Allometric analysis of brain cell number in Hymenoptera suggests ant brains diverge from general trends

Abstract

Many comparative neurobiological studies seek to connect sensory or behavioural attributes across taxa with differences in their brain composition. Recent studies in vertebrates suggest cell number and density may be better correlated with behavioural ability than brain mass or volume, but few estimates of such figures exist for insects. Here, we use the isotropic fractionator (IF) method to estimate total brain cell numbers for 32 species of Hymenoptera spanning seven subfamilies. We find estimates from using this method are comparable to traditional, whole-brain cell counts of two species and to published estimates from established stereological methods. We present allometric scaling relationships between body and brain mass, brain mass and nuclei number, and body mass and cell density and find that ants stand out from bees and wasps as having particularly small brains by measures of mass and cell number. We find that Hymenoptera follow the general trend of smaller animals having proportionally larger brains. Smaller Hymenoptera also feature higher brain cell densities than the larger ones, as is the case in most vertebrates, but in contrast with primates, in which neuron density remains rather constant across changes in brain mass. Overall, our findings establish the IF as a useful method for comparative studies of brain size evolution in insects.

1. Introduction

Brain size is commonly measured as a mass or volume, variables that do not necessarily correlate well with the behavioural ability [1], and cannot distinguish changes in cell size from changes in cell number. Because the absolute number of neurons and their connectivity better correlate with information processing capacity than relative brain size alone [2], analysing changes in cell density within and across clades may provide specific, testable hypotheses about brain size and cognitive evolution [3,4]. Additionally, while brain size evolution comprises ultimate (i.e. selection on behaviour or energy use) and proximate (i.e. cell division, growth or differentiation) components, evolutionary studies have overwhelmingly focused on ultimate hypotheses (e.g. diet [5], sociality [6,7], parental care [8]). Proximate questions addressing how differences in size, hence expenditure of metabolic energy, are achieved are less common (exceptions include [3,9–11]), but are central to understanding how selection acts on brain development to drive variation in brain size over evolutionary history [12]. Therefore, measuring brain cell number and density can advance studies seeking unifying patterns of brain and cognitive evolution [11].

Historically, a considerable barrier to estimating the total number of neurons in a brain or brain region has been the tedious nature of cell counting methods. Very few estimates for total brain cell number existed until Suzanna Herculano-Houzel [13] introduced the isotropic fractionator (IF) method, allowing rapid estimation of neuron and glia numbers in a brain. In the IF method brain tissue is homogenized and subsampled, with nuclei counts scaled to estimates of total cells in a brain or brain region. It has vastly expanded the taxonomic sampling of brain cell numbers in vertebrates [14–16]. Studies quantifying whole-brain and region-specific cell numbers have revealed that clade-specific changes in neuron density do not always correspond with volumetric differences [17]. In particular, birds [16] and primates [18] have especially cellularly dense brains composed of many small neurons, with forebrain neuron populations of some birds comparable in number to much larger primates [16].

We adapted the IF method to provide brain cell number estimates for Hymenoptera (sawflies, wasps, bees and ants), species of which vary widely in body size, diet, life history and behaviour, and span a vast spectrum of social structures. Hymenoptera have been the focus of neuroethological and comparative neuroscience for more than a century (e.g. [19,20]). However, while brain mass or volume measurements exist for numerous ant [19–23], bee [24–26] and wasp [27,28] species, an estimate of total brain cell number is available only for the European honeybee (Apis mellifera) [29]. Here, we report allometric brain-body scaling relationships, including estimates of total brain cell numbers for major Hymenoptera superfamilies using the IF method and establish the IF as a useful approach for studying brain size evolution in insects.

2. Material and methods

(a). Morphometrics

We examined female specimens, excluding queens from social species, from 40 hymenopteran species (electronic supplementary material, table S1). Total body mass (BM) and fixed, wet brain mass including optic lobes and suboesophageal ganglion (BrM) of specimens were measured to the nearest 0.01 mg using an analytical balance (Mettler Toledo AT261; Marshall Scientific, Hampton, NH, USA). Brains or whole heads were fixed, rinsed, stored in a buffer and then gently blotted dry for weighing. For collection and dissection details and a list of vouchered specimens, see the electronic supplementary material, Materials and Methods and tables S3 and S8.

(b). Isotropic fractionator

We adapted the IF method described by Herculano-Houzel & Lent [13] to estimate all of the cells in insect brains. To do this, we homogenized the entire brain in a glass tissue homogenizer with a sodium citrate detergent solution to dissociate nuclei from cells. We then diluted the sample, labelled cell nuclei with the DNA-specific fluorescent probe, DAPI (D9542 Sigma-Aldrich; St Louis, MO, USA) and counted them with a haemocytometer under epifluorescence. Ten subsamples of each homogenized brain were counted, providing a mean and standard deviation of total nuclei number (NN) for each specimen brain. A coefficient of variation (CV) was calculated for each brain to measure how well the brain was homogenized. To assess how methodological error compares with variation in population mean estimates, we calculated an error ratio (ER), defined as the ratio of variation in the IF method to variation in estimates of NN for each species. Cell density (CD) was obtained by dividing the mean number of nuclei for a given brain by its mass and assumes one nucleus per brain cell. Detailed histological and counting methods are included in the electronic supplementary material.

We assessed the quality of our adaptation of the IF method by comparing IF nuclei estimates with those from sectioned brains for the vinegar fly (Drosophila melanogaster) and a desert ant (Novomessor spp.). With the exception of D. melanogaster, a reliable nuclear glial marker does not yet exist for insects [30], so our method does not distinguish neurons and glia. However, the proportion of glia to neurons in the insect brain may be far lower than that of vertebrates (ca 5–10% in D. melanogaster; [31,32], but see [33]).

(c). Data analysis

Statistical analyses were carried out in R (v. 4.0.2). To assess the brain size scaling relationships in Hymenoptera we used species-level mean log-transformed values of BM, BrM and NN in phylogenetic least-squares (PGLS) regression. We used the SLOUCH package in R, which takes measurement error (ME) into account [34], with a Brownian motion model and the Hymenoptera phylogeny from Peters et al. [35]. Some genera included in our dataset were not included in this phylogeny, so we additionally performed ordinary least square (OLS) regression to get estimates for the entire dataset. Specific species included in OLS (n = 38 for BrM; n = 29 for NN) and PGLS (n = 26 for BrM; n = 20 for NN) are noted in the electronic supplementary material, table S1. To test for differences among superfamilies in BrM or NN based on BM or BrM, respectively, we performed ANOVA on linear models of log-transformed values that included superfamily. In cases where superfamily was significant, we followed with post hoc pairwise comparisons using Tukey's honest significant difference test.

Brain size scaling in Apoidea (bees and ‘sphecoid’ wasps) and Formicidae (ants) were also analysed separately using PGLS with ME. For PGLS, we used the Apoidea phylogeny from Sann et al. [36] and the Formicidae phylogeny available on AntWiki [37]. To test for differences in BrM, NN and CD among species, we performed ANOVA on linear models that included species and BM or BrM as independent variables. Generalized linear models using a Gaussian distribution with a log link function were used except for CD where a linear regression model on log-transformed variables showed better conformity with model assumptions. To test whether the IF method could be used to detect NN differences within species (i.e. for correlation with variables such as genetic background, behaviour, developmental environment, … etc), we used linear models on untransformed BM, BrM and NN values to describe scaling relationships in two species that show body size polymorphism in the worker caste, the common eastern bumblebee (Bombus impatiens) and a carpenter ant (Camponotus festinatus).

3. Results

We report the BM and BrM from 39 Hymenoptera species and provide total brain nuclei number (NN) estimates for 32 species in seven Hymenoptera superfamilies (electronic supplementary material table S1). Our samples span a body mass range from 0.22 mg to greater than 900 mg and brain mass range from 0.07 mg to greater than 10 mg (figure 1; electronic supplementary material, table S1).

Figure 1.

Phylogeny (adapted from [35]) of hymenopteran genera included in phylogenetically controlled brain-to-body mass regression analysis (* =not included). Symbols (diamond, circle, star, etc.) associated with family names are used in subsequent figures. The size of grey dots logarithmically represents the brain and body mass of the respective genera's specimen. Arrows point at images of some of the examined species. Note that the species vary considerably in body size as indicated by the length of the respective scale bars, each representing 2 mm. The inset demonstrates the size range of the sampled species, representing the largest (Pepsis thisbe) and the smallest (Leptopilina hetero) species and, for comparison, a honeybee (Apis mellifera), all reproduced at the same scale (scale bar, 10 mm). The images of Neodiprion and Calliopsis have been modified with permission from originals provided by Andrey Ponomarev (©Andrey Ponomarev, http://insecta.pro) and, respectively, Heather Horn (©Heather Horn; www.pollinatorsnativeplants.com). (Online version in colour.)

(a). Verifying the isotropic fractionator method for insect brains

The CV for specimen-level NN estimates ($\overline{x}$ = 0.203, s = 0.078 across all species) was often higher than reported for vertebrates [17], particularly for small brains (electronic supplementary material, figure S2) and when samples were diluted to less than 0.5 μg ml⁻¹ (electronic supplementary material, table S2 and figure S3). Aside from the comparisons of Acromyrmex versicolor with Apis mellifera or B. impatiens, CV does not differ significantly across genera (electronic supplementary material, figure S2A). There is a weak, negative correlation between CV and NN estimates, which may be owing to small brains being too diluted prior to counting (electronic supplementary material, figure S2B). The ER ranged from 0.44 to 3.15 but was not correlated with NN (electronic supplementary material, figure S2E), suggesting variability in the method is not biased with respect to variables used in our statistical analyses. Not surprisingly, both the variation in CV and ER decrease with increased sample size (electronic supplementary material, figure S2C,F). Indeed, some of our NN estimates comprise a very small number of individuals and larger samples sizes would improve estimates.

Our estimate of total brain cell number for the European honeybee (Apis mellifera; $\overline{x}$ ≈ 6.13 × 10⁵, s = 1.28 × 10⁵; electronic supplementary material table S1) was lower than the existing estimate from brain sections ≈ 8.5 × 10⁵ [29]. Similarly, our D. melanogaster estimate is slightly lower than the published estimates (ca 100 000 neurons ([33,34]; electronic supplementary material, figure S1). However, our estimates from sectioned brains of a desert ant (Novomessor spp. ≈ 9 × 10⁴, (electronic supplementary material, figure S1) and the vinegar fly (D. melanogaster ≈ 9.25 × 10⁴; electronic supplementary material, figure S1), were comparable with those from the IF ($\overline{x}$ ≈ 7.02 × 10⁴, s = 2.4 × 10⁴ and $\overline{x}$ ≈ 8.83 × 10⁴ s = 1.38 × 10⁴, respectively; electronic supplementary material, table S1).

(b). Brain size scaling in Hymenoptera

In the PGLS model of brain scaling across all Hymenoptera, BrM scaled with BM comparably across genera (BrM = −2.96BM^0.696; figure 2a, electronic supplementary material, table S4) with the exception of ants (Formicoidea), which have smaller brains than predicted for their body mass when compared with bees and related wasps (p < 0.001 post hoc comparison of Apoidea versus Formicoidea). We found that IF-estimated NN scaled with BrM across genera (NN = 12.89BrM^0.597; figure 2b,c; electronic supplementary material, table S4). Ants appear to have less cellularly dense brains than other clades. In post hoc comparisons following OLS regression, brain mass controlled NNs in ants (log-transformed, $\overline{x}$ = 11.8, s = 0.162) were lower than Apoidea ($\overline{x}$ = 13.0, s = 0.11, p < 0.001), Pompiloidea ($\overline{x}$ = 13.5, s = 0.406, p = 0.0142) and Vespoidea ($\overline{x}\hspace{0.17em}$ = 13.3, s = 0.223, p = 0.0016).

Figure 2.

Brain size scaling across Hymenoptera. (a) Brain-body mass regression analysis for Hymenoptera from PGLS. (b) Sample-level values used to calculate mean values for (c). (c) Brain mass and nuclei number regression from PGLS. Family indicated by shape, superfamily by colour (light blue, Tenthredinoidea, dark blue, Formicoidea; green, Apoidea; red, Pompiloidea; pink, Vespoidea; orange, Ichneumonoidea). Separate analyses of Apoidea (green, subscript A) and Formicidae (blue, subscript F) indicated by dotted lines. Saturated symbols represent species included in PGLS. For specific species used in each analysis see the electronic supplementary material, table S1; for samples sizes used in each analysis, see the electronic supplementary material, table S4. (Online version in colour.)

Because our data suggest Apoidea and Formicoidea evolved different brain allometries, we also analysed these clades independently. Within Apoidea, the BrM to BM allometry (BrM = −2.13BM^0.677; figure 2a; electronic supplementary material, table S4) was similar to that found for Hymenoptera overall (figure 2a). This is explained in part by a large number of genera from this superfamily spanning a large body size range in our dataset (figure 2c; electronic supplementary material, table S1). In comparison, Formicoidea had a much shallower BrM to BM allometry (BrM = −2.81BM^0.496; figure 2a; electronic supplementary material, table S4). The NN to BrM allometry for Apoidea (NN = 13.02BrM^0.420) and Formicoidea (NN = 12.07BrM^0.408) are more comparable, though the latter shows a relatively weak fit to the data (figure 2c; electronic supplementary material, table S4). Additionally, BM is more strongly associated with CD in Apoidea than Formicoidea (figure 3a,c), though variation in the Formicoidea data probably prevented us from detecting a relationship between CD and BM in ants.

Figure 3.

Brain cell density scaling with body size in Apoidea (a) and Formicoidea (c) and brain cell density differences among Apoidea (b) and Formicoidea (d) species. Apoidea family symbols in (a) and (b) from figure 1. Grey symbols in (a) and (c) represent individual data points. Pairwise comparisons from Tukey's HSD following least-squares regression for brain mass (b,d, ranked by body mass). For full species names, see the electronic supplementary material, table S1. For model statistics, see the electronic supplementary material, table S5. Species listed in grey in (a) not included in PGLS. Grey dots in (c) and (d) not included in statistical analyses because either estimate is from one specimen (c) or because BM was not measured (d). Letters denote statistically significant differences among groups (p < 0.05). (Online version in colour.)

Brain size scaled comparably across many bee and ant species, but there were a few species with particularly large or neuron-dense brains for their size (electronic supplementary material, figure S4; overall model statistics, electronic supplementary material, table S6). Within Apoidea, a halicitid bee (Augochlorella sp.) stood out as having particularly high NN for its BrM, and correspondingly high cell density for its body mass (figures 2a and 3b; electronic supplementary material, figure S4A). Across ants, a trap-jaw ant (Odotonmachus sp.) stood out as having greater NN compared with a number of similarly sized genera (electronic supplementary material, figure S4C,D).

Finally, we tested whether the IF method could be used to detect intraspecific differences in brain size scaling in species with pronounced body size polymorphism. We found that in the common eastern bumblebee (B. impatiens) BrM scaled linearly with body size (electronic supplementary material, figure S5A), whereas we could detect only a very weak correlation between NN and BrM (electronic supplementary material, figure S5B). In a carpenter ant (Camponotus sp.), BrM scaled linearly with BM (electronic supplementary material, figure S5C) and NN scaled linearly with BrM (electronic supplementary material, figure S5D).

4. Discussion

(a). Estimating brain cell numbers in insects

The advantage of the IF method is its relative simplicity and speed, making it particularly suited for comparative work, which is most robust with large sample sizes and broad taxonomic coverage. In vertebrate studies, unbiased stereological methods [38] yield cell numbers comparable to those generated using the IF [39]. Our data suggest this is also true for insects and that the adaptation of the IF method will be useful for studies of brain size evolution in insects. While we manually counted cells, the method could be substantially expedited using flow cytometry for automated cell counting [40]. Additionally, the use of nuclear markers could aid in distinguishing functionally distinct cell types and their population sizes, in particular glia.

Our adaptation of the IF method may not yet be suitable for within-species comparison in instances where differences in brain cell number across individuals are small. Our results from analyses of brain size scaling in polymorphic species (electronic supplementary material, figure S5) and assessment of error (electronic supplementary material, figure S2A–F) indicate that controlling sources of error in implementing the method are crucial to detecting subtle differences across individuals of a species. We suspect some of the variation reported for Bombus and other flying Hymenoptera is owing to the inclusion of damaged or incomplete peripheral visual structures (retina and lamina) in samples during dissection, and that the complete removal of these tissues, rather than attempting to include them in their entirety, could reduce sample variation. In many bees and wasps the retina can be removed post-fixation and we suggest it could be weighed and counted separately to quantify retina cell density.

(b). Cell density in Hymenoptera brains

Many studies in vertebrates and insects focus on relative brain size relationships and correlations often do not adequately explain ecological or behavioural differences. By contrast, neuron numbers may be better suited to appreciate information processing capacities in the context of comparative studies [11]. A plain example is the comparison of brain size versus neuron density in the mammalian and bird brains, in which birds have smaller brain masses (the proverbial ‘bird brain’), but much higher neuron densities than mammals. Neuron densities of corvids and parrots surpass those of some much larger primates, an attribute that correlates with the advanced cognitive capacities of these taxa. Overall, the highest neuron densities have been found in the smallest respective species examined (smoky shrews in mammals; 2.08 × 10⁵ neurons mg⁻¹ [14] and goldcrests in birds; 4.9 × 10⁵ neurons mg⁻¹ [16]). The Hymenoptera in our sample have on average higher cell densities than vertebrates (5.94 × 10⁵ cells mg⁻¹; n = 30 species).

In vertebrates, regional neuron densities can be exceptionally high (up to 10⁶ neurons mg⁻¹ brain mass in the cerebellum of small birds [16]). Coarse estimates of Kenyon cell density (neurons comprising the insect mushroom body) from sectioned bumblebees are 6.3 × 10⁶ cells mg⁻¹, a number comparable to bird cerebellar densities, but considerably higher than cellular densities seen in any other species. These numbers were extrapolated from [41], which was based on sectioned material without controlling for tissue shrinkage, suggesting those values may represent an overestimate. Owing to their small size and brain structure, we can only homogenize entire insect brains and therefore can only compare our IF data with the averaged cell densities of vertebrate brains. Even so, some of the smallest species in our sample have considerably higher total brain cell densities than even the cerebelli of small birds (e.g. the halictid bee Augochlorella: 2.0 × 10⁶ cells mg⁻¹; or the andrenid bee Calliopsis: 1.8 × 10⁶ cells mg⁻¹).

In the small Hymenoptera, neuron size may be a limiting factor for brain miniaturization, as shown for the smallest insects (the parasitoid wasp Megaphragma [42]), whose larval brains comprise less than 5000 cells, the cell bodies of which are lost during pupation. The brain of the smallest species in our sample (the parasitoid wasps Leptopilina; figure 1) comprised around 30 000 cells (electronic supplementary material, table S1). Similar neuron numbers (2.2 × 10⁵–3.7 × 10⁵ neurons; [43]) have been estimated for fairyflies (Hymenoptera: Mymaridae), which are smaller than the smallest of our samples, suggesting that our cell number estimates may be conservative.

Birds have particularly small and cellularly dense brains, probably owing to evolutionary pressure on small body mass associated with flight, which is energetically demanding and body mass-limited (reviewed in [44]). Hymenoptera, with the exception of ant workers, comprise mainly flying insects and under the logic of evolutionary pressure on body size accompanying flight, one might expect flying insects to have relatively smaller, dense brains. Instead, we found that while flying bees indeed have greater cell density for a given body size, it is the flightless ants which have relatively small brains (figures 2a and 3c,f). An intuitive hypothesis for ants' relatively small brains may be that flying insects have larger eyes and larger visual brain centres (optic lobes) than most ants because flight control relies on rapid visual information processing. Most ant species have small eyes, and accordingly their optic lobes are comparatively small (electronic supplementary material, figure S6). Additionally, the optic lobes of insects comprise many tiny neurons, which may explain why the flying Hymenoptera with brain mass comparable to ants showed higher cell density (figure 2c). Comparisons of non-flying ants and flying ants (e.g. males, virgin queens) and the inclusion of ant genera that feature pronounced visual worker behaviours (e.g. Gigantiops, Myrmecia, or Harpegnathos) are necessary to test this intuitive hypothesis. The optic lobes can be carefully removed from brains of flying Hymenoptera, and a comparison of central versus optic lobe brain cell densities could aid in testing whether mass limitations drive brain size as is observed in birds.

(c). Do ants have uniquely small brains?

Wehner et al. [22] found that ants have smaller brains than expected for vertebrates of comparable body size, with a regression slope similar to that of birds (figure 4c; electronic supplementary material, table S7). These authors propose that the overall smaller-than-expected brain size of insects may be based on the differences between the insects’ relatively heavier exoskeleton and vertebrates' more lightweight endoskeleton. However, instead of skeletal mass differences, our data suggest that, by including only ants in their analysis, Wehner et al. [22] may have happened to choose particularly small-brained insects (figure 4). When we plot our hymenopteran dataset against vertebrate data, insect brain mass (figure 4a) and neuron number (figure 4c) seem to match the vertebrate allometries quite well. However, ant data are situated below the OLS regression line for brain mass (figure 4a) and for neuron number (figure 4c). This is in line with observations of Eberhard & Wisclo [11], who discussed the unusual grade shift in ants in their meta-analysis of brain-body allometries across animals spanning 15 orders of magnitude. Our data support ‘Haller's rule’ [11,46], which shows that smaller animals in general have relatively larger brains (electronic supplementary material, figure S7; samples < 1 mg body mass omitted owing to extreme grade shifts in miniature arthropods [11]).

Figure 4.

Comparisons of insect brain allometries with estimates from vertebrate studies. Brain-body mass (a,b) and brain mass–neuron number comparison (c) of Hymenoptera and vertebrates; logarithmical plots. Vertebrate data in (a) are from a compilation by [45], in (b) compiled from ([23,25,45–51]; electronic supplementary material, table S7) and neuron data in (c) are from ([52]; mammals) and from ([53]; birds). In (a,b), we included a large sample of ant data (green crosses; 62 species from [19] and two species from [54]) to supplement our own data (magenta squares in a–c). Note that vertebrate data in (c) represent neurons only, whereas our hymenopteran data include glia (which are much less numerous in insects than in birds and mammals). Slope lines in (a,c) are linear regressions whereas slopes in (b) include more complex regression models (for details, see the electronic supplementary material, table S7). (Online version in colour.)

Our estimate of the slope for ant brain mass allometry (0.496, s.e. = 0.136; figure 4b) is lower than previous estimates (0.670 reported by [19] and 0.567 reported by [22]), an effect in part probably owing to the narrow body size range of the included genera. However, Wehner et al. [22] reported a 95% confidence interval of 0.329–0.931 (electronic supplementary material, table S7), which encompasses our estimate, so the addition of more genera is necessary to confidently compare this allometry with other insects or vertebrates.

(d). Conclusions

Most comparative neurobiological studies focus on narrower ranges of taxa, often to correlate certain sensory, behavioural or environmental aspects across different species, genera or subfamilies with their respective differences in brain composition. Such studies can only be interpreted in a meaningful way if the general brain-body relationships are known on a larger taxonomic scale, such as we have begun here for the order of Hymenoptera. More importantly, we want to promote the use of the IF in insects and other small invertebrates with this introductory study. Since its first publication [13], this method has dramatically increased our understanding of how brains are differently designed in different vertebrate orders. For instance, similar studies of cell numbers and density help explain primates' and birds’ advanced cognitive abilities as compared with rodents, but have also highlighted that human brains are not unique in these qualities when compared with other primates. While our cell counting data for Hymenotpera are still exploratory, we hope that other researchers will adapt and improve this tool for their respective invertebrate brain research. The IF technique has a high potential to shed light on current disputes in brain evolution and to address new questions that cannot be answered by comparing brain mass or volume alone.

Ethics

While experiences of pain and suffering in insects and other arthropods are somewhat contentious topics, we acknowledge that insects experience distress upon being handled. Therefore, we anaesthetized all specimens on ice prior to measurement and sacrifice for experiments. We also collected the minimum number of specimens we expected to need for experiments, and piloted methods on specimens available in large numbers from laboratory or local colonies.

Data accessibility

The dataset supporting this article has been uploaded as part of the electronic supplementary material and is also available from the Dryad Digital Repository: https://dx.doi.org/10.5061/dryad.3xsj3txdt [55].

Authors' contributions

R.K.G. participated in research design, collected experimental animals, performed cell counting experiments, analysed data and co-wrote the manuscript; W.G. designed research, collected experimental animals, analysed data and co-wrote the manuscript; M.S. collected experimental animals, performed cell counting experiments and critically revised the manuscript. All authors gave final approval for publication and agree to be held accountable for the work reported therein.

Competing interests

We declare we have no competing interests.

Funding

This work was supported by the National Science Foundation (ISO-1354191 to W.G.) and a fellowship from the University of Arizona Graduate Interdisciplinary Program (GIDP) to R.K.G.