Introduction

The establishment of a highly eusocial form of life occurred independently in two closely related lineages of bees, the honey bees and the stingless bees1,2, in two lineages of wasps3, and in an ancestor common to all ants4. This highly eusocial lifestyle, characterized by large colonies composed of at least one highly reproductive queen and hundreds of subfertile or sterile workers, represents an irreversible major evolutionarily transition5, as the morphologically distinct queen and worker castes have become mutually completely dependent on one another. The development of such physical castes not only strengthened the queen’s reproductive monopoly, but also promoted division of labor and task specialization in the worker caste, which enabled the highly eusocial Hymenoptera to establish very large colony sizes.

This transition in the level of social organization was accompanied by a change in the queens’ mating strategy that occurred independently in several lineages of the ants, in two lineages of wasps (Vespa and Vespula) and in the honey bees (genus Apis). While the females of the facultatively social or primitively eusocial bees and wasps generally mate with only a single male (monandry), the queens of several highly eusocial species are known to mate with more than two males (polyandry), which, in cases of extreme polyandry, can be 20 or more6. Polyandry has fitness consequences for the workers, as it diminishes the within-colony genetic relatedness, and thus, affects the structure of the genetic conflict over male production and sex ratio allocation7,8. As monandry is the ancestral condition9, the multiple mating systems and their associated male traits have drawn attention and generated controversial arguments about the adaptive significance of multiple mating in the highly eusocial Hymenoptera10. Among the eight hypotheses proposed for the evolution of polyandry in social insects11, the currently most accepted one is that it strengthens the colonies’ resistance against parasite or pathogen infections12. Polyandry, however, is not a simple categorical trait that stands in contrast with monandry, but there is a very strong skew in the number of recorded mating events9,13. Specifically, in only 0.2% of the known eusocial hymenopteran species the queen mates with more than three males, and mating frequencies with over 20 males are extremely rare6.

The Western honey bee, Apis mellifera, falls within the category of these extremely polyandrous species, with mean mating frequencies estimated from effective paternity calculations as varying between 13 and 20 (range 8–32)14,15. Hence, while on the one hand, A. mellifera is the best-established model organism for functional and structural aspects of the complex social organization in insects, the entire genus Apis strongly differs from all the other social bees in terms of its gonad morphology and mating behavior.

An adult honey bee queen has very large ovaries, each composed of 100–150 serial units (ovarioles), and referred to as a transgressive (exaggerated) ovary phenotype16,17, while the ovaries of an adult worker consist of only 2–12 ovarioles each18,19. In contrast, in all the other bees, the females have only 3–4 ovarioles per ovary20,21, including the primitive eusocial bumble bees (Bombini) and the highly eusocial stingless bees (Meliponini). Together with the solitary to facultatively eusocial orchid bees (Euglossini) and the honey bees (Apini), these four tribes form a monophyletic clade, the corbiculate bees22, and thus, the question is: why and how did the ovarian architecture of the honey bees become so different from all the other bees? Or put in other terms, what may have been the evolutionary conditions or drivers that favored the origin of the exaggerated ovary phenotype in the genus Apis?

To address the this question we decided to take an EvoDevo approach where we contrasted the larval gonad transcriptomes for three bee species that differ in gonad architecture and reproductive strategies within a well defined phylogenetic background. For this, we generated RNA-seq libraries of larval gonads of the bumbe bee, Bombus terrestris, and the stingless bee, Melipona quadrifasciata, for a comparativ transcriptomics approach against our prior RNA-seq data generated for the honey bee queen, worker, and drone larval gonads23.

Results

Comparative transcriptome analysis for honey bee, bumble bee, and stingless bee gonads

In terms of read numbers, the RNA-seq results for the larval gonads of the bumble bee B. terrestris (n = 3) and the stingless bee M. quadrifasciata (n = 3) were similar to those simultaneously obtained for the honey bee larval gonads (n = 3 each for workers, queens and drones)23, consisting of between 29,304 and 41,890 fragments per million, where more than 83% of these good quality reads could be mapped to the respective reference genomes of the three species. The alignment scores were high for A. mellifera, varying between 83 and 87.4%, and also for B. terrestris (79% to 86%), but were lower for M. quadrifasciata (60.8% to 63%). Hence, while the 15 RNA-seq libraries generated for this study had similar numbers in terms of good quality reads, they differed in their mapping rates. This difference is likely due to the better annotation quality of the honey bee24 and bumble bee genomes25, compared to that of the M. quadrifasciata genome26.

As a first step in the cross-species comparison, we performed a differential expression analysis for the A. mellifera reads (DESeq2 based on normalized reads, Padj < 0.05), and here it is important to note that we used a less stringent cutoff level for differentially expressed genes (DEGs) than in our prior study on these honey bee larval gonad transcriptomes, where we had set the cutoff at Padj < 0.0123. The reason for adopting the less stringent DEG level in the current study was that for the subsequent hierarchical clustering against the bumble bee and stingless bee transcriptomes and the evolutionary analysis we needed a sufficiently large gene set of clear orthologs. This less stringent analysis revealed 5,546 genes as differentially expressed among the three types of honey bee larval gonads, including a central set of 870 genes that was common to all three contrasts (Fig. 1).

Figure 1
figure 1

Venn-diagram representing differentially expressed genes (DEGs) in Apis mellifera queen, worker, and drone larval gonads. Three contrasts of comparisons were analyzed, resulting in a total of 5546 DEGs. A set of 870 DEGs is common to all contrasts.

Full size image

The 870 Apis-DEGs became the focal gene set for the comparative analysis with the bumble bee and the stingless bee larval gonads based on the reasoning that this DEGs set represents genes that are: (a) associated with the conservation of a high ovariole number in the queen caste, in contrast with the reduced ovariole number in the workers due to the massive programmed cell death in their larval gonads17,23, and (b) differentially expressed also between the two sexes, independent of caste in the female sex. A Gene Ontology (GO) analysis for these 870 DEGs returned 566 genes with GO information (Table S1), and enrichment was found for six categories.

For the bumble bee we obtained the orthologs of this gene set from the B. terrestris genome deposited in NCBI using the BLASTp tool. For M. quadrifasciata, the corresponding BLASTp searches were done in the Hymenoptera Genome Database (hymenopteragenome.org)27. In both cases we performed reciprocal BLAST searches against the honey bee genes. The RPKM counts (reads per kilobase per million mapped) for these orthologs were then retrieved from the respective RNA-seq data. We opted to use the RPKM normalization method because it eliminates the influence of gene length and variation in sequencing counts for comparisons between RNA-seq libraries28.

While our initial intention was to perform a hierarchical clustering analysis with all the 870 honey bee DEGs against their respective orthologs in the B. terrestris and M. quadrifasciata genomes, this was only possible for the comparison with the bumble bee, because many of the 870 honey bee DEGs showed unclear orthology relationships in the stingless bee genome. For B. terrestris, we obtained a total of 806 orthologs for the 870 A. mellifera DEGs, and the hierarchical clustering analysis performed in JMP Genomics software showed that the two species are clearly separated (Fig. S1) for this full gene set. This indicated already a possible connection between gonadal architecture and the mating system. Yet, including M. quadrifasciata in the analysis was fundamental, due to the fact that bumble bees differ from the honey bees and stingless bees with respect to their level of sociality (primitively vs. highly eusocial). Hence, we decided to perform the joint analysis for the three bee species with a reduced set of 45 DEGs that represent enriched GO categories (Table S2) and clear orthologies among the three species.

As shown in Fig. 2, all nine A. mellifera libraries are grouped together and are well separated from the cluster composed by the respective B. terrestris and M. quadrifasciata orthologs. In this set of 45 genes, we could identify 24 that showed divergent RPKM levels for the three species, thus setting apart the gene expression patterns in the larval gonads of the two monandrous bees with an ancestral gonad morphology from the polyandrous honey bee with its derived, exaggerated gonad morphology. In Fig. 2, these 24 genes are marked with an asterisk, and Table S3 presents detailed descriptions.

Figure 2
figure 2

Hierarchical clustering analysis of the RPKM counts of 45 selected A. mellifera gonad DEGs and of their orthologs in the Bombus terrestris and Melipona quadrifasciata RNA-seq libraries. Each gene is identified by its honey bee gene ID. The red-blue scale represents relative RPKM counts. Asterisks denote the 24 DEGs with divergent RPKM levels in the gonad transriptomes of the three bee species.

Full size image

A Principal Component Analysis (PCA) performed on the RPKM counts of these 45 genes in the 15 RNA-seq libraries produced a similar result (Fig. S2). The two first principal components (PC1 and PC2), which explain 96.25% of the total sample variance, not only separated the three species, but also placed the two highly eusocial species close to each other and clearly separated from the primitively eusocial bumble bee. This indicates that the factor “level of sociality” needs to be taken into account in transcriptome comparisons across social bees.

Evolutionary analysis of gonad development-related genes in Apoidea

We next retrieved the nucleotide and amino acid sequences of the respective orthologs for the 24 divergently represented genes shown in Fig. 2 in the genomes of 13 bee species for which annotated sequences were available (Table S4), and we tested these for evidence of positive (diversifying) selection using the Branch-site Unrestricted Statistical Test for Episodic Diversification (BUSTED)29. This test assumes that evolutionary patterns can change quickly and lead to independent evolutionary rates between branches. The analysis was done with the four fully sequenced Apis species serving as test branch against the other nine bee species from the families Apidae, Megachilidae, and Halictidae as background.

In this set of 24 divergently represented genes, four showed evidence of positive selection (Table 1), denoted by a dN/dS value > 1 and/or a p value < 0.05. These were as follows: CYP9Q3, CYP6AQ1, PHGDH, and ACADM, encoding, respectively, Cytochrome P450 9e2, Cytochrome P450 6AQ1 × 1, D-3 phosphoglycerate dehydrogenase, and the mitochondrial medium-chain specific acyl-CoA dehydrogenase. Among these genes, CYP9Q3 (GB43728) and CYP6AQ1 (GB52023) showed lower RPKM counts in the larval gonads of A. mellifera and M. quadrifasciata compared to B. terrestris (Fig. 2). The PHDGH gene (GB40503) showed low RPKM counts for A. mellifera queens and workers and higher counts in honey bee drones, as well as in the bumble bee and the stingless bee, but the opposite was the case for ACADM (GB42141), which showed higher counts in all honey bee samples and also in the stingless bee, contrasting with lower count numbers for the bumble bee.

Table 1 BUSTED test results for evidence of positive selection in 24 genes revealed in the hierarchical clustering analysis of honey bee, versus bumble bee and stingless bee larval gonad transcriptomes.
Full size table

To obtain a more in-depth information we performed gene tree analyses for the deduced amino acid sequences of these four genes in MEGA7 software30, applying the models and rate substitution parameters listed in Table S5. Among the two cytochrome P450 genes, CYP9Q3, which had the highest index for positive selection, showed only a minor divergence with respect to the Apoidea phylogenetic tree1,31, and this divergence is related to the positioning of A. florea (Fig. 3A). In contrast, the CYP6AQ1 gene tree showed a major divergence from the bee phylogenetic tree, with the four Apis species being completely separated from the representatives of the other Apoidea (Fig. 3B). The gene trees for PHDGH and ACADM (Fig. 3C and D), were largely consistent with the Apoidea phylogenetic tree.

Figure 3
figure 3

Gene trees for the four genes with evidence of positive selection with respect to bee gonad phenotype in 13 bee species. The gene trees were inferred using the Maximum Likelihood method based on the best model (Table S5), and the trees with the highest log likelihood (1000 bootstrap replicates) are shown. The trees were generated in MEGA7, with branch lengths representing the number of substitutions per site.

Full size image

This divergent result for the CYP6AQ1 gene tree prompted us to take a closer look at the multiple alignments for CYP6AQ1. While no apparent site-restricted pattern in terms of codon changes was apparent (Fig. S3), potential functional insights were obtained through BLASTP searches against the Drosophila melanogaster protein annotation database, which revealed CYP6G2 and CYP6G1 as the most similar CYPs (42.94% and 44.04% identity, respectively) to the A. mellifera CYP6AQ1 amino acid sequence. For A. mellifera CYP9Q3, the most similar D. melanogaster CYPs were CYP9F2 and CYP6A13 (34.85% and 33.27% identity, respectively). These Drosophila CYPs, especially CYP6G2 and CYP6G1, are involved in interesting biological functions, such as juvenile hormone biosynthesis in the case of CYP6G2, and for CYP6G1, in DDT and imidacloprid resistance32,33,34,35.

Discussion

Our study was driven by the hypothesis that the extremely polyandrous mating system in honey bees, not only in the Western honey bee, Apis mellifera, but in all species of the genus Apis36, and the high number of ovarioles and testioles in the gonads of both queens and drones may be evolutionarily and developmentally coupled. As this stands in contrast to all other social bees, including the highly eusocial stingless bees, which essentially have a monandrous mating system37, we proposed to test this by a comparative transcriptome analysis of the larval gonads of A. mellifera queens, workers, and drones23, against those of B. terrestris and M. quadrifasciata larvae as the ancestral state.

One may argue that we had missed an opportunity by pooling the gonads of the queens, workers, and males for the bumble bee and for the stingless bee, but this made sense because we had based our comparative analysis on the set of 870 A. mellifera genes that represented DEGs for all the three gonad contrasts (Fig. 1). Furthermore, and different from the honey bee, the larvae of the stingless bee M. quadrifasciata cannot be distinguished according to sex and caste, as they are all reared in brood cells of the same size and type, and in the bumble bee, B. terrestris, males and queens are only reared at the end of the colony cycle, separately from the worker brood.

The comparative transcriptome analysis on the set of 45 genes (Fig. 2) indicated evidence of positive (diversifying) selection in four genes that emerged as differentially represented in the gonads of the A. mellifera gonads compared to M. quadrifasciata and B. terrestris (Fig. 2). Two of these are cytochrome P450 genes, CYP9Q3 and CYP6AQ1 (Table 1). In the gene tree analyses, CYP6AQ1 presented a configuration that set the genus Apis well apart from all the other Apoidea (Fig. 3), and this gene tree is highly inconsistent with the phylogenetic tree of the corbiculate bees, which has Apini as sister group to the Euglossini1,31. The latter are the orchid bees, whose females have a solitary or incipiently eusocial life style38.

Insect cytochrome P450 enzymes are involved in many metabolic and biosynthesis processes, including the synthesis of pheromones, cuticular hydrocarbons, and hormones, as well as the degradation of lipid hormones, and the detoxification of plant secondary metabolites and pesticides33. The annotation of cytochrome P450 genes in the first published version of the honey bee genome had indicated a significantly reduced set of orthologs for the insect GST and CYP protein clades39, and this finding has since been confirmed40. But there are exceptions, for instance the CYP6AS family, which is unique to the Hymenoptera. This family appears to have undergone an expansion in gene number, and its members have been associated with the digestion of quercetin, a flavonol compound present in nectar and pollen41. This is likely an adaptive trait, because quercetin acts as a competitive inhibitor of P450 enzymes, and when quercetin was fed to adult workers, these became increasingly resistant to queen pheromone and showed ovarian activation42. A recent phylogenetic analysis of the gene trees for a total of 481 bee P450 genes furthermore revealed that the CYP6AS clade within the CYP3 family is fast evolving and exhibits evidence of positive selection43. While this is mainly discussed in the context of selection pressure on the need for detoxification of secondary plant compounds, our results now indicate that this high evolutionary rate and divergence in the CYP3 family may have also been co-opted in developmental pathways that generated novelties, such as the transgressive gonad phenotype in the genus Apis.

Currently, experimental evidence for a reproductive system function of the CYP6AQ1 gene is still lacking, but CYP6AQ1 was listed as expressed in honey bee antennae39 and, like CYP6AS, it is a member of the CYP3 clade. Recently, its expression in honey bee fat body was also documented as related to aggression, and it was found down-regulated as a result of immune system activation44. Hence, it is quite possible that CYP6AQ1 may be involved in a signaling function that regulates gonad development and activation in bees, and through diversifying selection it may have gained a novel role associated with the transition from the ancestral monandrous mode of reproduction in social bees to polyandry in the genus Apis. A current difficulty still is to understand how the evidenced positive selection may have resulted in functional changes in the honey bee CYP6AQ1 protein. In the CLUSTALW alignment of the bee orthologs of CYP6AQ1 (Fig. S3), the non-synonymous substitutions appeared spread out across the entire CDS, and there is currently no information available on the location and structure of the enzyme’s catalytic site. Most cytochrome P450 enzymes have broad catalytic properties34, and detailed mappings on substrate recognition sites have so far only been done for a handful of insect cytochrome P450 enzymes, such as certain CYP6B sequences of lepidopterans45 and for the housefly CYP6A146.

Our BLASTP searches against the Drosophila genome identified CYP6G1 and CYP6G2 as the most similar ones to the honey bee CYP6AQ1, while CYP9F2 and CYP6A13 were the most similar ones to the honey bee CYP9Q3. In the Drosophila genome, CYP6G1 and CYP6G2 map right next to each other47, possibly resulting from a duplication, but they are involved into very different functions. While CYP6G1 has long been identified as a gene related to insecticide resistance, especially against DDT and the neonicotinoid imidacloprid48, CYP6G2 is highly expressed in the ring gland of Drosophila larvae, where it is considered to be involved in juvenile hormone (JH) biosynthesis35. These functional annotations in Drosophila are of interest with respect to honey bee biology. On the one hand, JH is a major driver in honey bee caste development and also for the behavioral maturation of adult workers49. On the other hand, imidacloprid is a neonicotinoid that is strongly associated with colony losses50, including sublethal effects on the learning and memory of honey bee workers, which affect the homing of foragers as they return from nectar and pollen resources. Furthermore, JH has recently been shown to play a critical role in the brain-reproduction energetics tradeoff that distinguishes the honey bees from the bumble bees51.

JH is a fundamental hormone in the insects’ life cycles, orchestrating molting, metamorphosis, and reproduction, and it is a key driver for the adaptive expression of characters, including the castes of social insects52. Together with our results that indicate positive selection acting on the CYP6AQ1 gene in the Apis lineage, selection on JH signaling-related genes in the evolution of sociality has recently also been demonstrated in a comprehensive comparative genomics study in halictid bees53. These authors looked for evidence of positive selection in genes associated with lineages that have independently made the transition from a solitary to a eusocial lifestyle, versus relaxed selection in those lineages where the females reverted from eusociality to secondarily become solitary nesting. Under this premise, four genes showed such contrasting evidence. Two of these, apolipoprotein and hexamerin 110, are abundant insect hemolymph proteins related to JH binding and transport.

In conclusion, the fact that we and others53 could identify evidence of positive (diversifying) or relaxed selection in genes or gene families that are related to important transitions in the evolution of insect sociality shows the strength of comparative genomics or transcriptomics when performed under a well-defined phylogenetic background. Though not all these genes may be directly associated with important functions, such as JH signaling, they can heuristically guide future research efforts on how natural selection may have shaped developmental and behavioral plasticity54, which is key to the queen-worker polyphenism in social insects.

Methods

Sample preparation

Stingless bee larvae (M. quadrifasciata) were obtained during the Southern summer season (January/February) from brood combs of three hives kept in the Experimental Apiary of the University of São Paulo in Ribeirão Preto, Brazil. The larval stage equivalent to the fifth-instar early feeding stage (L5F1) of honey bees used in our prior study23 is the early third instar (L3.1) of Melipona bees55. Bumble bee (B. terrestris) larvae were obtained during the Northern summer season (June/July) from three commercial Koppert pollination hives (Behr Bestäubung Biologischer Pflanzenschutz Welle, Germany). These hives were kept in the Baden-Württemberg State Bee Research Institute at the University of Hohenheim, Stuttgart, Germany. The developmental schedule listed in56 was used for the identification and dissection of third instar B. terrestris larvae, which is the stage corresponding to the L5F1 stage of A. mellifera, for which RNA was extracted simultaneously and at the location23.

The larval gonads of all three species, A. mellifera23, M. quadrifasciata, and B. terrestris, were dissected in sterile insect saline and immediately transferred to TRIzol reagent (Life Technologies), and the manufacturer’s protocol for the use with small tissue quantities was used for RNA extraction: TRIzol (500 μL), chloroform (100 μL), glycogen (5 μg), and overnight precipitation (− 80 °C) with isopropanol (250 μL). After solvent evaporation, the RNA was resuspended in 20 μL RNase-free water. The number of gonads composing each A. mellifera sample is reported in23. For B. terrestris, each of the three biological samples consisted of 40 larval gonad pairs, and for M. quadrifasciata, the number of gonad pairs was 50 for each of the three samples. Sample quality control, library preparation, and high-throughput sequencing protocols were all performed simultaneously by CeGaT GmbH (Tübingen, Germany), as previously described23. Paired-end sequencing was carried out on a NovaSeq6000 (Ilumina) platform, with read lengths of 2 × 100 bp and an output of 6 Gb per sample.

RNA-seq bioinformatics analysis

The initial RNA sequencing data analysis was done using the Nextflow-based RNA-Seq pipeline release 1.2 (https://github.com/nf-core/rnaseq). It included the MultiQC v1.6 program57 and FASTQC v0.11.8 to determine FASTQ file quality, and after removal of low-quality reads, adapter sequences were trimmed with Trim Galore 0.5.0. The raw trimmed sequences of each of the three B. terrestris and M. quadrifasciata larval gonad RNA-seq libraries were submitted to NCBI (SRA, accession number: B. terrestris SRX8487631; M. quadrifasciata SRX8487632).

Next, using the HISAT2 v2.1.0 aligner58, the reads that had passed the quality control were mapped to the respective genomes of Bombus terrestris (fasta: Bombus_terrestris.Bter_1.0.dna.toplevel.fa, gff3: Bombus_terrestris.Bter_1.0.40.gff3), and the Melipona quadrifasciata genome (fasta: GCA_001276565.1_ASM127656v1_genomic.fna, gtf: Melipona_quadrifasciata_v1.1_on_GCA_001276565.1_ASM127656v1_complete_mar2_2017.gtf). An evaluation of the mapped RNA-seq reads was done with RSeQC v2.6.459, and for read counting of the features (e.g., genes), FeatureCounts v1.6.260 was used. The read counts were then normalized to RPKM to remove potential technical bias in the RNA-seq data, such as depth of sequencing and gene length. All programs were used with default settings for mismatch allowance.

Hierarchical clustering analysis

The starting point of the analysis were the 870 DEGs (Fig. 1) of the honey bee larval gonads, based on the previously published RNA-seq data set23. The focal genes for the comparison across the three bee species were defined by a Gene Ontology (GO) enrichment analysis on the 870 A. mellifera larval gonad DEGs performed using Blast2GO implemented in OmicsBox (biobam, Valencia, Spain). Enrichment significance was assessed using Fisher’s Exact Test. This resulted in a set of 45 genes, for which the B. terrestris orthologs were retrieved from GenBank using the NCBI BLASTP tool. For M. quadrifasciata, the respective BLASTP searches were done in the Hymenoptera Genome Database (hymenopteragenome.org)27. Ortholog tables with the respective gene_id (available in HymenopteraMine v1.4; http://hymenopteragenome.org/hymenopteramine/)61 were used to find and manually confirm the corresponding gene_ids. Their orthology status was confirmed by reciprocal BLASTP searches against the honey bee genome.

Using the respective gene_id, the RPKM counts for each gene were retrieved from the 15 transcriptomes (9 for A. mellifera and 3 each for B. terrestris and M. quadrifasciata). A principal component analysis and a hierarchical cluster analysis on the RPKM counts were performed in JMP Genomics 12.2.0 software (SAS).

Analysis of positive selection in candidate genes

For the 24 genes that showed clear contrasts for the honey bee gonad transcriptomes against those of the bumble bee and the stingless bee in the hierarchical clustering analysis, BLAST tools (BLASTP and then BLASTN) were used to extract the corresponding nucleotide sequences (CDS) of their orthologs from the genomes of 13 bee species deposited in GenBank and the Hymenoptera Genome Database27 (Table S4). These included four species of the genus Apis (A. mellifera, A. cerana, A. dorsata, and A. florea), two bumble bee species (Bombus terrestris and B. impatiens), the stingless bee Melipona quadrifasciata, the orchid bee Eufriesea mexicana, the carpenter bee Ceratina calcarata, and the anthophorid Habropoda laboriosa. These 10 species all belong to the family Apidae. In addition, we included Megachile rotundata (Megachilidae) and two Halictidae, Lasioglossum albipes and Dufourea novaeangliae. The respective nucleotide sequences were aligned using ClustalW and visualized in Aliview62. The software geneious (https://www.geneious.com) was used to convert the nucleotide sequences into consensus amino acid sequences, and a codon alignment analysis was performed using Pal2Nal (http://www.bork.embl.de/pal2nal/index.cgi?example=Yes#RunP2N).

To reveal whether genes or specific sites may have experienced episodes of positive selection, the codon alignment was used as input to an evolutionary test performed in BUSTED29, implemented in the DataMonkey server (https://datamonkey.org/busted). BUSTED checks for synonymous and non-synonymous substitutions and calculates the dN/dS value for each gene. Additionally, it computes the dN/dS value for the respective gene within each species and calculates a p value for the gene in the comparison between the test branch versus the background branches. With this, a dN/dS-value is attributed to each gene, and a result for evidence of selection (p < 0.05) is given. This avoids relying on dN/dS-values alone, as these have been shown to have limited power in detecting positive selection63. In our analysis, the four species of the genus Apis served as test branch against nine other bee species.

Evolutionary trees were generated for the four genes with evidence of positive selection using MEGA7 software30. Modeltest implemented in MEGA was used to identify the evolutionary model that best described the amino acid substitution pattern for each data set. The model with the lowest BIC scores (Bayesian Information Criterion) was the used for the construction of unrooted trees based on Maximum Likelihood. Models, BIC scores, as well as the G+ and I+ values used for gene tree construction are listed in Table S5. Bootstrap analyses with 1000 replicates were run for each tree.

Ethical approval and consent to participate

Not applicable. As no chordate species were used, the Committee for the Use of Animals in Experiments (CEUA) of FMRP-USP declared the study protocol as exempt from regulation.