Brain evolution is hypothesized to be driven by behavioral selection on neuroarchitecture. We developed a novel metric of relative neuroanatomical investments involved in performing tasks varying in sensorimotor and processing demands across polymorphic task-specialized workers of the leafcutter ant Atta cephalotes and quantified brain size and structure to examine their correlation with our computational approximations. Investment in multi-sensory and motor integration for task performance was estimated to be greatest for media workers, whose highly diverse repertoire includes leaf-quality discrimination and leaf-harvesting tasks that likely involve demanding sensory and motor processes. Confocal imaging revealed that absolute brain volume increased with worker size and functionally specialized compartmental scaling differed among workers. The mushroom bodies, centers of sensory integration and learning and memory, and the antennal lobes, olfactory input sites, were larger in medias than in minims (gardeners) and significantly larger than in majors (“soldiers”), both of which had lower scores for involvement of olfactory processing in the performance of their characteristic tasks. Minims had a proportionally larger central complex compared to other workers. These results support the hypothesis that variation in task performance influences selection for mosaic brain structure, the independent evolution of proportions of the brain composed by different neuropils.

Mature fully sclerotized workers collected from colonies Ac09, Ac16, Ac20, and Ac21 were decapitated immediately prior to brain dissection and fixation. Workers were sampled from five worker size groups identified by head width (HW): minims (0.6mm±0.1mm), medias (1.2mm±0.1mm, 1.8±0.1mm, or 2.4mm±0.1mm), and majors (3.0mm or larger). Brains (n=30) from workers sampled from Ac09, Ac20, and Ac21 were dissected in ice-cold HEPES Buffered Saline (HBS), placed in 16% zinc-formaldehyde (Ott 2008) and fixed overnight at room temperature (RT) on a shaker. Whole brains were processed to visualize the presynaptic protein synapsin. Fixed brains were washed in HBS six times, 10 minutes per wash, and fixed in Dent’s Fixative (80% MeOH, 20% DMSO) for minimally 1 hour. Brains were then washed in 100% methanol and either stored at -17°C or immediately processed. Brains were washed in 0.1M Tris buffer (pH=7.4) and blocked in PBSTN (5% neutral goat serum, 0.005% sodium azide in 0.2% PBST) at RT for 1 hour before incubation for 3 days at RT in primary antibody (1:30 SYNORF 1 in PBSTN; monoclonal antibody anti-synorf 3C11 obtained from DSHB, University of Iowa, IA, USA; 62). They were washed 6x10 minutes in 0.2% PBST and incubated in the secondary antibody (1:100 AlexaFluor 488 goat anti-mouse in PBSTN) for 4 days at RT. Brains were then washed a final time (6x10 minutes in 0.2% PBST) and dehydrated in an ethanol and PBS series (10 minutes per concentration, 30/50/70/95/100/100% ethanol in 1x PBS), then cleared with and immersed in methyl salicylate, and mounted on stainless steel glass windowed slides for imaging.

Brains were imaged with a Nikon C2 confocal microscope and images were manually annotated using Amira 6.0 software to quantify neuropil volumes (not including cell bodies). The individual who annotated all brains for the study did not have any expectation of specific outcomes and did not have knowledge of predictions generated by our model. The annotation process involved using paintbrush- or magic wand-style tools to select areas to be included in a given neuropil in a given single scan of a 3D stack. The margins of focal neuropil regions were identified visually (or automatically when using the magic wand tool) based on the presence of synapsin staining. The magic wand-style tool was used primarily to annotate the antennal lobe glomeruli. Every third frame was annotated manually (or every other frame in the case of the antennal lobes) and intervening frames were filled in using the interpolation function of Amira. Interpolated frames were also checked and edited for accuracy. Annotated slices were then used to calculate the 3D volume of each neuropil using Amira and these data were exported for analysis. We recorded the volumes of OL, AL, MB, CX, SEZ, and ROCB. We use the term ROCB for simplicity and to correspond with our ability to associate specific compartments with sensorimotor functions to describe the tissue composed of the superior neuropils, lateral horn, ventrolateral neuropils, inferior neuropils, and ventromedial neuropils, as designated in a fruit fly brain (Ito et al. 2014). For the ALs, only glomerular tissue was included (excluding aglomerular neuropil and all soma layers). For the OLs, we measured only the medulla and lobula neuropils, excluding surrounding cell bodies. Similarly, measurements of the SEZ did not include somata. We also measured and separately examined substructures of the MB: the medial calyces (MB medial calyces), lateral calyces (MB lateral calyces), and peduncle and lobes (MB peduncle). Our peduncle measurements incorporated vertical and medial lobes; these metrics are included in all discussions of the peduncle. The volumes of these components were combined to quantify total MB size (total MB) across worker size groups. For bilateral structures, one hemisphere was measured, and for compartments located along the brain midline (SEZ and CX), the whole structure was measured (Supplementary Table 1; Supplementary Table 2). When calculating total brain volume, we excluded all soma layers and used only neuropil volumes.