Evolutionary History of Chemosensory-Related Gene Families across the Arthropoda

Introduction

Chemosensation refers to the physiological responses of sense organs to chemical stimuli, including taste and odor, and is observed across a wide range of taxa from bacteria to humans (Bargmann 2006;Vosshall and Stocker 2007;Nei et al. 2008;Kaupp 2010). Chemosensory systems play critical roles in mediating behavioral responses such as feeding, mating, predator avoidance, and predation. Chemosensing in the phylum Arthropoda is particularly intriguing, given the extraordinary diversity of habitats and ecological niches that arthropods have been able to colonize, spanning marine, brackish, hypersaline, freshwater, terrestrial, and extremely arid environments (Cloudsley-Thompson 1975;Sømme 1989;Glenner et al. 2006;Kelley et al. 2014). These habitat colonizations would have imposed novel challenges and requirements for chemosensation, as the transmission and reception of chemical stimuli become altered in diverse environments, such as in aquatic versus aerial media. Such diverse transmission media would impose varying evolutionary pressures on genes underlying chemosensory responses. Interestingly, the three major subphyla within the Arthropoda, that is, the Pancrustacea (e.g., crustaceans and insects), Myriapoda (e.g., centipede and millipedes), and Chelicerata (e.g., spiders, mites, and scorpions) have colonized freshwater and terrestrial habitats independently (Giribet et al. 2001;Regier et al. 2010;von Reumont et al. 2012;Oakley et al. 2013). Thus, given these independent transitions to land, have chemosensing systems evolved through the same pathways during these parallel but independent colonization events?

Based primarily on the study of the fruit fly Drosophila melanogaster, arthropod chemoreception has been found to be mediated by three different multigene families of chemosensory receptors. These include two gene families of seven transmembrane receptors, namely the gustatory receptors (GRs) (Clyne et al. 2000) and the more derived odorant receptors (ORs) (Clyne et al. 1999;Gao and Chess 1999;Vosshall et al. 1999), which are unrelated to the vertebrate GRs and ORs (Gardiner et al. 2009). More recently, a third family of chemosensory receptors has been discovered in D. melanogaster, namely, the ionotropic receptors (IRs), which are a class within the ancient and highly conserved ionotropic glutamate receptor (iGluR) family of ligand-gated ion channels (Benton et al. 2009;Croset et al. 2010;Abuin et al. 2011;Benton 2015). In addition, two soluble binding protein families, the chemosensory proteins (CSPs) and insect-type odorant binding proteins (OBPs), are known to mediate the transport of ligands to the chemosensory receptors (Pelosi et al. 2006;Laughlin et al. 2008;Vieira and Rozas 2011;Pelosi et al. 2014). In this study, we refer to these five gene families (ORs, GRs, IRs, CSPs, and OBPs) collectively as the "Chemosensory-Related Gene families" (CRGs).

Although CRGs have been studied intensively since the 2000s, little information has been gained regarding these genes in arthropods beyond the insects (Hexapoda), until very recently. Thus, the evolutionary history of CRGs throughout the Arthropoda had remained largely unexplored and poorly understood. Emerging data are beginning to suggest that the major CRGs might have expanded, contracted, or become completely lost throughout the course of arthropod evolution (Robertson and Wanner 2006;Peñalva-Arana et al. 2009;Robertson and Kent 2009;Hansson and Stensmyr 2011;Vieira and Rozas 2011;Zhou et al. 2012;Pelosi et al. 2014;Robertson 2015;Saina et al. 2015).

Some hypotheses have posited a link between CRG family expansion and habitat colonizations. In particular, the expansion of the OR gene family had been hypothesized to be associated with the colonization of land by insects (Hexapoda), to enable the detection of volatile compounds in air (Robertson et al. 2003;Peñalva-Arana et al. 2009;Kra ˚ng et al. 2012). This hypothesis was consistent with the intriguing absence of ORs and OBPs in the water flea Daphnia pulex, belonging to the crustacean lineage (Branchiopoda) that forms a clade with the insects (Peñalva-Arana et al. 2009; Vieira and Rozas 2011) (fig. 1). Nevertheless, there is some debate regarding whether the expansion of the OR gene family was the result of a terrestrial adaptation (Missbach et al. 2014). In addition, prior studies had not sampled the crustaceans outside of the branchiopod/hexapod clade, preventing resolution on whether the ORs and OBPs are absent from the Daphnia lineage alone or instead absent from all crustaceans outside of the insect clade. Also, unresolved is whether independent colonizations of land in the other arthropod subphyla (i.e., Chelicerata and Myriapoda) also coincided with expansions of the OR gene family (Chipman et al. 2014).

More generally, the evolutionary histories of CRG families and hypotheses regarding which CRG gene families are the most ancient have been gaining some clarity only recently. An earlier hypothesis had posited that the IRs represent the most ancient arthropod chemoreceptors, dating back to the origin of the Protostomia (Croset et al. 2010). In contrast, more recent studies found GRs to be more ancient, originating early in the evolution of metazoans, given their presence in the eumetazoan phylum Placozoa (Trichoplax adhaerens) (Robertson 2015;Saina et al. 2015). In addition, the analysis of evolutionary histories of IR genes has been based mostly on studies of insects, with relatively little investigation of their presence or absence in other arthropod lineages (Croset et al. 2010).

Addressing the hypotheses above, regarding patterns of CRG evolution across the Arthropoda, requires the analysis of multiple members within the subphylum Pancrustacea beyond the insects (Hexapoda), the inclusion of the arthropod subphyla Myriapoda (e.g., centipedes and millipedes) and Chelicerata (e.g., spiders, mites, and scorpions), as well as the inclusion of outgroup phyla. However, until recently, genomic data beyond the hexapod/branchiopod clade (e.g., insects and Daphnia) had been lacking. Very few comparative analyses of CRG evolution had included the subphyla Chelicerata and Myriapoda (Chipman et al. 2014;Robertson 2015), and only a few molecular evolutionary studies of crustacean CRGs had been performed (Peñalva-Arana et al. 2009;Kra ˚ng et al. 2012;Corey et al. 2013). Within the Pancrustacea, the critical phylogenetic placement of the Copepoda enables the resolution of CRG family gain or loss in the insects (Hexapoda), as they are outside of the Allotriocarida (Hexapoda/ Branchiopoda/Remipedia) clade, yet are often found to be the closest sister group to this clade (von Reumont et al. 2012;Oakley et al. 2013;Sasaki et al. 2013;Eyun 2017). Thus, we focused much attention on the Copepoda, in order to explore patterns of CRG gain or loss in close evolutionary proximity to the clade containing the insects.

In addition to the crucial phylogenetic placement of the Copepoda, their chemoreception is inherently interesting from both ecological and evolutionary perspectives.

Copepods occupy an enormous range of habitats in the aquatic realm, from freshwater to hypersaline, and shallow pool to deep sea environments (Hardy 1956;Huys and Boxshall 1991;Martin and Davis 2001). They also form the largest biomass of all animals in the world's oceans, and possibly on the planet (Hardy 1956;Huys and Boxshall 1991;Humes 1994;Verity and Smetacek 1996). Copepods are particularly known to frequently exhibit cases of cryptic speciation, where large genetic distances and reproductive isolation are accompanied by morphological stasis (Burton 1990;Ganz and Burton 1995;Edmands 1999;Lee 2000;Lee and Frost 2002;Goetze 2003;Grishanin et al. 2006;Rynearson et al. 2006;Eyun et al. 2007;Chen and Hare 2011). In the absence of morphological cues and differentiation, it has been hypothesized that speciation in copepods occurs through rapid evolution of chemical sensing (Snell and Morris 1993).

Thus, the goals of this study were to address the hypotheses above on CRG family evolution across the Arthropoda. Our specific goals were to: 1) determine patterns of gains and losses of CRG families across the phylum Arthropoda, 2) infer the evolutionary origins of the arthropod CRG families, and 3) examine sex-specific differences in CRG family expression in copepods.

Arthropod Chemosensory-Related Genes . doi:10.1093/molbev/msx147

General Characteristics of the Copepod Eurytemora affinis Genome

We sequenced the full genome of the copepod Eurytemora affinis, as copepods provide a critical phylogenetic outgroup data point to the branchiopod/hexapod clade for analyzing patterns of CRG evolution. The E. affinis genome was sequenced as part of the i5K pilot at the Baylor College of Medicine Human Genome Sequencing Center, a pilot project to investigate large-scale genomic sampling of the arthropods and provide a framework for comparative arthropod genomics. Genome sequencing was performed on an inbred line (see Materials and Methods), with a genome size estimated at 0.6-0.7 pg DNA/cell ($587-685 Mb) based on Feulgen DNA cytophotometry (Rasch et al. 2004). The draft genome assembly is relatively compact at 495 Mb, smaller than the total genome size due to our inability to assemble highly repetitive heterochromatin from short read sequence data. It is larger than the Daphnia pulex genome ($200 Mb) (Colbourne et al. 2011), a species selected in part for its small genome size in the age of expensive Sanger sequencing. The genome size of E. affinis is on the lower end of the range observed for copepods (0.14-12 pg) (Gregory 2016), and smaller than most crustaceans, where the average genome size of 6.7 pg has slowed the adoption of genome sequencing of these taxa.

The contiguity was below average with a contig N50 of 5.7 kb with a scaffold N50 of 863 kb, giving us confidence for a high-quality automated annotation (see supplementary table S1 for additional statistics and public repository accession numbers, Supplementary Material online). Automated gene model annotation using a Maker 2.2 pipeline customized for arthropods (Cantarel et al. 2008) generated 29,783 gene models. This number is likely an overestimate due to gene model fragmentation across gaps within and between scaffolds, but is somewhat in-line with the 18,440 gene models in D. pulex (PA42) (Ye et al. 2017) and 29,121 gene models in Daphnia magna, relative to the lower number of $15,000 for insects. Of 1,977 control genes expected to be present in all arthropods (Simao et al. 2015), 91.5% were identified in the genome assembly and 86.3% were represented in the automated gene model set. Thus, gene families and most genes were present in the assembly and gene set, but the absence of any particular gene from the assembly could be due to the draft nature of the assembly.

Overview of Chemosensory-Related Gene (CRG) Family Evolution

We comprehensively examined gains and losses of CRGs (ORs, GRs, IRs, CSPs, and OBPs), using 33 distinct genomes and transcriptomes across the phylum Arthropoda, as well as multiple metazoan outgroup phyla (Onychophora, Nematoda, Mollusca, Cnideria, Placozoa, Porifera, Ctenophora) and additional fungal and protistan groups (see Materials and Methods; supplementary tables S2-S4, Supplementary Material online). We found fewer chemosensory receptor genes in arthropods ($12-500; fig. 1), relative to vertebrates ($1,391 olfactory receptors and >300 vomeronasal receptors in mouse) and nematodes (>1,200 serpentine receptors in Caenorhabditis elegans) (Niimura and Nei 2003;Chen et al. 2005;Bargmann 2006;Robertson and Thomas 2006). While this relatively low number had been known for insects (Nei et al. 2008), we now confirm that this pattern holds generally true across the phylum Arthropoda (fig. 1) (Chipman et al. 2014;Gulia-Nuss et al. 2016).

Our results on patterns of CRG evolution revealed the GRs to be the most ancient of all the eumetazoan CRGs, given the inferred presence of GR or GR-Like genes in the common ancestor between arthropods and the phylum Placozoa (fig. 1; see next section for details). This result was consistent with FIG. 1 Continued Branching resolution among earliest animal lineages were obtained from Parfrey et al. (2010), Moroz et al. (2014), andWhelan et al. (2015). The gray branches indicate the protistan phyla and the gray-dashed branches are the fungal phyla. Numbers of CRG genes obtained through our analyses are indicated by asterisks to the left of the columns above, whereas references are provided (below) for data obtained from other studies. Species shown in the figure above are as follows:

Placozoa

Trichoplax adhaerens a ( (Croset et al. 2010). This study clarified the evolutionary history of antennal IRs in the Arthropoda (figs. 3 and 4; see below) and revealed that the antennal IR76b, previously thought to be insect-specific (Croset et al. 2010), originated prior to the divergence of the insects (fig. 4, see below). Intriguingly, antennal IRs in E. affinis showed higher expression in males than in females, the first such finding for an aquatic animal (fig. 5). These male-biased genes also showed signatures of natural selection (see below). The CSPs were present only in the Arthropoda (figs. 1 and 6), as found in another study (Pelosi et al. 2014). Our analysis, which included more pancrustacean taxa and greater sampling of arthropod clades than prior FIG. 2. Phylogeny of the GR gene families from representatives of some major clades within the Arthropoda. Phylogenetic relationships among GR genes of the fruit fly Drosophila melanogaster (Hexapoda, Groups I and II, olive), the waterflea Daphnia pulex (Cladocera, within Branchiopoda, Groups VII-IX, blue), the copepod Eurytemora affinis (Copepoda, Group III, red), the centipede Strigamia maritima (Myriapoda, Group X, magenta), and the black-legged tick Ixodes scapularis (Chelicerata Groups IV-VI, orange). The phylogeny was constructed using maximum likelihood, based on alignments of 1,346 amino acids of GR genes (see Materials and Methods). The numbers at internal branches show bootstrap support values (%) for the maximum-likelihood reconstruction and posterior probabilities (%) for the Bayesian reconstruction. Support values on the major internal branches are shown for values higher than 60%. Groups I-X each represent lineage-specific expansions (of more than three genes) and are supported by > 0.70 posterior probability in the Bayesian reconstruction. The scale bar represents the number of amino acid substitutions per site. See supplementary figure S1, Supplementary Material online, for a more detailed GR amino acid phylogeny using additional taxa. FIG. 3. Phylogeny of the iGluR gene families from nine copepod species and four other invertebrate species. All amino acid sequences except for the copepod sequences were taken from Croset et al. (2010). Information on the copepod sequence assemblies are shown in table 1. The phylogeny was constructed using maximum likelihood (see Materials and Methods) based on sequence alignments of 3,211 amino acids. The numbers to the left of the nodes show the bootstrap support values (%) for neighbor-joining and maximum-likelihood reconstructions, and posterior probabilities Arthropod Chemosensory-Related Genes . doi:10.1093/molbev/msx147 MBE Origin of Gustatory Receptors (GRs)

Our results place the timing of the origin of GRs to the timing of the most recent common ancestor of the Cnideria/Protostomia clade and the phylum Placozoa (Trichoplax adhaerens) (fig. 1). This timing of the origin of the GRs was based on the presence of GR or GR-like genes in the placozoan T. adhaerens, and the absence of GR gene candidates in the outgroup lineage leading to the animal phylum Porifera (sponge Amphimedon queenslandica) and the more distantly related Ctenophora (comb jelly Mnemiopsis leidyi) (fig. 1). In addition, we did not find GR or GR-like genes in the genomes of any protistan or fungal taxa examined, including members of the protistan phylum Choanozoa (choanoflagellate Monosiga brevicollis), the fungal phyla Ascomycota (Saccharomyces cerevisiae) and Basidiomycota (Sporobolomyces roseus), and the protistan phyla Mycetozoa (slime mold, Dictyostelium purpureum), Percolozoa (amoeboflagellate, Naegleria gruberi), and Metamonada (Giardia intestinalis and Trichomonas vaginalis) (fig. 1; see supplementary table S3 for list of genomes sampled, Supplementary Material online). Although our study was based on sampling of taxa (see supplementary table S4, Supplementary Material online) that was more comprehensive than and distinct from those of two prior studies (Robertson 2015;Saina et al. 2015), our results were consistent with the previous findings.

We found three GR-like genes in the placozoan T. adhaerens, consistent with results from two previous studies (Robertson 2015;Saina et al. 2015). We also identified four GRL genes in the genome of the cnidarian Nematostella vectensis. These genes had been identified previously, two by Saina et al. (2015), NvecGrl1 (KP294348) and NvecGrl2 (KP294349) (located in scaffold_86:815817.816695 and scaffold_91:194748. 194002), and two additional genes by Robertson (2015) (jgijNemve1j198670 and jgijNemve1j214946) found in scaf-fold_11 (818242.819090) and scaffold_214 (150068.149415). We also found two GR-like genes in a data set of expressed sequence tags of the cnidarian Acropora millepora (fig. 1). On the other hand, our analyses failed to identify GR candidates in another cnidarian genome, that of the polyp hydra, Hydra magnipapillata, consistent with Saina et al. (2015) and Robertson (2015). Additionally, we found three GR fragments (data not shown) in the draft genome of the velvet worm E. rowelli (Onychophora) and GR genes in the Chelicerata (12 GRs in Centruroides exilicauda and 1 GR in Loxosceles reclusa), the Theocostraca (Pancrustacea, one GR in the purple acorn barnacle Amphibalanus amphitrite), and the Copepoda (Pancrustacea, ten GRs in E. affinis and ten GRs in Tigriopus californicus) (fig. 1). Our findings represent the first discovery of GR genes in the Multicrustacea (within the subphylum Pancrustacea) and add to what has been found for other taxa (see fig. 1).

Lineage-Specific Expansions and Contractions of GRs across the Arthropoda

We observed and confirmed the GR gene family to exhibit high levels of lineage-specific gene expansions across the Arthropoda (Chipman et al. 2014;Gulia-Nuss et al. 2016). Our phylogenetic reconstruction suggests that GR genes most likely experienced gene duplications and differentiation following lineage-splitting events, given that we could not resolve orthologous relationships among GR genes from different clades, even among different hexapod orders (fig. 2 and supplementary fig. S1, Supplementary Material online). Based on high-quality full genome sequence data, we found a general pattern of GR gene family expansions in representative members of most major arthropod clades (i.e., Chelicerata, Myriapoda, and Branchiopoda/Hexapoda), but not for the Multicrustacea (e.g., Copepoda and Amphipoda) (figs. 1 and 2, and supplementary fig. S1, Supplementary Material online). For instance, based on high-quality genome sequence data, the black-legged tick Ixodes scapularis (Chelicerata) (fig. 2, Groups IV-VI, orange branches), the centipede Strigamia maritima (Myriapoda) (fig. 2, Group X, magenta FIG. 3 Continued (%) for the Bayesian analysis, respectively. Support values for the major internal branches are shown only for those higher than 60%. Gene abbreviations for other iGluR members (NMDAR, NMDA receptors; AMPAR, AMPA receptors; KR, Kainate receptors) are adopted from Benton et al. (2009). The NMDAR gene family was used as the outgroup (see Materials and Methods). The inset illustrates a current consensus of the invertebrate phylogeny (Regier et al. 2010;Zwick et al. 2012). Species names were abbreviated according to the following four-letter codes: In contrast to most arthropod groups (previous paragraph), the multicrustaceans showed a relative lack of a GR gene family expansion (fig. 2, Group III, red branches), typically containing a few or no GR genes within species (fig. 2 and supplementary fig. S1, Supplementary Material online). Based on full genome sequences, we found 10 GR genes in the copepod E. affinis, 10 GR genes in the copepod T. californicus, and 0 GR genes in the amphipod Hyalella azteca (figs. 1 and 2). Likewise, based on transcriptome data of additional multicrustacean species (including Thecostraca and Eumalacostraca), which are not fully reliable as GR genes might not be expressed or data sets might be incomplete, we found only one GR gene in the purple acorn barnacle A. amphitrite and two GR genes in the copepod anchor worm Lernaea cyprinacea (supplementary table S5 andfile S1, Supplementary Material online). Determining whether the low numbers of GRs are specific to the multicrustaceans, or are also characteristic of other crustacean lineages (such as the Ostracoda), requires further investigation.

Low Homology among GR Gene Candidates GR sequences share extremely low sequence similarity, even among paralogs within a species and among GR genes of insect species (Robertson et al. 2003;Saina et al. 2015). For example, the amino acid sequence identity among D. melanogaster GR proteins alone drops to as low as 8% (Robertson et al. 2003). Also, there are absolutely no conserved domains among insect GR protein sequences. Because of these characteristics of GRs, homologous relationships are extremely difficult to infer for this protein family. In order to overcome this difficulty, we undertook several analyses (see Materials and Methods). First, we explored the positions of introns, because many intron positions are conserved over extremely long evolutionary time spans (Rogozin et al. 2003). We found that two intron positions were shared even among the highly divergent GR genes of Arthropoda, Cnidarian, and Placozoa (supplementary fig. S2, Supplementary Material online). One The taxa used for this analysis are listed in the Results section (in the section "Origins of Ionotropic Receptors Subfamilies"). The antennal IR genes that show significant differential expression between the sexes in the copepod Eurytemora affinis are underlined in magenta, whereas the IR genes that do not show significant differences are underlined in gray (see fig. 5 and supplementary table S8, Supplementary Material online).

Arthropod Chemosensory-Related Genes . doi:10.1093/molbev/msx147 MBE intron position (indicated by a pink triangle, supplementary fig. S2, Supplementary Material online) was shared only between Nematostella vectensis Grl1 (NvecGrl1) and Trichoplax adhaerens Grl3 (TadhGrl3), but was absent in arthropods. This finding was consistent with the observation that the ancestral introns have generally been lost in arthropods (Rogozin et al. 2003). In addition, sequence homology was supported by codon phases (supplementary fig. S2, Supplementary Material online). For instance, in the supplementary figure S2, Supplementary Material online, the first matching intron position (indicated by an orange arrow and an asterisk) has phase 0 in all GR sequences except for TadhGrl3, which has phase 2. In this position of TadhGrl3, a non-GT-AG intron was found, indicating either a noncanonical intron or more commonly an error.

In our second approach, we analyzed the domain composition of putative N. vectensis and T. adhaerens GR-like genes to computationally infer their protein family

Species Abbreviations

FIG. 6. Phylogeny of CSPs from 6 copepods and 11 other arthropod species. The phylogeny was constructed using maximum likelihood based on sequence alignments of 402 amino acids. Fourteen CSP sequences from six copepods are included (shown in red). In addition to copepods, 81 CSP sequences are included from 11 representative arthropod species. All amino acid sequences except for the copepod sequences are taken from Vieira and Rozas (2011) and Gu et al. (2012). The asterisks indicate branches with at least one of the phylogenetic reconstruction approaches (maximum-likelihood, neighbor-joining phylogenies, or Bayesian) showing bootstrap values or posterior probabilities greater than 70%. The black arrow on the phylogeny points to the node forming a clade within Pancrustacea, composed of a Daphnia pulex CSP, seven insect CSPs, and copepod CSPs. This clade supports a clear homologous relationship between copepod and insect/branchiopod CSPs. This node is supported by a maximum-likelihood bootstrap value of 62% and a Bayesian posterior probability of 0.78. The following representative species were used in this analysis: the fruit fly Drosophila melanogaster (Diptera, olive), the silkworm moth Bombyx mori (Lepidoptera, pink), the red flour beetle Tribolium castaneum (Coleoptera, brown), the honeybee Apis mellifera (Hymenoptera, dark green), the pea aphid Acyrthosiphon pisum (Hemiptera, cyan), the human body louse Pediculus humanus (Phthiraptera, slate-blue), the waterflea Daphnia pulex (Cladocera, blue), the six copepod species (Copepoda, red), the centipede Strigamia maritima (Myriapoda, magenta), and the black-legged tick Ixodes scapularis (Chelicerata, orange). All other arthropod species are shown in black. The tree is midpoint rooted due to the absence of obvious outgroups. The scale bar represents the number of amino acid substitutions per site. S4, Supplementary Material online). We found orthologs of IR93a in the genomes of four chelicerates (I. scapularis, C. exilicauda, Latrodectus hesperus, and L. reclusa), five copepods (Multicrustacea) (Caligus rogercresseyi, T. californicus, Calanus sinicus, Acartia fossae, and E. affinis), and two branchiopods (D. pulex and Artemia franciscana), but absent from the velvet worm Euperipatoides rowelli (phylum Onychophora, immediate outgroup phylum to the Arthropoda) and also absent from all other outgroup phyla.

We are the first to discover orthologs of antennal IR76b occurring in arthropod taxa outside of the insect clade (fig. 4). This antennal IR was previously thought to be insect-specific (Croset et al. 2010). We found orthologs of IR76b in the genomes of the chelicerate bark scorpion C. exilicauda and in two copepod genomes, of E. affinis and L. cyprinacea, and in the genome of the branchipod Daphnia pulex. We confirmed that D. pulex IR304 (EFX75437.1) is the ortholog of IR76b of D. melanogaster. IR76b was absent from the genome of the velvet worm Euperipatoides rowelli (phylum Onychophora) and those of other phyla outside of arthropods.

Of the arthropod-specific antennal IRs, we found that the distributions of IR40a, IR21a, and IR8a were less widespread within the Arthropoda, but still occurring outside of the insect clade (fig. 4). With respect to IR40a, we found an IR40a ortholog (known as SmarIR49) present in the genomes of the myriapod centipede S. maritima (fig. 4). Also, a prior study did find IR40a in a chelicerate (the hunter spider Dysdera silvatica) (Vizueta et al. 2017) (see Discussion). However, our analyses failed to identify IR40a in the Onychophoran velvet worm, chelicerates, and all 14 crustacean species including the branchiopods (D. pulex and A. franciscana) (listed in the supplementary tables S2 and S4, Supplementary Material online).

We found the IR21a genes to be present only in copepods (Caligus rogercresseyi and E. affinis) and hexapods (insects). However, a prior study did find IR21a in present in a chelicerate (the hunter spider D. silvatica) (Vizueta et al. 2017) (see Discussion). Orthologs of IR21a were absent from the genomes of a species of the outgroup phylum Onychophora (velvet worm E. rowelli), four species of Chelicerata (I. scapularis, C. exilicauda, L. hesperus, and L. reclusa), one species of Myriapoda (centipede S. maritima), and two branchiopod species (D. pulex and A. franciscana).

We found IR8a orthologs to be present in the Copepoda (Lepeophtheirus salmonis, C. rogercresseyi, L. cyprinacea, T. californicus, and E. affinis) (supplementary table S7, Supplementary Material online) and also in the Myriapoda (S. maritima) (fig. 4). However, IR8a orthologs were absent from the Chelicerata (D. silvatica, I. scapularis, C. exilicauda, L. hesperus, and L. reclusa) (Vizueta et al. 2017), two branchiopod species (D. pulex and A. franciscana), and the outgroup phylum Onychophora (E. rowelli) (fig. 4).

For the nine copepod species examined (table 1 and supplementary table S2, Supplementary Material online), we identified 33 IRs from seven of the copepod species (fig. 3 and supplementary table S7, Supplementary Material online). Based on sequence similarity and phylogenetic analysis, we were able to classify the 33 copepod IRs into five antennal IR subfamilies (IR25a, IR76b, IR93a, IR8a, and IR21a) and divergent IRs (fig. 3 and supplementary table S7, Supplementary Material online). Interestingly, we observed duplicated IR genes in several copepod species (fig. 3 and supplementary table S7, Supplementary Material online). For instance, two IR8a genes were identified in L. cyprinacea and three in T. californicus, and two IR93a genes were identified each in C. sinicus and E. affinis. These genes (IR93a and IR8a) had not previously been found as duplicated genes in arthropods (Croset et al. 2010). In this study, we were unable to determine the details of the origins of divergent IRs, because divergent IRs showed no one-to-one orthology among Diptera species and exhibited lineage-specific gene duplications (Croset et al. 2010;Chipman et al. 2014) (fig. 4).

Among the 14 crustacean species examined (table 1 and supplementary table S4, Supplementary Material online), we were unable to find IRs in two copepod species Mesocyclops edax and Calanus finmarchicus. The absence of IRs in these two species might have arisen from very low coverage of whole-genome sequencing. Mesocyclops edax (accession numbers SRX246444 and SRX246445) and C. finmarchicus (accession number SRX456026) were sequenced only to $0.22 and $0.55 gigabases ($0.4 and $ 3 million reads) by 454 GS FLX Titanium and the Ion Personal Genome Arthropod Chemosensory-Related Genes . doi:10.1093/molbev/msx147 MBE Machine sequencer, respectively (supplementary table S2, Supplementary Material online). The N50 length of de novo assemblies in M. edax and C. finmarchicus was shorter than that of other copepod assemblies (401 and 347 bp, respectively; table 1). Therefore, the depth of coverage of sequencing might not have been sufficient to detect any IR sequences.

Sex Differences in Expression Levels of IRs, CSPs, and GRs in the Copepoda

In order to compare expression levels of the GR, IR, and CSP genes between the sexes in the copepod E. affinis, we mapped Illumina RNA-Seq reads to each gene and normalized for sequencing depth and gene length by presenting them in RPKM (reads per kilobase per million mapped reads) values. Most notably, three of the antennal IR genes (EaffIR8a, EaffIR25a, and EaffIR93-1) showed significantly greater expression in the male RNA-Seq samples, relative to the female samples (P < 0.001) (fig. 5 and supplementary table S8, Supplementary Material online). This was the first study to discover IRs with male-biased expression in an aquatic animal.

In contrast to the significant sex-specific differences in expression of the three antennal IR genes, we found no sexspecific difference in five representative housekeeping genes of E. affinis (Cyclophilin-33, Actin 42A, Heat shock protein 83, Glyceraldehyde 3 phosphate dehydrogenase 1, and Ribosomal protein L32) (supplementary table S9, Supplementary Material online). The levels of expression were similar between the sexes for these housekeeping genes, in contrast to the large sex differences in expression we found for three antennal IR genes (EaffIR8a, EaffIR25a, and EaffIR93-1) and one CSP gene (EaffCSP1). Although we had only two replicate samples for each sex, we included $220 individual copepods per replicate, and found very low variance between the replicates for both the antennal IRs and CSP gene, as well as for the five housekeeping genes (see standard deviations in the supplementary tables S8 and S9, Supplementary Material online).

In contrast to the male-biased expression of some antennal IR genes, the expression of the E. affinis CSP gene EaffCSP1 was $30-fold higher in female RNA-Seq samples (in RPKM reads) than in male samples (P < 0.0001 by edgeR and Prob. ¼ 0.95% by NOISeq; fig. 5 and supplementary table S8, Supplementary Material online) (see Discussion). The other E. affinis CSP genes showed slightly higher, but not significant (P > 0.1643), expression levels in male than in female samples (fig. 5, supplementary table S8, Supplementary Material online).

Six E. affinis GR genes showed no difference in expression between the sexes (fig. 5 and supplementary table S8, Supplementary Material online). In the contrast to relatively high expression levels in antennal IRs, we found that crustacean GRs were generally expressed at very low levels, except for the copepod T. californicus GR7 (TcalGR7) and the barnacle A. amphitrite GR1 (AampGR1) (supplementary tables S5 and S8, Supplementary Material online). In E. affinis, the RPKM values of all six E. affinis GRs from all four samples were lower than 1 (fig. 5 and supplementary table S8, Supplementary Material online).

Signatures of Selection in Antennal IR Genes

When we tested for signatures of natural selection in antennal IR genes, we found significantly stronger signatures of purifying selection in the IR genes showing elevated expression in E. affinis males, relative to IR genes that showed no sex differences in expression (see fig. 5, supplementary fig. S3, Supplementary Material online). Based on expression levels of the antennal IR genes, we classified them into two groups (fig. 5), namely "male-biased expression IRs" (IR25a, IR93a-1, The number of contigs (> 300 bp). The NCBI accession numbers and sequencing platforms were summarized in the supplementary tables S1 and S2, Supplementary Material online. The transcriptomes and the genomes were assembled using the software package Trinity and Velvet, respectively (more details in Materials and Methods).

Eyun et al. . doi:10.1093/molbev/msx147 MBE and IR8a), which displayed significantly elevated expression in males, and "unbiased expression IRs" (IR76b and IR21a), which showed no difference in expression between the sexes.

To compare patterns of molecular evolution in the two sets of IR genes, we used the branch model in codeml in the software package PAML (Yang 2007). All the male-biased expressed IR genes (IR25a, IR93a-1, and IR8a) showed significantly stronger signatures of purifying selection relative to the unbiased IR genes (IR76b and IR21a) (supplementary fig. S3, Supplementary Material online). When comparing the average x (the ratio of nonsynonymous to synonymous substitutions, x or d N /d S ) between the two groups, they both showed signatures of purifying selection (d N /d S < 1) (supplementary fig. S3, Supplementary Material online). However, the x (d N /d S ) of unbiased IRs (x ¼ 0.0249) was 1.9 times higher than that of the male-biased IRs (x ¼ 0.0131), and the difference was significant (P ¼ 0.0387; supplementary fig. S3, Supplementary Material online). This lower value of x (d N /d S ) in male-biased IRs indicated that purifying selection has acted more strongly in these genes.

Chemosensory Proteins (CSPs), a Class of CRGs Unique to the Arthropoda Our results indicated that CSPs are an arthropod-specific gene family that emerged after the divergence between the phyla Arthropoda and Onychophora (698.5 Ma) (fig. 1). CSPs were found in all arthropod taxa we examined (fig. 1 and6), except for the transcriptome assembly of the barnacle A. amphitrite (Thecostraca) and the genome sequence of the brown recluse spider L. reclusa (Arachnida). In contrast, CSPs were absent in the draft genome of the velvet worm E. rowelli, a member of the outgroup phylum Onychophora, and all other nonarthropod genomes (fig. 1).

CSP gene numbers tended to be low within arthropod genomes, relative to other arthropod CRG families (fig. 1). Within crustaceans, we identified 14 CSPs in six copepod species and seven CSPs in three other crustacean species (Hyalella azteca, Penaeus monodon, and Artemia franciscana) (fig. 1 and supplementary table S10, Supplementary Material online). Our phylogenetic analyses of CSPs showed that subgroups could not be resolved for most of the major nodes due to the low bootstrap values (below 50%) (fig. 6). This result was reflected in the low levels of sequence similarity among all arthropod CSPs (as low as 15.9% among four Drosophila CSP proteins) and the short sequence lengths of CSPs (average length of $127 amino acid residues). Our phylogenetic analysis revealed that CSPs from all six copepod species formed a well-supported monophyletic clade (fig. 6, red branches). The copepod CSPs formed a larger clade with a D. pulex CSP and seven insect CSPs (indicated by the arrow in fig. 6, and supported by Bayesian posterior probability of 0.78 and maximum-likelihood bootstrap value of 62%), supporting homology between them.

We found that all arthropod CSPs we examined, including those of D. pulex (Cladocera), S. maritima (Myriapoda), and three chelicerates (I. scapularis, C. exilicauda, and L. hesperus), contained a highly conserved four cysteine motif that is found in insects (Forêt et al. 2007;Liu et al. 2012). Interestingly, copepod CSPs contained this motif (CX 6-7 CX 16-19 CX 3-4 C) and two additional cysteines (supplementary fig. S4, Supplementary Material online). Although this motif in copepods was conserved, it was slightly less conserved than that of insects (CX 6 CX 6-18 CX 2 C) (supplementary fig. S4, Supplementary Material online).

Protein Structural Homology-Modeling and a Potential Conserved Role of IRs and CSPs

To understand the spatial distribution of the ligand-binding amino acid residues, we performed homology-modeling of copepod IR25a and CSP proteins (see Materials and Methods). In IR25a, three ligand-binding amino acid residues (corresponding to the positions 489R, 654A, and 739D in T. californicus IR25a) were proposed (Benton et al. 2009) (supplementary fig. S5, Supplementary Material online). These three potential ligand-binding amino acid residues were located in the extracellular domain, which might play critical roles in ligand recognition (supplementary fig. S6, Supplementary Material online). Furthermore, the potential ligand-binding amino acid residues that we found are identical to those of D. melanogaster (DmelIR25a, ADU79032.1), the waterflea D. pulex (DpulIR25a, EFX86214), and the mollusc Aplysia californica (AcalIR25a, XP_005102425.1) (supplementary fig. S5, Supplementary Material online).

The predicted 3D structural model we constructed for the copepod E. affinis CSP2 protein (EaffCSP2) comprised six ahelices and two pairs of disulphide bridges (supplementary fig. S7, Supplementary Material online). This 3D model was concordant with the X-ray structure of the CSP protein from the cabbage moth (Mamestra brassicae) MbraCSPA6, which appears in a globular shape composed of six amphiphatic ahelices that surround an internal hydrophobic binding pocket (Campanacci et al. 2003). Also, we found that all copepod CSPs, except for one incomplete CSP (LsalCSP: 76 aa), possess the typical six a-helices from the sequence-based secondary structure prediction, but do not have ancient 5-helical structure in arthropods (supplementary table S11, Supplementary Material online) (Kulmuni and Havukainen 2013). Based on the presence of a conserved four-cysteine motif and protein structure similarity, copepod CSPs might have similar functions to those of insects (fig. 6

Discussion

This study provides the most comprehensive analysis to date of CRG family evolution of the Arthropoda, as well as of some outgroup animal phyla. Our phylogenetic and molecular evolutionary analyses offer a more lucid and comprehensive view of the evolutionary histories of the arthropod CRGs by including more in-depth sampling of arthropod and related taxa. Our study included all the major subphyla within the Arthropoda and representatives from a range of metazoan and protistan phyla. Most notably, this study was the first to include multiple crustacean genomes outside of the branchiopod/insect clade, allowing more detailed inference of evolutionary patterns proximate to the insects. Thus, this Arthropod Chemosensory-Related Genes . doi:10.1093/molbev/msx147 MBE more comprehensive analysis provided the strongest case thus far to infer that the ORs and OBPs are unique to insects (Hexapoda) and that CSPs are specific to the Arthropoda, and clarified the evolutionary histories of antennal IR subfamilies (see below). Moreover, we gained insights into general principles governing patterns of multigene family evolution of the CRGs (see below on "Birth-and-Death Model of Multigene Family Evolution").

Gustatory Receptors (GRs) Are the Most Ancient of the Arthropod CRG Families Our study confirmed that GRs arose early in the course of metazoan evolution and are the most ancient of the CRGs found in arthropods. Our results revealed the presence of GRs in the metazoan phyla Placozoa and Cnidaria, but not in the phyla Porifera and Ctenophora, or in fungal and protistan phyla (fig. 1). Given that recent phylogenetic studies indicate that lineages leading to the phyla Ctenophora and Porifera branched earlier during metazoan evolution (fig. 1) (Moroz et al. 2014;Whelan et al. 2015), we can infer that the GRs evolved after the emergence of metazoans (850-550 Ma) and during the early stages of animal evolution (fig. 1).

Our finding that places the origin of GRs at the early stages of animal evolution was consistent with results from recent studies (Robertson 2015;Saina et al. 2015). Our results, along with those of Saina et al. (2015) and Robertson (2015), placed the origin of GRs earlier than prior studies, which had placed the origin of GRs dating back either to the Cnideria (Nordstro ¨m et al. 2011) or to the Ecdysozoa (Robertson et al. 2003;Croset et al. 2010). Our study sampled multiple chelicerates and crustaceans, an immediate outgroup phylum Onychophora, and other basally-branching phyla (including single-cell eukaryotes), making the placement of the evolutionary origins of CRG gene families more certain. This study included five protists and two fungi (supplementary table S3, Supplementary Material online), whereas Robertson (2015) examined two protist species (choanoflagellates Monosiga brevicollis and Salpingoeca rosetta). Our sampling of protists and fungi was similar to that of Saina et al. (2015), but our study included additional invertebrate animal phyla (fig. 1 and supplementary table S4, Supplementary Material online).

The gustatory roles of GRs in noninsect arthropod taxa are poorly understood and require functional studies. In insect models, GRs are known to be typically expressed at low levels in only a few gustatory or olfactory sensory neurons (Wang et al. 2004;Thorne and Amrein 2008). Thus, the low expression levels of E. affinis GRs we found (fig. 5) were consistent with the low levels of expression found in insect GRs. Some Drosophila GR genes are known to be involved in proprioception, hygroreception, light sensing, and other sensory modalities (Thorne and Amrein 2008). The functional roles of GR (or GR-like) genes from the placozoan Trichoplax and cnidarian Nematostella as GRs are still inconclusive, as they have not been confirmed to have obvious chemosensory roles. Our computational protein family classifications strongly support the inference that Trichoplax and Nematostella GRs are homologous to those of arthropods (supplementary table S6, Supplementary Material online). Interestingly, the cnidarian homolog to the insect GR gene, NvecGrl1 (KP294348) in Nematostella, has been found to play a role in early developmental body patterning, rather than in external chemosensation (Saina et al. 2015).

Evolutionary Origins of Ionotropic Receptors (IRs)

The IRs had previously been hypothesized to be most ancient of the arthropod CRGs, dating back to the Protostomia, based on their presence in arthropods, nematodes, and molluscs, but absence in the basally branching metazoan phyla, such as Cnidaria, Placozoa, and Porifera (Croset et al. 2010). Consistent with Croset et al. (2010), our analysis also found IRs present in the protostomes, including arthropods, an onychophoran (velvet worm Euperipatoides rowelli) and a mollusc (California sea slug Aplysia californica), and absent in the basally branching metazoan phyla outside of the protostomes (figs. 1, 3, and 4). In contrast to Croset et al.'s (2010) postulation, however, the evolutionary history of IRs is considerably more recent than that of GRs, given that GRs have since been found in several basally branching metazoan phyla (see previous section; fig. 1).

Until recently, only the antennal IRs IR25a and IR93a were thought to occur outside of the insect clade, whereas four others (i.e., IR40a, IR21a, IR76b, and IR8a) were considered to be insect specific (Croset et al. 2010). In addition, more recent studies also have found IR25a and IR93a in the Caribbean hermit crab Coenobita clypeatus (Pancrustacea) (Groh et al. 2014;Groh-Lunow et al. 2015) and in the spider mite Tetranychus urticae (Chelicerata) (Ngoc et al. 2016), and IR93a in the tick Ixodes scapularis (Chelicerata) (Gulia-Nuss et al. 2016). However, recent studies have also uncovered three of the "insect-specific" antennal IRs outside the insect clade, namely IR8a and IR40a in the centipede Strigamia maritima (Myriapoda) (Chipman et al. 2014) and IR21a and IR40a in the hunter spider D. silvatica (Vizueta et al. 2017). With the inclusion of our study we now know that none of the six arthropod antennal IRs are unique to insects (see next paragraph).

Our more comprehensive analysis, including 14 crustacean taxa, revealed patterns of gains and losses of antennal IR subfamilies across the arthropod clades (fig. 4). This study discovered an additional antennal IR gene subfamily occurring outside of the insect (Hexapoda) clade, namely IR76b, which previously had been considered insect-specific (Croset et al. 2010). Our finding of IR76b in the genomes of the chelicerate bark scorpion C. exilicauda and in two copepods, E. affinis and L. cyprinacea, but absent in the velvet worm or some other phyla outside of arthropods, revealed this antennal IR to be more widespread within the Arthropoda than previously thought (fig. 4).

Our results suggest that IR40a emerged in the common ancestor of arthropods, but was subsequently lost from the genomes of some chelicerates (C. exilicauda and L. reclusa) and from all 14 crustacean species we examined, including the multicrustaceans and branchiopods (fig. 4). The insect clade is nested within pancrustaceans, yet they do possess IR40a (fig. 4). Likewise, our finding of IR21a present in copepods and insects and the prior finding of this IR subfamily in a Eyun et al. . doi:10.1093/molbev/msx147 MBE spider (Chelicerata) (Vizueta et al. 2017) suggest that IR21a arose in the common ancestor of arthropods, but was lost in myriapods and branchiopods (fig. 4), although sampling in the myriapods is scant. However, the absence of this gene in two species of branchiopods (D. pulex and A. franciscana) suggests a loss in this clade.

Likewise, we also found IR8a orthologs occurring outside the insect clade, in the Copepoda (see Results; fig. 4), and previous studies found this gene present in the genome of the centipede (Myriapoda) Strigamia maritima (known as SmarIR8a) (Chipman et al. 2014). This antennal IR8a gene was absent in the Onychophora (E. rowelli), five chelicerate species including the spider (Chelicerata) D. silvatica (Vizueta et al. 2017), and two branchiopod species (see Results). Our results suggest that IR8a arose in the myriapod and pancrustacean lineages after their split from the chelicerates, but was subsequently lost in the branchiopods (fig. 4).

IRs of arthropods have been found to be associated with a variety of sensory functions, including taste, olfaction, thermosensation, and hygrosensation (Benton et al. 2009;Croset et al. 2010;Abuin et al. 2011;Zhang et al. 2013;Stewart et al. 2015;Knecht et al. 2016). For example, Knecht et al. (2016) demonstrated that IR93a/IR25a mediates thermosensation and hygrosensation and IR21a/IR25a responds to cool temperatures. In addition, IR76b was found to be expressed in gustatory neurons of D. melanogaster, implicating this IR group in taste detection (Zhang et al. 2013). Antennal IR25a and IR93 have been found to be expressed in the olfactory neurons of antennules of the terrestrial hermit crab Coenobita clypeatus (Pancrustacea) (Groh-Lunow et al. 2015). For the spider Dysdera silvatica (Chelicerata), a homolog of the antennal IR25a/IR8a protein family was found to be overexpressed in the first pair of legs and the palps, which are thought to be olfactory appendages in spiders (Vizueta et al. 2017). These results suggest that some IRs mediate olfactory signaling in a wide range of arthropods. Furthermore, the function of the antennal IR84a might be related to male courtship behavior in D. melanogaster (Grosjean et al. 2011). However, elucidating the functional roles of IRs is still in the very early stages of discovery, and much more remains to be discovered.

Ionotropic Receptors (IRs) Mediating Copepod Chemodetection during Mating?

Examining differences in CRG gene expression profiles between males and females could provide clues regarding the roles of chemical perception in mate-searching. However, few studies have elucidated the molecular mechanisms linking specific genes to specific sexual behaviors (Kopp et al. 2008;Zhou et al. 2009). In D. melanogaster, the expression of OR, GR, and OBP genes is more extensive in males than in females, but other receptors (4 GRs and 12 ORs) show altered expression in females after mating (Zhou et al. 2009). In this study, several intriguing patterns emerged regarding the expression and incidence of the antennal IRs in copepods, suggestive of a role in mating. In particular, we found that three antennal IR genes (IR8a, IR25a, IR93a-1) showed significantly greater expression in males of the copepod E. affinis, relative to females (fig. 5 and supplementary table S8, Supplementary Material online). In contrast, expression levels of all six GR genes in E. affinis showed no difference between the sexes (supplementary table S8, Supplementary Material online). Our findings are notable in being the first to find sex-specific differences in expression of CRGs in an aquatic organism.

Interestingly, two of the antennal IR genes that exhibited male-biased expression (IR8a and IR93a) have also experienced gene duplications in several copepod species (supplementary table S7, Supplementary Material online). These gene duplicates of male-biased antennal IRs might serve to increase expression of antennal IR proteins even further. The duplications of IR8a and IR93a we found in copepods are notable, given that the IR93a and IR8a subfamilies have generally not been found as duplicated genes in arthropods (Croset et al. 2010).

Most notably, the same antennal IR genes showing malebiased expression (fig. 5, IR8a, IR25a, IR93a-1) also exhibited stronger purifying selection than the unbiased IR genes (supplementary fig. S3, Supplementary Material online). This result suggests that the antennal IR genes showing elevated expression in males are subjected to greater functional evolutionary constraints. Such functional conservation is consistent with our protein structure model of the copepod T. californicus IR25a, where the potential ligand-binding amino acid residues were found to be identical to those of D. melanogaster, D. pulex, and the mollusc A. californica (see Results; supplementary fig. S6, Supplementary Material online). This result suggests that ligand-binding functions of IR25a are conserved across protostomian species (Benton et al. 2009;Liang et al. 2016). Whether these conserved ligand-binding regions serve an important role in male behavior or other functions would be worth investigating.

Our results, as well as results from other studies, suggest that antennal IRs might have functions related to the chemically mediated mate-recognition behavior observed in male copepods. For instance, in the fruit fly D. melanogaster, mutational knockdown of the antennal IR84a markedly reduces male courtship behavior (Grosjean et al. 2011). In three Drosophila sibling species, IR genes are differentially expressed among species and between the sexes (Shiao et al. 2015). IR76a shows significantly higher expression in female D. simulans, but no significant difference between the sexes in D. melanogaster and D. sechellia (Shiao et al. 2015). Also, IR25a shows slightly greater expression in the females than males in all three Drosophila sibling species (supplementary table S4 in Shiao et al. 2015). This result differs from ours, as we found no antennal IR gene where female expression was significantly higher than that of males (fig. 5). Interestingly, in the hover fly Scaeva pyrastri, only one IR gene (SpyrIR84a) exhibits significant sex differences in expression, with male-biased expression in the antennae (Li et al. 2016).

The male-biased elevated antennal IR expression we found (fig. 5) might possibly be localized in the antennal tissue, and might be involved in functions related to mating. We speculate that the expression of these antennal IRs is localized in the antennae based on anatomical studies of this species, where chemosensory palps are localized heavily in the Arthropod Chemosensory-Related Genes . doi:10.1093/molbev/msx147 MBE antennae, especially of the male copepod (Katona 1973;Griffiths and Frost 1976;Snell and Morris 1993). IR25a, which we found to be highly expressed in males (fig. 5), was also found localized in olfactory organs of a hermit crab (Groh-Lunow et al. 2015) and a spider (Vizueta et al. 2017).

In copepods, studies have shown evidence of chemosensation by males during initial mating, such as the detection and tracking of females from a distance (Gauld 1957;Katona 1973;Friedman and Strickler 1975;Snell and Morris 1993;Doall et al. 1998;Heuch et al. 2007;Yen et al. 2011). During mating, the male copepod grips the female with his first antenna (see supplementary movies S1 and S2, Supplementary Material online) (Katona 1973;Snell and Morris 1993), consistent with the potential importance of antennal IRs in mating. The possible roles of antennal IRs in mediating the mating behavior of males might have imposed functional evolutionary constraints, possibly imposed by coevolution between female ligand/pheromone and male IRs. Such coevolutionary constraints might be reflected in the signatures of purifying selection we found in the male-biased antennal IR genes (supplementary fig. S3, Supplementary Material online). Elucidating the actual functions of these male-biased antennal IRs, and whether they are localized in the copepod antennae and are involved in mating, requires further investigation.

Chemosensory Proteins (CSPs) Occur in the Arthropoda Only

Our analysis revealed that CSPs are unique to the phylum Arthropoda, and are present in all the major arthropod lineages, including in the chelicerates, myriapods, and pancrustaceans (crustaceans and insects) (fig. 1). Our results were consistent with a prior study that found CSPs only in arthropods (Vieira and Rozas 2011;Pelosi et al. 2014). However, our study differed from this prior study in that we included 14 crustacean taxa beyond the Branchiopoda/Hexapoda (Allotriocarida) clade (table 1 and supplementary tables S2 and S4, Supplementary Material online), and also incorporated many additional invertebrate animal phyla (fig. 1; supplementary table S4, Supplementary Material online), making the conclusion more robust. Given that we did find CSP genes in all the major crustacean taxa examined, the widespread occurrence of CSPs across the Arthropoda is more strongly substantiated. Also, the lack of CSPs in the other invertebrate phyla (fig. 1) strengthened the conclusion that CSPs occur in arthropods only.

Insect CSP genes have been linked to a variety of feeding, mating, and other behaviors (Gu et al. 2012;Liu et al. 2012;Pelosi et al. 2014). For example, in the Oriental migratory locust Locusta migratoria manilensis, the CSP gene LmigCSP91 was highly expressed only in adult male testicles and adult female accessory glands, but was absent in male accessory glands and ovaries, as well as in sensory organs (Zhou et al. 2013). In the tsetse fly, Glossina morsitans morsitans, GmmCSP2 was proposed to be associated with female host-seeking behavior, because this gene was mainly expressed in the female antennae and their transcript levels increased markedly after a blood meal (Liu et al. 2012). In addition, in the alfalfa plant bug, Adelphocoris lineolatus, three antennae-biased CSPs might mediate host plant recognition (Gu et al. 2012). These genes showed higher expression levels in the antennae than in the head, legs, and wings.

Interestingly, the copepod E. affinis CSP gene EaffCSP1 showed significantly higher expression in female RNA-Seq samples relative to male samples (P < 0.0001 by edgeR and Prob. ¼ 0.95% by NOISeq) (fig. 5). Based on this pattern, we speculate that the EaffCSP1 gene might be involved in mate recognition. It would be worth exploring the functions of this gene in future studies, especially with respect to its role in mating behavior and interaction with sex pheromone compounds.

Odorant Receptors (ORs) and Odorant Binding Proteins (OBPs) Are in Insects Only

We found ORs and OBPs present only in the insects (Hexapoda), and completely lacking in all other arthropod taxa, including the nonhexapod pancrustaceans, chelicerates and myriapods (fig. 1). Our study more conclusively revealed that ORs and OBPs are specific to the insects alone (fig. 1), given that our analysis was the first to examine genomes of multiple crustacean taxa outside of the Branchiopoda/ Hexapoda clade, including the genomes and transcriptomes of 13 crustacean species (supplementary table S4, Supplementary Material online). This inclusion of multiple crustacean taxa was critically important for discerning the uniqueness of ORs and OBPs to the insects, because the insects are nested within the pancrustacean clade (von Reumont et al. 2012;Oakley et al. 2013;Sasaki et al. 2013). With our more intensive sampling within the Arthropoda and of outgroup phyla (fig. 1 and supplementary tables S2-S4, Supplementary Material online), our study showed more definitively than prior studies that the ORs and OBPs are present in the Hexapoda alone. In addition, our results indicated that OR genes are not universally associated with terrestrial invasions by arthropods, given the absence of these genes in the terrestrial chelicerates and myriapods (fig. 1).

Although, Vizueta et al. (2017) found two novel candidate chemosensory gene families in the hunter spider D. silvatica, one of them being distantly related to the canonical insect OBPs (i.e., three copies of OBP-like proteins) and the other encoding 12 copies (not related to OBPs). Some of these genes are expressed in the putative chemosensory appendages of these spiders, and show typical characteristics of secreted chemosensory proteins, such as a conserved cysteine pattern and the presence of a clear signal peptide. However, the specific functional roles of these putative chemosensory related genes are unknown, and further studies are required to determine whether they do function similarly as insect OBPs.

Our more comprehensive analysis is consistent with, and considerably extends, results from prior studies, which did not include the crustaceans beyond the branchiopod/hexapod clade (Peñalva-Arana et al. 2009;Missbach et al. 2014). Our analysis was consistent with the hypothesis, first stated by Robertson et al. (2003), that the ORs arose after the emergence of the Hexapoda from within the Pancrustacea ($470 Ma), and expanded greatly in the hexapod lineage. Prior studies found that the genome of the water flea D. pulex Eyun et al. . doi:10.1093/molbev/msx147 MBE (Branchiopoda) and the transcriptome of the Caribbean hermit crab Coenobita clypeatus (Pancrustacea, Malacostraca) lacked ORs and OBPs (Peñalva-Arana et al. 2009;Vieira and Rozas 2011;Groh et al. 2014). Recent studies also found ORs and OBPs to be lacking in the genomes of several species from the arthropod subphyla Chelicerata and Myriapoda, such as the myriapod centipede (S. maritima) (Chipman et al. 2014) and three chelicerate spider mites (Tetranychus urticae, Tetranychus evansi, and Tetranychus lintearius) (Phuong 2013). We confirmed the absence of ORs and OBPs in four additional chelicerates (black-legged tick I. scapularis, bark scorpion C. exilicauda, black widow spider L. hesperus, and brown recluse spider L. reclusa).

Existing data from the literature indicate that OBPs evolved earlier in the evolution of the insects, whereas ORs are thought to have emerged long after the establishment of a terrestrial lifestyle, with their appearance correlated with the emergence of winged insects (Missbach et al. 2014). For instance, recent studies focusing on basally branching insects, such as members of the orders Archaeognatha, Zygentoma, and Phasmatodea, demonstrate that the jumping bristletail Lepismachilis y-signata (Archaeognatha, wingless insects) possesses OBPs, but does not have ORs (Missbach et al. 2014(Missbach et al. , 2015)). In contrast, OR repertoires (including Orco) were found in the firebrat Thermobia domestica (Zygentoma) and the leaf insect Phyllium siccifolium (Phasmatodea) that do have wings, indicating that they arose after the emergence of wings (Missbach et al. 2014). However, these studies examined transcriptome sequences of insects, and more thorough analyses of comprehensive genome data would clarify the evolutionary history of the emergence of OR and OBP gene families within the insects.

Although our study provided much added support for the exclusivity of ORs and OBPs to the insects (Hexapoda), one taxonomic group remains to be examined. No study has yet examined the other member of the Allotriocarida clade, the class Remipedia, which are also crustaceans closely related to the Hexapoda. Thus, we cannot yet conclude definitively that ORs and OBPs are exclusive to the Hexapoda (fig. 1).

The absence of ORs and OBPs in noninsect arthropod clades raises the interesting question of what chemosensing system the noninsect terrestrial arthropods (i.e., Chelicerata, Myriapoda) are using to detect volatile ligands in air. Insect ORs respond to various volatile odorants and pheromonal molecules that diffuse in air (Hallem et al. 2004). So then have the terrestrial chelicerates and myriapods co-opted an existing system that has been described to perform this function? Or are they using some other gene family that has not yet been discovered? The newly discovered CCPs and OBP-like genes found in a spider (chelicerate) (Vizueta et al. 2017) might fulfill this role in terrestrial habitats, though the functions of these genes are not yet known. The chemosensing systems of the noninsect terrestrial arthropods would be worth exploring.

Most Arthropod CRG Families Follow the Birth-and-Death Model of Multigene Family Evolution

The patterns we found of frequent gene losses and gains by the GR gene families and the lack of orthologous GR genes among different arthropod orders (supplementary fig. S1, Supplementary Material online) suggest that these genes have been evolving according to the "birth-and-death" model of multigene family evolution (Nei and Hughes 1992;S anchez-Gracia et al. 2011). A few prior studies have also found patterns of CRG evolution consistent with this model (see below) (Vieira et al. 2007;S anchez-Gracia et al. 2009S anchez-Gracia et al. , 2011)). Under this model, new genes are created by gene duplication. Then, after the divergence of major lineages, some of the genes are retained in the genome for a long time as functional genes, whereas others become nonfunctional through deleterious mutations or are eliminated from the genome (Nei and Rooney 2005). This model was first proposed as an alternative to the previously well-accepted model of concerted evolution (Nei and Rooney 2005), in order to explain the high degree of polymorphism found at MHC loci in mammals (Nei and Hughes 1992).

One line of support for the "birth-and-death" model of gene family evolution would be that different lineages have undergone unique gene family expansions or contractions. We see such patterns of lineage-specific expansions or contractions in multiple arthropod lineages (fig. 1 and2 and supplementary fig. S1, Supplementary Material online). For instance, most of the insect, copepod, and chelicerate GRs generally formed distinct clades without clear orthology to one another (Groups I-IX in fig. 2, >0.73 posterior probability in the Bayesian phylogeny). Likewise, there was an expansion of 61 GRs in the myriapod (S. maritima) genome, forming a distinct monophyletic clade (Group X in fig. 2 and supplementary fig. S1, Supplementary Material online) (Chipman et al. 2014). Within the Hymenoptera, the wasp Nasonia vitripennis genome had an expansion of 58 GRs (Robertson et al. 2010). In contrast, the honeybee (Apis mellifera) genome had only ten GRs (Robertson and Wanner 2006), suggesting a lineage-specific GR gene family contraction in this lineage (supplementary fig. S1, Supplementary Material online).

Also in support of the "birth-and-death" model is the fact that we observed large genetic divergences between GR gene families in closely related clades (supplementary fig. S1, Supplementary Material online). For the GR proteins within the purported Allotriocarida (Branchiopoda/Hexapoda) clade, the closest sequence similarity between D. melanogaster and D. pulex was 43.4% (by local alignment between DmelGR64b and DpulGR56). A consequence of the large divergences between the clades is the fact that different orders of arthropods lack truly orthologous GR genes (fig. 2 and supplementary fig. S1, Supplementary Material online). For example, D. melanogaster and the silkworm moth Bombyx mori represent two closely related orders (see the inset of supplementary fig. S1, Supplementary Material online). However, GR orthologs cannot be identified between the two species, except for the carbon dioxide receptors and sugar receptors (Wanner and Robertson 2008), which are relatively conserved within insects (supplementary fig. S1, Supplementary Material online).

We also observed patterns consistent with the birth-anddeath model in other CRG members. For instance, divergent IRs displayed patterns consistent with this model, such as large genetic divergences and no orthology between divergent IRs of D. melanogaster and B. mori (Croset et al. 2010). Additionally, insect ORs formed a large and highly divergent Arthropod Chemosensory-Related Genes . doi:10.1093/molbev/msx147 MBE gene family with no close orthologs, such as between ORs of D. melanogaster and B. mori, except for Orco (Hansson and Stensmyr 2011). Patterns consistent with the birth-and-death model have also been reported for CSPs and OBPs (Vieira et al. 2007;S anchez-Gracia et al. 2009;Hansson and Stensmyr 2011).

In contrast, antennal IRs are quite conserved in sequence within the Arthropoda (fig. 3), and did not conform to the birth-and-death model. The antennal IRs showed clear orthologous relationships even among distantly related species, such as between D. melanogaster and copepod species (fig. 3). Many of the antennal IRs have generally remained as single-copy genes (fig. 3). These genes have remained highly conserved and retained homologous structures across all protostomian species (fig. 4 and supplementary figs. S5 and S6, Supplementary Material online).

Overall, the evolutionary patterns we observed here are consistent with the birth-and-death model being a major mechanism of molecular evolution in all the CRG families except for the antennal IRs. Thus, we speculate that such a birth-and-death process of CRG evolution might reflect a common process of rapid diversification associated with adaptation to diverse environments (Hayden et al. 2010), resulting in high rates of gene family gains and losses.

Conclusions and Future Studies

Elucidating patterns of CRG family evolution provides an important step toward understanding the interactions between organisms and their environments, as CRGs are fundamental to sensing the environment and adapting to various ecological niches (Hayden et al. 2010). For instance, the fact that most CRG families appear to be evolving under the birth-anddeath model, with rapid species-specific gene duplications, suggests rapid species-specific adaptations to their environments. Several of our results are suggestive of some CRGs found in this study playing functional roles in mating. For instance, this study is the first to find sex differences in expression of CRGs in an aquatic organism. Most notably, we found that three of the antennal IRs were highly expressed in male copepods of the species Eurytemora affinis, and that these male-biased antennal IRs showed significantly stronger signatures of purifying selection than nonsex-biased IRs. It would be worth exploring the role of antennal IRs in mating behavior, and how molecular evolutionary changes in antennal IR proteins might correspond to changes in mating behavior. In addition, prior studies on CRGs' roles in mating have focused predominantly on terrestrial organisms. As the physics of chemosensing and the diffusion of ligands would differ between water and air, the role of CRGs in mating and other functions would be worth exploring in the aquatic realm.

Overall, our results have generated several intriguing hypotheses that should be further explored with functional studies. A comparative functional evolutionary approach that included diverse arthropod taxa, especially beyond the insect clade and from multiple habitat types, would provide key insights into the evolutionary history and functions of CRG families.

High Throughput Genome and Transcriptome Sequencing and Genome Assembly

The copepod E. affinis was one of 30 arthropod species sequenced as a part of the pilot project for the i5K Arthropod Genomes Project at the Baylor College of Medicine Human Genome Sequencing Center (https://www.hgsc. bcm.edu/arthropods; last accessed May 4, 2017). Supplementary table S1, Supplementary Material online, provides details of all sequences generated for the copepod E. affinis and their National Center for Biotechnology Information (NCBI) accession numbers, as well as assembly and annotation statistics and their NCBI accessions. The primary NCBI BioProject for the genome sequencing, annotation, and assembly of E. affinis is PRJNA203087.

An enhanced Illumina-ALLPATHS-LG sequencing and assembly strategy enabled multiple species to be approached in parallel at reduced costs. For E. affinis, four libraries of nominal insert sizes 180 bp, 500 bp, 3 kb, and 8 kb at genome coverages of 28.1Â, 21.2Â, 16.6Â and 9.0Â, respectively, for a total of 75Â genome coverage were sequenced. Libraries were prepared using standard methods as described previously (Anstead et al. 2015). Sequencing was performed on Illumina HiSeq2000 platforms generating 100-bp paired-end reads. These raw sequences have been deposited in the NCBI SRA, accessions are shown in the supplementary table S1, Supplementary Material online, BioSample ID: SAMN02302763. Additionally, three RNAseq libraries were prepared from separated samples of adult males, females, and mixed sex juvenile stages and sequenced using standard techniques (Anstead et al. 2015). These Eyun et al. . doi:10.1093/molbev/msx147 MBE transcriptome sequences were used to support automated and manual annotation.

The genomic sequence of the copepod E. affinis was assembled using ALLPATHS-LG (release 3-35218) (Gnerre et al. 2011) and further scaffolded and gap-filled using in-house tools Atlas-Link (v.1.0) and Atlas gap-fill (ver.2.2) (https:// www.hgsc.bcm.edu/software; last accessed May 4, 2017). This yielded an assembly of size 494.8 Mb including gaps within scaffolds, with contig N50 of 5.7 kb and scaffold N50 of 862.6 kb. The assembly has been deposited in the NCBI (BioProject PRJNA203087).

Automated Gene Annotation Using a Maker 2.0 Pipeline Tuned for Arthropods

The copepod E. affinis was one of 30 i5K pilot genome assemblies subjected to automatic gene annotation using a Maker 2.0 annotation pipeline tuned specifically for arthropods. The pipeline was designed to be systematic, providing a single consistent procedure for the species in the pilot study. Also, the pipeline was scalable to handle hundreds of genome assemblies, evidence-guided (using both protein and RNA-Seq evidence to guide gene models), and targeted to utilize extant information on arthropod gene sets. The core of the pipeline was a Maker 2 instance, modified slightly to enable efficient running on our computational resources (Holt and Yandell 2011). The genome assembly was first subjected to de novo repeat prediction and CEGMA analysis to generate gene models for initial training of the ab initio gene predictors. Three rounds of training of the Augustus (Stanke et al. 2008) and SNAP (Korf 2004) gene predictors within Maker were used to bootstrap to a high-quality training set. Input protein data included 1 million peptides from a nonredundant reduction (90% identity) of Uniprot Ecdysozoa (1.25 million peptides), supplemented with proteomes from 18 additional species (S. maritima, Tetranychus urticae, C. elegans, Loa loa, T. adhaerens, A. queenslandica, Strongylocentrotus purpuratus, N. vectensis, Branchiostoma floridae, Ciona intestinalis, Ciona savignyi, Homo sapiens, Mus musculus, Capitella teleta, Helobdella robusta, Crassostrea gigas, L. gigantea, and Schistosoma mansoni), leading to a final nr peptide evidence set of 1.03 million peptides. RNA-Seq reads from E. affinis adult males and females were used judiciously to identify exon-intron boundaries, but with a heuristic script to identify and split erroneously joined gene models. CEGMA models for QC purposes were used: for E. affinis, of 1,977 CEGMA single-copy ortholog gene models, 1,808 were found in the assembly, and 1,707 in the final predicted gene set. Finally, the pipeline used a nine-way homology prediction with human, Drosophila and C. elegans, and InterPro Scan5 to allocate gene names. The automated gene set is available at the BCM-HGSC website (https://www.hgsc. bcm.edu/arthropods/bed-bug-genome-project) and at the National Agricultural Library (https://i5k.nal.usda.gov).

Sample Preparation for Transcriptome Sequencing

Eurytemora affinis transcriptomes were generated using RNA-Seq strand-specific paired-end Illumina sequencing with one sample per HiSeq2000 channel at the Institute for Genome Sciences in the University of Maryland School of Medicine. To compare relative expression of CRG families in males versus females of the copepod E. affinis, three different types of samples were sequenced: Female, male, and mixed female þ male samples, with two replicates each and $220 individual copepods per replicate sample. Females and males from an inbred line (line VA-30-1, 30 generations of full-sib mating) derived from the same population of E. affinis used for genome sequencing (described above) were used. Inbred copepods were reared under controlled laboratory conditions, at 13 C, 15 PSU (practical salinity unit % parts per thousand) salinity, and on a 15L:9D photoperiod, until the copepods reached adulthood. The copepods were fed with saltwater algae Rhodomonas salina. To prevent bacterial infection, copepods were treated with the antibiotics Primaxin (20 mg/l), D-amino acid cocktail (10 mM of D-methionine, D-leucine, D-trypto- phan, and 5 mM D-tyrosine), and Voriconazole (0.5 mg/l) every 3-4 days. The D-amino acids we used (D-methionine, Dtryptophan, D-leucine, and 5 mM D-tyrosine) were found to induce negligible responses in the insect chemoreceptors tested (D. melanogaster IRs) (Croset et al. 2016).

Two days prior to RNA extraction, males and females were separated into different beakers, and treated them with an antibiotic cocktail in order to minimize contamination of copepod RNA with bacterial RNA. The separated female and male samples (two replicates per sex) received a full antibiotic cocktail (described below), whereas the mixed males þ female samples (two replicates) received the regular antibiotic cocktail used in culturing (with only Primaxin, D- amino acids, and Voriconazole). qPCR of the bacterial 16S rRNA gene on the copepod RNA samples was performed before and after administering varying combinations of antibiotic recipes to devise a full antibiotic cocktail that effectively removed nearly all bacterial contamination. All the antibiotics used had been tested for toxicity on E. affinis in prior experiments. Our full antibiotic cocktail consisted of: Primaxin (20 mg/l), Voriconazole (0.5 mg/l), D-amino acids (10 mM Dmethionine, D-tryptophan, D-leucine, and 5 mM D-tyrosine), Sitofloxacin (10 mg/l, increased to 20 mg/l in last 24 h), Rifaximin (3 mg/l, increased to 10 mg/l in last 24 h), Phosophomycin (20 mg/l), Daptomycin (3 mg/l), and Metronidazole (15 mg/l). In order to clear the guts, the copepods were starved and treated with 120 ml/l of 6.0-mm copolymer microsphere beads (Thermo Scientific cat# 7505A, Fremont, CA) for the last 24 h before RNA extraction. Total RNA was extracted with Trizol reagent (Ambion RNA, Carlsbad, CA) and then purified with Qiagen RNeasy Mini Kit (Qiagen cat# 74104, Valencia CA), following the protocol described by Lopez and Bohuski (2007).

Differential Gene Expression Analysis of CRGs between the Sexes

To compare relative expression of CRG families in males versus females of the copepod E. affinis, sex-specific expression levels of the GR, IR, and CSP genes were determined in transcriptome sequences of female and male samples, with two replicates each and $220 individual copepods per replicate sample. The approaches used for transcriptome sequencing are described above. As controls, we also examined differential expression of five representative housekeeping genes (Cyclophilin-33, Actin 42A, Heat shock protein 83, Glyceraldehyde 3 phosphate dehydrogenase 1, and Ribosomal protein L32) (supplementary table S9, Supplementary Material online) in males and female samples. These controls were used to verify that there was no general sex-specific bias in gene expression in these samples.

Single-end reads were mapped onto our assembled transcriptomes using Bowtie (ver. 1.0.1) with 0 mismatches (Langmead et al. 2009;Katz et al. 2010). We checked the raw Illumina sequences corresponding to the IR genes and confirmed their identities using Integrative Genomics Viewer (Thorvaldsd ottir et al. 2013). Numerical count data were transformed into RPKM to normalize for the number of sequencing reads and total read length (Mortazavi et al. 2008). RPKM values above 0.3 were used as the threshold for gene expression (Ramsko ¨ld et al. 2009). The statistical differences in gene expression levels between male and female samples were determined using both parametric (edgeR) (Zhou et al. 2014) and nonparametric (NOISeq) (ver. 2.8.0) (Tarazona et al. 2011) approaches in the R Bioconductor package (http://www.bioconductor.org) (ver. 3.1.2). Genes were considered to be differentially expressed if they had Pvalues less than 0.05 using edgeR or had probability values Arthropod Chemosensory-Related Genes . doi:10.1093/molbev/msx147 MBE