Recent host-shifts in ranaviruses: signatures of positive selection in the viral genome

Abstract

Ranaviruses have been implicated in recent declines in global amphibian populations. Compared with the family Iridoviridae, to which the genus Ranavirus belongs, ranaviruses have a wide host range in that species/strains are known to infect fish, amphibians and reptiles, presumably due to recent host-switching events. We used eight sequenced ranavirus genomes and two selection-detection methods (site based and branch based) to identify genes that exhibited signatures of positive selection, potentially due to the selective pressures at play during host switching. We found evidence of positive selection acting on four genes via the site-based method, three of which were newly acquired genes unique to ranavirus genomes. Using the branch-based method, we identified eight additional candidate genes that exhibited signatures of d_N/d_S (non-synonymous/synonymous substitution rate) >1 in the clade where intense host switching had occurred. We found that these branch-specific patterns of elevated d_N/d_S were enriched in a small group of viral genes that have been acquired most recently in the ranavirus genome, compared with core genes that are shared among all members of the family Iridoviridae. Our results suggest that the group of newly acquired genes in the ranavirus genome may have undergone recent adaptive changes that have facilitated interspecies and interclass host switching.

Introduction

Since the 1980s, amphibian populations have increasingly declined on a global scale, and by 2004, at least 32 % of all amphibian species were categorized as threatened by the International Union for Conservation of Nature (Collins & Storfer, 2003; Stuart et al., 2004). In addition to several biotic and abiotic factors, two emerging pathogens, a chytrid fungus (Batrachochytrium dendrobatidis) and a virus (ranavirus), have been implicated in this ecological crisis (Daszak et al., 1999). The genus Ranavirus is a member of the family Iridoviridae, which consists of large dsDNA viruses that are currently divided into five genera (Williams et al., 2005). Members of two of these genera (Iridovirus and Chloriridovirus) infect invertebrates and members of three (Lymphocystivirus, Megalocytivirus and Ranavirus) infect heterothermic vertebrates. Members of the genera Lymphocystivirus and Megalocytivirus solely infect fish, but ranaviruses infect fish, amphibians and reptiles, presumably due to recent host-switching events (Jancovich et al., 2010). Direct contact and ingestion of contaminated water are known transmission modes, and infection leads to necrosis of the liver, spleen, skin and haematopoietic tissue (Lesbarrères et al., 2012). The bait, pet and food industries are major factors in the spread of these pathogens geographically to naïve localities and species (Jancovich et al., 2005; Picco & Collins, 2008). Although both pathogens pose a lethal threat to amphibians, ranaviruses tend to cause recurrent population die-offs that target larval-stage amphibians (Chinchar, 2002). Numerous ranavirus species/strains have contributed significantly to amphibian die-offs on five continents, and are known to infect over 70 amphibian species (Daszak et al., 1999; Miller et al., 2011).

Ranaviruses are emerging pathogens, meaning that amphibian populations have only recently been affected due to increased incidence, geography or host range (Rachowicz et al., 2005). There is significant evidence for host switching of ranaviruses between vertebrate classes (Jancovich et al., 2005, 2010; Bandín & Dopazo, 2011), as well as documented instances of simultaneous infections of sympatric fish and frogs with the same viral isolate (Mao et al., 1999; Bayley et al., 2013). In viral host switching, the virus of one species evolves such that it can replicate and spread in a new species. There is great interest in understanding this process, not just for conservation reasons but also because it underlies the emergence of new human diseases (Wain et al., 2007; Parrish et al., 2008; Choe et al., 2011; Demogines et al., 2012a, b, 2013). In general, host switching is thought to occur via a series of steps that involve both geographical movement of viruses (leading to exposure to new hosts) and, in most cases, the evolution of viral genomes for compatibility with new host species (Herfst et al., 2012; Kaelber et al., 2012). With regard to host switching in ranaviruses, the process probably began with the exposure of a novel amphibian host to a viral strain that originated in a fish host (Fig. 1). New mutations, including point mutations, insertions, deletions and even the acquisition or loss of whole genes, may have made a certain strain of the fish virus compatible with the new amphibian host species. Following initial infection, additional mutations or gene acquisitions may have increased viral fitness in a new host, thus allowing local transmission within species and between classes. Indeed, ranavirus genomes have diverged in sequence, and are known to have acquired several new genes that are not found in the broader family of the Iridoviridae (Eaton et al., 2007). These new genes may have been acquired via the processes of horizontal gene transfer or recombination. The presence of several virus-encoded cellular homologues in ranavirus genomes supports the acquisition of new viral genes by horizontal transfer from host genomes (Tidona & Darai, 2000). In addition, the presence of several ranavirus genes without corresponding cellular homologues, coupled with the high rate of recombination seen in these viruses, suggests that recombination is also a significant source of new genes (Eaton et al., 2007; Chinchar et al., 2009). The situation with ranaviruses may be analogous to the evolution of simian immunodeficiency viruses, where the continued acquisition and evolution of new viral genes led to expansion of these viruses to new primate hosts and ultimately to humans, giving rise to HIV (Gifford, 2011).

Fig. 1.

A model for ranavirus host switching within and between vertebrate classes. During cross-species transmission, evolution of a viral genome makes the virus increasingly compatible with the genetic background of new host species, and may facilitate interclass transmission. Mutations (black bars) and the acquisition of new genes (open rectangles) may work to progressively increase the compatibility between the ranavirus genome and new host species. Two separate scenarios that include point mutations occurring throughout the original and newly acquired regions of the genome can lead to transmission within species (2a) or between vertebrate classes (2b) over time. The acquisition of novel genes in the ranavirus genome and subsequent mutations within those genes most likely play a fundamental role in the ranavirus host-switching process.

The main factor limiting our ability to understand the broad host range of ranaviruses is the paucity of information regarding the viral genes that govern host specificity. Ranavirus genomes encode approximately 100–140 putative gene products, most of which have unknown functions (Grayfer et al., 2012). Computational tools that can detect episodes of rapid adaptation of specific genes, in conjunction with the increased availability of genomic data, have provided new avenues for identifying candidate genes potentially involved in dynamic processes such as host switching (Sawyer & Elde, 2012). In many cases, genes that are involved in host–pathogen interactions exhibit episodes of intense positive selection during the process of establishing a new host species (Shackelton et al., 2005; Hoelzer et al., 2008; Meyerson & Sawyer, 2011; Bhatt et al., 2013). Genes identified via these methods are often involved in immune evasion, replication, reproduction, gene expression, host–pathogen co-evolution and host defences (Endo et al., 1996; Yang et al., 2000; McLysaght et al., 2003; Harrison & Bonning, 2004; Shackelton et al., 2005; Sabeti et al., 2006; Kosiol et al., 2008; Elde et al., 2012). However, the absence of positive selection does not necessarily mean that genes are not key factors in host range and immune evasion, just as the presence of positive selection does not automatically assign genes to these roles.

Methods used to study adaptive molecular evolution typically involve the identification of gene regions or specific codons where the non-synonymous substitution rate (d_N) exceeds the synonymous substitution rate (d_S). This is expected only when non-synonymous mutations offer a selective advantage and become fixed at a higher rate than expected under neutral theory (Goldman & Yang, 1994). In practice, d_N/d_S>1, d_N/d_S = 1 and d_N/d_S<1 are inferred to reflect positive selection, neutral evolution and purifying selection, respectively. Two widely used selection-detection methods identify individual codon sites (site-based model) and specific branches along a phylogeny (branch-based model) that exhibit signatures of positive selection (Nielsen & Yang, 1998; Yang, 1998). The availability of several complete genomes of members of the family Iridoviridae from viruses infecting vertebrates provides the opportunity to assess the prevalence of positive selection within specific genes and across the ranavirus phylogeny.

Here, we describe a genomic-scale characterization of ranavirus evolution. The goals of this study were to detect viral genes that were under positive selection, particularly on branches in the phylogeny where potential host-switching events have occurred. Using eight fully sequenced ranavirus genomes, we analysed 46 genes for evolutionary signatures. We found evidence of positive selection acting on four genes via the site-based method, three of which were newly acquired genes unique to ranavirus. Using the branch-based method, we identified eight additional candidate genes that exhibited signatures of d_N/d_S>1 along at least one branch in the clade where intense host switching had occurred. We also found evidence of significantly higher d_N/d_S values for ranavirus-specific genes along individual branches when compared with core genes shared among all members of the family Iridoviridae. We concluded that adaptive evolution is occurring in newly acquired viral genes, and propose that both the acquisition and the subsequent evolution of these genes has been key in facilitating host switching of these viruses.

Results

In previous studies, the full-genome sequences of 15 vertebrate viruses of the family Iridoviridae have been determined (Table 1). Ten of these are ranaviruses, and represent viruses isolated from bony fish, salamander, frog and turtle species (Figs 2 and S1, available in JGV Online). These viral genomes vary somewhat in length and gene content, and contain approximately 100–140 genes. The genes examined in this analysis were identified previously as orthologues via a whole-genome comparative analysis (see Methods; Eaton et al., 2007). In total, we were able to align 75 orthologues from these genomes, and assigned each of these 75 genes to four groups depending on when each gene was acquired in the phylogenetic history of these viruses (Fig. 2). The ‘core’ gene group included 26 genes found in all 15 sequenced iridovirus genomes. The ‘RV’ (ranavirus) gene group comprised 23 genes specific to the ten ranavirus genomes but not found in the other five genomes. This group of genes was gained prior to the initial host switch from fish to amphibians (Jancovich et al., 2010). The ‘F/A/R’ (fish/amphibian/reptile) gene group included six genes found exclusively in the target clade, which is the monophyletic group of ranaviruses, excluding SGIV and GIV. The target clade included the group of viruses that has been newly transmitted beyond fish. This F/A/R gene group was particularly interesting, because these genes were specific to the clade of viruses that has experienced the most recent host switching. Lastly, the remaining 20 genes were assigned to the ‘other’ gene group, which included genes found in ranaviruses and at least one other iridovirus genus (i.e. Ranavirus and Chloriridovirus but no other genera).

Table 1. Fifteen genomes utilized in the present study and their host species.

Viral species/strain	Genus	Host species	GenBank accession no. (reference)
Frog virus 3 (FV3)	Ranavirus	Rana pipiens	AY548484 (Tan et al., 2004)
Soft-shelled turtle iridovirus (STIV)	Ranavirus	Trionyx sinensis	EU627010 (Huang et al., 2009)
Rana grylio virus (RGV)	Ranavirus	Rana grylio	JQ654586 (Lei et al., 2012)
Tiger frog virus (TFV)	Ranavirus	Rana tigrina	AF389451 (He et al., 2001)
Common midwife toad virus (CMTV)	Ranavirus	Alytes obstetricans	JQ231222 (Mavian et al., 2012a)
Ambystoma tigrinum virus (ATV)	Ranavirus	Ambystoma tigrinum	AY150217 (Jancovich et al., 2003)
Epizootic haematopoietic necrosis virus (EHNV)	Ranavirus	Oncorhynchus mykiss	FJ433873 (Jancovich et al., 2010)
European sheatfish virus (ESV)	Ranavirus	Silurus glanis	JQ724856 (Mavian et al., 2012b)
Singapore grouper iridovirus (SGIV)	Ranavirus	Epinephelus tauvina	AY521625 (Song et al., 2004)
Grouper iridovirus (GIV)	Ranavirus	Epinephelus awoara	AY666015 (Tsai et al., 2005)
Lymphocystis disease virus 1 (LCDV1)	Lymphocystivirus	Platichthys flesus	L63545 (Tidona & Darai, 1997)
Lymphocystis disease virus China (LCDVC)	Lymphocystivirus	Paralichthys olivaceus	AY380826 (Zhang et al., 2004)
Infectious spleen and kidney necrosis virus (ISKNV)	Megalocytivirus	Siniperca chuatsi	AF371960 (He et al., 2002)
Orange-spotted grouper virus (OSGIV)	Megalocytivirus	Epinephelus coioides	AY894343 (Lü et al., 2005)
Rock bream iridovirus (RBIV)	Megalocytivirus	Oplegnathus fasciatus	AY532606 (Do et al., 2004)

Fig. 2.

Phylogenetic relationships among vertebrate iridovirus genomes included in the study. A phylogenetic tree of viral species/strains isolated from fish, amphibians and reptiles based on a maximum-likelihood analysis of a concatenated dataset of 26 core iridovirus genes. The target clade within the ranaviruses is highlighted in black. Arrows indicate the origin of genes shared among all iridovirus genomes (core gene group; n = 26), among all ranavirus genomes (RV gene group; n = 23) and among the target clade (F/A/R gene group; n = 6). Tick marks indicate branches along which the initial host switch from fish to salamanders (point A) and the host switch to frogs and reptiles (point B) are thought to have occurred (Jancovich et al., 2010). Bootstrap values are shown. Note that each fish, amphibian and reptile symbol denotes multiple species as indicated in Table 1. Bootstrap values are indicated on the branch nodes.

Four genes under positive selection identified using site-based tests

We next analysed a subset (46/75 genes) of the individual gene alignments for patterns of non-synonymous and synonymous substitution using the NSsites codon models in paml 4 (Yang, 2007). The 29 genes excluded from the analysis failed to meet the outlined criteria based on sequence alignment and phylogenetic reconstruction (see Methods; Table S1). Of the 46 analysed, we identified four genes that exhibited signatures of positive selection under the NSsites models (Table 2). This group of genes included a single member of the core gene group (ORF53R), one gene from the RV gene group (ORF79R) and two genes from the F/A/R gene group (ORF40R and ORF71R). Evidence of selection acting on RV and F/A/R genes might be explained by the fact that these genes have been recently acquired and are unique to a single genus where host switching is evident. Relative to the newly acquired RV and F/A/R genes, evidence of selection acting on core genes was unexpected due to the highly conserved nature of this group of genes within the family Iridoviridae. The conservation of core genes may indicate that, in general, these genes are essential to the viral life cycle and that they are subject to purifying selection pressures.

Table 2. Four genes under positive selection identified using site-based tests.

Gene name	Gene group	M8a-M82Δl	P value (α = 0.05)	Tree length	Amino acid positions of dN/dS>1*†‡
ORF53R/myristylate membrane protein	Core	3.92	0.0478	0.16	387G (0.86)
ORF79R/putative ATPase-dependent protease	RV	5.19	0.0227	1.06	81L (0.88), 82A,† 84Q (0.89), 86L (0.81), 87V (0.83), 91S (0.88), 123L (0.85), 137R,† 140N (0.87), 146L (0.83)
ORF40R	F/A/R	4.74	0.0294	0.82	29M (0.84), 131Q,† 134A (0.88), 164T (0.83), 165V (0.81), 168L (0.83), 191T (0.88), 192P (0.86)
ORF71R	F/A/R	12.9	0.0003	0.38	19T,‡ 50M (0.94), 58I,‡ 70I†

^*Amino acid coordinates correspond to the position in the FV3 amino acid sequence. Values in parentheses refer to the PP of a site fitting into a d_N/d_S>1 class.

^†PP >0.95.

^‡PP >0.99.

In addition to identifying individual genes that have experienced positive selection, specific codons under positive selection were also identified (Table 2). At least one codon with a posterior probability (PP) of >0.80 was identified for each gene. To determine whether the positively selected sites fell into structurally flexible regions of the protein, we utilized the Phyre2 server to predict the secondary structure of the proteins encoded by the four genes that exhibited signatures of positive selection under the NSsites model. We then mapped the amino acid sites onto the secondary structure to determine whether any of the residues resided in areas of conformational flexibility, described as unstructured regions with high disorder (Fig. S2). We found that one out of one sites for ORF53R, ten out of ten sites for ORF79R, three out of eight sites ORF40R and one out of four sites for ORF71R fell into unstructured (neither α-helix nor β-strand) regions of the protein. Unstructured regions of proteins may be more tolerant to conformational and adaptive changes (Nilsson et al., 2011).

Eleven genes with d_N/d_S>1 identified using branch-based tests

Based on the evolutionary relationships of the viral strains and the species from which they were isolated, we can support hypotheses about where host-switching events have occurred in the ranavirus species phylogeny (Jancovich et al., 2010). For instance, we can support the hypothesis that the original host switch involved a jump from fish to salamanders (Fig. 2, point A). Moreover, we can also support a host switch from fish to frogs or from fish to frogs and turtles on the lineage leading to the (FV3, STIV, RGV, TFV and CMTV) clade (Fig. 2, point B). To examine evidence for accelerated gene evolution of any genes on these lineages, we next analysed branch-specific patterns of d_N/d_S for each of our 46 genes. Whereas the previous test for selection identified specific codons that had d_N/d_S>1, in this analysis each gene was assigned a global d_N/d_S value along every branch of the tree.

We identified 11 genes that exhibited signatures of d_N/d_S>1 along at least one branch under the free-ratio test. These included three genes from the RV and core gene groups (Fig. 3), four genes from the F/A/R gene group (Fig. 4) and four genes from the ‘other’ gene group (Fig. S3). In some cases, d_N/d_S values were much higher than 1, as for gene ORF40R on the branch leading to TFV, where d_N/d_S = 3.7 (Fig. 4b). In other cases, d_N/d_S values were closer to 1 and may reflect neutral evolution rather than positive selection. Likelihood ratio tests were performed to determine whether the likelihood values obtained with individual branches assigned d_N/d_S>1 were statistically greater than likelihood values obtained when the same branch was fixed at d_N/d_S = 1. Based on this analysis, on no branch was d_N/d_S significantly greater than 1; however, identification of genes with elevated d_N/d_S signatures was notable considering d_N/d_S was averaged across the entire gene. There is reason to believe that some of these signatures could still be meaningful, because three of these same genes (ORF79R, ORF40R and ORF71R) were also identified previously as evolving under positive selection in the site-based analysis described above (Table 2).

Fig. 3.

Three core and RV genes with elevated d_N/d_S values identified using branch-based tests. Phylogenies with free-ratio results for genes with branches exhibiting signatures of d_N/d_S>1 are presented. The branch values of d_N/d_S are shown, along with the estimated numbers of non-synonymous and synonymous (N : S) mutations that are predicted to have occurred along each branch (in parentheses). Branches with values of d_N/d_S>1 are shown in bold. Branches with incalculable d_N/d_S values (due to S = 0) are highlighted in bold if the ratio of N : S was greater than or equal to 4 : 0. Genes indicated with an asterisk were also found to exhibit signatures of positive selection under site-based tests (Table 2).

Fig. 4.

Four F/A/R genes with elevated d_N/d_S values identified using branch-based tests. Phylogenies with free-ratio results for genes with branches exhibiting signatures of d_N/d_S>1 are presented. Branch values of d_N/d_S are shown, along with the estimated numbers of non-synonymous and synonymous (N : S) mutations that are predicted to have occurred along each branch (in parentheses). Branches with d_N/d_S values >1 are shown in bold. Branches with incalculable d_N/d_S values (due to S = 0) are highlighted in bold if the ratio of N : S was greater than or equal to 4 : 0. Genes with an asterisk were also found to exhibit signatures of positive selection under site-based tests (Table 2). Note variation in the topologies of ORF71R and ORF93L due to slightly different gene trees.

Despite the fact that F/A/R represented the smallest gene group analysed, four of the six F/A/R genes analysed (ORF39R, ORF40R, ORF71R and ORF93L) exhibited d_N/d_S>1 on at least one branch (Fig. 4). When compared with the percentage of genes with d_N/d_S>1 branches for the core (5 %), RV (18 %) and ‘other’ (36 %) groups, genes from the F/A/R group (67 %) exhibit a relatively higher percentage of genes with elevated d_N/d_S values. Based on this result and the identification of two out of the six F/A/R genes via the site-based method (Table 2), we conducted an enrichment analysis to determine the probability that four randomly selected genes would belong to the F/A/R gene group. We calculated probabilities of P = 0.00009 when 46 genes were included and P = 0.00001 when 75 genes were included in the analysis. These results are consistent with selection in ranaviruses having been biased towards viral genes that are new to this clade.

Significant differences in median and mean d_N/d_S values of the core, RV and F/A/R gene groups detected using branch-based tests

We wished to examine further whether there were significant differences in branch d_N/d_S values among the core, RV and F/A/R gene groups. For each of the 13 branches in the target clade (the clade where host switching is pervasive; Fig. 2), median and mean d_N/d_S values for each gene group were compared. For each branch, we performed a Kruskal–Wallis (K-W) test (α = 0.05) to compare the median d_N/d_S values for each gene group, and a one-way ANOVA (α = 0.05) to compare the mean values. On four of the 13 branches, significant differences were noted (Fig. 5). We detected significant differences in the medians and means for branch B (K-W, P = 0.001; ANOVA, P = 0.0001) and the ATV branch (K-W, P = 0.0037; ANOVA, P = 0.0005) (Fig. 5b, d). In both cases, the median and mean d_N/d_S values were highest for the F/A/R gene group. However, whilst F/A/R d_N/d_S values were significantly different from the core gene group values, they did not differ significantly from the RV gene group based on post-hoc tests. We also detected a significant difference in the medians for the TFV branch (P = 0.0381) using the K-W test (Fig. 5c); however, the ANOVA was not significant (P>0.05). Lastly, branch A was marginally significantly different in the ANOVA analysis (P = 0.0446), but there was no significant difference in the medians based on the K-W test (P>0.05). Higher d_N/d_S values for the ranavirus-specific gene groups along branches A and B could indicate selection acting prior to a potential host switch between amphibians and reptiles, which was the most recent putative host-switching event examined in this study. Moreover, this interpretation might also apply to the ATV branch, which corresponded to selection following a host switch from fish to amphibians. The trend of higher mean and median values for the F/A/R gene group for three of the branches (Fig. 5b–d) is consistent with the idea that F/A/R genes are involved in ranavirus host switching.

Fig. 5.

Significant differences in median and mean d_N/d_S values of the core, RV and F/A/R gene groups detected using branch-based tests. The inset phylogenies show the significant branches in bold. Boxplots for statistically significant branches are shown (branches without significant support are not shown). K-W tests (medians) were performed on the transformed dataset (square root of d_N/d_S, raw data shown). One-way ANOVA tests (means) were performed on the transformed dataset (square root of d_N/d_S, raw data shown). An asterisk indicates that the mean of the F/A/R gene group was statistically different from the core gene group based on Bonferroni post-tests.

Discussion

The phylogeny of viruses in the family Iridoviridae, particularly in the ranavirus clade, is consistent with extensive host switching by these viruses (Fig. 2). The adaptation of a virus to a novel host species is often paired with intense episodes of positive selection acting on genes that increase the compatibility between the virus and host. We conducted a genome-wide scan for genes and lineages under positive selection to gain further insights into ranavirus host switching. We identified a total of 12 genes exhibiting signatures of d_N/d_S>1, and six of these were in a special category of genes, which have been newly acquired by these viruses (RV and F/A/R gene groups). We expected that genes in the RV and F/A/R gene groups may play a fundamental role in the adaptation of ranaviruses to new hosts, because the acquisition of these genes appears to have been coincident with the onset of host switching in this virus family (Jancovich et al., 2010). In support of this hypothesis, several approaches showed that d_N/d_S values tended to be elevated in these gene families.

Regarding the six ranavirus-specific genes that we identified as adaptively evolving, several aspects are notable. ORF79R (RV gene group) was the only gene to have d_N/d_S>1 on the branch that separates the two major ranavirus clades (Fig. 3c), which suggested gene evolution as the viruses spread to new frog species (Fig. 2). Thus, ORF79R may represent a gene that was fundamental to the initial host switch from fish to salamanders and possibly subsequent jumps. The identification of four genes specific to the F/A/R gene group with d_N/d_S>1 on at least one branch supports the hypotheses of intensified positive selection acting on these newer genes. Three of the four identified F/A/R genes are of unknown function and have no known homologues. It is possible that these genes encode novel virulence factors, which are typically proteins that counteract host-specific immune defences. For instance, examinations of genomes of members of the family Poxviridae and the order Herpesvirales have revealed the presence of virus-encoded homologues of cellular cytokines that have evolved to modulate the host immune response (Slobedman et al., 2009). In fact, the F/A/R gene ORF26R (α-subunit of eukaryotic initiation factor 2) is a virus-encoded cellular homologue (Eaton et al., 2007). Although no evidence of positive selection was detected in ORF26R, a recent study using a knockout approach showed that impairment of the gene in FV3 resulted in decreased virulence in vivo (Chen et al., 2011). We also found that mean and median d_N/d_S values for several branches in the ranavirus phylogeny were significantly higher for the F/A/R group when compared with the core gene group (Fig. 5). Considering this result and the low probability (P = 0.00009) of finding d_N/d_S>1 branches in four of the six F/A/R genes by random chance, we can conclude that F/A/R genes represent a group of newly acquired viral genes that have undergone adaptive changes that may have facilitated interspecies and interclass host switching (Fig. 1).

There are other examples where adaptive evolution of virus genomes has accompanied host-switching events. For instance, parvoviruses experienced adaptive evolution of the gene encoding the surface-exposed viral protein, the capsid, after this virus first became established in dogs in the 1970s (Hoelzer et al., 2008). These evolutionary changes improved the interaction between the capsid and the dog-encoded cell-surface receptor for virus entry, the transferrin receptor. Also, a recent study of the adaptive evolution of influenza viruses showed that elevated adaptive evolution was associated with the establishment of avian influenza viruses in swine populations (Bhatt et al., 2013). In this case, signatures of selection were located across the genome and so the specific host–virus interactions that drove this evolution are unknown. With regard to ranaviruses, this situation is unique from these other examples because the cross-species transmission of these viruses appears to have coincided with the addition of new genes to the viral genome, as well as their evolutionary refinement for infection of new host species.

Although we did not explicitly test the hypothetical model presented in Fig. 1, the relatively recent acquisition of new genes (RV and F/A/R genes) coupled with an excess of non-synonymous mutations in these genes supports the idea that gene gain followed by gene-specific mutation may influence ranavirus host switching. Moreover, evidence of simultaneous infections of sympatric fish and amphibian species with the same viral isolates (Mao et al., 1999) supports the idea that both interspecies and interclass transmission may be a common occurrence. However, additional studies of ranavirus spread in natural environments are needed to specifically test the model. Furthermore, estimates of viral divergence times and the timing of gene acquisitions are needed to fully understand the emergent nature of ranaviruses. Although humans have influenced the movement of aquatic species for millennia, it is more likely that recent human activities (globalized food, bait and pet trades) have enhanced the spread of these pathogens to naïve species. Unfortunately, as the rates of evolution in iridoviruses are currently unknown, uses of molecular clock methods are not possible (Ridenhour & Storfer, 2008). In the absence of these data, the genes identified in this study are prime candidates for knockout and site-directed mutagenesis analyses to determine gene function and the effect of the genes on viral host range in this unique genus.

Methods

Data collection and recombination analyses.

Sequences for 75 ORFs from the 15 complete vertebrate Iridoviridae genomes were downloaded from GenBank (Table 1). These ORFs were placed into specific gene groups based on the genomic analysis of Eaton et al. (2007), which identified orthologous genes throughout the family using a combination of blast searches and Viral Orthologous Clusters software (Eaton et al., 2007). Genes from viral genomes not analysed by Eaton et al. (2007) were identified via blast searches and analyses conducted by the authors of the published genome sequences (Huang et al., 2009; Jancovich et al., 2010; Lei et al., 2012; Mavian et al., 2012a, b). The first group, the core gene group, included 26 ORFs that have orthologues in all members of the family Iridovirdae (including the two invertebrate genera Iridovirus and Chlorididovirus). The second group, the RV gene group, included 23 ORFs that have orthologues only found in members of the genus Ranavirus. The third group, the F/A/R gene group, included six ORFs that have orthologues only found in the target clade, which included FV3, STIV, RGV, TFV, CMTV, ATV, ESV and EHNV species/strains. The remaining ORFs were identified as ‘other’ and included 20 ORFs that are shared between ranavirus species/strains and at least one other genus (including those that infect invertebrates).

The sequences for each ORF were aligned using clustal w (Larkin et al., 2007). Four ORFs with alignments that were judged to be unreliable were excluded from further analyses (Table S1). CodonTest (Delport et al., 2010) was then used to determine the best nucleotide substitution model for the remaining 71 alignments. The nucleotide substitution models were then used to calibrate the program gard (genetic algorithm for recombination detection) (Kosakovsky Pond et al., 2006), which was used to test for the presence of recombination in each ORF. A total of 25 ORFs with significant evidence of recombination were split based on the location of the recombination breakpoints, and each component was tested individually in subsequent analyses.

Phylogenetic reconstruction.

Phylogenetic trees were reconstructed for individual ORF alignments using the rapid bootstrap command in RAxML, which automatically applies the GTR+Γ model (500 bootstrap replicates were used to assess node support). Preliminary analyses of alignments that included sequences from the 15 members of the family Iridoviridae and ten ranavirus species/strains resulted in highly divergent phylogenies that fell outside the ideal range for the selection analyses (with some divergences of more than 20 nt substitutions per codon). To analyse divergences closer to 1 nt substitution per codon, which is a commonly used standard (Anisimova et al., 2001), only alignments using the eight sequences from the target clade were used in subsequent analyses. Please note that the apparent topological differences in the 15-taxon and the eight-taxon trees are due to the inclusion of additional data and the placement of the root. Fig. S4 illustrates that the relationship between ATV, and EHNV and ESV is consistent between the unrooted topologies. Moreover, the rooted topology seen in Fig. S4(b) can be recovered in Fig. S4(a) when the root of the 15-taxon tree is moved to match that of the eight-taxon tree. Thus, the phylogenetic relationships are consistent among the phylogenies.

The resulting eight-taxon phylogenies were accepted based on the criteria of a fully resolved topology and bootstrap support values that were ≥50 for the following clades: FV3, STIV and RGV; FV3, STIV, RGV, TFV and CMTV; and ATV, EHNV and ESV. A total of 25 ORFs that did not meet these requirements were excluded from the analyses, and the remaining core (18), RV (11), F/A/R (six) and ‘other’ (11) ORFs were further analysed. The resulting topologies for each ORF, without the inclusion of branch lengths, were then used as input topologies for the selection-detection analyses in the paml 4 (phylogenetic analysis by maximum likelihood) software package (Yang, 2007).

To determine the overall phylogenetic relationships of the vertebrate iridoviruses, a phylogenetic tree was reconstructed via maximum likelihood using a concatenated dataset of the 26 core group genes (Figs 1 and S1). The phylogeny was reconstructed using RAxML (Stamatakis, 2006) under the GTR+Γ model with the data partitioned by gene, and 500 bootstrap replicates were used to assess node support.

Detection of positive selection

Identifying selection at specific sites.

Site-based models were used to identify specific sites (codons) under positive selection via the NSsites models implemented in paml 4. The NSsites models allow d_N/d_S to vary among sites, and use likelihood ratio tests and Bayesian analyses to identify positively selected sites. Each test involves the comparison of a null model (M8a) with a model that incorporates positive selection (M8). M8a versus M8 models were run for each ORF, and the significance of the likelihood ratio scores was determined via the use of a χ² distribution (α = 0.05). The models were run under the f61 codon frequency models. Sites that fell into positively selected classes were then subjected to a Bayes empirical Bayes analysis, which calculates a PP of a site fitting into a d_N/d_S>1 class. The sites detected were reported as being under positive selection if they had a PP of >0.80 under the Bayes empirical Bayes method.

Identifying selection along branches.

Branch-based models seek to identify specific branches along a phylogeny that exhibit signatures of positive selection. To test for positive selection along specific branches in the phylogeny, the free-ratio model that is implemented in paml 4 was used to estimate branch-specific d_N/d_S ratios. The free-ratio model allows a unique d_N/d_S value to be calculated for each branch in the phylogeny. The free-ratio model was run for each ORF under the f61 codon frequency model to account for codon usage bias. To maximize topological consistency in our analyses, we constrained the topology of gene trees with alternative topologies to the topology supported by Fig. 2 (accepted topology). We then ran the branch-based analysis and used likelihood ratio tests to determine whether the results were significantly different between the two trees. The accepted topology is presented in cases where no significant difference was detected. Likelihood ratio tests were performed to determine whether the likelihood value for each branch that was assigned d_N/d_S>1 was statistically significant when compared with the likelihood value of the same branch when fixed at d_N/d_S = 1. The d_N/d_S of the target branch was fixed at 1, and the likelihood was calculated (null model). The null model was then compared with the alternative model, which fixed the target branch at the d_N/d_S value calculated under the free-ratio model.

To determine whether there were significant differences in d_N/d_S values among the core, RV and F/A/R gene groups, we compared both the median and mean d_N/d_S values for each group for each branch using K-W and ANOVA tests, respectively. The d_N/d_S values of several branches were undefined due to zero observed synonymous changes along the branch (d_N/0). However, branches with S = 0 are not uninformative with respect to selection; they are likely to be evolving near neutral if N is small, and under positive selection if N is large. Therefore, to include these branches in our analyses, we assigned a value of d_N/d_S = 0.95 (near neutral) for branches with observed N : S ratios greater than 1 : 0 and less than 4 : 0. Likewise, we assigned a value of d_N/d_S = 2.0 to branches with observed N : S values greater than 4 : 0 to account for likely positive selection. We also obtain consistent results where a d_N/d_S value of 1.5 is assigned, instead of 2.0, to this latter class. These assignments were consistent among all gene groups to keep the relative differences of the median and mean d_N/d_S values consistent for each dataset.

Protein structure prediction analyses.

The secondary protein structure for each of the ORFs found to be under positive selection under the NSsites method was determined using the Phyre2 server (Kelley & Sternber, 2009). The full-length FV3 amino acid sequence for each ORF was used as the input sequence for the analysis. The predicted domains, secondary structure and disorder were determined and a relative confidence value for each component was calculated. The positively selected sites identified by the NSsites methods were then mapped onto the predicted secondary structure of the protein.