Concerted developmental programming may constrain changes in component structures of the brain, thus limiting the ability of selection acting on individual brain compartments to form an adaptive mosaic independent of total brain size or body size. Measuring patterns of gene expression underpinning brain scaling in conjunction with anatomical brain atlases can aid in identifying influences of concerted and/or mosaic evolution. Species exhibiting exceptional size and behavioral polyphenisms provide excellent systems to test predictions of brain evolution models by quantifying brain gene expression. We examined patterns of brain gene expression in a remarkably polymorphic and behaviorally complex social insect, the leafcutter ant Atta cephalotes. Approximately ~50% of differential gene expression observed among three morphologically, behaviorally, and neuroanatomically differentiated worker size groups was attributable to body size, but we also found strong evidence of differential brain gene expression unexplained by worker morphological variation. Transcriptomic analysis identified patterns of gene expression not linearly correlated with worker size but rather, in some cases, mirroring neuropil scaling. Additionally, we observed enriched gene ontology terms associated with nucleic acid regulation, metabolism, neurotransmission, and sensory perception, further supporting a relationship between brain gene expression and worker social role. These findings demonstrate that differential brain gene expression among polymorphic workers is linked to behavioral and neuroanatomical differentiation underpinning complex agrarian division of labor in A. cephalotes.

Brains of mature, fully sclerotized Atta cephalotes workers categorized by size group as minims (0.5–0.7mm in head width), medias (1.7–1.9mm in head width), and majors (≥3mm in head width) were sampled from three mature colonies for gene expression analyses. We prepared between nine and 11 samples from each worker size group, composed of three to 10 pooled brains depending on worker size group, distributed across three colonies of origin, for a total of 30 samples.

Total RNA was extracted from worker brains using a ThermoFisher PicoPure kit. Sample quality and quantity, as well as lack of protein or DNA contaminants, were assessed using a Thermo Scientific Nanodrop spectrophotometer and an Agilent Bioanalyzer 2100, respectively.

Libraries were sequenced using a combination of Illumina NextSeq and MiSeq with SE 75 reads. RNAseq unstranded libraries with mRNA poly-A selection were prepared by Harvard BioPolymers using a KAPA mRNA HyperPrep kit. mRNA sequence libraries were individually barcoded and multiplexed in equal proportions and all libraries were sequenced across four lanes.

Kallisto (Bray et al., 2016) was used to pseudoalign sequenced reads to the available transcriptome and to quantify transcript abundance. The kallisto index file was created using the A. cephalotes version 1.0 cDNA set accessed through Ensembl Metazoa Genomes (Howe et al., 2020). DESeq2 (Love et al., 2014) was used to statistically assess the significance of differential gene expression based on transcript abundance counts generated by kallisto.

Weighted gene coexpression network analysis (WGNCA) was performed using the WGCNA R package (Langfelder & Horvath, 2008). biomaRt was used to assign GO categories to all expressed genes in our data set (Smedley et al., 2009).