Comparative transcriptomics reveals the conserved building blocks involved in parallel evolution of diverse phenotypic traits in ants

Abstract

Background: Reproductive division of labor in eusocial insects is a striking example of a shared genetic background giving rise to alternative phenotypes, namely queen and worker castes. Queen and worker phenotypes play major roles in the evolution of eusocial insects. Their behavior, morphology and physiology underpin many ecologically relevant colony-level traits, which evolved in parallel in multiple species.

Results: Using queen and worker transcriptomic data from 16 ant species we tested the hypothesis that conserved sets of genes are involved in ant reproductive division of labor. We further hypothesized that such sets of genes should also be involved in the parallel evolution of other key traits. We applied weighted gene co-expression network analysis, which clusters co-expressed genes into modules, whose expression levels can be summarized by their 'eigengenes'. Eigengenes of most modules were correlated with phenotypic differentiation between queens and workers. Furthermore, eigengenes of some modules were correlated with repeated evolution of key phenotypes such as complete worker sterility, the number of queens per colony, and even invasiveness. Finally, connectivity and expression levels of genes within the co-expressed network were strongly associated with the strength of selection. Although caste-associated sets of genes evolve faster than non-caste-associated, we found no evidence for queen- or worker-associated co-expressed genes evolving faster than one another.

Conclusions: These results identify conserved functionally important genomic units that likely serve as building blocks of phenotypic innovation, and allow the remarkable breadth of parallel evolution seen in ants, and possibly other eusocial insects as well.

Fig. 1

Phylogenetic relationships of 16 ant species studied shown with their pictures (source http://www.antweb.org/) and three biological traits: worker sterility (grey square, can lay unfertilized eggs; black square, completely sterile), colony queen number (grey square, single queen; black square, multiple queens), and invasiveness (grey square, not invasive; black square, can be invasive). The phylogenetic tree was constructed using OGG alignments with the software RAxML (v. 8) [87]. The data set contained 1427 genes and 3.59 Mb of sequence, and the analysis was partitioned by gene and conducted under a GTRGAMMAI model

Fig. 2

Correlation between module eigengenes and the biological traits (caste, worker sterility, colony queen number and invasiveness). Modules were clustered based on GO term similarities obtained with GOSemSim [93], which computes semantic similarity among sets of GO terms (Additional file 8). Expression of most modules is strongly associated with caste phenotypes. In addition, expression of several of these modules was also associated with other phenotypes, such as obligate worker sterility, colony queen number, and invasiveness. This shows that modules likely play multiple roles, and that their constituent genes have many functions

Fig. 3

Box plots showing the distribution of d _N/d _S ratios before accounting for OGG connectivity and expression levels for OGGs in non-caste-associated modules (NTA), OGGs in queen-associated modules (Queen) and OGGs in worker-associated modules (Worker), and calculated using PAML. The median d _N/d _S values are indicated above the boxplot. OGGs in worker-associated modules had significantly higher d _N/d _S than OGGs in queen-associated modules, and OGGs in non-caste-associated modules. * p < 0.05, ** p < 0.01

Fig. 4

Only a single gene is consistently differentially expressed between queens and workers. The plot shows the number of caste differentially expressed genes (DEGs) in common in a variable number of randomly selected species (bootstrap resampling 100 times). This pairwise analysis shows either that few genes are consistently caste-biased across species or that comparison of differentially expressed genes lacks power to detect these biases. By contrast, network analysis manifested significant underlying regulatory structure, suggesting that it is a more powerful approach (Fig. 2). A similar analysis was conducted at the level of GO terms (Fig. 5)

Fig. 5

No overlap was found in the number of enriched GO terms for caste-biased genes across all 16 species. The plot shows the number of enriched GO terms for caste-biased genes in common in a variable number of randomly selected species (bootstrap resampling 100 times). The results of this analysis parallel findings at the level of individual differentially expressed genes (Fig. 4)