Evolution of androgen receptors contributes to species variation in androgenic regulation of communication signals in electric fishes

Abstract

Hormones and receptors coevolve to generate species diversity in hormone action. We compared the structure and function of androgen receptors (ARs) across fishes, with a focus on ARs in ghost knifefishes (Apteronotidae). Apteronotids, like many other teleosts, have two ARs (ARα and ARβ). ARβ is largely conserved, whereas ARα sequences vary considerably across species. The ARα ligand binding domain (LBD) has evolved under positive selection, and differences in the LBD across apteronotid species are associated with diversity in androgenic regulation of behavior. The Apteronotus leptorhynchus ARα LBD differs substantially from that of the Apteronotus albifrons ARα or the ancestral AR. Structural modeling and transactivation assays demonstrated that A. leptorhynchus ARα cannot bind androgens. We propose a model whereby relative expression of ARα vs. ARβ in the brain, coupled with loss of androgen binding by ARα in A. leptorhynchus might explain reversals in androgenic regulation and sex differences in electrocommunication behavior.

1. INTRODUCTION

Hormones often coevolve with the enzymes that synthesize and/or metabolism them and with the receptors that mediate their actions This coevolution can result in new functions for hormones and their associated enzymes and receptors (Bonneton et al., 2003; Eick et al., 2012; Li et al., 2005; Ogino et al., 2018a; Prokop et al., 2012). For example, all vertebrate nuclear steroid hormone receptors evolved from a single ancestral steroid receptor gene via multiple gene/genome duplications and sub- and neo-functionalization (Eick and Thornton, 2011; Thornton, 2001). One of these duplication events gave rise to the immediate progenitors of androgen receptors (ARs) and progesterone receptors approximately 400 million years ago (Okafor et al., 2021; Thornton, 2001; Thornton and Kelley, 1998). Androgen receptors then coevolved with their ligands to produce variation in androgen function across taxa. For example, some ARs in teleost fishes are more potently activated by 11-ketotestosterone (11-KT) than by testosterone (T) or 5α-dihydrotestosterone (5α-DHT), which supports 11-KT’s role as the primary masculinizing androgen in fishes (Hossain et al., 2008; Olsson et al., 2005). In contrast, 5α-DHT is often a more potent ligand than 11-KT for mammalian ARs, (Olsson et al., 2005; Yazawa et al., 2008). Co-evolution between ligands and receptors on shorter evolutionary time scales, such as between closely related species, is often more subtle. For example, variation in nine amino acids in the ligand binding domain of glucocorticoid receptor paralogs in midshipman fishes, trout, and cichlids is linked to variation in their relative affinities for mineralocorticoids (Arterbery et al., 2011); and variation in AR sequence across birds has been hypothesized to influence androgenic regulation of demanding courtship displays in manakins and hummingbirds and of monogamy, biparental care, and sibling competition in seabirds (Schuppe et al., 2020).

Species vary in how potently androgens regulate traits. Some species variation in androgen sensitivity results from tissue-specific variation in AR expression (Fuxjager et al., 2015; Johnson et al., 2018). Variation in AR coding sequence, structure, and function, however, may also affect androgen sensitivity and responsiveness. For example, sequence variation in the human and rat AR genes is linked to variation in male sexual development (McPhaul et al., 1993; Yarbrough et al., 1990). Species-specific coding sequence variation in ARs likely contributes to variation in ligand binding and selectivity, interaction with coactivators and co-repressors, and transcriptional regulation of target genes (Schuppe et al., 2020). The role of androgens in regulating sexually differentiated traits thus likely evolves as the expression, structure, and function of ARs diversify.

Androgen receptors are comprised of four interacting functional domains. (Eisermann et al., 2013; Schuppe et al., 2020). The N-terminal domain (NTD) facilitates protein-protein interactions, such as AR dimerization and co-regulator interaction. The DNA binding domain (DBD) binds to androgen response elements (AREs) in target gene promoters to influence which genes are regulated by androgens. The unstructured hinge domain mediates myriad functions, including nuclear localization, DNA binding, and ARE selectivity (Clinckemalie et al., 2012). The ligand binding domain (LBD) forms a binding pocket to interact with androgens and upon ligand binding undergoes conformational changes that allow AR dimerization and interactions with cofactors required to regulate gene transcription (Bennett et al., 2010; Davey and Grossman, 2016; Eisermann et al., 2013; Nadal et al., 2017). AR DBD and LBD sequences are largely conserved across vertebrates, possibly reflecting constraints on AR ligands and targets (Schuppe et al., 2020; Thornton and Kelley, 1998).

Teleost fishes are a good model for studying coevolution of androgens and their receptors. Hormone function varies widely across teleosts, even among closely related species (Dijkstra et al., 2012; Forlano and Bass, 2011; Gonçalves and Oliveira, 2011; Oliveira et al., 2005). Early in teleost evolution, a whole genome duplication created an opportunity for neofunctionalization of many traits (Glasauer and Neuhauss, 2014), including the evolution of electric organs and sonic muscles from skeletal muscle precursors and the evolution of the bulboarteriosus from cardiac muscle (Arnegard et al., 2010; Moriyama et al., 2016; Moriyama and Koshiba-Takeuchi, 2018; Ravi and Venkatesh, 2008; Thompson et al., 2014; Volff, 2005; Zakon et al., 2006). The duplication of genes for hormone receptors and synthesizing enzymes likely allowed for the evolution of novel hormone functions and contributed to the remarkable diversity and plasticity of reproductive physiology across teleost fishes (Desjardins and Fernald, 2009; Douard et al., 2008; Godwin, 2010; Lorin et al., 2015; Mank and Avise, 2009; Ogino et al., 2009; Oliveira et al., 2005).

The teleost genome duplication generated two AR paralogs, ARα and ARβ, that likely facilitated diversification of androgenic regulation (Douard et al., 2008; Ogino et al., 2016; Ogino et al., 2018a; Ogino et al., 2018b). For example, different androgens can have distinct functions in teleosts. Unlike most mammalian species, females in many teleost species can have testosterone (T) levels equal to or greater than those of males; and the primary masculinizing androgen in fishes is 11-ketotestosterone (11-KT) (Borg, 1994; Brantley et al., 1993; Desjardins et al., 2006; Dunlap et al., 1998; Ho et al., 2010; Lorenzi et al., 2012; Mank, 2007; Petzold and Smith, 2016). In species with alternative male reproductive morphs, “sneaker” males often have higher T concentrations than territorial males, whereas territorial males often have elevated 11-KT concentrations, which allows T and 11-KT to have distinct roles even within males (Brantley et al., 1993; Knapp and Neff, 2007; reviewed in Oliveira et al., 2005). In three-spined sticklebacks (Gasterosteus aculeatus), 11-KT is elevated in males and is far more potent than T in inducing the hypertrophy of the kidney to produce a male-specific, nest building protein (spiggin), whereas T concentrations are similar between the sexes in some G. aculeatus populations and are linked to individual variation in body size (Jakobsson et al., 1999; Kitano et al., 2011). Similarly, in the electric fish Brachyhypopmus guaderio, both 11-KT and T masculinize the duration of their electric communication signals, but only 11-KT masculinizes the amplitude of these signals (Goldina et al., 2011).

In this study, we examined the sequence of AR genes across teleosts, with a focus on the ghost knifefishes (Apteronotidae), the most speciose family of South American electric fishes (Gymnotiformes). Apteronotid fishes vary in androgenic regulation of sex differences in their communication behavior and thus provide an interesting opportunity to investigate the coevolution of hormone receptors and hormonal regulation of behavior. Apteronotids have an electric organ that produces a weak electric discharge for communication and electrolocation. The frequency of the electric organ discharge (EODf) can communicate species identity and sex (Hagedorn and Heiligenberg, 1985; Hopkins, 1988; Turner et al., 2007). In some species, EODf differs between males and females, and the direction and the magnitude of sex differences in EODf vary across apteronotid species (Smith, 2013; Zakon and Dunlap, 1999). Sex differences in EODf are regulated by androgens and/or estrogens, and species and population differences in the sexual dimorphism of EODf are caused by changes in how these hormones affect EODf (Dunlap et al., 2017; Ho et al., 2013; Petzold and Smith, 2016; Zakon and Dunlap, 1999). EODf is controlled by a single, dedicated brain region, the pacemaker nucleus (Pn), and sex differences in EODf are caused by steroid-induced changes in the excitability of Pn neurons (Schaefer and Zakon, 1996; Smith, 1999; Szabo and Enger, 1964).

The two apteronotid species in which hormonal regulation of EODf has been most studied, A. albifrons and A. leptorhynchus, differ both in the direction of EODf sex differences and in the androgenic regulation of EODf. In A. albifrons, males have a lower EODf than females, whereas A. leptorhynchus males have a higher EODf than females (Dunlap et al., 1998). Correspondingly, 11-ketotestosterone (11-KT) robustly masculinizes EODf in both species by decreasing EODf in A. albifrons, but by increasing EODf in A. leptorhynchus (Dunlap et al., 1998). Additionally, the efficacy of another androgen, 5α-DHT, differs between these species. 5α-DHT masculinizes EODf in A. albifrons but does not affect EODf in A. leptorhynchus (Dulka, 1997; Dunlap et al., 1998; Meyer et al., 1987). EODf is not affected by androgens in apteronotid species and populations with sexually monomorphic EODf (Ho et al., 2010; Ho et al., 2013; Petzold and Smith, 2016; Smith, 2013). Thus, apteronotid fishes provide a good model to investigate how variation in androgen receptors might contribute to species differences in androgenic regulation of sexually dimorphic behavior.

We compared AR sequences across species to examine the evolution of AR paralogs across teleosts and to determine whether species variation in the AR sequence across apteronotid fishes could underlie species variation in androgenic regulation of EODf. We used three different approaches to test this hypothesis: (1) comparing sequences of AR domains across teleost species and testing for diversifying selection across teleosts in general and specifically within gymnotiform species; (2) analyzing structural models to examine how AR LBDs in A. leptorhynchus vs. A. albifrons interact with androgens; and (3) employing a luciferase assay to quantify how androgens act on the LBDs of the A. albifrons and A. leptorhynchus ARs to regulate gene transcription. We find diversification in the structure and function of one of the AR paralogs in apteronotids that might contribute to species diversity in the sexual dimorphism of electrocommunication signals.

2. METHODS

2.1. Androgen receptor phylogenies

To examine the evolution of ARs across teleosts, we constructed an amino acid sequence-based phylogeny by using AR sequences from teleost fishes (35 species, 51 ARs) and from non-teleost fishes (3 species, 3 ARs) as outgroups (Table S1). We used all teleost AR sequences we could obtain from previously generated transcriptomes (Smith et al., 2018) or from BLAST searches of the NCBI and electric fish genomics databases (https://efishgenomics.integrativebiology.msu.edu/) with AR sequences from Astatotilapia burtoni, Danio rerio, and A. albifrons as templates. Although some teleost AR sequences were annotated specifically as ARα or ARβ within the source databases, others were unannotated or were annotated just as “androgen receptor.” Assignments as ARα or ARβ were made post-hoc based on phylogenetic relationships, and largely match assignments in other phylogenetic analyses of teleost ARs (Hoadley et al., 2022b; Ogino et al., 2016). An earlier phylogenetic study of fish ARs (Douard et al., 2008), however, inverted this nomenclature, and the ARα or ARβ in the present study correspond to AR-B and AR-A, respectively, of Douard et al. (2008). Sequences were trimmed to include only the base pairs from the start codon to the stop codon and were translated from nucleotide to amino acid sequences with EXPASY as needed (Artimo et al., 2012). In four cases, the full AR sequence was found not in a single transcript, but across two transcripts. In all four of these cases, the two AR transcripts had significant overlapping sequences that were identical between the two transcripts, and which together corresponded to a complete AR sequence. All four cases were from de novo transcriptome assemblies, where partial assemblies are common (particularly when the species used for a reference transcriptome is not very closely related to the target species). All four cases were assumed to be a single transcript and are denoted in Table S1 with two transcript IDs. For any mRNA sequences that did not already have a corresponding amino acid sequence in the source database, the mRNA sequence was translated to protein sequence with EXPASY (Artimo et al., 2012; Gasteiger et al., 2003). Protein sequences were aligned by using the MUSCLE algorithm (version 3.8.31) on Indiana University’s server cluster (KARST). The alignment was imported into jalview (Waterhouse et al., 2009), and amino acid sequences were simultaneously trimmed to include only the AR.

A phylogeny was constructed from full length AR amino acid sequences, with RAxML v. 7.4.2 (Stamatakis, 2006; Stamatakis, 2015) on Indiana University’s KARST server cluster. Maximum likelihood trees, with 1,000 bootstrap replicates, were used on amino acid sequences with the JTT and gamma models of substitution. Non-teleost actinopterygian (Acipenser ruthenus, Erpetoichthys calabaricus, and Lepisosteus oculatus) AR sequences were used as outgroups. The output from RAxML was visualized with dendroscope (v.3.5.10) (Huson and Scornavacca, 2012), which reports the bootstrap values at each node. Figtree was used to visualize the phylogeny (Rambaut, 2014).

2.2. Tests for diversifying selection

To test whether AR genes have undergone diversifying selection, codon-based multiple sequence alignments of AR genes were constructed with codon-msa in hyphy (https://github.com/veg/hyphy-analyses/tree/master/codon-msa) on Indiana University’s Quartz computing cluster and with MUSCLE (Edgar, 2004) running on a PC. Episodic diversifying selection was assessed separately for the two AR paralogs (ARα and ARβ) by using adaptive branch-site random effects likelihood (aBSREL) in hyphy (Smith et al., 2015). In addition, aBSREL was used to separately identify episodic diversifying selection in each AR domain. Site-specific episodic selection was assessed on ARα and ARβ by using a mixed effects model of evolution (MEME) in hyphy (Murrell et al., 2012), with a significance threshold of 0.05. These analyses were all conducted with the previously generated amino acid-based phylogeny and with the three non-teleost ARs as outgroups. Statistical tests were confined to the terminal branches.

2.3. Calculating similarity between A. albifrons and A. leptorhynchus AR sequences

ARα and ARβ amino acid sequences from A. albifrons and A. leptorhynchus were aligned by using the MUSCLE algorithm running on KARST. To determine the percent similarity of AR amino acid sequences between A. leptorhynchus and A. albifrons, pairwise sequence comparisons were made with the SIAS (Sequence Identity and Similarity) server (http://imed.med.ucm.es/Tools/sias.html). To determine percent similarity for each domain of the AR (i.e., NTD, DBD, hinge, and LBD), the AR alignment was imported into Jalview version 2.11.1.4 (Waterhouse et al., 2009), and an alignment with ARs from other taxa with well characterized domains was used to determine the boundaries of each domain in the apteronotid ARs (Hossain et al., 2008; Ogino et al., 2016). Sequences for each domain were exported and pairwise comparisons of similarities were calculated as described above for the whole AR.

2.4. ARα and ARβ expression in the pacemaker nucleus

AR expression in the Pn was quantified by using the reads per kilobase million (RPKM) values from a previous transcriptomic study of the Pn of A. albifrons (n=4 males, 4 females) and A. leptorhynchus (n=3 males, 3 females; for details see Smith et al., 2018). We also used quantitative PCR (qPCR) to confirm the expression levels of ARα and ARβ in the Pn.

To collect samples for qPCR to study expression of ARα and ARβ, pacemaker nuclei were dissected from 12 A. leptorhynchus (6 males, 6 females) and 12 Colombian A. albifrons (6 males, 6 females). All subjects were adult, wild-caught fish obtained from commercial suppliers (EMark Tropical, Brooklyn, NY: Ruinemann’s, Miami, FL: East Coast Transship, Baltimore, MD). Fish were housed individually in 38-L tanks in a recirculating aquarium system, and were maintained on a 12:12 light:dark cycle at 25–27°C, pH 5.5–6.5, and a conductivity of 150–350uS/cm.

Fish were deeply anesthetized by immersion in 2-phenoxyethanol (1 mL/L, Sigma-Aldrich) in tank water for several minutes. Fish were then placed on ice, and brains were removed rapidly and placed in RNAlater (Thermo-Fisher Scientific, Waltham, MA). The dura was removed from the ventral surface of the medulla to expose the Pn, which is readily visible as a protrusion on the ventral surface of the hindbrain. The entire Pn was cleanly dissected from the rest of the hindbrain by using Vannas scissors. The gonads were also removed, and the mass of the gonads and body of the fish were measured to calculate gonadosomatic index (GSI) as an index of the reproductive condition (GSI = 100 * gonad mass/body mass). Pn tissue was stored in RNAlater at 4°C overnight and then at −20°C. Whole brains (minus the already harvested Pn) were collected for use as a cross-plate calibrator and were stored with the same method as the Pn. Animal usage was performed with procedures covered by the National Research Council’s Guide for the Care and Use of Laboratory Animals and under a protocol approved by the Indiana University Bloomington Institutional Animal Care and Use Committee.

Primers for ARα, ARβ, and two reference genes (HECTD3 and Slc25a5) were validated for both A. albifrons and A. leptorhynchus (Table S2a) by confirming the presence of a single PCR amplicon. To compare the relative expression of the two ARs in the Pn of A. albifrons and A. leptorhynchus, we extracted RNA from each Pn by using Maxwell RSC simplyRNA Tissue Kits (Promega Corporation, Madison, WI). RNA quality (RIN>7.2, 8.0 ± 0.04, avg±SE) and quantity were assessed with an Agilent Tape Station 2200 (Agilent Technologies, Santa Clara, CA). cDNA was produced with iScript gDNA Clear cDNA Synthesis Kits (Bio-Rad USA, Hercules, CA) using 10 ng of RNA input for each 20 μL cDNA reaction. qPCR was performed with PrimeTime gene expression master mix (Integrated DNA technologies, Inc. San Diego, CA) on a QuantStudio 6 flex (Applied Biosystems, Waltham, MA), with TaqMan probes and primers (Integrated DNA technologies, Inc. San Diego, CA). Efficiencies for each primer set fell between 97–101% for both species (supplemental Table S2a).

ARα and ARβ expression in the Pn was quantified relative to the expression of two reference genes. qPCR for ARα and ARβ was run on one plate, and the reference genes (HECTD3 and Slc25a5) were run on another plate. The whole-brain calibrator was run with the reference gene primers on both plates. The intraplate variation, quantified as the difference in the calibrator reference gene CT values between the two plates, was very low (1.6%). Normalized AR gene expression in each Pn sample was quantified by calculating 2^(−Δ CT), using the CT of each AR gene relative to the geometric mean of the two reference genes’ CT values.

2.5. Modeling the structure of the A. albifrons and A. leptorhynchus AR LBD

Computer models were used to predict the 3D structure of the LBD of ARα. Structures of ARα LBDs from A. albifrons and A. leptorhynchus were modeled using the Phyre2 homology modeling (Kelley et al., 2015) with the ancestral androgen receptor (AncAR1) LBD (https://doi.org/10.2210/pdb7RAE/pdb) as the template. Figure 7 was generated using ClustalW and ESPript.

Figure 7: Structural Models of ARα LBDs.

(A) Sequence alignment of the ancestral androgen receptor (AncAR1) LBD and ARα LBDs from A. albifrons and A. leptorhynchus. Identical residues are highlighted with a red background, while conserved residues are displayed in red text. Secondary structure elements of AncAR1 LBD are indicated at the top. Asterisks highlight residues that directly interact with androgens. Below/next to asterisks are residue numbers in the corresponding A. albifrons and A. leptorhynchus ARα, respectively. The AF-2 region is indicated by blue boxes and is comprised of portions of helices 3, 4, and 12 (AF-H). (B) Structural superimposition of the AncAR1 LBD (cyan) with ARα LBDs from A. albifrons (green) and A. leptorhynchus (orange). Tif2 coactivator peptide is shown in magenta, and 5α-DHT ligand is shown in blue. (C-E) Amino acids surrounding the ligand binding pocket and interactions with 5α-DHT. Hydrogen bonds are formed between 5α-DHT and the AncAR1 LBD (C), A. albifrons ARα LBD (D), and A. leptorhynchus ARα LBD (E), but steric hindrances created by Trp 352 and Leu 519 likely block 5α-DHT occupation in the ligand binding pocket of the A. leptorhynchus ARα (E).

2.6. Reporter gene assay for interaction of androgens with ARα LBDs

To assess the functional integrity of the apteronotid ARα LBDs, we used ARLBD-Gal4DBD/UAS-driven reporter assays. ARαLBD-Gal4DBD constructs were generated by replacing the LBD from an existing mineralocorticoid receptor (MR) LBD-Gal4 DBD construct with the LBDs of either A. leptorhynchus or A. albifrons ARα by using Gibson assembly cloning. Briefly, the plasmid and coding sequences for wild-type (WT) ARα LBD from A. albifrons and A. leptorhynchus were PCR-amplified by using primers listed in Table S2b. These PCR products were cloned into the above-mentioned PCR-amplified pSG5 expression vector following a Gal4 DNA-binding domain. The resulting construct fully replaced the MR LBD in the original MRLBD-Gal4DBD construct with the apteronotid ARα LBD. Mutant A. leptorhynchus ARα LBDs were generated by QuikChange site-directed mutagenesis kit (Agilent) using primers listed in Table S2b and were verified by Sanger sequencing prior to use. U-2 OS human osteosarcoma cells were maintained and passaged in α-minimal essential medium (Life Technologies) supplemented with 10% stripped FBS (Invitrogen). Cells were grown in 96-well plates were transfected at 70% confluence with 10 ng of AR LBD-Gal4 DBD, 50 ng of upstream activator sequence-driven firefly luciferase reporter, and 1 ng of Renilla luciferase reporter under the control of the constitutively active pRL-TK promoter, with FuGene HD (Roche Applied Science) in OptiMEM (Invitrogen). Twenty-four hours after transfection, cells were treated with different concentrations of 5α-DHT, 11-KT, or vehicle (DMSO) in triplicate. Renilla and firefly luciferase activities were measured 24 hours after androgen or vehicle treatment (to allow sufficient time for transactivation and luciferase protein synthesis) by using the DualGlo kit (Promega) read on a BioTek Neo plate-reader (Winooski, VT). Data were fit with log(agonist) vs response curve in GraphPad Prism v8 (GraphPad, Inc).

3. RESULTS

3.1. AR Phylogeny

Most gymnotiform knifefishes have two AR paralogs corresponding to ARα and ARβ in other teleosts (Fig. 1; Ogino et al., 2016). ARα sequences have diversified more than ARβ sequences in apteronotids, as they have in other teleosts (Ogino et al., 2016). Although a short, 129-base pair fragment with ~80% amino acid similarity to portions of ARα LBD was identified in the E. electricus genome, no full length ARα sequences were found in the genome or in transcriptomes from multiple tissues from this species. Similarly, although two transcripts corresponding to short, 1848 base pair, regions of the ARβ NTD were found in B. gauderio, the combined transcripts lacked more than the first 800 base pairs of the NTD. We thus did not include the E. electricus ARα fragments or the B. gauderio ARβ fragments in our phylogenetic analysis. Additionally, no sequence corresponding to ARα was found in Siluriformes (catfishes, e.g., Ictalurus punctatus and Trichomycterus areolatus), Salmoniformes (Onchorhynchus mykiss), or Cypriniformes (e.g., zebrafish, Danio rerio; goldfish, Carrasius auratus).

Figure 1: Phylogeny of actinopterygian AR amino acid sequences.

AR in non-teleost actinopterygians (A. ruthenus, E. calabaricus, and L. oculatus) were used as outgroups. The support values listed at each node are the bootstrap probabilities based on the JTT model and the GAMMA model in the maximum likelihood analysis. Red branches indicate AR sequences in apteronotids. ARα has longer branch lengths than ARβ in most lineages, suggesting that ARα diversified more than ARβ.

3.2. ARβ sequence conservation

With the exception of ARβ in Sternarchorhynchus sp., ARβ sequence is largely conserved across gymnotiform species (Fig. 2). Pairwise similarities of ARβ range from 77% (Gymnotus ormararum vs. Electrophorus electricus) to 96% (A. albifrons vs. A. leptorhynchus). Most variation in the gymnotiform ARβ is in the NTD (pairwise similarities 62% to 97%) and the hinge region (pairwise similarities 48% to 100%). In contrast, the DBD and LBD of ARβ are highly conserved (pairwise similarities 99% to 100% for DBD and 93% to 100% for LBD for all species except Sternarchorhynchus).

Figure 2: Pairwise comparison of ARα and ARβ amino acid sequences across gymnotiform species.

Each cell in the matrix shows the pairwise sequence similarity of the entire AR gene (A,B), the N-terminal domain (NTD, C,D), DNA-binding domain (DBD, E,F), hinge region (G,H), and ligand-binding domain (LBD, I, J) between gymnotiform species. Species are organized based on phylogenetic relationships. Each comparison matrix is split into three quadrants (separated by white lines): the lower left quadrant shows comparisons among the non-apteronotid gymnotiform species; the upper right quadrant shows comparisons within the family Apteronotidae; and the upper left quadrant compares apteronotid with non-apteronotid gymnotiform species. Note the lower similarities between species for ARα than for ARβ, particularly in the NTD and LBD.

ARβ in Sternarchorhynchus sp. differed significantly from ARβ sequence in other gymnotiforms (Fig. 2). The lack of conservation in the Sternarchorhynchus ARβ results primarily from a stop codon after the residue corresponding to human residue 759 (Sternarchorhynus residue E735; Fig S1), in the middle of the LBD. The resulting truncation of the LBD likely renders the Sternarchorhynchus ARβ unable to bind to androgens.

3.3. ARα sequence divergence

Based both on branch lengths in the AR sequence phylogeny (Fig. 1) and on pairwise sequence comparisons (Fig. 2), ARα sequence has diversified more than ARβ sequence in gymnotiforms. Pairwise similarities of ARα sequence range from 59% (A. leptorhynchus vs. Eigenmannia virescens) to 100% (Sternopygus macrurus vs. Brachyhypopomus gauderio). In particular, ARα sequence in Sternarchorhynchus and A. leptorhynchus differs markedly from that of other apteronotids. Sequences both of the entire ARα and of individual ARα domains are less similar between species than the corresponding domains of ARβ (Fig. 2). As with ARβ, the DBD of ARα was relatively well-conserved across species (pairwise similarities 86% to 100%), whereas the ARα NTD (pairwise similarities 31% to 100%) and hinge region (pairwise similarities 48% to 100%) were more variable across species, particularly between apteronotid and non-apteronotid gymnotiform species (Fig. 2). Unlike ARβ, where the LBD is highly conserved (except for the truncation of ARβ in Sternarchorhynchus, Fig. 2I), the LBD of ARα varies substantially across gymnotiform species, including across apteronotids (pairwise similarities 69% to 100%, Fig. 2J). Sternarchorhynchus and A. leptorhynchus ARα LBDs are particularly dissimilar to ARα LBD in other gymnotiform species (70–75% and 69–79% similarity, respectively).

3.4. Diversifying Selection on ARα and ARβ

Branch-site models of nonsynonymous and synonymous substitutions in both full-length ARα and ARβ as well as in individual domains of these genes indicate differential positive selection. Although ARβ shows significant evidence of positive selection in sixteen of the 31 teleost species tested, only one of those species (Sternopygus macrurus) was a gymnotiform (Fig. S2a, Table S3). Positive selection on ARβ is concentrated in the NTD. The ARβ NTD was under positive selection in all sixteen of these lineages, whereas none of the other ARβ domains had significant diversifying selection on any terminal branch (Fig. S3, Table S3). ARα was under positive selection in thirteen of twenty-one teleost species tested, and three of those species (Eigenmannia virescens, A. leptorhynchus, and Sternarchorhynchus sp.) were gymnotiforms (Fig S2b, Table S4). As with ARβ, positive selection on ARα often occurred in the NTD (6/21 species, including E. virescens, Table S4, Fig S4a); and the DBD was highly conserved (no significant positive selection on any branch, Table S4). Unlike ARβ, however, positive selection on ARα also occurred in some taxa in the hinge region (Fig S4b, Table S4) and the LBD. Two non-gymnotiform teleosts (Astatotilapia burtoni and Astyanax mexicanus) had significant positive selection on the ARα hinge region. The ARα LBD of two apteronotids (A. leptorhynchus and Sternarchorhynchus sp.) has undergone positive selection (Fig 3).

Figure 3: Adaptive branch-site random effects likelihood (aBSREL) model for episodic diversification on ARα LBD.

The phylogeny is topological, and branches are not to scale. Red species labels on the terminal branches indicate significant episodic diversifying selection (Table S4). Colors on the terminal branches indicate omega values from the aBSREL model, with redder values (ω >> 1) indicating stronger diversifying positive selection, grey values (ω ≈ 1) indicating neutral evolution, and bluer values (ω << 1) indicating more constraint. The relative lengths of colored segments on terminal branches indicates the proportion of sites fit with the omega values indicated by the colors.

The domains of ARα and ARβ differed in the proportion of codons that have undergone episodic positive selection (Fig 4). Forty-seven of 1015 (4.6%) of ARβ codons and 49/964 (5.1%) of ARα codons showed significant positive episodic selection in at least some lineages, and these sites were not randomly distributed. The NTD and hinge regions of both ARα and ARβ contained numerous sites that had undergone positive episodic selection. In contrast, the DBD was highly conserved in both ARα and ARβ; only 1/100 ARβ DBD sites and 0/98 ARα DBD sites were under positive selection. Interestingly, the LBD was highly conserved in ARβ (2/380 sites (0.5%) under positive selection), but much more subject to positive selection (19/266 sites (7.1%)) in ARα.

Figure 4: ARα and ARβ sites under diversifying selection.

Scale along top indicates codon position in each gene. Red lines indicate sites found by the mixed effects model of evolution (MEME) to have undergone diversifying selection in at least one branch of the phylogeny for ARα or ARβ. The color-coded bars below indicate the domains and numbers in parentheses indicate the percentage of sites under diversifying selection. Note the relatively larger proportion of sites under selection in the NTD and hinge region of ARβ, and the relatively larger proportion of sites under selection in the NTD, hinge, and LBD of ARα. Raw MEME results available in Supplementary Materials.

3.5. Comparing sequence and expression of ARα and ARβ in A. albifrons and A. leptorhynchus

Because A. leptorhynchus and A. albifrons differ markedly in how androgens regulate electrocommunication behavior (Zakon and Dunlap, 1999), we focused on comparisons of ARs in these two species. As in other gymnotiform species, ARβ is well conserved between A. albifrons and A. leptorhynchus (Table 1, Figs. 5, S1). In contrast, ARα differs much more between these two species, particularly in the LBD. Forty-four of 251 to 263 amino acids differed between the A. albifrons and A. leptorhynchus ARα LBD (Table 1, Figs. 2J and 5). Two of these non-synonymous substitution sites in the ARα LBD (Gly350/ Try352 and Thr517/Leu519) interact directly with androgens (Fig. 5). Interestingly, the Gly/Try substitution at a ligand-interacting site is also present in the ARα LBD of another apteronotid, P. hasemani (site 39 in Fig. S1).

Table 1.

Protein sequence similarity of ARα and ARβ domains between A. albifrons and A. leptorhynchus

	1. Whole AR	2. N-terminus domain (NTD)	3. DNA binding domain (DBD)	4. Hinge	5. Ligand Binding Domain (LBD)
ARα	87.6%	93.6%	98.8%	88.0%	79.1%
ARβ	96.2%	93.9%	100.0%	94.1%	99.6%

Figure 5: Comparison of amino acid sequence of the ligand binding domain of ARs in humans (Hs), the teleost Astatotilapia burtoni (Ab), and apteronotids A. albifrons (Aa) and A. leptorhynchus (Al).

AR sequences from humans and A. burtoni are provided for outgroup comparisons. Numbers above each residue correspond to the amino acid number in the human AR. Dashes indicate gaps in the alignment. AF-2 region indicated by the blue bars. Black arrowheads indicate highly conserved residues of ARα in other teleosts that differ in the A. leptorhynchus ARα, with open arrowheads signifying conservative substitutions and filled arrowheads indicating non-conservative substitutions. The five boxes with white asterisks indicate conserved residues that interact directly with androgens. Numbers beneath the asterisks indicate residue numbers in the A. albifrons and A. leptorhynchus ARα, respectively (see also Fig 7). A. leptorhynchus ARα differs from A. albifrons ARα in two of these ligand-interacting residues, indicated with arrowheads beneath the asterisks (Gly350/Trp352 and Thr517/Leu519). The consensus sequence is based on sequence of 60 vertebrate ARs, including teleost ARα (n=22) and ARβ (n=31), and non-teleost fish and tetrapod AR (n=7) (Table S1). The bar plot above each residue indicates how conserved (using percent identity) each residue is across these 60 ARs.

The brain region that controls EOD frequency expressed both ARs. ARα was more abundant than ARβ in transcriptomes from the Pn of A. albifrons and A. leptorhynchus (Smith et al., 2018; Table 2). qPCR also confirmed that ARα was more highly expressed than ARβ in the Pn of both species (Table 2). GSIs were indicative of robust reproductive condition and did not differ significantly between the two species either in the transcriptomic study (Smith et al., 2018; females: A. leptorhynchus 7.7 ± 2.6 vs. A. albifrons 6.2 ± 1.7; males: A. leptorhynchus 0.37 ± 0.07 vs. A. albifrons 0.47 ± 0.08; ANOVA, main effect of species F_1,10 = 0.22, p = 0.65) or in the present qPCR experiment, (females: A. leptorhynchus 7.6 ± 0.8 vs. A. albifrons 8.1 ± 1.8; males: A. leptorhynchus 0.42 ± 0.04 vs. A. albifrons 0.62 ± 0.11; ANOVA, main effect of species, F_1,20 = 0.07, p = 0.79). Correlations between ARα expression and ARβ expression in the Pn differed between the two species. ARα and ARβ expression in the Pn were not correlated with each other in A. albifrons (Spearman correlation, S=197, ρ=0.31, p= 0.3), but the expression of these two genes in the Pn was strongly correlated in A. leptorhynchus (Spearman correlation, S=17, ρ=0.94, p<0.001) (Fig. 6).

Table 2.

Expression of ARα and ARβ mRNA in the pacemaker nucleus of A. albifrons and A. leptorhynchus as measured by RNAseq and qPCR

	Transcriptome1		qPCR
	ARα (RPKM±SEM)	ARβ (RPKM±SEM)	ARα ((2^[-Δct])±SE)	ARβ ((2^[-Δct])±SE)
A. albifrons	26.54 ± 3.0	8.73 ± 0.8	0.28 ±0.01	0.13 ±0.01
A. leptorhynchus	38.35 ± 3.2	5.73 ± 0.4	0.25 ±0.02	0.13 ±0.01

¹From Smith et al., 2018

Figure 6:

Spearman correlations between expression of ARα and ARβ (as measured by qPCR) in the medullary pacemaker nucleus (Pn), which controls electric organ discharge frequency in A. albifrons (A) and A. leptorhynchus (B). ARα and ARβ expression in the Pn were not correlated in A. albifrons (S=197, ρ=0.31, p= 0.3), but were tightly correlated in A. leptorhynchus (S=17, ρ=0.94, p<0.001).

3.6. Structural model of ARα and interactions with 5α-DHT

Because the LBD of ARα differed substantially between A. leptorhynchus and A. albifrons, we further investigated differences in the structure and function of the ARα LBD in the two species. Primary sequence alignment of the LBDs of ARα from A. leptorhynchus and A. albifrons with that of a reconstructed ancestral vertebrate androgen receptor (AncAR1; Colucci and Ortlund, 2013; Okafor et al., 2021) showed that the AncAR1 LBD is more similar to the A. albifrons ARα LBD than to the A. leptorhynchus ARα LBD (Fig. 7A). These differences are most pronounced after residue 198 of the AncAR1 sequence, located in the middle of helix 10. Overall, ARα LBDs from both A. leptorhynchus and A. albifrons adopt the classical LBD structure, consisting of a helical sandwich with three layers, including eleven α helices and four β strands (Fig. 7B). The predicted structure of A. leptorhynchus ARα LBD stops in the middle of helix 12 (AF-H), largely due to the high sequence divergence (Fig. 7A,B). AR ligands, such as 5α-DHT, occupy the ligand binding pocket (LBP) at the base of LBD. Detailed inspection of the LBPs of the three ARs shows extensive hydrogen bonds formed between the LBD and 5α-DHT. The carbonyl oxygen on the A-ring of 5α-DHT participates in a hydrogen bond network with residues Gln43 and Arg84 of the AncAR1. Residues Asn37 and Thr209 make hydrogen bonds with the hydroxyl oxygen on the D-ring. (Fig. 7C). Similar hydrogen bonds are formed between 5α-DHT and the ARα LBD from A. albifrons (Fig. 7D). Three of the four hydrogen bonds formed by the AncAR LBD with 5α-DHT are also available in the A. leptorhynchus ARα LBD. However, critical substitutions in the A. leptorhynchus ARα LBD likely interfere with ligand binding. Specifically, Leu519 in the A. leptorhynchus LBD is in the position analogous to Thr517 in the A. albifrons ARα LBD and Thr209 of the AncAR1 LBD, rendering the A. leptorhynchus ARα incapable of forming a hydrogen bond at this position. Most importantly, Gly40 in the AncAR1 LBD (and Gly350 in A. albifrons ARα LBD) is replaced by Trp352 in the A. leptorhynchus ARα LBD. The bulky side chain of Trp352, as opposed to a small Gly residue, creates steric hindrances with 5α-DHT that potentially block 5α-DHT binding. Modeling binding of 11-KT with the A. albifrons and A. leptorhynchus ARα LBD similarly reveals that these changes in the A. leptorhynchus ARα LBD likely also prevent 11-KT binding (Fig. S5). Indeed, the 11-keto group of 11-KT likely exacerbates the steric hindrance created by Trp352 in the A. leptorhynchus ARα LBD.

Additionally, the activation function-2 (AF-2) domain, which is comprised of portions of helices 3, 4, and 12 of the LBD and which is required for ligand-dependent interactions of the AR LBD with coregulators and the NTD (Tan et al., 2015), is not well conserved in the A. leptorhynchus ARα LBD (Figs. 5, 7A). In particular, the helix 12 portion of AF-2 is well-conserved in ARβ of both apteronotids, with 89% identity to the corresponding segment of the human AR AF-2. The helix 12 component of AF-2 in the A. albifrons ARα is somewhat less conserved (55.6% identity, 77.8% similarity to human AR AF-2), but is this sequence is poorly conserved in the A. leptorhynchus ARα (25% identity, 37.5% similarity to human AR AF-2).

3.7. ARα reporter gene assay

We used a luciferase reporter assay to examine androgen-dependent transactivation mediated by the ARα LBD. The ARαLBD-Gal4DBD constructs contained the LBD of either A. leptorhynchus or A. albifrons ARα. Without ligand (DMSO only), neither ARα LBD construct increased luciferase activity compared to a mineralocorticoid receptor LBD in the same pSG5 construct, suggesting no constitutive activation without androgens (data not shown). Both 5-αDHT and 11-KT induced strong luciferase responses in cells transfected with a construct that had the A. albifrons ARα LBD, but neither androgen stimulated luciferase activity in cells transfected with the construct containing the A. leptorhynchus ARα LBD (Figs. 8A and B). Both 5-αDHT and 11-KT were strong agonists for A. albifrons ARα LBDs with EC₅₀ values of 4 nM and 10 nM, respectively (Fig. 8C). To investigate the role of key residues in the LBD (particularly Trp352 and Leu519 in the A. leptorhynchus ARα LBD), we mutated these two residues in A. leptorhynchus ARα LBD to those found in the analogous position in the A. albifrons ARα (i.e., W352G and L519T). These single or double mutations (W352G/L519T), however, did not rescue the ability of the A. leptorhynchus ARα LBD construct to induce a luciferase response when incubated with 5-αDHT or 11-KT (Figs. 8D, E).

Figure 8: Different responses of A. albifrons and A. leptorhynchus ARα LBDs to androgens.

Chimeric constructs used in this assay fused the LBD from the apteronotid ARα with a Gal4DBD construct. (A,B) Receptor activation for wild type ARα LBD was measured by luciferase reporter gene activation in the presence of 1 uM 5α-DHT (A) or 11-KT (B) normalized to DMSO only as a control. Aa- A. albifrons, Al- A. leptorhynchus. Error bars indicate S.D. from three replicates and from three independent experiments. (C) Relative luciferase activity across different concentrations of androgens (5α-DHT or 11-KT) with the A. albifrons ARα LBD construct. (D,E) Receptor activation for variants containing point mutations in A. leptorhynchus ARα LBD compared to wild type A. albifrons ARα when exposed to 1 uM 5α-DHT (D), or 1 uM 11-KT (E). Theoretically, the double point mutation (Al-W352G L519T) might have been expected to restore ligand binding in A. leptorhynchus ARα, since these residues were the ones creating steric hindrance with 5α-DHT (Fig. 7) and were mutated to be the same as residues as those found in the A. albifrons ARα. Androgenic activation of the A. leptorhynchus ARα, however, was not restored by these mutations.

4. DISCUSSION

4.1. Evolution of teleost AR paralogs

Sexual dimorphism and reproductive characteristics vary substantially across teleost fishes (Desjardins and Fernald, 2009; Godwin, 2010; Mank and Avise, 2009; Oliveira et al., 2005). One factor that might contribute to this species diversity is the whole-genome duplication (WGD) event that occurred in the ancestor of all teleosts and that generated paralogous copies of all genes, including those that regulate hormone action and sex differences (Glasauer and Neuhauss, 2014). Gene paralogs created by WGD typically undergo one of three fates that can influence evolution of traits regulated by these genes: (1) pseudogenization and loss of one of the gene copies; (2) subfunctionalization, in which the ancestral functions of the original gene are divided between the two gene copies, breaking up pleotropic functions of the ancestral genes and allowing each paralog to evolve to optimize its respective subfunctions; and (3) neofunctionalization, in which one paralog retains ancestral functions, freeing the other paralog to evolve and adopt novel functions. Lineage-specific loss, subfunctionalization, and neofunctionalization of genes like the AR that regulate gonadal steroid action likely contributed to the diversification of the sex-related traits in fishes (Douard et al., 2008; Lorin et al., 2015; Ogino et al., 2023).

Indeed, each of these processes has likely influenced androgen signaling in different teleost lineages. Our phylogenetic analysis of teleost ARs as well as those of others (Douard et al., 2008; Hoadley et al., 2022b; Ogino et al., 2016) revealed numerous independent losses of ARα. Cypriniformes, Siluriformes (catfishes), and Salmoniformes all have ARβ, but lack ARα, which suggests that ARα was lost in these orders. A previous analysis (Douard et al., 2008) suggested that the loss of ARα might have occurred at the base of the Otophysi (Cypriniformes + Charciformes + Siluriformes + Gymnotiformes). Our findings that some gymnotiform and charciform (e.g., Pygocentrus nattereri and Astyanax mexicanus) species have both ARα and ARβ, however, indicate that Siluriformes and Cypriniformes lost ARα independently, while Gymnotiformes and Characiformes retained ARα. Similarly, within the knifefishes, we found that ARα has apparently been lost in electric eels, and ARβ may be severely truncated and potentially non-functional in B. guaderio. In some teleost lineages where both AR paralogs remain, they have apparently undergone subfunctionalization. Studies in which ARα and/or ARβ have been knocked out revealed that ARα and ARβ regulate distinct male-typical traits in cichlids (Astatotilapia burtoni) and medaka (Oryzias latipes), which indicates that the two AR paralogs have assumed different masculinizing functions of the ancestral AR (Alward et al., 2020; Hoadley et al., 2022a; Ogino et al., 2023). Patterns of diversification of ARα vs. ARβ are also consistent with neofunctionalization in some lineages. As in other teleost groups (Ogino et al., 2016), we found that ARα diversified more than ARβ in gymnotiforms, which is consistent with ARβ retaining ancestral AR function and thereby freeing ARα to evolve novel functions. We also found that the two paralogs differed in conservation of functional domains of the AR gene across teleosts. The DBD of both ARα and ARβ has few nonsynonymous substitutions and is likely highly constrained. The LBD, on the other hand, is highly conserved with few nonsynonymous substitutions in ARβ, but in ARα has been subjected to diversifying selection in at least some teleost lineages. This suggests evolutionary lability and neofunctionalization of ARα’s interactions with androgens, whereas ligand interactions with ARβ have likely been largely conserved to preserve ancestral AR function.

4.2. ARβ sequence conservation and function in A. albifrons and A. leptorhynchus

Most of the sequence variation in ARβ in gymnotiforms and other teleosts was in the N-terminus domain (NTD) and the hinge region of ARβ, and these were the domains most subject to diversifying selection. Like ARs in other vertebrates, the DBD and LBD of gymnotiform ARβ were highly conserved and have undergone little or no diversifying selection across teleosts. The conservation of the AR DBD and LBD reflect stabilizing selection based on the functional importance of these domains and suggests that the interactions of ARβ with androgens and target gene promotors is conserved across teleosts (Schuppe et al., 2020; Thornton and Kelley, 1998). As in other teleosts, the LBD of ARβ in A. albifrons and A. leptorhynchus is also well conserved, with only a single amino acid difference (Table 1, Fig. 5). The conservation of ARβ across apteronotids, including A. albifrons and A. leptorhynchus, suggests that differences in androgen effects on EODf in these two species are unlikely to result from ARβ sequence variation.

4.3. ARα sequence variation and function in A. leptorhynchus and A. albifrons

ARα sequence varied more than ARβ sequence across gymnotiform species. Similar to ARs in other vertebrates, gymnotiform ARα varied substantially in the NTD and was more conserved in the DBD than in other domains (Fig. 2; Blázquez and Piferrer, 2005; Schuppe et al., 2020; Thornton and Kelley, 1998). Additionally, we found that the NTD was the domain of ARα most likely to undergo positive selection in most teleosts, whereas selection on the ARα DBD was highly constrained. Unlike in the gymnotiform ARβ or ARs in other vertebrates, however, the ARα LBD varied substantially across gymnotiform species. The LBD of ARα varied more among the four apteronotid species than among the non-apteronotid gymnotiform species, and two apteronotids (A. leptorhynchus and Sternarchorhynchus) were the only teleosts tested in which the ARα LBD has undergone significant positive selection.

The LBD was also the domain of ARα that differed most between A. albifrons and A. leptorhynchus (Table 1). LBD residues that directly interact with androgens in the A. albifrons ARα LBD were similar to those of other vertebrates. In contrast, the A. leptorhynchus ARα LBD differed in at least two amino acids that interact directly with 5α-DHT (Fig. 5). The sequence differences in the ARα LBD between A. albifrons and A. leptorhynchus likely contribute to species differences ARα-mediated androgenic regulation. These substitutions likely impair androgen binding to the A. leptorhynchus ARα. Structural models of the LBD predicted that androgens are stabilized in the ligand binding pocket of the A. albifrons ARα as it is in other vertebrate ARs but would be hindered from binding to the A. leptorhynchus ARα (Fig. 7).

The inability of A. leptorhynchus ARα LBD to bind androgens was further confirmed with transactivation assays. As expected, the A. albifrons ARαLBD-Gal4DBD construct induced robust gene transcription with 11-KT or 5α-DHT, demonstrating that the A. albifrons ARα can bind these androgens to induce gene transcription. In contrast, the A. leptorhynchus ARαLBD-Gal4DBD construct did not induce gene transcription with either 5α-DHT or 11-KT. Moreover, reverting the two residues predicted to constrain androgen binding in the A. leptorhynchus ARα LBD to those found in the A. albifrons ARα LBD still did not restore androgen-mediated gene transcription of the AR construct. This suggests that these mutations were not sufficient to make A. leptorhynchus LBD androgen-responsive, even though the ligand binding might have been restored. The failure of androgens to induce transcription even in the partially reverted A. leptorhynchus ARα LBD could be due to additional mutations in the A. leptorhynchus ARα LBD, particularly in the AF-2 region. AF-2 plays key a role for ligand-dependent recruitment of transcriptional coactivators and interactions between the LBD and NTD that then drive target gene transcription (Davey and Grossman, 2016; Jin et al., 2019; Tan et al., 2015). AF-2, particularly the portion from helix 12, is poorly conserved between the A. leptorhynchus and A. albifrons ARα LBDs, with only one of the nine amino acids in common and with little similarity to this portion of AF-2 in humans (Fig. 5 and Fig. 7A). Indeed, the lack of conservation in this portion of the LBD rendered structure of the LBD after the middle of helix 10 of A. leptorhynchus ARα LBD unable to be reliably predicted in the structural model and thus might create a conformation disfavoring ligand-dependent coactivator binding or N-C interactions. Thus, ARα in A. leptorhynchus likely is unable to act as a functional androgen receptor both because its LBD is incapable of binding androgens and because additional mutations in the LBD coregulator binding site might interfere with recruitment of transcriptional co-activators to the receptor complex.

The luciferase transactivation assay using LBD-Gal4/UAS constructs is a well-established tool to assess the function of nuclear receptor LBDs (Eick et al., 2012; Liu et al., 2010; Markov et al., 2017). This assay clearly showed the loss of ligand-dependent function of the A. leptorhynchus ARα LBD. A potential limitation of this assay, however, is that the ARLBD-Gal4 constructs contained only the LBD of ARα. Thus, the effects of sequence variation in the ARα NTD, hinge, or DBD would not be revealed in this assay, and additional mutations outside the LBD of the A. leptorhynchus ARα could also impact the function of the native receptor. For example, in most vertebrates, a nuclear translocalization signal in the hinge region and DBD of the AR allows the AR to be translocated to the nucleus when it binds to androgens (Zhou et al., 1994). However, unlike other vertebrate ARs, ARα in many teleosts is constitutively localized to the nucleus independent of androgen binding, at least in transfected cells and fish embryos (Bain et al., 2015; Ogino et al., 2009; Ogino et al., 2016). Moreover, variants of AR in the gilthead seabream (Sparus aurata L.) and humans (homo sapiens) with truncations in the LBD translocate to the nucleus and are constitutively active (Sánchez-Hernández et al., 2014; Watson et al., 2010). If A. leptorhynchus ARα were also localized to the nucleus and/or were constitutively active like the sea bream AR, this might not have been revealed in the luciferase assay used in this study because the nuclear localization signal is in the AR DBD and hinge region, which were not included in the chimeric ARαLBD-Gal4DBD construct.

Similarly, our finding that the A. albifrons ARαLBD-Gal4DBD construct more strongly activated gene transcription in the presence of 5α-DHT than 11-KT might be confounded by the lack of other ARα domains in the construct. Native ARα in other teleosts binds 5α-DHT with greater affinity than 11-KT, but 11-KT is nevertheless as or more potent than 5α-DHT at inducing gene transcription via these receptors (Bain et al., 2015; Hossain et al., 2008; Ogino et al., 2016; Olsson et al., 2005). The paradoxically enhanced potency of 11-KT in stimulating ARα-mediated gene transcription despite its lower binding affinity than 5α-DHT results from complex interactions of the LBD and DBD (Ogino et al., 2016). Thus, although 5α-DHT was a more potent ligand than 11-KT for the A. albifrons ARαLBD-Gal4DBD construct, the native A. albifrons ARα (with its own NTD, hinge region, and DBD) might induce gene transcription as or more effectively when bound to 11-KT than when bound to 5α-DHT, like ARα in some other teleosts. Evaluating these confounds will require future studies to assess the structure and transactivation potential of other AR domains and of the full-length native ARs in apteronotids.

4.4. Species variation in AR expression, sequence, and function might underlie variation in androgenic regulation of behavior:

Our findings combined with studies of AR function in other teleosts lead us to propose a model for how species differences in the expression, structure, and function of ARα and ARβ might contribute to differences between A. albifrons and A. leptorhynchus in androgenic regulation and sexual dimorphism of EODf (Fig. 9). The reversal in the sex difference in EODf between these two species is caused by reversed effects of androgens. 11-KT raises EODf in A. albifrons but lowers EODf in A. albifrons (Dulka, 1997; Dunlap et al., 1998; Meyer et al., 1987; Schaefer and Zakon, 1996; Zakon and Dunlap, 1999). The brain region that controls EODf, the Pn, expresses more ARα than ARβ in both A. albifrons and A. leptorhynchus (Table 2). We found that ARα likely binds androgens in A. albifrons, but not in A. leptorhynchus. In other teleost fishes, ARα more strongly activates target gene expression than ARβ (Ogino et al., 2009; Ogino et al., 2016).

Figure 9:

Proposed conceptual model of how differences in expression and function of ARα and ARβ might underlie species differences in androgenic regulation and sexual dimorphism of electric organ discharge frequency (EODf) in A. albifrons and A. leptorhynchus. (A) Androgens lower EODf in A. albifrons but raise EODf in A. leptorhynchus (Zakon and Dunlap, 1999). (B) Androgen action in the pacemaker nucleus (Pn), which controls EODf, in A. albifrons. Androgens interact with both ARα and ARβ. ARα is expressed more (Table 1) and more potently activates gene expression than ARβ (Ogino et al., 2009; Ogino et al., 2016). If ARα and ARβ antagonize each other’s actions (as they do in cichlids; Hoadley et al., 2022a), then ARα-mediated effects of androgens might dominate in A. albifrons to lower EODf. (C) In A. leptorhynchus, ARα cannot bind androgens, and androgen actions are mediated primarily by ARβ. ARα may enhance the effects of ARβ in A. leptorhynchus via heterodimerization and enhanced nuclear localization. We hypothesize that ARβ-mediated androgen effects dominate in A. leptorhynchus to raise EODf. The model also incorporates greater potency of 11-KT than 5α-DHT as ligands for ARβ, but similar potency of these androgens as ligands for ARα, as has been found in other teleosts (Bain et al., 2015; Hossain et al., 2008; Ogino et al., 2009; Ogino et al., 2016; Olsson et al., 2005). The dominance of ARα (11-KT potency ≈ 5α-DHT potency) in mediating androgen action in the A. albifrons,Pn, but of ARβ (11-KT potency > 5α-DHT potency) in mediating androgenic effects in A. leptorhynchus might explain why 5α-DHT potently masculinizes EODf in A. albifrons but not in A. leptorhynchus (Dulka, 1997; Dunlap et al., 1998; Meyer et al., 1987).

The higher potency and Pn expression of ARα compared to ARβ, combined with the loss of ARα androgen binding in A. leptorhynchus might explain the reversed EODf sex differences in A. leptorhynchus vs. A. albifrons (Fig. 9). Because ARα is expressed at higher levels in the Pn and more strongly activates target genes, the masculinizing effects of androgens on EODf in A. albifrons are likely mediated predominantly through activation of ARα. In contrast, in A. leptorhynchus, ARα cannot bind androgens, and androgen-induced gene transcription is likely mediated only through ARβ. In other teleosts, ARα and ARβ may have distinct, and potentially antagonistic, effects on physiology and behavior. Specifically, in the cichlid, Astatotilapia burtoni, ARβ (but not ARα) mediates androgenic regulation of male-specific coloration, whereas ARα (but not ARβ) mediates androgenic regulation of male reproductive and aggressive behaviors (Alward et al., 2020). Further, ARα can antagonize actions of ARβ; ARβ facilitates testis growth, whereas ARα inhibits testis growth (Hoadley et al., 2022a). Similarly, in medaka (O. latipes), ARβ regulates male-typical fin morphology, head pigmentation, and sexual motivation, whereas ARα regulates tooth and sperm duct morphology. Moreover, the downstream genes whose transcription is activated in the O. latipes brain by ARα and ARβ are also significantly non-overlapping (Ogino et al., 2023). We hypothesize that ARα and ARβ might similarly have opposing effects on EODf. Specifically, we posit that androgens acting via ARα in the Pn might lower EODf, whereas androgenic action via ARβ in the Pn might raise EODf. If this assumption is correct, androgens would be expected to have a net effect of lowering EODf in A. albifrons because ARα is expressed more in the Pn and is more potent than ARβ. This would result in A. albifrons males having lower EODf than females. In contrast, in A. leptorhynchus, ARα can no longer bind androgens, and thus androgens likely act solely via ARβ, which might raise EODf and cause males to have higher EODf than females. Thus, the loss of androgen binding by ARα in A. leptorhynchus could have shifted androgen action from relying primarily on ARα in A. albifrons to relying on ARβ in A. leptorhynchus, resulting in a reversal in the direction of androgenic regulation and sexual dimorphism of EODf. Testing this model will require confirming that ARα and ARβ have divergent effects on EODf in apteronotids as they do on testis mass, coloration, and behavior in cichlids.

The loss of androgen binding of the A. leptorhynchus ARα might also explain species variation in the effects of different androgens on EODf (Fig. 9). In A. albifrons, both 5α-DHT and 11-KT robustly masculinize EODf, whereas in A. leptorhynchus, 11-KT masculinizes EODf, but 5α-DHT has little or no effect (Dulka, 1997; Dunlap et al., 1998; Meyer et al., 1987; Zucker, 1998). In several teleost species, 11-KT is a more effective ligand than 5α-DHT for ARβ, whereas 11-KT and 5α-DHT often have similar potencies to induce ARα-mediated gene expression (Bain et al., 2015; Hossain et al., 2008; Ogino et al., 2009; Ogino et al., 2016; Olsson et al., 2005). The greater expression and potency of ARα than ARβ in the A. albifrons Pn might explain why both 11-KT and 5α-DHT masculinize EODf. In contrast, because the A. leptorhynchus ARα cannot bind androgens, masculinization of EODf in A. leptorhynchus depends on ARβ, which in other teleosts is less activated by 5α-DHT than by 11-KT. This might explain why 5α-DHT is less effective in masculinizing EODf in A. leptorhynchus.

4.5. Putative function of an ARα that cannot bind androgens

Although ARα in A. leptorhynchus cannot bind androgens, ARα expression in the Pn is still robust and is tightly correlated with ARβ expression and with EODf in A. leptorhynchus but not in A. albifrons (Fig. 6; Proffitt and Smith, in prep). This correlation suggests that ARα’s function in A. leptorhynchus might be tied to that of ARβ. Hormone receptors losing their ability to bind ligands, but retaining a regulatory role on their paralogous receptors has occurred in other systems. In the cephalochordate Branchiostoma floridae, the estrogen receptor BfER lost its ability to induce gene expression in response to estrogens, but instead acts as a competitive repressor of estrogen response element binding by another steroid receptor, BfSR, the ortholog of vertebrate androgen, progesterone, and corticosteroid receptors (Bridgham et al., 2006). Because ARs function as dimers, the A. leptorhynchus ARα might modulate ARβ action via heterodimerization, which is common for steroid receptors. For example, the estrogen receptors ESRα and ESRβ form heterodimers that regulate expression of different suites of genes than ESRα or ESRβ homodimers (Monroe et al., 2005; Powell and Xu, 2008). Moreover, mammalian ARs can form heterodimers with orphan receptors or with NTD-truncated AR splice variants to inhibit AR-mediated gene transcription (Ahrens-Fath et al., 2005; Ikonen et al., 1998; Lee et al., 1999; Palvimo et al., 1993); with LBD-truncated AR splice variants to promote AR nuclear localization and gene 38ranscription (Palvimo et al., 1993; Roggero et al., 2021; Xu et al., 2015); or even with estrogen receptors (ESRα) to inhibit AR-mediated and ESRα-mediated gene transcription (Panet-Raymond et al., 2000).

If ARα heterodimerizes with ARβ in A. leptorhynchus, it could influence ARβ’s cellular localization and/or its ability to regulate gene transcription. Unlike ARβ or most ARs in non-teleosts, ARα expressed in transfected cultured cells or in fish embryos can localize constitutively to the nucleus even in the absence of ligand and can more strongly stimulate androgen-regulated gene transcription than ARβ (Ogino et al., 2009; Ogino et al., 2016), although a recent study suggests that in some tissues ARα can have a cytoplasmic localization and thus that that ligand-independent nuclear localization is not universal (Ogino et al., 2023). If ARα can constitutively localize to the nucleus and more potently stimulate gene transcription, then its heterodimerization to ARβ might therefore induce ligand-independent nuclear localization and enhanced androgen-stimulated gene expression, similar to AR LBD variants in mammals (Palvimo et al., 1993; Xu et al., 2015). Determining whether and how ARα interacts with and influences the function of ARβ will require confirmation that ARα and ARβ are expressed in the same cells in the Pn and further studies in which the two ARs are co-expressed to assess their influence on each other’s ability to bind androgens and regulate gene transcription. Because one of the substitutions that likely renders ARα incapable of binding androgens in A. leptorhynchus also occurs in another apteronotid, P. hasemani (Fig. S1), comparing how ARα and ARβ interact to influence androgen actions across species provides an exciting opportunity to investigate the evolution of synergies between hormone receptors and their ligands.

4.6. Loss of androgen binding by ARβ may promote evolution of ARα in Sternarchorhynchus sp.

In most other taxa, the DBD and LBD of ARs are highly conserved, whereas the NTD often varies more across taxa (Schuppe et al., 2020; Thornton and Kelley, 1998). ARβ in apteronotids followed this pattern with one exception, the Sternarchorhynchus ARβ. A stop codon in the middle of the Sternarchorhynchus ARβ LBD likely renders it incapable of binding to androgens. The truncation of the LBD of ARβ in Sternarchorhynchus is accompanied by a high degree of diversification of the LBD of ARα in this species. As in A. leptorhynchus, ARα in Sternarchorhynchus differs substantially from that in other gymnotiforms; and Sternarchorhynchus is one of two apteronotid species in which the ARα LBD has undergone diversifying selection. The truncation of the LBD and loss of androgen binding in Sternarchorhynchus ARβ, the more “conserved” AR paralog in teleosts, might have created selective pressure for ARα to refunctionalize and restore some of the functions of ARβ. The diversification and positive selection on ARα in Sternarchorhynchus might reflect this refunctionalization. However, the changes in the Sternarchorhynchus ARα sequence are not simple reversions to ARβ sequences (Fig. S1). Because ARα was likely already undergoing neofunctionalization when the truncation of the ARβ LBD occurred, the sequence of ARα might not have been able to revert directly back to its ancestral sequence to recover ancestral function. At least one other study has found that when neofunctionalized genes are selected to reacquire their ancestral functions, they do so by novel mechanisms (Ben-David et al., 2019). Further studies of the structure and function of the Sternarchorhynchus ARs will be needed to test this hypothesis and to investigate how the function of paralogous hormone receptor genes may influence each other’s evolution.

5. Conclusion

We found that, as in many other teleost lineages, ARβ, one paralog of an ancient duplication of the androgen receptor, was highly conserved in most South American ghost knifefishes (Apteronotidae), but that the other paralog, ARα, diversified. We documented specific changes in the structure and function of ARα in A. leptorhynchus and proposed a model for how these differences might underlie species differences in how androgens regulate a sexually dimorphic behavior. Under this model, the duplication of the AR gene may have allowed neofunctionalization whereby the two AR paralogs interact in ways that differentially affect EOD frequency, and the subsequent loss of androgen binding to one of these paralogs could then cause a reversal in the direction of androgenic regulation and sexual dimorphism of this behavior. The diversity both of AR sequence and of and hormonally regulated sexually dimorphic behavior in apteronotids allow them to serve as a promising system for comparatively investigating how the evolution of steroid receptor expression, sequence, and function contributes to diversification of hormone-regulated behavior.

HIGHLIGHTS.

• One teleost androgen receptor paralog (ARα) has evolved more than the other (ARβ).

• The ligand binding domain (LBD) of ARα has been particularly evolutionarily labile.

• Mutations in Apteronotus leptorhynchus ARα LBD led to loss of androgen binding.

• Species variation in AR function underlies diversity of behavioral sex differences.