Comparative Genomics Reveals the Origins and Diversity of Arthropod Immune Systems

1All animals must defend themselves against a battery of natural enemies, ranging from pathogens such as viruses, bacteria and fungi, to macroscopic parasites such as parasitic worms or insects. The immune defenses that have evolved in response to this challenge must distinguish self from nonself, and produce effectors that target and kill these invaders. All the major groups of animals possess an innate immune system, where immune receptors are genetically hard-coded and the response is typically relatively nonspecific with respect to individual pathogen strains or previous exposure. The innate immune system originated early in animal evolution before the split between protostomes and deuterostomes, as some components of the vertebrate innate immune system show clear homology to insect immune molecules. In addition to an innate immune system, vertebrates possess an adaptive or acquired immune system, where receptor diversity is generated somatically and there is immunological memory. Although key components of the adaptive immune system are not found much beyond vertebrates, some invertebrates have independently evolved both immune memory and mechanisms to generate receptor diversity somatically.

2Arthropods have a powerful innate immune response, our understanding of which comes largely from insects, especially Drosophila and mosquitoes. In these species, pathogen-associated molecular patterns (PAMPs) such as bacterial peptidoglycan or fungal beta-1,3 glucan are recognized by pattern recognition receptors such as peptidoglycan recognition proteins (PGRPs) and beta-1,3 glucan recognition proteins (βGRPs). Following recognition, these receptors then activate the Toll and Imd signaling pathways, leading to the translocation of Nf-κb transcription factors into the nucleus and a humoral response characterized by the expression of antimicrobial peptides (AMPs). In addition, there is a melanization response that kills parasites by depositing the dark pigment melanin along with the production of toxic molecules. Alongside the humoral response, there are cellular responses where blood cells called plasmatocytes phagocytose pathogens and specialized flattened cells called lamellocytes can encapsulate larger targets such as parasitoid wasp eggs. The main defense against viruses is RNAi, where short RNAs are generated from double-stranded viral RNA and loaded into Argonaut proteins to guide the degradation of viral RNA.

3Whole-genome analyses have revealed much conservation of key immune pathways and gene families between insect species. The Toll, Imd, JAK/STAT, and JNK signaling pathways are remarkably well conserved, often in 1:1 orthologous relationships between species. A notable exception to this pattern is the pea aphid, which appears to have lost the Imd pathway. Despite this, much variation in presence/absence, copy number, and sequence divergence is observed in other genes, particularly those encoding recognition and effector molecules. For example, mosquitoes show extensive duplications in gene families associated with the response to the malaria parasite Plasmodium.

4Beyond the insects, Toll-like receptors (TLRs), their associated signaling components, and Nf-κb transcription factors all have mammalian homologs, suggesting that the origin of these genes predates the protostome/deuterostome split over 600 Ma. The same is true for the PGRPs and thioester-containing proteins (TEPs), which show similarities to vertebrate alpha-2 macroglobulins and complement factors. Components of the Imd pathway resemble the tumor necrosis factor receptor pathway of mammals.

5The evolution and diversity of innate immune systems across the arthropods remain poorly understood, despite the importance of arthropods as disease vectors, pests, and components of biodiversity. As yet, the only detailed whole-genome analysis of a noninsect arthropod investigated the crustacean Daphnia, which is the sister group to the insects. This found a repertoire of immune genes that is remarkably insect-like, with the notable absence of PGRPs. However, there is a lack of genome-level studies of the more divergent myriapods and chelicerates. As such the timing and nature of many key innovations in arthropods remain unresolved, and we lack an overview of the immune system in major arthropod groups. The large variation in the copy number of some immunity gene families in insects would suggest that great variation in arthropod immune systems might occur when looking over larger phylogenetic distances.

6The recent sequencing of multiple whole arthropod genomes, some of which are unpublished, provides an opportunity to examine the arthropod immunity gene repertoire in a systematic and consistent fashion. In this study, we have used these genomes to characterize the evolution of the innate immune system across all the main arthropod taxa. Our results show both remarkable diversification of the immune response across the arthropods, and unexpected conservation and similarities to mammalian genes.

7To investigate the evolution and origins of the arthropod innate immune system, we identified homologs of insect immunity genes in species that diverged early in the evolution of the arthropods. The arthropods are a phylum that contains four extant subphyla, with the chelicerates, the myriapods, and the crustaceans sequentially diverging from the lineage leading to the insects. This allows us to identify which components of the immune system were present in the ancestral arthropod, and which have been gained or lost later in evolution.

8The timing of events early in the evolution of the arthropods is highly uncertain, but it is clear that these four major groups all diverged very early in the evolution of animals. The common ancestor of the arthropods existed an estimated 543 Ma, and by 511 Ma all four of the subphyla had formed (; see figures below). In our analysis we included the genomes of the insect Drosophila melanogaster, the crustacean Daphnia pulex (water flea), the myriapod Strigamia maritima (coastal centipede) and five species of chelicerate: Mesobuthus martensii (Chinese scorpion), Parasteatoda tepidariorum (house spider), Ixodes scapularis (deer tick), Metaseiulus occidentalis (western orchard predatory mite), and Tetranychus urticae (red spider mite).

9To identify homologs of immunity genes across these great phylogenetic distances, we combined methods based on sequence similarity with the predicted cellular location of proteins and analyses of domains, motifs, and residues that are known to be essential for the immune function of the encoded proteins (supplementary table S1, Supplementary Material online). These features can often be identified across very distantly related species, which allows us to guard against the inevitable loss of power to detect sequence similarity when looking at distantly related species.

10TLRs are transmembrane proteins that play a central role in the immune response of insects and vertebrates. They have an extracellular leucine-rich repeat (LRR) region at the N-terminal, and a cytosolic Toll/interleukin-1 receptor (TIR) domain at the C-terminal. In mammals, different TLRs directly recognize a variety of PAMPs. In Drosophila, Toll-1 plays key roles in both development and immunity with the immune function relying on Toll-1 binding to an endogenous cytokine rather than directly to pathogen molecules (see below; ). Several other Drosophila TLRs have been suggested to have immune functions, although these are poorly characterized, and it is likely that most have primarily developmental roles.

11In all the arthropod genomes we find multiple TLRs, with a TIR domain separated from LRRs by a transmembrane helix. On a tree reconstructed from the sequence of the TIR domain, Toll-1 clusters with three other Drosophila TLRs (albeit with weak support; fig. 1 and supplementary fig. S1, Supplementary Material online), so there are no clear 1:1 homologs of Toll-1 outside of the insects. Therefore, it cannot be predicted which if any of the TLRs beyond the insects are likely to have a role in immune signaling.

12The TLRs have been frequently lost and duplicated during arthropod evolution, as there is extensive variation in copy number and little congruence between the gene and species trees (fig. 1). Within the myriapods there have been two large copy number expansions, resulting in a total of 27 TLRs in the Strigamia genome (fig. 1). The spider Parasteatoda and scorpion Mesobuthus also have a high number of TLRs (16 and 14, respectively), whereas at the other extreme the mite Tetranychus has just two (fig. 1, supplementary table S1, Supplementary Material online).

13There are two major structural classes of TLRs and both are widespread in arthropods. The single cysteine cluster TLRs (sccTLRs) have a single cysteine cluster at the end of the LRRs adjacent to the cell membrane, whereas the multiple cysteine cluster TLRs (mccTLRs) have multiple cysteine clusters. The vertebrate TLRs are largely sccTLRs, whereas Drosophila Toll receptors other than Toll 9 are mccTLRs. The immune function the Drosophila sccTLR (Toll 9) is unclear, with overexpression of Toll-9 upregulating AMPs, whereas a Toll-9 knock out did not affect the immune response. We found that although mccTLRs are most common in arthropods, vertebrate-like sccTLRs are also widespread (fig. 1).

14The division between sccTLRs and mccTLRs is reflected in their evolutionary relationships, with the two structural classes forming two major clades on the TLR phylogeny (fig. 1). In most cases, the arthropod sccTLRs are more closely related to vertebrate sccTLRs than they are to arthropod mccTLRs (fig. 1, yellow shading). The mccTLR clade (fig. 1, pink shading) contains most of the arthropod TLRs and the Caenorhabditis elegans TOL1 (nematodes are protostomes like arthropods). This pattern of the structure of the TLRs being reflected in their phylogeny is especially striking as the phylogeny is reconstructed using the sequence of the intracellular TIR domain, whereas the structural classification is based on extracellular sequences. The structural classification thus provides corroboration for this phylogenetic division, despite the tree being poorly resolved (supplementary fig. S1, Supplementary Material online). Therefore, throughout the arthropods there are a small number of TLRs that are more similar to vertebrate TLRs than other arthropod TLRs.

15The presence of arthropod TLRs clustering with vertebrates shows that mccTLRs and sccTLRs diverged early in animal evolution, but it is unclear whether the common ancestor of protostomes and deuterostomes had both types of TLR. Our tree places arthropod sccTLRs sequences interspersed among vertebrate TLRs (fig. 1). However, this may simply reflect error in the tree reconstruction, as bootstrap support for these relationships is low (supplementary fig. S1, Supplementary Material online) and we were unable to reject a tree where all the deuterostome and protostome TLRs were monophyletic (Shimodaira–Hasegawa Test: 2Δl = 6, P > 0.05). As we cannot reliably root the tree (fig. 1), the most parsimonious interpretation of the data is that the common ancestor had sccTLRs, with mccTLRs appearing later in protostome evolution. We also find a few cases of unusual extracellular domain structures, such as short truncated TLRs in the mccTLR clade that have only have a single cysteine cluster and TLRs with more than two cysteine clusters (fig. 1 and supplementary table S7, Supplementary Material online).

16The humoral immune response of Drosophila and other insects centers on the Toll and Imd pathways, both of which result in Nf-κb transcription factors being activated and translocated into the nucleus, where they upregulate the expression of AMPs and other genes. In Drosophila, recognition of Gram-positive bacteria and fungi by βGRPs and short-chain PGRPs causes cleavage of Spätzle, which subsequently binds to and activates Toll-1, initiating the Toll pathway. Following binding by cleaved Spätzle, the Toll-1 TIR domain initiates a signaling cascade that culminates in the translocation of the Nf-κb transcription factors Dorsal, and Dorsal-related immunity factor (Dif) to the nucleus.

17We found Toll pathway members to be highly conserved across the arthropods (fig. 2), with homologs of Spätzle, Myd88, pelle, cactus, and dorsal in all species. We failed to find a Tube homolog in any of the species studied except Mesobuthus, but this is likely to be a lack of power to detect the gene, as a previous analysis of Tube suggested that it is homologous to IRAK-4, which occupies an equivalent position in the vertebrate Toll pathway. Indeed we find multiple proteins across species with the IRAK-like death domains that characterize Tube. Most genes are present in single copies, although Dorsal (encoding the Nf-κb transcription factor) has been duplicated twice in the spider Parasteatoda (as well as being duplicated in Drosophila), and Cactus (encoding its inhibitor) is duplicated in Daphnia and Mesobuthus.

18In Drosophila the systemic humoral immune response to Gram-negative bacteria is controlled by the Imd pathway, which is initiated by the binding of the transmembrane protein PGRP-LC to peptidoglycan. The intracellular RHIM motif of PGRP-LC interacts with Imd, initiating the signaling cascade. Imd in turn activates TAK1, which together with the IkB kinase complex (IKK)b/y complex, Fadd and DREDD (a caspase-8 homolog) activate the Nf-κb transcription factor Relish. Relish is translocated into the nucleus, upregulating AMPs, and other genes. Imd also activates the JNK pathway through Tak1. The Imd pathway is especially important in the epithelial immune response, where it regulates genes such as Drosomycin, which are controlled by the Toll pathway in the systemic response. Furthermore, some components of the pathway also have roles outside the immune response, such as the elimination of mutant or unfit cells.

19Relish is an unusual and easily identifiable gene, as the Nf-κb transcription factor and its inhibitor are combined in a single protein. We find clear Relish homologs in all species but Metaseiulus, suggesting that a Relish-based immune response may have been present in the common ancestor of the arthropods (fig. 3). The IKK is required for the cleavage and activation of Relish, and we find homologs of both the catalytic subunit, IKKb, and the regulatory subunit, IKKg, in most species (fig. 3). The Relish homologs all have an N-terminal relish-like domain that shows closest sequence similarity to Drosophila Relish; however, some lack the distinctive C-terminal ankyrin repeat region that plays the role of the Nf-κb inhibitor (fig. 3). In Daphnia the ankyrin repeat region is absent from the public gene model due to an error in the automated gene model prediction, as manual annotation identified an additional portion of the gene with ankyrin repeats. It is unclear whether the three chelicerate sequences lacking the ankryin repeats are also annotation errors or reflect true loss of this region.

20Relish appears to have been lost entirely in the mite Metaseiulus (fig. 3), where no homologues could be discerned by sequence similarity or conserved domain searches (RHD-n relish domain). This is especially striking as Metaseiulus has a small and well-assembled genome so this is unlikely to be an artifact of an incomplete genome sequence. Overall, this species is missing more Imd pathway components than any of the other species (fig. 3), supporting the hypothesis that this branch of the immune response may have been lost. This would not be a unique occurrence, as Relish and other Imd pathway components have also been independently lost in the pea aphid (an insect).

21Despite the conservation of Relish in most species, many other key components of the Imd pathway were only found in the mandibulates and were absent from the chelicerates (fig. 3). In the mites and ticks (Tetranychus, Metaseiulus, and Ixodes), we fail to find any likely Imd, Fadd or Dredd homologues. In arachnids (Parasteatoda and Mesobuthus) we find possible Fadd and Dredd homologues, although they apparently lack N-terminal DED domains, so they may not have the same function as in Drosophila. As Fadd homologues are known to be widespread in the animal kingdom, these results suggests that secondary losses of Fadd and Dredd may have occurred in the chelicerates.

22The absence of an intact Imd pathway in the chelicerates is supported by the distribution of transmembrane PGRPs, which are found at the start of the pathway in Drosophila. The PGRP domain can clearly be detected across these phylogenetic distances (see below), and transmembrane proteins can be robustly predicted. Transmembrane PGRPs were entirely absent from the chelicerates (figs. 3 and 4).

23Although the Imd pathway is largely intact in both the myriapods and crustaceans, transmembrane PGRPs are not. Daphnia does not possess any PGRPs (figs. 3 and 4; ), suggesting that the Imd pathway is either not functional or is activated in a different way in this species. In the myriapod Strigamia there are two transmembrane PGRPs, but we failed to identify the Imd-interacting RHIM domain using HMMER. Furthermore, the transmembrane PGRPs identified in Strigamia possess only one external recognition domain—unlike PGRP-LF in Drosophila which has two. Therefore, the activation of the Imd pathway by a transmembrane PGRP could either represent an innovation acquired in the insect lineage or it may have been lost in the crustaceans.

24In contrast to the absence of many Imd pathway components in chelicerates, the JNK pathway is highly conserved across the arthropods with Basket and Jun universally present. This may be a consequence of this pathway playing a role in many key cellular processes in addition to its role in Imd-related signaling.

25The JAK/STAT pathway plays a role in the immune response of both mammals and Drosophila. In Drosophila and Anopheles mosquitoes, following immune challenge activated STAT translocates to the nucleus where it alters the expression of many genes, including upregulating the Drosophila immunity protein TEP1. We find clear homologues of Domeless, Hop, and Stat92e in most species (table 1). Parasteatoda and Mesobuthus and Metaseiulus lack a clear Hop homologue however and Domeless would also appear to be absent from Mesobuthus.

26PGRPs bind to bacterial peptidoglycan and can act as pathogen recognition receptors, negative regulators of the immune response, or effectors that kill bacteria. They fall into two main groups. The noncatalytic PGRPs can function as pattern recognition receptors, and play a key role in activating the Toll and Imd pathways of Drosophila following infection (PGRP-LC/LE in the Imd pathway and PGRP-SA in the Toll pathway). The catalytic PGRPs possess amidase activity that allows them to enzymatically break down peptidoglycan (in Drosophila: PGRP-SC1/2, SB1/2, and LB), and can they function either as negative regulators of the immune response by removing immunogenic peptidoglycan or as effectors that kill bacteria by degrading their peptidoglycan bacteria. PGRPs are also found in mammals, and were therefore presumably present in the common ancestor of the arthropods.

27The copy number of PGRPs varies greatly across the arthropods (fig. 4; table 1). They have been entirely lost from both the crustacean Daphnia (see also ) and the chelicerate Mesobuthus, whereas Metaseiulus and Tetranychus each have just a single PGRP. At the other extreme, the myriapod Strigamia has 20 PGRPs. All bar one of the Strigamia genes cluster together on the gene tree, which suggests that they have probably resulted from one or two ancestral PGRPs that duplicated extensively within the myriapods (fig. 4).

28It is possible to predict whether a PGRP has catalytic activity from the sequence of its PGRP domain. The catalytic (amidase) activity of insect PGRPs and bacteriophage T7 lysozyme is zinc-dependant, and the noncatalytic PGRPs have lost residues required for zinc binding. To predict the function of the PGRPs, we aligned their PGRP domains and identified a cysteine and two histidine residues (see fig. 3 of ) that are required for zinc binding. In support of the functional relationship between these sites, we found that the presence of these three residues was strongly correlated—18 of the 21 sequences with the cysteine also had the two histidines, whereas only one of the 16 sequences without the cysteine had both histidines (supplementary table S2, Supplementary Material online).

29Catalytic and noncatalytic PGRPs are scattered across the arthropods, and every species with more than one PGRP has both types (fig. 4, catalytic forms marked*). There have been frequent gains or losses of catalytic activity during evolution, as the two forms are interspersed on the gene tree, and closely related pairs of sequences can differ in whether they have predicted catalytic activity (fig. 4). Together these results suggest that in other arthropods, as in Drosophila, PGRPs are likely to play a variety of roles. However, as catalytic or noncatalytic PGRPs are absent from taxa across the tree, it seems likely that these functions may be replaced by other molecules in these groups.

30In chelicerates and myriapods, many of the PGRPs do not have a signal peptide and are therefore predicted to be intracellular (fig. 4). In Drosophila, intracellular isoforms of PGRP-LE are important in responses to intracellular bacteria, and regulating gut immune responses to both pathogens and the microbiota. It is unlikely the intracellular PGRPs that we identified perform similar roles, as they differ from PGRP-LE in that they are predicted to be catalytic and do not contain a predicted RHIM motif that could interact with Imd. Therefore, these intracellular catalytic PGRPs may have a novel function such as killing intracellular bacteria.

31βGRPs, which are also known as Gram-negative binding proteins (GNBPs), bind β-1,3 glucan in microbial cell walls (particularly fungi), and can act as coreceptors with PGRP-SA to recognize Gram-positive bacteria and initiate the Drosophila Toll pathway through Spätzle. Similar to pattern recognition receptor PGRPs, βGRPs originate from enzymes that degrade glucan (β-1,3-glucanases) that have lost their enzymatic activity. Many insects have both β-1,3-glucanases and βGRPs, although the Drosophila genome encodes only βGRPs. We identified β-1,3-glucanases and βGRPs using the diagnostic O-Glycosyl hydrolase 16 domain (GH16-superfamily) that they share. Glucanases can be separated by the presence of two amino acid residues that are essential for catalytic activity.

32We find that proteins bearing the GH16 domain are limited to the Mandibulata (insects, crustaceans, and myriapods) and entirely absent from all of the chelicerates (table 1; fig. 5). This suggests a single loss event on the branch leading to chelicerates, as βGRPs and β-1,3-glucanases are present in other invertebrates such as mollusks. Drosophila and Strigamia both possess three GH16-bearing proteins, whereas Daphnia has ten. From phylogenetic analysis, it would appear that these are the result of lineage-specific expansions (fig. 5).

33Of the 13 GH16-bearing proteins that we identified, ten of them contain the two glutamate residues that are essential for enzymatic activity, suggesting that they are β-1,3-glucanases that may function in digestion or pathogen killing (supplementary table S5, Supplementary Material online, and fig. 5). There are two proteins in Daphnia and one in Strigamia that lack one or both of these residues and therefore presumably lack glucanase activity. The nonenzymatic proteins are not related to the Drosophila βGRPs, and, in a pattern similar to the PGRPs, the loss of glucanase activity has occurred on multiple occasions. Therefore, if these proteins function as pattern recognition receptors, our results would suggest that they have arisen independently by convergent evolution.

34The TEPs include the vertebrate complement factors C3, C4, and C5, the insect TEPs, and a family of vertebrate protease inhibitors called alpha-2-macroglobulins. We found members of the TEPs family in all species, confirming that they were present in the common ancestor of the arthropods and have been retained in all the major arthropod lineages (fig. 6).

35In Drosophila and mosquitoes, TEPs can covalently bind to the surface of pathogens and parasites, and mark them for destruction by phagocytosis or melanotic encapsulation. The TEPs have a characteristic thioester motif, which once the TEP has been cleaved into its active form can covalently bind to pathogens. We found TEPs bearing the thioester motif in all species (fig. 6).

36All the arthropod genomes also encode TEPs that lack the canonical thioester motif GCGEQ, and therefore presumably lack the ability to form covalent thioester bonds to microbial surfaces (fig. 6 and supplementary table S3, Supplementary Material online). Importantly, most of these proteins lack the critical cysteine required for the formation of thioester bonds (supplementary table S4, Supplementary Material online). All but two of these fall into a single clade, all the members of which lack this motif (fig. 6, highlighted green). This clade includes the Drosophila protein MCR (macroglobulin complement related or Tep VI), so we have named these MCR proteins.

37Despite lacking the thioester motif, in Drosophila MCR can bind to the fungus Candida albicans and promote phagocytosis. In addition to this, MCR may also play a role in epithelial barrier formation preventing the spread of pathogens across tissues. Our results show that MCR proteins were present in the common ancestor of arthropods and all eight of the species that we studied have at least one copy, with Daphnia having four copies and Strigamia and Parasteatoda having three copies (fig. 6; table 1).

38Phylogenetic analysis revealed proteins related to vertebrate C3 complement factors in the myriapod Strigamia and the chelicerates Tetranychus and Parasteatoda (fig. 6). The complement factors fall into a monophyletic group containing a single clade of vertebrate complement factors, a clade of arthropod sequences, and a basal lineage found in corals (fig. 6). The relationships of these clades therefore mirror the phylogeny of these three groups, and indicate an ancient origin of the C3 complement factors. These results support a previous finding of a sequence from the chelicerate Carcinoscorpius that was most similar to C3 (; fig. 6). Therefore C3-like proteins are widely scattered across the arthropods, but have been lost from most lineages including Drosophila.

39The remaining two clades include the vertebrate alpha2 macroglobulins and the Drosophila TEPs (fig. 6, pink and yellow). Members of the Drosophila TEP clade are found in all the arthropod genomes we analyzed, although species other than Drosophila only have one or two copies (fig. 6, yellow). The function of these proteins has been best studied in Anopheles mosquitoes where they can bind bacteria and eukaryotic parasites, promoting phagocytosis and encapsulation. Interestingly C. elegans Tep2 is a sister to this monophyletic group of arthropod TEPs. Sister to this group of insect TEPs we found a clear monophyletic grouping of vertebrate alpha-2 macroglobulins and arthropod sequences (fig. 6, pink). Of our sequences, we found five Parasteatoda, two Ixodes, one Metaseiulus, and one Daphnia sequence to be alpha-2 macroglobulin-like. An insect sequence—TEP7 annotated in the honey bee genome project—also fell into this clade.

40Dscam (Down syndrome cell adhesion molecule) is a protein with a well-established function in neuronal patterning that is also thought to play a role in immunity. In Drosophila it is expressed in hemocytes, can bind to bacteria and knocking down its expression greatly reduces the rate at which bacteria are phagocytosed. In Anopheles gambiae, silencing Dscam compromises both the mosquito’s bacterial and malarial resistance and affects phagocytosis. Furthermore, in a number of different arthropods Dscam is upregulated after infection. The Dscam gene has a remarkable level of alternative splicing—generating up to a possible 38,000 isoforms in Drosophila—which has led to speculation that this diversity contributes to a pathogen pattern-recognition repertoire in arthropods resembling that of vertebrate antibodies. However, little evidence has emerged to support this hypothesis, so it is unclear whether this extreme diversity is important for its immune function.

41There are multiple Dscam copies in all the species we studied, and frequent duplications or losses of the gene (fig. 6). Lineage-specific duplications of Dscam have resulted in 60 copies in Strigamia (a similar number were previously reported in this species; ) and 35 in the spider Parasteatoda, whereas in the other species we find 4 to 14 copies (fig. 6). Although we did not characterize the diversity generated through alternative splicing, these results suggest that Dscam diversity is an important trait that is generated in different ways across the arthropods. The Dscam paralogs we identified are an excellent system to test whether Dscam diversity plays a role in immunity or is important only to neuronal patterning.

42Fibrinogen-related proteins (FREPs) are found in taxa ranging from arthropods to mammals, and have been implicated in several antipathogen processes. These include vertebrate complement activation and binding to pathogens to act as pattern recognition receptors. In mosquitoes, FREPs are upregulated following infection and key to effective anti-plasmodium and antibacterial responses. They are also thought to contribute to the specificity of pathogen recognition in snails, where different FREPs bind to different classes of pathogens. The copy number of FREPs in insects varies widely—in the genus Drosophila species have between 14 and 43 FREPs, whereas A. gambiae has 59. Based on the conserved fibrinogen domain, we found all species to have FREPs, with copy number being highly variable. Notably, the mites Metaseiulus and Tetranychus contained only one and two fibrinogen domain containing proteins, respectively, whereas all other genomes contain between 19 and 36 (table 1).

43Genes of the Nimrod superfamily are characterized by the presence of NIM-repeats, a specialized EGF-domain, and have roles in phagocytosis in insects, nematodes and humans. In A. gambiae, Nimrods are induced by bacteria and are important in bacterial killing. In addition to this immune role, in Drosophila the Nimrod draper is involved in the phagocytosis of apoptotic cells.

44We found multiple Draper-like proteins in all genomes except Metaseiulus and Mesobuthus, which may represent secondary losses or a lack of detection power. We failed to identify any B-type or C-type (e.g., Eater in Drosophila) Nimrod proteins in any species, perhaps due to the HMM-approach used being insufficiently sensitive.

45Melanin is essential in the encapsulation of parasites, wound healing and the hardening of cuticle. Melanin is produced when the inactive zymogen prophenoloxidase (PPO) is cleaved into its active form, phenoloxide (PO), by a serine protease cascade and catalyses the hydroxylation and oxidation of phenols producing quinones. Quinones are toxic and play an important role in killing pathogens, and are the precursors of melanin, which physically encapsulates parasites. The three D. melanogaster PPOs each bear Hemocyanin N, Hemocyanin M, and Hemocyanin C domains in that order. Hemocyanin M additionally has a tyrosinase motif, which may distinguish it from closely related proteins such as hexamarins and larval storage proteins. We find proteins bearing the three syntenic domains in Daphnia, Strigamia, Mesobuthus, and Parasteatoda (although conserved domain searches only predict tyrosinase activity in the Parasteatoda and Mesobuthus homologs; table 1). This protein family was greatly expanded in the arachnids, with 13 copies in Parasteatoda and 8 in Mesobuthus. These proteins may be functioning as PPOs or oxygen-carrying hemocyanins.

46As PO activity generates toxic compounds, it is tightly regulated by a serine protease cascade and associated serine protease inhibitors (serpins). Serine proteases and serpins are also important in other aspects of the immune response, such as the Toll signaling pathway. The main group of serine proteinases involved in immunity are the CLIP-domain serine proteinases, and these are found in all the species (table 1). The family appears to be greatly expanded in Drosophila, which has 24 copies, compared with the three to eight copies that we found other arthropods. Serpins are also found in all the species, with the myriapod Strigamia and the crustacean Daphnia having a lower copy number than Drosophila and the chelicerates (table 1).

47Dual oxidase (DUOX) was found to be present in all taxa (table 1). DUOX generates reactive oxygen species (ROS) that are important intestinal microbicidal agents in Drosophila that maintain homeostasis between infectious and commensal bacteria. The expression of DUOX and ROS production is controlled by the stress-related p38 pathway. Beyond generating ROS to kill pathogens, DUOX plays a role in wound repair, the renewal of intestinal epithelial cells and signal transduction. We found DUOX in all the species, with two copies in Daphnia, Tetranychus, and Mesobuthus, and three copies in the spider Parasteatoda.

48Lysozymes are enzymes that can cleave peptidoglycan in bacterial cell walls, and function both as immune effectors and digestive enzymes. They are found across the metazoa, and in insects are typically expressed in the gut and salivary glands. We found lysozymes in all taxa studied, with between one and five genes in the different species (table 1). This is less than the 13 found in Drosophila, which may be because fruit flies feed on bacteria-rich rotting fruit and require lysozymes for digestion (although some other arthropods have additional genes with the lysozyme-like domain, cl00222, which we conservatively did not include in our list).

49AMPs are short peptides that are produced by animals and plants and kill a wide range of pathogens. Many AMPs have been identified, and they are often restricted to a single clade. Furthermore, the small size of AMPs makes their identification by sequence similarity problematic across large phylogenetic distances. We searched for the functional domains associated with all the main Drosophila AMPs, and only found proteins bearing the Defensin2 (cl03093) functional domain in Ixodes and Mesobuthus.

50The primary antiviral defense of arthropods and many other animals is RNAi. Double-stranded viral RNA is cleaved into short interfering RNAs (siRNAs) by Dicer, and then loaded into Argonaute proteins to direct the degradation of viral RNA. There are three RNAi pathways in insects. The piRNA pathway is Dicer-independent and involves PIWI family Argonautes, whereas the siRNA and miRNA pathways both rely on Argonaute-like Argonautes and Dicer proteins. It is the siRNA pathway that acts as a defense against viruses and somatic transposable element activity, and this involves the Argonaute Ago2 and the Dicer Dcr2 in Drosophila.

51Across the arthropods there has been considerable diversification of the antiviral siRNAi pathway, whereas the paralogous miRNA pathway is conserved (fig. 8A). This is most striking in Ago2, which has a highly variable copy number—Drosophila, Daphnia, and Metaseiulus have a single copy, whereas in all the other species there are multiple paralogs, with Tetranychus and Mesobuthus having six copies (copy numbers in four of the species have been reported previously by. Furthermore, these duplications of Ago2 have occurred independently in the different species (fig. 8A). In common with previous work, we also found that the rate of molecular evolution of Ago2 is dramatically higher than its paralog Ago1 (fig. 8A).

52Dicer 2 has a single copy in most species, but it has been duplicated in Daphnia and in the mite Metaseiulus there are five copies (fig. 8B, supplementary table S5, Supplementary Material online, table 1). The absence of Dicer in Mesobuthus is likely an error in the genome assembly or annotation, as we were able to find a truncated Dicer-like gene.

53Our results confirm an ancient origin for the innate immune system, predating the split between protostomes and deuterostomes. We find striking examples of conservation between vertebrates and arthropods, despite these two groups having diverged before the Cambrian explosion some 543 Ma. These include a group of arthropod TLRs that share structural similarity with vertebrate TLRs and cluster with them phylogenetically. Similarly, several arthropods have C3-complement like factors that have been lost from Drosophila.

54Despite such conservation, we also find remarkable diversity in the immune system of different arthropods. The Imd pathway—essential for recognition and response to Gram-negative bacteria in Drosophila—is almost entirely absent from the chelicerates. We also observe extensive copy number variation in recognition and effector genes, suggesting complex evolutionary dynamics in these functional categories. Detailed analysis of PGRPs suggests that this change in gene copy number is accompanied by changes in function.

55Several factors may be driving the design and diversification of arthropod immune systems. Evolution involves tinkering with existing processes, and this is clear in the immune system. For example, the ability of enzymes to specifically bind molecules associated with pathogens has led to them losing their enzymatic activity and becoming pattern recognition receptors such as PGRPs and βGRPs. Melanin is produced by many species for various functions, and has been recruited to the immune system in arthropods. Here, the production of toxic compounds during melanin synthesis has been exploited for pathogen killing. Immune pathways may have arisen by combining modules of proteins with different functions—the Toll pathway in Drosophila looks more like a classical Nf-κb pathway combined with an extracellular serine protease cascade analogous to that controlling PPO activity. Similarly, the downstream part of the Imd pathway resembles a classical Nf-κb pathway, whereas the upstream part is similar to the mammalian tumor necrosis factor-receptor pathway.

56Many components of the immune system have diversified very rapidly, and this likely reflects the need to counter an ever-changing array of parasites, as existing pathogens evolve to evade or sabotage immune responses, and entirely new pathogens appear in populations. For example, the widespread production of suppressors of RNAi by insect viruses may explain why the antiviral siRNA pathways evolve and diversify much faster than paralogous proteins in the micro-RNA pathway. Functional constraints are also likely to be important. For example, signaling pathway members relatively rarely show much copy number variation, despite natural selection driving rapid sequence evolution. This may be because these pathways have many functions, so any change in copy number would have damaging pleiotropic effects.

57The noninsect arthropods comprise a significant proportion of the earth’s biodiversity and include many species of economic and medical importance. Ours is the first detailed genome-wide analysis of arthropod immune systems outside of insects, and characterizing the function of the immune genes that we described remains an important challenge for the future. It is likely that many arthropods have immune defenses that are absent in Drosophila and therefore missing from our analysis, and these can only be discovered by experimental work on these species.

58We retrieved the complete predicted-peptide sets for D. melanogaster (r5.54, Flybase) and seven additional noninsect arthropods: S. maritima (v1.20, Ensembl Genomes), Met. occidentalis (v1.0, NCBI Refseq), M. martensii (v1.0, Science Data Sharing Platform Bioinformation; ), T. urticae (v1.2, Ensembl Genomes; ), I. scapularis (v1.2, Vectorbase), Daphnia pulex (r20, Ensembl Genomes; ), Parasteatoda tepidariorum (Augustus 3, Spiderweb) and performed all analyses on these data. A set of key immunity genes in D. melanogaster was compiled from Immunodb, IIID, Flybase and, to be used as query sequences for homology searches (supplementary table S5, Supplementary Material online).

59Broadly, to identify sequence homologues of a query set of D. melanogaster immunity genes (supplementary table S4, Supplementary Material online) we used multiple BLAST (Basic Local Alignment Search Tool)-based and hidden Markov model based approaches, to compile a redundant list of candidate homologues in each of the seven additional noninsect arthropod species. This list was then filtered by similarity, quality, e value, best reciprocal Drosophila hit, presence/absence of conserved domains known to be essential to function, and additionally by tree-based similarity measures, producing a final nonredundant list of high confidence predicted Drosophila innate immunity functional homologs in each peptide set.

60To identify clusters of orthologous genes, we performed all-versus-all BLAST-based clustering of protein sequences using Orthomcl V1.4 and the predicted peptide sets from all eight arthropod species. Default parameters were used, and homologues from each of the seven noninsect arthropods were assigned to immune homology groups based on clustering with a Drosophila innate immunity peptide from our query set (supplementary table S5, Supplementary Material online).

61For a subset of genes, we searched for homologs in each full peptide-set using the hidden Markov model based HMMER. HMMER aims to be more sensitive than a traditional whole-sequence based search such as BLAST, by utilizing profile HMMs. Profile HMMs are statistical models of multiple sequence alignments whereby each residue in the alignment is determined to be more or less relevant to homology based on its conservation between sequences. Hits with an e value of less than 10−5 were considered significant and retained.

62We built profile HMMs for each of the 18 signaling pathway peptides (Toll, Imd, JNK, and JAK/STAT), using multiple alignments of the D. melanogaster gene and its high-confidence orthologs in the immune-annotated insects Bombyx mori, Tribolium castaneum, and Apis melifera. Multiple alignments were built using MAFFT, and orthologous sequences were retrieved from (B. mori), (T. castaneum), and (Ap. melifera).

63For the highly variable and divergent Nimrod genes (for which BLAST/orthomcl does not find significant homologues, and no single diagnostic domain exists), we built a profile HMM of the conserved NIM motif CXPXCXXXCXNGXCXXPXXCXCXXGY, using the multiple alignment of and searched the complete peptide sets using HMMER.

64We also searched all seven non-Drosophila peptide-sets for homologues of Drosophila immune genes using BLASTp, retaining hits with an e value less than 10−6, greater than 20% identity, and a bit score greater than 80. We additionally queried each species’ complete peptide-set against a BLAST database of all D. melanogaster peptides, in order to identify the best Drosophila BLASTp hit for each gene in each other species.

65Some gene classes rely on a conserved protein domain to function in the manner described in D. melanogaster, without which the protein was assumed to be nonfunctional and therefore discarded. We searched each whole non-Drosophila peptide-set for domains known to function in innate immune pathways using the NCBI BatchConservedDomain tool and identified proteins containing domains listed in the query set (supplementary table S1, Supplementary Material online).

66We finally compiled results from all the above techniques into a single list of potential immunity homologs. Examining these results individually, we filtered out hits where either an essential protein domain was absent (supplementary table S1, Supplementary Material online) or the reciprocal top D. melanogaster blast hit was to a nonimmunity-related gene.

67For PGRPs, GNBPs/βGRPs, and TLRs, we scanned putative homologues for transmembrane helices and signal peptides using the TMHMM Server v. 2.0 and the TargetP 1.1 Server, respectively. We also built a profile HMM of the PGRP RHIM (Imd-binding) domain after and, and scanned putative PGRPs in all species to predict presence/absence of the RHIM domain.

68Toll-like proteins/receptors are characterized by an N-Terminal TIR domain, a transmembrane helix, and a variable number of LRRs extending to the C-terminal. TIR domains and transmembrane helices were identified as above, whereas LRRs were identified using the web-interface of LRR-finder.

69We additionally built trees of gene families where duplications were suspected to be assembly errors (e.g., haplotypes annotated as separate contigs), or to distinguish orthologs and paralogs. Multiple alignments were assembled using MAFFT, with phylogenetic tree construction performed in PHYML using the WAG+G model. Shimodaira–Hasegawa tests to compare the log-likelihood of differing tree topologies were performed in RAxML using the WAG+G model and multiple alignments created as above in MAAFT.

70Supplementary alignments, figures S1–S4, and tables S1–S7 are available at Molecular Biology and Evolution online (http://www.mbe.oxfordjournals.org/).