Evolutionary Transitions in Social Behavior are Associated With Convergent and Partially Reversible Expansions of Transcription Factor Binding Sites

Abstract

The evolution of sociality involves shifts in physiology and behavior, most notably the emergence of a reproductive division of labor. Within social colonies, these distinct behavioral phenotypes arise from differential regulation of a shared genome. Changes in transcription factor (TF) binding motifs are one potential mechanism underpinning this plasticity, with prior studies suggesting that social species exhibit expansions in TF binding sites. However, it remains unclear whether motif expansions are reversed when sociality is lost. Here we analyze predicted TF motif occurrences across gene promoters in 42 bee species spanning millions of years of evolutionary divergence and multiple independent gains and losses of sociality. We compare motif presence across species to test whether motif expansions are a convergent feature of social evolution and whether their presence secondarily decreases when social behavior is lost. Our findings are consistent with previous research, demonstrating an expansion of TF motifs in lineages which have gained sociality. However, contrary to expectation, we do not observe genome-wide motif contractions in lineages which have secondarily lost social behavior. Despite this overall pattern, we still identify several motifs, promoter regions, and specific motif-promoter pairs which exhibit complementary changes with both gains and losses of sociality. These regulatory targets are enriched for similar organismal functions, providing strong candidates for further study. Our results lend additional support to the hypothesis that novel phenotypes may arise through modification of existing gene regulatory networks.

Significance.

Gene regulation shapes the development of phenotypic variation across taxa. However, we still have limited knowledge about how gene regulatory changes influence evolutionary processes and the emergence of novel phenotypes. Social evolution involves the development of phenotypically differentiated castes and is thought to require an increase in regulatory complexity. Our study across a wide range of bees exhibiting both social and solitary behavior, including species that have recently reverted to a solitary state, provides evidence that increased regulatory complexity is involved in the transition to sociality. We also find evidence that when social behavior is secondarily lost, only some of this regulatory complexity is reversed. These findings provide new insights into transcriptional control points involved in the evolution of sociality.

Introduction

Social animals have captured the attention of biologists for centuries, yet the molecular basis of social trait evolution remains unresolved. The shift from solitary to social living involves a diverse suite of behavioral and physiological adaptations, most notably the emergence of a reproductive division of labor. Within social groups, some individuals reduce their direct reproductive output and instead help their relatives reproduce (Bourke 2011). This reproductive division of labor is especially striking in many species of social insects. Highly specialized reproductives (“queens”) can lay hundreds to thousands of eggs per day, and in contrast, sterile nonreproductives (“workers”) specialize in essential tasks such as nest construction, foraging, and other nonreproductive tasks critical for success of the group (Wilson 1985). The cooperation of queens and workers within social colonies represents an evolutionary departure from a solitary ancestral state, where a single female completed all reproductive and nonreproductive tasks alone. Therefore, the evolution of queen and worker castes, and particularly the evolution of nonreproductive workers, represents a core innovation of social living.

One emerging, generalizable theory about the molecular basis of social evolution is that it involves an increase in gene regulatory complexity (West-Eberhard 2003; Linksvayer and Wade 2005). The development of queen and worker castes arises through environmentally induced plasticity that is influenced by cues such as pheromones (Wheeler 1986; Matsuura et al. 2010) and nutrition experienced during larval development (Wheeler 1986; Kamakura 2011; Linksvayer et al. 2011). Due to the prominent role of developmental plasticity in caste determination, increased context-dependent regulation of gene expression is predicted to be associated with the origin of social behavior (West-Eberhard 2003; Linksvayer and Wade 2005). This prediction has been supported across a range of insect lineages including bees (Sinha et al. 2006; Kapheim et al. 2015; Jones et al. 2020, 2023), ants (Bonasio et al. 2012; Simola et al. 2013, 2016; Gospocic et al. 2021), wasps (Weiner et al. 2013), and termites (Glastad et al. 2016). Studies have linked changes in levels of DNA methylation to phenotypic plasticity in social insects (Linksvayer et al. 2011; Li et al. 2018; Chen et al. 2021). However, other studies have called into question the role of DNA methylation in sociality more broadly, as they failed to find associations between DNA methylation and caste determination (Libbrecht et al. 2018; Harris et al. 2019; Behrends et al. 2025) [reviewed in {Oldroyd and Yagound 2021}]. Other epigenetic processes, such as histone modifications, have also been linked to caste determination in some social insects (Simola et al. 2016; Wojciechowski et al. 2018; Choppin et al. 2021). These studies suggest that in extant social species, many epigenetic factors are involved in mediating the environmental signals that influence caste determination.

Despite growing research associating epigenetic changes and social phenotypes within individual species, few consistent molecular mechanisms underlying caste evolution have been identified across social insect lineages. One notable exception to this is the addition of transcription factor binding sites, which has emerged as an apparently more ubiquitous pattern linked with transitions in social evolution (Simola et al. 2013; Kapheim et al. 2015; Jones et al. 2023). Transcription factors (TFs) are proteins that regulate gene activity, acting as “molecular switches” that can activate or repress gene expression, by binding to short DNA sequences known as motifs. These motifs are often located in noncoding regions of DNA, such as promoter or enhancer regions. The binding of TFs to these motif regions influences where, when, and how strongly genes are expressed (Maston et al. 2006; Lambert et al. 2018). Each TF belongs to a larger family of TFs that are defined by a conserved DNA binding domain (e.g. homeobox, zinc-finger, winged helix, etc.) which determines the types of motif sequences it can recognize (Lambert et al. 2018). The binding of TFs to these motifs drives and shapes the activity of genes within broader gene regulatory networks that underlie development and behavior. Since TF binding can activate or repress gene expression, changes in motif sequence, the number of motifs, or accessibility of motifs within a cellular context can lead to differences in gene regulation (Gilbert 2000; Schmidt et al. 2010). These cis-regulatory modifications have been increasingly recognized as influential drivers of phenotypic evolution (Carroll 2005, 2008; Wray 2007). In addition, expansions and diversifications of TF families and their binding sites have been associated with the evolution of phenotypic complexity and diversity (Levine and Tjian 2003; Halfon 2017). In insects, comparative studies suggest that social evolution is accompanied by greater regulatory complexity, reflected by increased predicted motif binding (Simola et al. 2013; Kapheim et al. 2015; Jones et al. 2023). In an initial study spanning multiple bee families, social species were found to consistently exhibit greater predicted TF motif binding sites compared to solitary species (Kapheim et al. 2015). Expansion of binding sites in social species may reflect the need for increased context-dependent regulation of gene expression required to support distinct phenotypes. In a more recent study within the family Halictidae (sweat bees) that includes secondarily solitary species (species with social ancestors), this pattern persisted, with social species exhibiting a greater number of predicted TF motif binding sites than solitary species (Jones et al. 2023). However, previous analyses have not distinguished secondary reversions from ancestral solitary lineages, leaving unclear whether motif expansions are reversed when sociality is lost.

Here, we take advantage of a growing number of high quality genomic resources for bees, which span millions of years of evolutionary divergence and multiple independent gains and losses of sociality (Michener 1974; Danforth et al. 2013; Branstetter et al. 2017), to more comprehensively test whether TF motif expansion is a convergent feature of social evolution in bees. Using a database of known insect motifs, we ask whether transitions both to and from sociality are driven by shared changes in gene regulation. Leveraging reversions of sociality (social to solitary transitions) in sweat bee lineages, we specifically ask whether motif presence secondarily decreases when social behavior is lost. Our results provide new insights into how gene regulatory changes, both genome-wide and for specific regulatory targets, may shape evolutionary transitions in bee sociality.

Results

Evolutionary Gains of Social Behavior are Associated With Widespread TF Motif Expansions

To investigate whether transitions in sociality are associated with TF motif expansions, we used phylogenetic generalized least squares (PGLS) models to test associations between predicted TF motif counts and social behavior across 42 bee species (Fig. 1). Focusing first on transitions from an ancestrally solitary to a social life history strategy, we identified 12,763 motif-orthogroup (OG) pairs significantly correlated with gains of sociality (P < 0.001, adjusted R² > 0.25; 3,084 with FDR < 0.05), 0.6% of all pairs tested (Table S1). Of 8,171 OGs tested, 5,516 (67.5%) had a change in normalized motif count for at least one motif based on our PGLS analysis, with 3,126 (38.3%) OGs significant for two or more motifs. For genome-wide assessments, a motif was considered biased if the difference between social and solitary-correlated OGs was at least 10 (i.e. a social-biased motif has at least 10 more social-correlated OGs than solitary-correlated OGs and vice versa for solitary-biased motifs). With this threshold, 149 motifs were social biased (i.e. more OGs had higher motif counts for social species compared with ancestrally solitary species) (Fig. 2a and b), compared with only 30 solitary-biased motifs.

Fig. 1.

Phylogenetic and motif-based framework for identifying motif changes associated with social transitions. (a) 42 species were analyzed with OrthoFinder, resulting in the identification of 13,168 OGs and a species phylogeny. Sociality classifications are indicated (red = social, blue = ancestrally solitary, yellow = secondarily solitary), along with three independent gains (red triangles) and six recent and independent losses (yellow Xs) (Michener 1974; Wcislo and Danforth 1997; Danforth et al. 2003; Wcislo and Fewell 2017), enabling comparative approaches for analyzing motif changes with social transitions. (b) Promoter regions of identified OGs were scanned for predicted transcription factor binding motifs. Two separate PGLS models were performed for gains and losses of sociality, respectively, and motif-OG pairs were classified as social biased or solitary biased based on changes in motif count for each OG. (c) Calculations for motif biases were completed for both gains and losses per motif across all 13,168 OGs. (d) Final results for all 286 motifs show overall social and solitary bias by aggregating OG correlations for separate transitions in gains and losses. Note: panels b–d illustrate methods used but do not show actual results.

Fig. 2.

Patterns of motif bias in gains and losses of sociality. (a) Motif-level bias in gains of sociality (ancestrally solitary to social). Each bar represents a single motif tested in PGLS1 with a minimum difference of 10 between social-correlated and solitary-correlated OGs. The number of social-biased OGs associated with the motif is in red, while the number of solitary biased OGs is in blue. (b) Results of motif biases (minimum difference of 10 correlated OGs) in 100 permutations of PGLS1 for social gain. The observed value of 149 social-biased motifs is significantly more than expected by chance based on permutations results (P = 0.01, FE = 4.41), while the observed 30 solitary-biased motifs is not more than expected by chance. c) Results of motif biases (minimum difference of 10 correlated OGs) in 100 permutations of PGLS2 for losses of sociality. Neither social- nor solitary-biased observed values were more than expected by chance based on permutation results. (d) Raw motif-level biases in losses of sociality (social to secondarily solitary). Each bar represents a single motif tested in PGLS2 with a minimum difference of 10 between its social-correlated and solitary-correlated OGs. The number of social biased OGs associated with the motif is in red and the number of solitary-biased OGs is in yellow.

To evaluate the significance of this social-biased pattern, we performed PGLS permutations where the phenotypic classifications were randomized among species tips, while preserving the overall number of ancestrally solitary and social members for each motif-OG test. Only one permuted PGLS run resulted in a signal as strong as the observed 149 motif bias [Fig. 2b, Table S2; mean permuted bias: 33.81, empirical P = 0.01, fold enrichment {FE} = 4.41]. In contrast, the 30 solitary-biased motifs we observed were not more than expected by chance, with several randomly permuted PGLS results finding similar numbers of solitary-biased motifs when shuffling social phenotypes (Fig. 2b, Table S2). These results are consistent with previous findings (Kapheim et al. 2015; Jones et al. 2023), demonstrating that social bee genomes are associated with an overall expansion in predicted TF binding sites in gene promoters relative to solitary bee genomes.

To better compare our results to Kapheim et al. (2015) and Jones et al. (2023), where analyses were restricted to single-copy OGs (i.e. all species contained only one gene within these OGs), we also examined our results after filtering OGs with more than one gene member per species. Using single-copy OGs only, we found 19 social-biased compared with only three solitary-biased motifs (Tables S3 and S4, Fig. S1), significantly more social-biased motifs than expected by chance (Table S5; mean permuted bias: 2.59, empirical P < 0.01, FE = 7.34).

Losses of Social Behavior are Not Associated With Contractions in TF Motif Presence

After identifying expansions in motif presence with social gains, we next investigated whether secondary losses of sociality are associated with TF motif contractions. Our second PGLS model identified 5,283 motif-OG pairs significantly correlated with losses in sociality (P < 0.001, adjusted R² > 0.25; 290 with FDR < 0.05), 0.2% of all pairs tested (Table S6). Of the 8,482 OGs tested, 2,734 (32.2%) had a change in normalized motif count for at least one motif based on our PGLS analysis, with 1,025 (12.1%) OGs significant for two or more motifs. As before, a motif was counted as biased if the difference between social and solitary-correlated OGs was at least 10 (i.e. a social-biased motif has at least 10 more social-correlated OGs than solitary-correlated OGs and vice versa for solitary-biased motifs). In contrast to our predictions, we did not observe widespread contractions of predicted motif presence in secondarily solitary lineages. Instead, only 9 motifs showed evidence of contraction, while 25 motifs showed overall solitary-biased correlations. However, neither of these biases were more than expected by chance based upon our permutations of phenotypic labels, suggesting that losses of sociality are not associated with genome-wide changes in motif presence for any TF (Fig. 2c and d). When restricting to only single-copy OGs, no motifs were found to be social biased or solitary biased genome-wide when comparing social and secondarily solitary species (Tables S4 and S7).

Regulatory Motifs Implicated in Bee Social Evolution Target Genes With Shared Organismal Functions

To determine whether motif expansions that are most strongly associated with gains of sociality are targeting similar gene functions, we examined OGs associated with the top 25 most social-biased motifs in gains of sociality (Fig. 3a). Of these top 25 social-biased motifs, eight were previously reported to be expanding with gains of sociality (Kapheim et al. 2015; Jones et al. 2023), including lbe, h, ara, and Stat92E (Fig. 3a; Table S8). GO enrichment analysis on social-biased OGs targeted by these motifs (i.e. all unique OGs to the right of x = 0 for black bars in Fig. 3a, n = 1065) revealed enrichment for multiple biological processes (16 terms with Fisher weighted P < 0.05; Table S9). Enriched terms included functions related to “glycolytic process”, “negative regulation of TOR signaling”, “synaptic vesicle priming”, and “proton transmembrane transport” (Fig. 3b; Table S9). While the level of social motif bias we identified is significantly more than expected by chance (Fig. 2b), we applied an additional correction for multiple testing on motif-OG pair correlations using a Benjamini–Hochberg false discovery rate (FDR) (Benjamini and Hochberg 1995). Applying an FDR threshold of 0.05 resulted in a reduced set of 29 social-biased motifs significantly associated with gains of sociality (Fig. 3c). This filtered set includes many (13) of the same TFs identified in the less conservative (raw P < 0.001, adjusted R² > 0.25) analysis. In addition, seven social-biased motifs in this conservative set were previously reported to exhibit expansions in social species with gains of sociality (Kapheim et al. 2015; Jones et al. 2023), including h and Stat92E (Fig. 3c; Table S8). GO enrichment on social-biased OGs targeted by these motifs for gains (i.e. all unique OGs to the right of x = 0 for black bars in Fig. 3c, n = 414) revealed significant enrichment for 17 biological processes (Fisher weighted P < 0.05; Table S10), including the same “glycolytic process” and “synaptic vesicle priming” terms enriched in the top 25 motif target set using less conservative (raw P < 0.001, adjusted R² > 0.25) thresholds (Fig. 3d; Table S10).

Fig. 3.

Targets of top social-biased motifs show enrichment for multiple biological processes (a) top 25 most social-biased motifs for gains of sociality. Motifs are ordered by the greatest social bias with black bars representing their correlations with OGs for gains and gray bars representing their correlations with OGs for losses. Motif sequence logos to the right of the bar chart indicate motifs previously found to be significantly social biased in either Kapheim et al. (2015) or Jones et al. (2023), and those significantly social biased in both studies are noted with double asterisks. (b) Enriched biological process GO terms (Fisher weighted P < 0.05) for social-biased OGs for gains in (a), plotted in two-dimensional semantic space and grouped by functional similarity. Bubble size reflects the number of unique GO terms per grouping, and color intensity reflects statistical significance. (c) 29 social-biased motifs for gains in sociality after filtering motif-OG correlations for FDR < 0.05. Motifs are ordered by the greatest social bias with black bars representing their correlations with OGs for gains and gray bars representing their correlations with OGs for losses. Motif sequence logos to the right of the bar chart indicate motifs previously found to be significantly social-biased in either Kapheim et al. (2015)or Jones et al. (2023), and those significantly social biased in both are represented by double asterisks. (d) Enriched biological process GO terms (Fisher weighted P < 0.05) for social-biased OGs for gains in (c), plotted in two-dimensional semantic space and grouped by functional similarity. Bubble size reflects the number of unique GO terms per grouping, and color intensity reflects statistical significance.

Gene Regulatory Hotspots of Social Evolution

To complement our motif-centered analyses and identify downstream target genes repeatedly associated with transitions in sociality, we identified OGs with evidence of TF motif changes in the promoter regions associated with both gains and losses of sociality. To reduce false positives and highlight OGs with the strongest bias, we only included OGs with a minimum difference of three motifs between social and solitary-correlated relationships (Table S11). 24 OGs showed consistent social-biased signal (Fig. 4), significantly more than expected by chance [permutation mean = 3.5, P < 0.01, FE = 6.86 {Table S12}], with greater numbers of social-correlated motifs in the promoter regions of these genes for both gains and losses of social behavior. Of those, only 10 showed changes involving the same motif, while the remaining 14 had social-biased motif counts but for distinct motifs when comparing gains to losses. Consistent social-biased hotspots included the promoters of voltage-gated potassium channel subunit beta-2 (KCNAB2), mitochondrial transcription factor A (TFAM), and ubiquitin carboxyl-terminal hydrolase 31 (USP31) (Table S11). GO enrichment analysis on these 24 OGs revealed significant enrichment for 11 biological processes including “cellular response to NGF”, “neurotrophin receptor signaling pathway”, and “t6A tRNA modification” (Fisher weighted P < 0.05; Table S13). Interestingly, four social consistent OGs, homeobox protein ARX, mitochondrial transcription factor A (TFAM), myoneurin, and LOC102655824 (uncharacterized), are annotated with the “DNA binding” GO term (GO:0003677), a marginally significant enrichment relative to expectations based upon the number of total OGs annotated with DNA binding (2.7-fold enrichment; P = 5.57e−2, hypergeometric test of overlap) (Table S14). These results suggest that some of the targets of regulatory change associated with social evolution are themselves transcriptional regulators.

Fig. 4.

Gene regulatory hotspots of social evolution. 24 social consistent OGs (pink) exhibit greater numbers of social correlated motifs in their promoters for both gains and losses of social behavior. 13 solitary consistent OGs (blue) exhibit greater numbers of solitary correlated motifs in the promoter regions of these genes for both gains and losses of social behavior. All OGs with annotations in Apis mellifera are labeled, while unlabeled circles indicate OGs annotated as uncharacterized loci.

13 OGs showed consistent solitary-biased motif signal (Fig. 4), marginally more than expected by chance [permutation mean = 4.7, P = 0.05, FE = 2.8 {Table S12}], with greater numbers of solitary-correlated motifs in the promoter regions of these genes for both gains and losses of social behavior. Of those, only 3 showed convergent signal with the same motif, while the remaining 10 had consistently solitary-biased motif counts but of distinct motifs when comparing gains to losses. Consistent solitary-biased hotspots included the promoter regions of dynein regulatory complex protein 9, akirin-2, and GPN-loop GTPase 2 (GPN2) (Table S15). GO enrichment analysis on these 13 OGs revealed significant enrichment for six biological processes (Fisher weighted P < 0.05; Table S16), including functions associated with immune system processes, transcriptional regulation, and cellular structure. Similar to the social signal, the appearance of the TF kayak among OGs with solitary-biased motif presence in their promoter regions suggests that these regulatory changes could have multilayered and hierarchical effects on gene network activity.

Convergent and Reversible Changes in Gene Regulation Associated With Sociality

After identifying motifs and promoter regions with consistent regulatory changes linked with gains and losses of social behavior, we next asked if specific motif-OG pairs were convergently and reversibly modified in both transitions. If motif changes are tightly linked to sociality and incur a cost when no longer needed after sociality is lost, we might expect reversions where promoter regions experiencing expansions in motif binding for gains subsequently experience contractions in motif binding for losses or vice versa. 149 motif-OG pairs, 7.26 more than expected by chance [permutation mean = 20.53, P = 0.01 {Tables S17 and S18}], were identified from overlaps in the 12,763 motif-OG pairs significantly correlated with gains in sociality and the 5,283 motif-OG pairs significantly correlated with losses in sociality (Table 1; 8 motif-OG pairs with FDR < 0.05, all of which are social-biased for both gains and losses, Table S19). These pairs represent the subset of motif and promoter associations that exhibit significant regulatory changes from both ancestrally solitary to social life history strategies and from ancestral social to secondarily solitary life history strategies. Strikingly, 146 out of the 149 motif-OG pairs were consistent in their direction (i.e. either social-biased for both gains and losses or solitary-biased for both gains and losses), indicating complementary changes with gains and losses of social behavior (Table 1). 104 pairs were social-biased in both gains and losses [permutation mean = 4.57, P < 0.01, FE = 22.76 {Table S18}], while 42 pairs were solitary-biased in both transitions [permutation mean = 5.33, P = 0.03, FE = 7.88 {Table S18}]. These results suggest that not only are motif expansions associated with the evolution of social behavior, but specific regulatory targets may be convergently modulated in a reversible manner across independent gains and losses. This pattern was robust to considering only single-copy OGs, with 40 pairs social-biased in both gains and losses (permutation mean = 1.15, P < 0.01, FE = 34.78) and 13 pairs solitary-biased in both transitions (permutation mean = 1.49, P = 0.02, FE = 8.72) (Tables S20 and S21).

Table 1.

Motif-OG pairs overlapping in both PGLS models.

	Solitary-biased for gains	Social-biased for gains
Social-biased for losses	0	104**
Solitary-biased for losses	42*	3

Counts of motif-OG pairs identified as significantly correlated in both gains (transition from ancestrally solitary to social state) and losses (transition from ancestral social to secondarily solitary state), separated by direction of correlation for both transitions. The majority of social-biased motif-OG pairs in gains were also social-biased in losses. Social-biased overlap (104): permutation mean = 4.57, **P < 0.01, FE = 22.76. Solitary-biased overlap (42): permutation mean = 5.33, *P = 0.03, FE = 7.88.

We visualized these convergent regulatory changes using the normalized motif counts across all 149 motif-OG pairs (Fig. 5). Unsupervised hierarchical clustering revealed a clear structure where social and solitary biased motif-OG pairs naturally grouped into distinct row clusters, and columns clustered social species separate from both solitary groups. In addition, a majority of social convergent motif-OG pairs showed regulatory changes across both families associated with gains (Apidae and Halictidae; see star in Fig. 5), while a smaller subset of the social convergent motif-OG pairs showed stronger motif associations within the Halictidae family (see asterisk in Fig. 5). These results suggest both conserved and lineage-specific motif associations are involved in social evolution.

Fig. 5.

Convergent motif-orthogroup pairs across social transitions. Heatmap rows represent all 149 motif-OG pairs significantly correlated in both gains and losses of social behavior, while columns represent species. Motif count values were normalized and scaled per row. The left-side bar indicates direction of bias (pink = social-biased in both gains and losses, blue = solitary-biased in both gains and losses, gray = mixed or not concordant). Species are annotated along the top of each column with color indicating taxonomic family, and along the bottom of each column with 4-letter species codes (Table S22) and text color indicating social status. Cluster of rows marked with a star show consistent motif signal across multiple families, while the cluster marked with an asterisk reflects motif signal primarily within the family Halictidae. Rows annotated with cartoon promoter-gene rectangles (left rectangle contains the name of the motif, right rectangle contains the name of the gene) to the right of each row are for hotspot OGs previously implicated as showing consistent social or solitary bias (Fig. 4) and series of motifs with similar normalized counts to the same promoter region are indicated with vertical pink and blue lines to the right of grouped rows.

Gene ontology enrichment analysis on the OGs within the 104 social convergent motif-OG pairs identified enrichment for 20 biological processes, including terms associated with steroid hormone biosynthesis and RNA processing (Fisher weighted P < 0.05; Table S23). OGs within the 42 solitary convergent motif-OG pairs were enriched for 18 processes associated with fatty acid synthesis and neuronal development, along with additional metabolic functions (Fisher weighted P < 0.05; Table S24). These results highlight particular functional groups which may harbor especially important regulatory changes for social behavior to evolve but which may also incur costs when social behavior is lost.

Discussion

In this study, we leveraged high-quality genomic resources spanning over 40 bee species representing multiple independent gains and losses of sociality to test whether the evolution of social behavior in bees is accompanied by consistent and reversible changes in gene regulation. By distinguishing ancestrally solitary species from secondarily solitary species in our analysis of transitions to and from sociality, we directly tested whether regulatory expansions associated with social gains are reversed when sociality is lost. We found evidence of convergence related to sociality at the level of TF motifs, gene promoters, and even specific motif-promoter regulatory relationships which are modified during transitions in social behavior. Across independent transitions from solitary to social life histories, we observed broad patterns of motif expansions in promoter regions, consistent with previous findings of increased regulatory complexity in social lineages. However, we did not find evidence that these widespread motif expansions are lost following reversions to solitary living, suggesting the loss of social behavior does not require loss of these regulatory changes. At the same time, we identified a smaller set of specific gene regulatory connections between TFs and target genes with complementary patterns of expansion and contraction associated with evolutionary gains and losses of social behavior. Together, our results provide evidence that increased regulatory complexity is consistently associated with gains in social behavior, and additionally highlight regulators and regulatory targets which may be key mediators of social evolution in bees.

Genome wide we found strong support for bee social evolution involving regulatory expansions. Transitions from ancestrally solitary to social living were consistently associated with increased numbers of predicted TF motif occurrences in gene promoter regions for over half of motifs tested, similar to previous findings (Kapheim et al. 2015) despite methodological differences. In addition to the inclusion of over 30 additional species, the use of a different database of motif binding probabilities and distinct bioinformatics tools, a primary difference between our analyses and that of Kapheim et al. is that we retained all OGs identified by OrthoFinder rather than restricting the analysis to single-copy OGs. This inclusion allowed us to capture signals in paralogous genes, which can serve as sources of new genetic material (Ohno 1970) and can be driving forces in insect morphological evolution (Loker and Mann 2022; Singh and Krumlauf 2022). Paralogous genes may have similarly important roles in social evolution and including them in our analyses allowed us to identify regulatory shifts in genes with duplications or more complex evolutionary histories. However, this may also introduce errors if genes are poorly annotated, and the approach may increase noise or obscure real signals. For example, our averaging of motif counts across paralogs may mask asymmetric changes in regulatory content among duplicate copies (e.g. when one copy gains and another loses motifs), and differences in gene duplication rates among social and solitary lineages (Chau and Goodisman 2017) could confound differences in motif abundance. To assess the potential impact of these effects, we also analyzed our data using only single-copy OGs, which yielded the same patterns of regulatory motif expansion associated with sociality as we observed with all genes. These results suggest our conclusions are robust to the inclusion of OGs with more complicated evolutionary histories, and that differences in gene copy number or annotation errors among species are not likely to explain our findings.

While our results show a strong association between social evolution and increased regulatory complexity in bees, it is possible that expansions in regulatory motifs are not driving sociality but instead reflect changes in selection pressures accompanying social evolution (Hunt et al. 2011; Rubin et al. 2019; Ferger and Tsutsui 2025). For example, relaxed selection on worker-biased genes is predicted to influence molecular evolution in social insects, and supportive evidence has been demonstrated across social insect lineages (Linksvayer and Wade 2016; Barkdull and Moreau 2023; Ewart et al. 2024). More broadly, genes involved in social phenotypes with plastic expression during development or across castes are expected to be shielded from selection, which could lead to accumulation of mutations and novel variants, including in regulatory regions. These changes may be adaptive if they result in improved regulatory control among phenotypes, which could lead to further selection and canalization associated with social complexity. Patterns of evolutionary rates in noncoding regions of bee genomes support this hypothesis, particularly for lineages with elaborations of social organization that demonstrate convergent rate changes in predicted regulatory regions (Rubin et al. 2019). In addition, the greatest number of clade-specific regulatory regions was identified in corbiculate bees, which contains the most socially complex lineages of honey and stingless bees, suggesting that the elaboration of social behavior was secondary to an expansion of regulatory regions in this lineage (Rubin et al. 2019). Collectively, these results all support the role of gene regulatory complexity in social evolution, but whether these changes were involved in driving social transitions or more often secondary consequences of changes in selection pressures requires further study.

When examining transitions from social living to secondarily solitary living, we found no evidence that these losses are associated with an equivalent contraction in motif occurrence. Instead, secondarily solitary species retain many previously expanded motifs. These results differ from those of a previous study which examined motif binding within the family Halictidae (Jones et al. 2023), although in that study both ancestrally solitary and secondarily solitary lineages were compared as one group to social lineages. With the inclusion of additional lineages and families of bees, as well as distinguishing between solitary species with distinct evolutionary histories, our results suggest that the pattern observed by Jones et al. may have been driven primarily by the large increase in motif presence during social gains. It is important to also point out that solitary reversions have occurred more recently in time than social gains. Thus, the lack of overall motif contraction we observed may reflect that difference in timing, and future research should investigate whether lineages with more or less time since phenotypic transition points show evidence of this time limitation in regulatory evolution. That said, regulatory elements have been shown to evolve quite rapidly. For example, in the ∼11 million year divergence between Drosophila melanogaster and D. yakuba, nearly 1,000 enhancers have been gained across the two species (Arnold et al. 2014). Losses of social behavior among our lineages occurred approximately 10 to 20 mya (Danforth et al. 2003; Jones et al. 2023), suggesting there has been sufficient time for at least some regulatory elements to turnover.

The asymmetry we observed between gains and losses of sociality is consistent with a predicted pattern in evolution where gains of complex traits are thought to require multiple, coordinated regulatory changes, whereas losses may occur through fewer, large-effect mutations (Prud’homme et al. 2007). For example, dramatic phenotypic losses such as stickleback spines (Chan et al. 2010) and loss of larval trichomes in Drosophila (Sucena and Stern 2000) can result from a single or few mutational changes. In contrast, gain of complex traits such as the complex wing color patterns in Heliconius butterflies arise from multiple regulatory modifications, including at least five essential enhancers of the major patterning gene optix (Lewis et al. 2019) and nearly 60 additional regulatory loci under positive selection (Lewis et al. 2020). Although gain of complex traits can involve few large-effect mutations (Martin and Orgogozo 2013), this asymmetric framework may explain why regulatory expansions associated with gains of sociality do not rapidly break down with losses of sociality, persisting due to functionality in solitary contexts, low cost and corresponding lack of selection to purge the additional binding sites, or perhaps lack of evolutionary time.

Among the gene targets of our most strongly social-biased motifs, enrichment for “negative regulation of TOR signaling” highlights a molecular theme that has repeatedly emerged in studies of social evolution. TOR (target of rapamycin) is a highly conserved eukaryotic nutrient signaling pathway that integrates nutritional and hormonal cues to regulate cell growth, metabolism, and lifespan (Colombani et al. 2003; Oldham and Hafen 2003). In insects, TOR signaling is tightly coupled with the IIS (insulin/insulin-like signaling) pathway and with the endocrine regulator juvenile hormone (JH) (Maestro et al. 2009), and in honey bees TOR signaling has been implicated as a key regulator in caste determination (Patel et al. 2007; Wolschin et al. 2011). Effects of TOR signaling on caste is mediated through JH signaling, with developmental suppression of TOR leading to reduced JH titers, shifting larval gene expression and methylation patterns and resulting in worker-like morphological and reproductive phenotypes (Wheeler 1986; Mutti et al. 2011). In addition to the role of TOR signaling in caste determination, loss of prepupal diapause, a life history trait shift thought to precede social evolution in both bees and wasps (Hunt and Amdam 2005; Santos et al. 2019), coincides with relaxed selection on anti-longevity genes and purifying selection on pro-longevity genes within the IIS/TOR pathway (Santos and Kapheim 2024). Changes in selective pressures on TOR pathway genes thus likely facilitate lifespan extension and altered nutrient responses, key components of caste differentiation and reproductive division of labor in social insects (Michener 1974; Santos and Kapheim 2024). In this context, targeting of TOR signaling genes by our most strongly social-biased motifs may reflect repeated fine tuning of nutrient and hormonal networks through gene regulatory network modulation during the evolution of social living.

In addition to TOR signaling, promoters of genes involved in glycolysis and glucose homeostasis were enriched among targets of expanding motifs, even after stringent filtering and FDR correction. Carbohydrate metabolism has been identified as a repeated target of selection in eusocial lineages, particularly in the more complex social bee groups where metabolic demands of perennial nests and extreme lifespan extension of queens may require innovative metabolic solutions (Woodard et al. 2011). Several key enzymes in the glycolysis pathway show evidence of accelerated evolution in their coding regions in solitary to social transitions as well as in elaborations of social complexity (Woodard et al. 2011). Our finding that promoters of genes involved in carbohydrate metabolism exhibit expansions of regulatory motifs suggests that coding and noncoding changes may work in tandem to modify metabolic pathways during evolutionary transitions in social behavior.

Several of the TF motifs we identified as contributing most strongly to the observed regulatory expansions associated with gains of sociality have known roles in social evolution and the regulation of social phenotypes (Kapheim et al. 2015; Jones et al. 2023). These motifs include those of the TFs lbe, ara, and h (hairy), which regulate developmental, neural, and hormonal signaling pathways. Lbe (ladybird early) is a homeobox TF involved in neuronal differentiation and muscle identity during embryonic development in Drosophila and is essential for specification of interneurons and motor neurons involved in locomotor activity (Jagla et al. 1997; De Graeve et al. 2004). Similarly, ara (araucan) is a homeobox TF that regulates dorsal-ventral patterning and sensory organ development in Drosophila (Gómez-Skarmeta et al. 1996), and hairy encodes a basic helix–loop–helix (bHLH) transcriptional repressor that acts downstream of the Notch pathway to control temporal patterning during neurogenesis (Skeath and Carroll 1991). In honey bees, hairy motifs are enriched among the promoters of genes differentially expressed between worker behavioral subtypes and in response to treatment with a JH analog (Sinha et al. 2006; Jones et al. 2020). JH signaling is linked with several aspects of social behavior, including the regulation of dominance, reproductive status, and division of labor (Hartfelder 2000). In addition, recent evidence suggests that JH binding proteins may enable JH to cross the blood brain barrier in bees (Brankatschk and Eaton 2010; Jones et al. 2023) where it could have direct effects on brain gene expression, and convergent selection has repeatedly acted on coding regions of these binding proteins when sociality is both gained and lost (Jones et al. 2023). Our finding that hairy motifs are among the most extensively expanded in promoter regions of social bees provides additional evidence for the central role of JH signaling modulation in evolutionary transitions of sociality. In addition, similar to glycolytic process genes exhibiting both coding and noncoding changes in association with social evolution, our results together with previous work suggest that JH signaling is modulated by both coding and noncoding changes during social evolution.

One caveat to our study is the reliance on Drosophila TF motifs and their binding specificities to identify putative motifs in our bee genomes. The database of motifs we used was characterized from comprehensive experimental data and represents a functionally validated subset (286) of the 628 TFs reported for D. melanogaster (Rauluseviciute et al. 2024). Although bees and flies diverged over 300 mya (Kumar et al. 2017), we feel the use of Drosophila-based binding specificities is justified given the high degree of evolutionary conservation in TF binding (Nitta et al. 2015). Even when comparing organisms as divergent as flies and humans, the binding specificities of most orthologous TFs are nearly identical (Nitta et al. 2015). Still, some TF orthologs are known to diverge in their binding probabilities, especially following gene duplication events which lead to initially redundant copies of the genes encoding TFs (Rosanova et al. 2017; Gera et al. 2022). With this in mind, our results reflect potential binding site changes in bee promoters that should be functionally validated. In addition, there are many additional TFs for which we did not analyze motif presence due to lack of experimental motif binding data. As information becomes available, particularly for nonmodel insects such as bees, the role of these additional TFs in social evolution should be examined.

Despite the diversity of social phenotypes included in our target species and potential for independent routes to sociality, we identified specific pairs of TF motifs and gene promoters with convergent and reversible changes in gene regulation associated with social behavior. Of 149 motif-OG pairs with significant correlations across both gains and losses, nearly all (98%) showed directional concordance. The majority of these convergent pairs (70%) were in the direction of increased motif presence in social species, supporting the hypothesis that social species require increased regulatory complexity. Intriguingly, a recent comparison of regulatory activity among populations of a facultatively social bee found that in social lineages, derived alleles were disproportionately predicted to increase TF binding (through the addition of a TF motif at the affected locus) (Jones et al. 2024). Surprisingly, despite the long divergence times between social origins in Halictidae and Apidae, many of the motif-OG pairs were conserved across these families, with a smaller subset restricted to within Halictidae. These findings suggest both conserved changes as well as more lineage-specific refinements of gene regulation are involved in the evolution of sociality. It is worth noting that the two gains of social behavior and repeated losses within halictids may increase the power to detect regulatory changes in this group relative to apids. As resources become available for additional groups representing additional independent gains and losses, this possibility should be explored. Still, the evidence implicating these convergent regulatory changes across lineages in association with social behavior provides strong candidates for future mechanistic studies.

The transition from solitary to social living involves the evolution of phenotypically differentiated castes, which is predicted to require increased regulatory complexity (West-Eberhard 2003; Linksvayer and Wade 2005). Our findings provide evidence that transitions to sociality are consistently accompanied by expansions in TF motif presence in promoters both genome wide and among specific regulatory targets. These motif expansions likely support context-dependent regulation of gene expression necessary to enable a single genome to produce specialized reproductive and nonreproductive castes. At the same time, we find that when social behavior is secondarily lost, promoter regions are generally not purged of these motifs, except for at specific loci that may be important for social behavior but costly in solitary contexts. Our results provide new insights into how expansions in gene regulatory complexity, particularly at key transcriptional control points, contribute to the evolution of sociality in bees.

Methods

Data Sources

We obtained 42 bee genome assemblies and their associated annotations from the National Center for Biotechnology Information (NCBI), the Darwin Tree of Life project (DToL), the Halictid Genome Browser out of Princeton University (https://beenomes.princeton.edu/), Ensembl Genome Browser, and from Dr. Priscila K. F. Santos (https://github.com/pkfsantos/Tetrapedia_diversipes_genome). Detailed information about genome assemblies, including accession numbers and access locations are available in Table S22. We used BUSCO [version 5.8.3, {Simão et al. 2015}] with the hymenoptera_odb12 database to evaluate assembly and annotation completeness of all genomes. Mean proportion of complete and single copy BUSCO orthologs was 94.65% across assemblies, and no pattern of quality difference was observed between social categories or across the phylogeny (Fig. S2). All GFF annotation files used in this study are available on the project GitHub repository, https://github.com/joneslabuky/beeTFevolution. We classified species as “ancestrally solitary”, “social”, or “secondarily solitary” based on published literature (life history references are provided in Table S25). Due to considerable variation in specific social biology across bees and lack of detailed natural history for many species, we focused on the presence or absence of a reproductive division of labor and cooperative brood care as the defining features of sociality. We excluded species with social polymorphism (i.e. both social and solitary nests observed), such as Lasioglossum albipes (Plateaux-Quenu 1993) and Megalopta genalis (Wcislo et al. 2004). In all, 13 species were classified as “social”. Note that all species exhibiting reproductive division of labor and cooperative brood care were grouped into the social category regardless of whether they are typically described as “primitively eusocial” or “highly eusocial” (Michener 1974). Species lacking those criteria (e.g. those with shared nests, or nests containing overlapping generations but no reproductive division of labor) were classified as solitary, then split into two groups based on the presumed ancestral state of the lineage. Twenty two species were classified as “ancestrally solitary”, with no history of the lineage expressing social behavior (Michener 1974; Wcislo and Fewell 2017). Seven species were classified as “secondarily solitary”, which we defined as species for which the ancestral state was social, but the lineage has secondarily reverted to a solitary life history strategy (Danforth 2002; Danforth et al. 2003). Note that while our genomic resources are concentrated on species with solitary life histories, the broader phylogeny includes many additional species that support the repeated losses of sociality among these groups (Wcislo and Danforth 1997; Danforth 2002; Danforth et al. 2003, 2013). Species selected for analysis include as many high-quality (low fragmentation, high completeness based on BUSCO; Fig. S2) bee genomes as were available that also represent as many independent gains and losses of sociality across lineages as possible.

Orthogroup Characterization

To facilitate cross-species comparisons, we used OrthoFinder 3.0.1 (Emms and Kelly 2019) to infer orthologous sets of genes (i.e. genes descended from a single gene in the last common ancestor of our 42 species (Fitch 2000; Gabaldón and Koonin 2013; Emms and Kelly 2019). We first determined the longest isoform for each gene using agatagat_sp_keep_longest_isoform.pl), then extracted nucleotide sequences for the coding regions of these isoforms from each assembly. These sequences were translated into peptide sequences, filtered to remove sequences with open reading frame errors, trimmed to remove stop codons (Dainat 2024) (agat_sp_extract_sequences.pl), and then used as the input for OrthoFinder. OrthoFinder analysis resulted in 13,168 total OGs, of which 2,087 were universal single-copy OGs. Because incomplete or incorrect annotations may hinder the identification of universal single-copy OGs, we also assessed the number of OGs that were single copy and present in >90% of species (at least 38 species), resulting in 5,127 OGs. These universal and relaxed single-copy OG sets are similar to previously published orthogroup analyses in bees (Santos and Kapheim 2024). In addition to generating OG sets, we also used OrthoFinder 3.0.1 (Emms and Kelly 2019) to generate a phylogeny of the 42 species for use in downstream analyses (Fig. 1a). General relationships among species within the OrthoFinder-generated phylogeny agree with previously published phylogenies (Gibbs et al. 2012; Branstetter et al. 2017; Henríquez-Piskulich et al. 2024).

Motif Prediction in Promoter Regions

We obtained TF binding motif sequences from the JASPAR 2024 CORE Insect motif database (Rauluseviciute et al. 2024), a curated and nonredundant set of 286 motif profiles derived from experimentally defined TF binding sites in D.melanogaster (Rauluseviciute et al. 2024). To account for variation in nucleotide composition across our different genomes, we converted the JASPAR-formatted motif profiles to species-specific probabilities of motif occurrences using background levels of nucleotide frequencies from each genome using PWMScan 1.1.9 (Ambrosini et al. 2018) (pwm_convert). This resulted in “tag” lists for each motif representing all sequences predicted to be potential TF binding sites in a given species, thresholded by a significance value of 0.00001 using the matrix_prob tool of PWMScan (Ambrosini et al. 2018). Species-specific tags for each motif were then aligned to promoter regions (defined as 5 kb upstream and 2 kb downstream of transcription start sites (Kapheim et al. 2015; Jones et al. 2023)) of all genes present in the 13,168 OGs using Bowtie v1.2.2 (Langmead et al. 2009), allowing no mismatches (bowtie –n0). All exact matches of tags aligning to promoter regions were considered motif occurrences for a given gene. Since our OGs were not all single-copy, we averaged the numbers of motif occurrences within a promoter region across all the genes within each OG for each species. This resulted in a single value for each combination of motif and OG, representing the average number of motif occurrences across genes in that OG for each species. These values were normalized within species using median absolute deviation (MAD) scaling and min–max normalization, which was performed for each motif across OGs, preserving the overall distribution of motif counts and enabling better comparison across species.

Correlation Between Predicted Motif Occurrences and Sociality

Using the phylogenetic tree generated by OrthoFinder (Emms and Kelly 2019) and the normalized motif counts, we performed PGLS analyses using Caper 1.0.3 (Orme et al. 2023) to evaluate associations between social behavior and motif occurrence while accounting for phylogenetic nonindependence among the 42 species. PGLS was performed for each OG and all 286 motifs using the model: normalized motif count ∼ sociality, where the response variable was the MAD normalized value, and the independent variable was sociality, which was treated as a categorical factor representing the social behavior of the species (i.e. ancestrally solitary, social, or secondarily solitary). We estimated the lambda parameter for our model using maximum likelihood, with the following bounds scenarios: (1e−5,1), (1e−3,1), (0.1,1), (1,1). For each OG-motif combination, we retained results from the first (broadest) lambda bounds scenario that successfully converged and documented the bounds used in the final analysis.

To analyze transitions both to and from sociality, we ran two separate PGLS models (Fig. 1b). The first PGLS model evaluated correlations between normalized motif counts for each OG and social status with respect to gains of sociality. This first PGLS only included ancestrally solitary and social species from the phylogeny to evaluate changes in motif occurrence in transitions from an ancestrally solitary to a social state. An OG was only included in this first model if it had data for at least 30% of the social species (>3 of 13 total possible) and at least 30% of the ancestrally solitary species (>6 of 22 total possible) for a given motif. This filtering resulted in 8,171 OGs tested in PGLS1. The second PGLS model evaluated correlations between normalized motif counts for each OG and social status with respect to losses of sociality. This second PGLS only included social and secondarily solitary species to evaluate changes in motif occurrence in transitions from a social ancestral state to a secondarily solitary state. An OG was only included in this second model if it had data for at least 30% of the social species (>3 of 13 total possible) and at least 30% of the secondarily solitary species (>1 of 7 total possible). This filtering resulted in 8,482 OGs tested in PGLS2.

Motif occurrence was considered associated with sociality if P < 0.001 and adjusted R² > 0.25 for a given motif-OG pair, with direction of the correlation coefficient determining the direction of the relationship with sociality. We summarized the extent to which each motif was social or solitary biased across promoters by aggregating the number of significant correlations within each PGLS (Fig. 1c). We additionally visualized the overall patterns of motif changes by plotting the number of social- and solitary-biased OG correlations for each motif for both gains and losses of sociality (Fig. 1d). All PGLS results are available at the project repository: https://github.com/joneslabuky/beeTFevolution.

To evaluate whether our findings were robust to potential biases introduced by gene copy number variation, we investigated just the universal single-copy OGs. The results of these universal single-copy OG analyses were consistent with trends found in the multicopy orthologs dataset, confirming the inclusion of multicopy OGs did not qualitatively alter our conclusions.

To control the false discovery rate, we also applied a Benjamini–Hochberg FDR correction across all PGLS results. Motif-OG pairs with FDR < 0.05 were considered significant after FDR correction. Our primary analyses and figures are based on the original significance thresholds (P < 0.001 and adjusted R² > 0.25) to maintain comparability with previous studies (Kapheim et al. 2015; Jones et al. 2023), but we additionally report results obtained with the FDR-correction to strengthen our conclusions.

To evaluate whether the number of biased motifs and motif-OG associations we observed with our significance thresholds and bias criteria exceeded those expected by chance, we performed 100 permutation replicates of our PGLS models. In these permutations, sociality labels were randomly reassigned across species for each PGLS while maintaining phylogenetic relationships and the same PGLS modeling structure. For each permutation, the complete set of PGLS analyses was rerun and the number of significant motif-OG correlations was counted under the same significance thresholds (P < 0.001 and adjusted R² > 0.25). We then compared the observed numbers of social and solitary biased motifs to the null distributions derived from these permutations. Empirical P-values were calculated as the proportion of permutation replicates with counts equal to or greater than the observed value. This framework also provided null expectations for consistently biased OGs and the overlap of significant motif–OG pairs.

Gene Ontology Enrichment Analyses

To assess functional enrichment among different sets of OGs of interest, we used the R package topGO 2.54.0 (Alexa and Rahnenfuher 2025) with gene ontology annotations obtained using biomaRt 2.58.2 (Durinck et al. 2009) from the Apis mellifera gene set (GCA_003254395.2, Amel_HAv3.1, EnsemblMetazoa Database version 114.3) (Dyer et al. 2025). OGs were mapped to genes, where only OGs with at least one A. mellifera gene were included, which were then mapped to unique GO associations. GO annotations were then aggregated at the OG level so that each OG could be treated in further analysis as a “gene” with all associated GO terms. For each test set, functional enrichment was performed separately for biological process and molecular function aspects. The default algorithm of weight01 and Fisher statistic were used while running topGO to account for GO topology and reduce any influence of highly connected terms (Alexa and Rahnenfuher 2025). Our gene universe included all orthogroups with available GO annotations. To visualize our enriched GO terms for our OGs of interest, we used GO-Figure! 1.0.2 (Reijnders and Waterhouse 2021), a semantic similarity clustering tool. Significant GO terms (Fisher weighted P < 0.05) from our topGO output were visualized using a similarity cutoff 0.5 (--similarity_cutoff 0.5), with bubble sizes reflecting the number of unique GO terms per cluster (-s members) and color intensity representing statistical significance (-c pval).

Supplementary Material

Supplementary material is available at Genome Biology and Evolution online.

Funding

This work was supported by a grant from the National Institutes of Health NIGMS (grant no. R35GM156952) to B.M.J.

Data Availability

Scripts, workflows, and additional data files are available at the following GitHub repository: https://github.com/joneslabuky/beeTFevolution.