and species are due to variations in neuronal density in the presence of only slight variations in microglial cell density.
Similar microglial proportions and densities across brain structures and species
We next turn to the question of how microglial cell density compares across brain structures in different mammalian species. Relative to the total number of cells in the structure, microglial cells amount to 5.77 ± 0.43% in the Cx as a whole, 1.50 ± 0.15% in the Cb, and 6.39 ± 0.51% in the RoB (Fig. 5A,B; Table 5). These average percentages, calculated among all species and clades, are significantly higher in both Cx and RoB compared with Cb (Wilcoxon, p < 0.0001), but not significantly different between the former two structures (Wilcoxon p = 0.4283) (Fig. 5B; Table 5). The percentage of all cells that are microglial in each major brain division varies both across structures in the same clade and within the same structure across clades (Fig. 5C). Variation in the percentage of all cells that are microglial correlates strongly with the neuronal density in the respective structure, as shown in Figure 5D (Spearman correlation: ρ = −0.8058, p < 0.0001 for all clades): the higher the neuronal density, the lower the percentage of all cells that is microglial.
However, within the non-neuronal cell population in each structure, microglial cells are a much less variable subpopulation in all brain structures, species, and clades examined, representing ∼7% of all non-neuronal cells in any structure (Table 5; Fig. 5F). Even if the exact percentage per structure varies from 2.18% in the Tasmanian devil's RoB to 24.56% in the pig Cb (Fig. 5E), the distribution of the percentages of non-neuronal cells that are microglial spans a very similar range across different structures and clades (Fig. 5E), and average percentages do not vary much across structures and clades (Fig. 5F,G). Moreover, the percentage of microglial cells among non-neuronal cells does not correlate significantly with the density of neurons in each structure as shown in Figure 5H (Spearman ρ = −0.2190, p = 0.0525). Thus, as brain structures vary in size, number of neurons, and neuronal density across mammalian species and clades, microglial cells appear to remain a fairly constant percentage of all non-neuronal cells.
Similar densities of microglial cells across brain structures and mammalian species
Like the percentage of non-neuronal cells that are microglial, average microglial cell densities are remarkably similar across brain structures and mammalian species of all clades: 4374 ± 341 microglia/mg in the Cx (gray and white matter combined, including hippocampus), 5021 ± 406 microglia/mg in the Cb (which also combines gray and white matter), and 4664 ± 464 microglia/mg in the RoB, with an overall average of 4667 ± 230 microglia/mg (all pairwise comparisons, Wilcoxon p > 0.1; Fig. 6A, middle, bottom; Table 5). In each of these structures, the coefficient of variation of microglial cell density is <0.10, both across species and across clades (Table 5). Whereas neuronal densities vary by >1000-fold across brain structures and mammalian species combined, microglial cell density varies by only 5-fold to 8-fold (Fig. 6E; Tables 1, 5), similar to densities of non-neuronal cells as a whole (which include astrocytes, oligodendrocytes, and vasculature-associated cells) (Herculano-Houzel and Dos Santos, 2018). When all brain structures, species, and clades are pooled together, there is no systematic variation in the density of microglial cells in the major brain divisions analyzed with variation in the total number of microglial cells in them (Spearman ρ = 0.1435, p = 0.2071; Fig. 6B). Similarly, no correlation is observed between those variables across species and clades when each structure is analyzed individually (Spearman, p = 0.3649 for Cx, p = 0.3076 for Cb, and p = 0.1351 for RoB). Thus, microglial cells appear to be added in highly variable numbers but fairly similar densities to all major brain structures, within and across brains and species.
Further support of no significant variation in microglial density across brain structures and species is the linear relationship we find between structure mass and the number of microglial cells (Fig. 6C; exponent, 0.962 ± 0.024, p < 0.0001). This linear relationship approximates that of the distribution of non-neuronal cells as a whole for all mammalian species analyzed previously (exponent, 1.056 ± 0.013, p < 0.0001) (Dos Santos et al., 2017; Jardim-Messeder et al., 2017). The linearity of this relationship implies that brain structures of similar size are composed of similar numbers of microglial cells across different mammalian brains, whatever the structure identity, species, and even clade. Similar linear relationships between structure mass and number of Iba+ cells are also found across structures and species within each clade (p < 0.0001; Fig. 6C).
Similarly, there is no significant correlation between microglial cell density and structure mass across species and clades for any particular brain structure (Spearman, p = 0.9117, p = 0.8753, and p = 0.6103 for Cx, Cb, and RoB, respectively), and also no correlation across species within or across clades when all structures are considered together (Fig. 6D). Despite the >1000-fold variation in neuronal density in the dataset, microglial cell density in each major brain structure is not correlated with neuronal density across structures and species neither within nor across clades (Fig. 6E). Thus, as brain structures vary in size, number of neurons, and neuronal density across mammalian species and clades, microglial cells both remain a fairly constant percentage of all non-neuronal cells and occur at fairly constant densities in each brain and its various structures (Fig. 6D,E).
Microglia/neuron ratio scales universally with neuronal density across structures and species
As a consequence of the very large variation in neuronal densities with little variation in microglial density across brain structures in the different species, the number of microglial cells found per neuron in each structure (the I/N ratio) is enormously variable, from >100 neurons per microglial cell in the Cb of several species to ∼3 microglial cells per neuron in the Cx or RoB of large nonprimate brains (Fig. 7A; Tables 1, 5). Typically, there are more neurons than microglial cells in most brain structures, with an exact I/N ratio that correlates strongly with the local density of neurons (Fig. 7B). In contrast, I/N ratios vary only loosely with structure mass across species and clades taken together (Fig. 7A). Finally, and as expected, the I/N ratio is highly correlated with the total numbers of non-neuronal cells per neuron (O/N; Spearman ρ = 0.9479, p < 0.0001), and consistently with the finding that microglial cells are a fairly constant percentage of all non-neuronal cells in any brain structure and species, the ratio I/N can be described as a power function of O/N with an exponent that is indistinguishable from linearity (1.020 ± 0.031, p < 0.0001, r2 = 0.932; Fig. 7C). These findings thus demonstrate that, with a relatively constant density of microglial cells, brain structures with low neuronal densities (and thus large neurons) (Mota and Herculano-Houzel, 2014) have more microglial cells per neuron, whereas brain structures with high neuronal densities (and thus small neurons) have few microglial cells per neuron.
Microglial cell numbers and densities in the whole brain
Across the 23 species examined for which all brain structures were available, total brain mass varies by 5038-fold, and the total estimated number of microglial cells in the brain varies by a fairly similar 7749-fold (Table 1). In contrast, and as previously reported, numbers of brain neurons vary by only 2300-fold across the same species. Overall, microglial cells amount to 4.42 ± 1.65% of all cells and to 7.60 ± 2.06% of non-neuronal cells in the whole brain.
Microglial cell densities in the brain as a whole vary <3-fold across species of all clades (Table 1), with an overall average of 4658 ± 371 microglia/mg. We find no differences in average microglial cell density across clades nor across species within each clade (Wilcoxon, all p > 0.05). However, because neuronal densities are enormously variable, the I/N ratio calculated for the whole brain varies in our dataset between 3 neurons per microglial cell in the giraffe brain and 45 neurons for every microglial cell in the elephant shrew brain (Table 1).
Discussion
Here we find that microglial cells occur in fairly similar and low densities across brain structures as different as multiple sites along the anteroposterior axis of the cerebral cortical gray and white matter; the Cb; the remaining structures in the hindbrain, midbrain, and forebrain; and across a wide sample of mammalian species that diverged as much as 218 million years ago (Murphy et al., 2004; Meredith et al., 2011). Finding consistent microglial cell densities across brain structures in such a wide range of mammalian species is indicative of little variation in the average size of these cells, as found previously for macrophages across species (Haley et al., 1991; Krombach et al., 1997). The unlikelihood that such consistent microglial cell densities occur by chance across so many brain structures and species indicates that microglial cell size and distribution have been conserved in mammalian evolution, whether due to biological, physiological, or physical reasons. An example of biological constraint would be the existence of fundamental morphologic, genetic, or regulatory mechanisms that constrain cell size, which may have been lifted only in some cell types, such as neurons (Savage et al., 2007; Herculano-Houzel and Dos Santos, 2018). There might also be physiological constraints in place, for instance, due to the volume-based macrophage-like functions of microglial cells, that render unviable any major variation in their size and/or distribution. Additionally, if microglial cell size is constrained and microglial cell proliferation is self-controlled such that it is arrested by contact inhibition, when confluence is reached in any tissue during development, a physical (mechanical) constraint to microglial cell distribution will follow, resulting in fairly constant microglial cell densities, as observed here. In support of such paired biological-physical constraints to microglial cell densities, contact inhibition has been reported during microglial progenitor cell proliferation (Gomez-Nicola and Perry, 2015; Askew et al., 2017) as well as during active scanning of the parenchyma, maintaining the mosaic distribution of microglial cells in the tissue (Nimmerjahn et al., 2005). Such a constant mosaic (tiled) distribution of microglial cells is also consistent with our finding of fairly constant densities of microglial cells across brain structures and species. Consequently, each microglial cell seems to take care of a territory of fairly similar volume across all brain structures and mammalian species.
Importantly, the remarkable constancy in microglial density reported here does not imply that microglial cells are functionally similar or equivalent across structures, and there is evidence to the effect that they are not (Xavier et al., 2015). It remains to bedetermined whether microglial cell densities are also constant across brain structures and species during development. While developmental trajectories appear to differ across mammalian species (Ueno and Yamashita, 2014; Matcovitch-Natan et al., 2016; Hanamsagar et al., 2017), the process of embryonic colonization by microglia is highly conserved across vertebrate species (Perry et al., 1985; Ashwell, 1991; Herbomel et al., 2001; Verney et al., 2010; Schlegelmilch et al., 2011; Swinnen et al., 2013). In mice, total numbers of microglial cells were found to peak at 3 weeks after birth and to decrease thereafter, showing similar densities at 1 and 9 months of age (Nikodemova et al., 2015). Aging has been shown in several species to affect microglial cell morphology and soma size, but not microglial cell density (Davies et al., 2017; Chan et al., 2018; Ali et al., 2019), which is consistent with our findings of relatively similar microglial densities across a wide range of species whose chronological age we cannot ascertain due to the opportunistic nature of our specimen collection.
Because microglial densities are similar while neurons are distributed unevenly and variably across brain structures and species, it follows that the number of neurons in the territory of each microglial cell is widely dissimilar across structures and species. Here we find that there is only rarely more than one microglial cell to each neuron in the brain; typically, there are one or more neurons per microglial cell in any brain tissue, in any mammalian species. As a consequence, to the extent that the physiological load on microglial function depends on the number of neurons served or monitored, microglial function should be more strained where there are more neurons per microglial cell (i.e., where neuronal density is higher), for instance, in the granular layers of the hippocampus and Cb, and in the densest regions of the Cx, such as entorhinal cortex. It is thus possible that the high vulnerability of these regions to neurodegenerative diseases, such as Alzheimer's disease (Braak et al., 1993; Larner, 1997; Mu and Gage, 2011; Jacobs et al., 2018), is at least partly associated with a high susceptibility to inflammation given their heavy neuronal load per microglial cell. Given the evidence that aging affects microglial function and morphology but not densities, as mentioned above, compensatory alterations are unlikely, such that different neuron/microglia ratios across brain sites probably have a lasting role in establishing susceptibility of different brain regions to insults over the lifespan. Similarly, one could expect the Cx to be more susceptible to inflammation in mammalian species with higher neuronal densities, such as mice and rats, compared with the human Cx. In that case, the laboratory rabbit, with neuronal densities closer to those found in humans (McMullen et al., 1994), would be a better model to study inflammation in the human cortex.
We report microglial percentages among non-neuronal cells in mammalian cortices that are comparable with the 5% found previously in the human cortical gray matter (Pelvig et al., 2008). Strikingly, not only the density of microglial cells is fairly constant across brain structures and species, but they also represent a fairly constant 7% of all non-neuronal cells. The latter include vasculature-related cells, which are possibly also a fairly constant (and small) percentage of all non-neuronal cells, given that capillary density appears to be consistently low across the few species sampled so far (Lauwers et al., 2008; Ventura-Antunes and S.H.-H., manuscript in preparation). A constant microglial percentage of all non-neuronal cells is consistent with neural tissue being composed of fixed percentages of astrocytes and oligodendrocytes as well (Herculano-Houzel and Dos Santos, 2018). Importantly, constant densities of each non-neuronal cell type in any brain structure and species would follow naturally if the average cell size of each cell type were invariable (Savage et al., 2007). Alternatively, the lack of much variation in cell density might be particular to microglial cells, since they do not originate from the same precursors that give rise to neurons and macroglial cells (astrocytes and oligodendrocytes) during development (for review, see Herculano-Houzel and Dos Santos, 2018). Our ongoing studies estimating directly the densities of astrocytes and oligodendrocytes will address whether or not macroglial cells also occur in similar densities across brain structures and species. Importantly, microglial cells are no exception in occurring at invariant densities and thus presumably constant average cell size: this seems to be the case of most somatic cell types other than neurons (with the possible exception of alveolar macrophages and adipocytes) (Savage et al., 2007). Neurons might thus be one of the rare cell types that escape the limitations that constrain cell size that have applied to other cell types over 200 million years of mammalian evolution.
We found that microglial densities do vary somewhat between cortical gray and white matter, with higher densities found in the latter, which might be simply a direct consequence of the scarcity of neuronal cell bodies in the cortical white matter. A higher density of microglial cells in white than in gray matter agrees with previous results in the adult rat (Savchenko et al., 2000), mouse (Askew et al., 2017), and human brain (Mittelbronn et al., 2001; Askew et al., 2017). Our finding that similar proportions of all cells are microglial between the gray and white matter is in conflict with previous results where authors found either higher proportions of microglia in the gray matter in the mouse Cx (Lawson et al., 1990), or in the white matter in the human Cx (Mittelbronn et al., 2001). While these differences in proportions might be due to the different technique applied in our study (the isotropic fractionator, instead of stereological methods) (see von Bartheld et al., 2016), they are most likely explained by the use in previous studies of antibodies that are now known to not stain all the microglial cell population (F4/80) (Lawson et al., 1990) or CD68, LCA, LCF, and MHC-II (Mittelbronn et al., 2001).
The highly variable ratio of neurons served per microglial cell that results from the fairly constant microglial cell densities uncovered here might have a very fundamental consequence for brain tissue function due to the role of microglial cells in synaptic pruning (Paolicelli et al., 2011; Schafer et al., 2012; Ueno et al., 2013; Li and Barres, 2018). While it is possible that constant microglial densities might be imposed by constant synaptic densities (which has been largely assumed to be the case, but for which solid, systematic evidence is still lacking) (O'Kusky and Colonnier, 1982; Miki et al., 1997), we find it much more likely that it is the fairly constant microglial densities reported here that contribute to imposing fairly constant synaptic densities across brain structures and species. One reason is the very early invasion of the developing parenchyma by microglial progenitors, well before formation of the first synapses (Ginhoux et al., 2010). Similarly, we expect that fairly constant densities of astrocytes would also induce fairly constant densities of synapses across brain structures and species. To date, the only systematic survey of synaptic densities was described recently for the adult mouse brain (Zhu et al., 2018). Only a systematic analysis of the distribution and densities of synapses in a large set of mammalian species will allow these fundamental questions about brain tissue organization to be addressed.
Footnotes
The authors declare no competing financial interests.
This work was supported by the Carlos Chagas Filho Foundation for Research Support of the State of Rio de Janeiro-FAPERJ (SDS fellowship), generous crowdfunding support by Brazilian supporters to the former S.H.-H. laboratory at Universidade Federal do Rio de Janeiro, the James S. McDonnell Foundation Grant 220020232 to C.C.S. and Grant 220020293 to S.H.-H., and Vanderbilt University laboratory startup funds to S.H.-H. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. We thank Anderson Miller and Louise Botelho for help in experiments; Mary Ann Raghanti for providing marsupial brains; and Laura Trice for sectioning the brains used for immunohistochemistry.
- Correspondence should be addressed to Suzana Herculano-Houzel at suzana.herculano{at}vanderbilt.edu