Evolution of the Vertebrate Resistin Gene Family

Resistin (encoded by Retn) was previously identified in rodents as a hormone associated with diabetes; however human resistin is instead linked to inflammation. Resistin is a member of a small gene family that includes the resistin-like peptides (encoded by Retnl genes) in mammals. Genomic searches of available genome sequences of diverse vertebrates and phylogenetic analyses were conducted to determine the size and origin of the resistin-like gene family. Genes encoding peptides similar to resistin were found in Mammalia, Sauria, Amphibia, and Actinistia (coelacanth, a lobe-finned fish), but not in Aves or fish from Actinopterygii, Chondrichthyes, or Agnatha. Retnl originated by duplication and transposition from Retn on the early mammalian lineage after divergence of the platypus, but before the placental and marsupial mammal divergence. The resistin-like gene family illustrates an instance where the locus of origin of duplicated genes can be identified, with Retn continuing to reside at this location. Mammalian species typically have a single copy Retn gene, but are much more variable in their numbers of Retnl genes, ranging from 0 to 9. Since Retn is located at the locus of origin, thus likely retained the ancestral expression pattern, largely maintained its copy number, and did not display accelerated evolution, we suggest that it is more likely to have maintained an ancestral function, while Retnl, which transposed to a new location, displays accelerated evolution, and shows greater variability in gene number, including gene loss, likely evolved new, but potentially lineage-specific, functions.


Introduction
Resistin (Retn) was first identified in mice about 15 years ago, where it was identified as "adipose-tissue-specific secretory factor" (ADSF) [1] and "found in the inflammatory zone 3" (FIZZ3) [2], but acquired the name Resistin as it was associated with "resistance to insulin" [3].Evidence for a role of resistin in glucose metabolism and insulin resistance included the observation that circulating levels of resistin in the blood of mice was down-regulated by anti-diabetic drugs, and that experimental modification of circulating resistin levels affected blood glucose and insulin function [3].However, when human resistin (RETN) was characterized, it was found to be expressed in macrophages rather than adipose tissue [4], and the levels of circulating resistin are associated with inflammation rather than the amount of adipose tissue [5].Further studies over the past 10 years have confirmed that rodent and human resistin have distinct expression patterns, and while the circulating levels of resistin may be increased in diabetes, these increases are likely mediated by different mechanisms in each species [6][7][8].Currently the identity of the resistin receptor is unknown [9], although evidence has been presented that resistin can bind to three distinct receptors, an isoform of decorin [10], the ROR1 receptor [11], and the toll-like receptor 4 (TLR-4) [12].If resistin interacts with multiple receptors, this might lead to multiple physiological targets.In addition to diabetes, elevated levels of resistin are associated with other human diseases, including cardiovascular diseases such as arteriosclerosis and heart failure [8,13,14] and cancer [9].
During the initial characterization of resistin, additional resistin-like genes were identified, with two members found in mice (originally named Fizz1 and Fizz2, but also called Relmα and Relmβ (gene symbols: Retnla and Retnlb, respectively)) and one additional gene in humans (originally named FIZZ1, but also RELMβ (gene symbol: RETNLB)) [1,15].Here we use RETNL to refer to the mammalian resistin-like peptide genes (i.e., Retnla, Retnlb, and Retnlg), while Retn-like refers to duplicated Retn genes found in non-mammalian vertebrates.A phylogenomic study showed that the human and mouse Retn genes were orthologous, with human RETN residing on chromosome 19, while the rodent Retnl genes (Retnla and Retnlb) were clustered and resided on a genomic region orthologous to the human RETNLB gene on chromosome 3 [16].This study also identified a second genomic sequence in the human genome (originally named hRELM-homo seq) with limited similarity to the human RETNLB gene sequence; however, it appeared that this sequence was not a functional gene as splicing signals were missing and no evidence for expression was found [16].A latter study identified a third Retnl gene in rodent genomes (Relmγ, gene symbol: Retnlg), with this gene residing in the genomic region containing the Retnla and Retnlb genes [17].The biological roles of the Retnl peptides are not well characterized, but have been reported to be involved in pulmonary remodeling [18,19], hyperlipidemia and arteriosclerosis [20,21], and intestinal adaptive immunity [22,23].
Despite the likely role of RETN and RETNL in human health, little is known about these genes except in humans, mice, and rats [6][7][8][9].The evolutionary relationships among these genes are unresolved; except the orthology of the Retn genes [16].For example, the phylogenetic relationships among the multiple rodent Retnl genes and human RETNLB is unclear (e.g., are rodent and human Retnlb more closely related to each other than to the rodent Retnla and Retnlb genes).It is also unknown when the Retnl gene family originated.Intriguingly, Retn was lost from the chicken and zebra finch genomes, along with a number of other adipokine genes, which might explain some of the differences in energy metabolism, and its regulation, between birds and mammals [24].Here we investigated the evolution of Retn and Retnl genes in vertebrates and conclude that Retn is located and the locus of origin for the duplications and that Retnl originated via duplication and transposition to a new genomic location.In contrast to the Retn gene, which is generally single copy in mammals, duplication and inactivation frequently occurred during the evolutionary history of Retnl, which may reflect changing functions for the Retnl peptides.

Database searches
Resistin (RETN) and resistin-like (RETNL, e.g., RETNLB) coding and genomic sequences were identified in genomic databases maintained by the National Center for Biotechnology Information (NCBI, www.ncbi.nlm.nih.gov/projects/mapview/) and Ensembl (www.ensembl.organd www.pre.ensembl.org) in December 2014, using approaches we have used in previous studies [25,26,27].Initial sequences were identified through searches of the Ensembl database using the gene symbols for resistin (RETN, human resistin; Retn, mouse resistin) and the resistin-like peptides (Retnla, mouse resistin-like alpha; RETNLB, human resistin-like beta; Retnlb, mouse resistin-like beta; Retnlg, mouse resistin-like gamma).Additional sequences were then identified in the NCBI and Ensembl databases via similarity searches with the tblastn algorithm [28] using a variety of diverse resistin and resistin-like protein sequences as queries.All sequences that had E-scores below 0.01 were examined.Sequences identified by the BLAST searches were used in reciprocal blastx searches of the human and mouse proteomes to ensure that their best matches were resistin or resistin-like peptide sequences.To identify coding sequences for Retn and Retnl genes from species where the gene had not been, or was incompletely, annotated, long genomic sequences that included the Retn or Retnl genes were aligned to the human or mouse genomic sequences with MultiPipMaker (pipmaker.bx.psu.edu/pipmaker/)[29,30].Human and mouse sequences were used as masters for these alignments, with the locations of the exons and coding regions for these genes obtained from the Ensembl database.The identity and locations of repetitive elements in the human and mouse genomic sequences were identified using RepeatMasker [31].Genomic alignments were used to refine the predicted potential coding regions of the genes.Potential pseudogenes were identified as sequences that failed to predict an open reading frame due to the presence of base changes that introduced stop codons, created frame shifts that disrupted the coding sequence, or disrupted splicing consensus sequences.
Sequences similar to Retn and/or Retnl genes were not found in several species.For these species, genomic sequences for genes in the predicted conserved genomic neighborhood were included in this analysis, an approach that can be used to identify genes with limited sequence similarity [32,33].Genomes were searched for orthologs of the genes that flank the Retn or Retnl genes in diverse species (see results for details) and the genomic sequences adjacent to these genes were searched for similarity to Retn or Retnl genes.Genes were named (see S1 Table ) to reflect their orthology-paralogy relationships, based on the phylogenetic analysis (see below) and sequence similarity (for incomplete sequences and pseudogenes).Multiple genes within a species were numbered arbitrarily.

Phylogenetic analysis
Phylogenies of Retn and Retnl coding sequences were generated using full-length resistin and resistin-like sequences from diverse vertebrate species (see S1 Table and S1 Fig) .Retn and Retnl coding sequences were aligned using MAFFT [34] as implemented at the Guidance web server site (http://guidance.tau.ac.il/) [35], using default parameters.Similar results were obtained if ClustalW [36] was used as the alignment program.DNA sequence alignments were based on codons to retain protein alignments.The reliability of the alignments was examined using Guidance [35] and trimmed alignments using sites that had values above the default cut-off of 0.93 were generated.
Phylogenetic trees of the Retn and Retnl sequences were generated using Bayesian methods with MrBayes 3.2.2[37], maximum likelihood with PhyML 3.0 [38], and neighbor-joining distance approaches with MEGA6.0 [39].Bayesian trees were generated from the coding sequences using parameters selected by hierarchical likelihood ratio tests with ModelTest version 3.8 [40], as implemented on the FindModel server (www.hiv.lanl.gov/content/sequence/findmodel/findmodel.html).MrBayes was run for 2,000,000 generations with four simultaneous Metropolis-coupled Monte Carlo Markov chains sampled every 100 generations.The average standard deviation of the split frequencies dropped to less than 0.02 for all analyses.The first 25% of the trees were discarded as burn-in with the remaining samples used to generate the consensus trees.Trace files generated by MrBayes were examined by Tracer (tree.bio.ed.ac.uk/software/tracer/) to verify if they had converged.Bootstrapped maximum likelihood trees, 100 replications, were generated with PhyML [38] on the PhyML webserver (www.atgcmontpellier.fr/phyml/)using parameters for the substitution model suggested by ModelTest.Maximum likelihood searches was initiated from trees generated by BIONJ and the best tree was identified after heuristic searches using the nearest neighbor interchange (NNI) algorithm.MEGA6.0 [39] was used to construct bootstrapped (1000 replications) neighbor-joining distance trees, using either Maximum Composite Likelihood distances for the DNA sequences or JTT distances for the protein sequences.Similar results were obtained if alternative outgroups (e.g., sequences from Coelacanth or Sauria) were used (results not shown).
With respect to orthology-paralogy issues, choice of outgroup, alignment method (MAFFT [34] or Clustal W [36]), or the use of full-length or trimmed (based on Guidance scores [35]) alignments had little influence on the key findings of these analyses.Methods that relied on shorter sequences (i.e., trimmed alignments or protein sequences) or simpler models of sequence evolution (i.e., neighbor-joining) tended to yield weaker support for the earlier diverging lineages, but none of our analyses were in significant conflict with the key inferences of our inferred phylogenies.

Distribution of Retn and Retnl genes
Previously characterized Retn and Retnl genes contain 3 coding exons (see S2 Fig) , with introns interrupting the coding region at homologous locations [41].All of the Retn and Retnl genes identified here (S1 Table ) have similar gene structures.Some differences in the exon organization of the untranslated regions of Retn genes have been reported [42], however, we did not attempt to identify untranslated exons in the Retn and Retnl genes, as these sequences typically evolve faster than coding sequence, thus are more difficult to detect using sequence conservation.Our searches of the genomes of 116 vertebrate species that were available in the Ensembl and NCBI genome databases in December 2014 resulted in the identification of 206 Retn or Retnl genes from a total of 84 species, of which 136 (from 78 species) possessed intact coding sequences that encoded potentially functional proteins (Table 1, S1 amino acids, except the Chinese alligator Retn1, see below) with most of the cysteine residues in these cysteine-rich peptide sequences being at conserved locations.A total of 76 mammalian (Class Mammalia) genomes were available in the Ensembl or NCBI databases, and at least one sequence that encoded a Retn or Retnl sequence was found in 75 of these species (Table 1 and S1 Table).The only mammalian species where we failed to identify a sequence with similarity to Retn or Retnl was the sloth (S1 Table ), a species with a low coverage (2.05X) genome assembly, thus it is possible that Retn and/or Retnl sequences exist in gaps in this genome sequence.A total of 185 putative Retn or Retnl genes were found in the 75 mammalian species that had at least one gene, with most species having two or more sequences.Of the 185 genomic sequences, 128 predict full-length coding sequences (representing 71 species), 28 are predicted to be pseudogenes as they contain mutations that prevent formation of an intact protein (see below for more details), and 29 are incomplete genes, where part of the gene sequence was missing (Table 1 and S1 Table).Most of the incomplete genes are from species with lower sequence coverage, thus the missing sequences potentially are in sequence gaps and therefore we expect that many of these genes are functional.
Distinct Retn and Retnl genes were found only in placental and marsupial mammals, and not in the platypus, the only monotreme mammal examined (S1 Table ).Within placental and marsupial mammals, a total of 69 Retn genes were identified from a total of 67 species, with two closely related species (alpaca and camel) each having two Retn genes (S1 Table ).We failed to find a Retn gene in 8 mammals (S1 Table ).Among the 67 species that have Retn genes, 61 have intact coding sequences and 6 were incomplete.Retn pseudogenes were only found in the camel and alpaca, which are closely related (S1 Table ).In both of these species an intact Retn gene was also found with the Retn pseudogenes being processed pseudogenes (see S3 Fig) .In contrast to Retn, the number of Retnl genes in the genomes of placental and marsupial species varied from 0 genes (in 9 different species) to 9 genes in the prairie deer mouse (S1 Table ).A total of 27 Retnl pseudogenes, from 26 species, were identified, with 13 of these coming from 12 species that did not otherwise have a potentially functional Retnl gene.Pseudogenes were generated by mutations that introduced in frame stop codons in 5 of these 12 species, including the marmoset that has two near identical Retnl pseudogenes that share the same inactivating in frame stop mutation (see S1 Table ).Retnl genes from all 6 ruminant artiodactyls examined (cow, yak, river buffalo, chiru, sheep, and goat) shared the same disrupting mutation, the insertion of a Bov-tA2 repetitive DNA element into exon 1 (see S4 Fig) .The last species that only had a Retnl pseudogene is the gibbon.While the gibbon was missing an intact Retnl gene, it did share the processed pseudogene (homologous to the hRELM-homo seq from humans [16]) that was inserted downstream of the Retnl gene in an early primate lineage (see S1 Retn-like genes were found in the genomes of the 7 non-avian species of clade Sauria (reptiles and birds) that have available genome sequences, with the number of genes ranging from 1 in the Chinese softshell turtle, to 4 in the Chinese and American alligators and the green sea turtle (Table 1 and S1 Table).Of the 7 species, intact Retn-like coding sequences were found in 6 of them.The two genes identified in the Burmese python, a species with a low coverage genome, were incomplete (S1 Table ).Most of the Retn-like genes identified in these species were incomplete, and suggested that they were partial gene duplicates.Since most of the Retn-like sequences within each genome in these genomes were most similar to each other, this suggests that the majority of them were recent lineage-specific duplicates.A Retn-like gene, Retn2 from the American alligator, contained a base change that produces a pseudogene (S1 Table ).Intriguingly, the only Retn-like gene with an altered gene structure was found in the Chinese alligator, where a duplicated exon 2 sequence was found within the Retn2 gene, which results in a larger (but in frame) predicted protein product of 138 amino acid residues (see S2 Fig).
In contrast to the above, Retn-like sequences were not found in the 8 bird (Class Aves, within clade Sauria) genomes examined (Table 1 and S1 Table).This observation is consistent with a recent report of the loss of adipokine genes, including resistin, from the chicken and zebra finch genomes [24].A single Retn-like gene was found in the one available amphibian (Class Amphibia) genome (Xenopus tropicalis), although it was incomplete as it was located at one end of a short genomic contig (Table 1 and S1 Table).The genome of the coelacanth (Subclass Actinistia), a lobe-finned fish that is more closely related to tetrapods (Superclass Tetrapoda) than to other fish [43,44], contained a single intact Retn-like gene (Table 1 and S1 Table).Searches of the genomes of 17 ray-finned fish (Class Actinopterygii), 1 cartilaginous fish (Class Chondrichthyes), and 1 jawless fish (Superclass Agnatha) (Table 1 and S1 Table) failed to identity any sequences similar to Retn, although this may simply reflect the sensitivity of the searches as the coelacanth sequence showed limited similarity.

Phylogenetic relationships of Retn and Retnl genes
As described above, multiple Retn and/or Retnl gene sequences were identified in many vertebrate species (see S1 Table ).To better understand the relationships among these genes, a phylogeny of these sequences was estimated (Fig 2).To obtain a phylogenetic tree with maximal support, only intact Retn and Retnl sequences (see S1 Fig) , and codons that could be confidently aligned, were used.An alignment of 136 intact Retn and Retnl sequences was generated and trimmed using Guidance [35], and the phylogenetic relationships among the sequences was assessed using multiple approaches.The phylogeny inferred by the Bayesian approach using MrBayes 3.2 is shown in Fig 2 .Similar phylogenies were obtained using other methods,  ).

The mammalian Retn gene resides at the locus of origin for the gene family
The phylogenetic analysis of the Retn and Retnl sequences ( Fig 2 and S6 Fig) showed that there were two types of Retn-like genes in mammals, Retn and Retnl (which include the previously named Retnla, Retnlb, and Retnlg genes).While the divergence of the Retn and Retnl genes occurred prior to the divergence of placental mammals from marsupials, the Bayesian and maximum likelihood analyses yielded different conclusion concerning the existence of a Retn ortholog in marsupial mammals.The Bayesian analysis suggests that both Retn and Retnl exist in marsupials; while the maximum likelihood analysis suggests that marsupials have ancient diverging Retnl genes and that the ortholog of the Retn gene was lost.To resolve this difference we examined the genomic neighborhoods of the Retn and Retnl genes.
The human RETN and RETNL genes reside on different chromosomes with a third RETNlike sequence, a truncated processed pseudogene (see below for description), adjacent to RETNL (see S1 Table ).The STXBP2 and MCEMP1 genes flank the human RETN gene on chromosome 19 (Fig 3).Putative RETN orthologs were inferred from the phylogenetic analysis shown in Fig 2 and their flanking genes were identified (Fig 3).As previously reported [24], and shown in Fig 3, many genes near the cow and mouse Retn genes are orthologous to the human genes flanking RETN, with similar results found for several other placental mammalian species that have well assembled genomes (results not shown).The putative opossum Retn ortholog was also found to be in a genomic neighborhood homologous to that of human RETN (Fig 3), demonstrating that the human and opossum Retn genes are indeed orthologous.A similar analysis of putative mammalian RETNL orthologs, including the opossum, yielded similar results, with orthologs inferred by the phylogenetic analysis shown in  Genomic organization of genes near Retn genes of representative vertebrate species.Retn genes are labeled in red.Genes that share genomic location with human genes are labeled in yellow, while genes labeled in green are either lineage-specific genes or are found at a different genomic location in the human genome (genes without name do not have a human ortholog).The painted turtle genomic neighborhood is composed of two scaffolds that are likely adjacent.Chromosome, genomic scaffold, or sequence accession numbers, with approximate coordinates and size, of the displayed fragment are shown.See S1 Table for details on genomic locations of Retn genes.Gene sizes and distances between genes are not to scale.Arrowheads indicate direction of transcription.Gene symbols are: Retn, resistin; Xab2, XPA binding protein 2; Pet100, PET100 homolog; Pcp2, Purkinje cell protein 2; Stxbp2, Syntaxin binding protein 2; Mcemp1, Mast cell-expressed membrane protein 1; Trappc5, Trafficking protein particle complex 5; Fcer2, Fc fragment of IgE, low affinity II, receptor for (CD23); Clec4g, C-type lectin domain family 4, member G; Kiss1r, KISS1 receptor; Jrk, Jerky; Pglyrp6, Peptidoglycan recognition protein 6; Rasal3, RAS protein activator like 3; Emr-like, Egf-like module containing, mucin-like, hormone receptor-like; Wiz, Widely interspaced zinc finger motifs; and Fam102b, Family with sequence similarity 102, member B. doi:10.1371/journal.pone.0130188.g003 Since the mammalian Retn and Retnl reside on different chromosomes, it is possible that the genomic neighborhoods may also help determine whether one of these two genes is located at the locus of origin.Retn-like genes in species that do not have both a Retn and a Retnl gene should have the gene at the original genomic locus of origin.Our phylogenetic analysis indicated that the platypus Retn gene diverged prior to the Retn-Retnl duplication event (Fig 2  Genomic organization of genes near Retnl genes of representative vertebrate species.Retnl genes are labeled in red.Genes that share genomic location with human genes are labeled in black, while genes labeled in green are either lineage-specific genes or are found at a different genomic location in the human genome (genes without names do not have a human ortholog).Chromosome, genomic scaffold, or sequence accession numbers, with approximate coordinates and size, of the displayed fragment is shown.See S1 Table for details on genomic locations of Retnl genes.Gene sizes and distances between genes are not to scale.Arrowheads indicate direction of transcription.Gene symbols are: Retnl, resistin-like; Morc1, MORC family CWtype zinc finger 1; Guca1c, Guanylate cyclase activator 1C; Trat1, T cell receptor associated transmembrane adaptor 1; Dzip3, DAZ interacting zinc finger protein 3; Kiaa1524, KIAA1524; Myh15, Myosin, heavy chain 15; Hhla2, HERV-H LTR-associating 2; Hjurp, Holliday junction recognition protein; Sh2d1b, SH2 domain containing 1B; Ift57, Intraflagellar transport 57; and Mrps23, Mitochondrial ribosomal protein S23. .The Xenopus Retn gene was on a short genomic contig (<11 kb) that did not contain any additional genes thus provided no information on the ancestral gene order.Gene order near the Retn gene in coelacanth differed greatly from that of other species with Retn genes, with only the Trappc5 being in the same genomic neighborhood (Fig 3).Similarly, a conserved genomic neighborhood could not be found in other fish, as the closely linked genes tended to be dispersed among different chromosomes (e.g., in zebrafish: Stxbp2, Chromosome 3; Trappc1, Chromosome 1; Kiss1r, Chromosome 5).Our analyses indicates that a conserved genomic neighborhood is shared between mammals and species of clade Sauria, with the neighborhood surrounding the non-mammalian Retn-like genes supporting the conclusion that the mammalian Retn gene is located at locus of origin and that Retnl was transposed to a new genomic location (see Fig 5 for evolutionary model).

Evolution of Retn and Retnl gene function
Since Retnl originated from Retn via duplication and transposition to a new genomic location, while Retn remained in the locus of origin, raises the possibility that Retn retained an ancestral function while Retnl gained new function (Fig 5).Since Retnl transposed to a new genomic location, its expression pattern may have changed due to the loss of regulatory elements that were not duplicated with the gene or due to the gain of new regulatory elements from its new genomic neighborhood (see Fig 5).A change in the site, time, or level of expression of the gene provides opportunities to change gene function.If Retnl evolved a new function, then one might expect an increase in the rate of Retnl sequence evolution after the initial gene duplication event, as selection should have been relaxed on this redundant copy [47,48].Intriguingly, the phylogenetic trees of the Retn and Retnl sequences suggest that the lineage leading from the common ancestor of the Retn and Retnl genes (i.e., duplication event) to the common ancestor of marsupial and placental Retnl genes is longer than the lineage form the common ancestor of Retn and Retnl genes and that common ancestor of marsupial and placental Retn genes (see Fig 2 and S6 Fig).Since both lineages represent the same amount of time (gene duplication to species divergence), the difference in branch lengths suggests more rapid evolution on the Retnl lineage, which could be due to relaxed selection [47,48].
Additional evidence for Retnl evolving new roles is based on the variability of the copy number of the Retnl gene in diverse mammalian genomes (see Fig 5).As mentioned above, searches of mammalian genomes for Retn and Retnl genes found that Retn is typically found as a single copy gene, suggesting that maintaining a single copy is essential.In contrast, the number of Retnl genes is much more was variable (S1 Table ).While we failed to find Retnl genes in 9 mammalian species (S1 Table ), some of these may have genes that exist in gaps in their genome assemblies.In addition to these 9 species, 12 other mammalian species (including species within primates, rodents, artiodactyls, and bats, see S1 Table ) only have Retnl pseudogenes, which suggests that this gene has been lost in multiple species, and thus is not essential.In rodents, however, multiple intact Retnl genes were found in many species (S1 Table ).The multiple rodent Retnl genes originated on the rodent lineage (see  ), suggesting they were generated by tandem gene duplication.Initial duplication of the Retnl gene in rodents must have been early in rodent evolution, with multiple subsequent duplications, and their phylogenies (S6 and S7 Figs) suggest the possibility of concerted evolution [49,50].Maintenance of divergent Retnl genes (e.g., Retnla and Retnlg in mice and rats) may suggest that duplicate Retnl genes have differing functions, consistent with these genes adapting to new, and potentially lineage-specific, Model for the evolution of the resistin and resistin-like genes.Retn and Retnl genes are indicated by the red arrows, with the arrow pointing in the direction of transcription.Other genes are shown in yellow (locus of origin) or blue (location of inserted Retnl gene).Curved arrows indicated gene duplications that were either a transposition (to generate Retnl) or tandem, on the rodent and primate lineages.X's indicate inactivating mutations in the primate and artiodactyl genes that generate pseudogenes.(A).In the ancestor to mammals, Retn was located in the locus of origin.(B).On an early mammalian lineage, prior to the placental mammal-marsupial divergence, a copy of Retn was transposed to a new genomic location to generate the Retnl gene.The transposition likely allowed Retnl to acquire a novel expression pattern.(C).Retn and Retnl genes have different fates on divergent mammalian lineages.While Retn remained as a single copy intact gene on different mammalian lineages, Retnl had different fates, raising the possibility that it acquired lineage-specific functions.Retnl remained as a single copy gene in perissodactyls, duplicated intact (potentially functional) copies in rodents, duplicated and generated a pseudogene in primates, or was inactivated in artiodactyls.[6,7,8].The more rapid evolution of rodent Retn sequences imply that rodent resistin, rather than human resistin, changed its function, thus the role of resistin in maintaining blood glucose levels and diabetes [3] may be a rodent-specific trait.The ancestral role for resistin therefore is likely among its other characterized functions in the cardiovascular system [8,13,14] or in cancer [9].

Conclusions
Our characterization of Retn and Retnl genes in diverse vertebrates allowed us to develop a model for the evolution of this gene family (Fig 5).Retn is located at the locus of origin for the duplicated genes, while the Retnl gene was transposed to a new genomic location during, or soon after, duplication.Since Retn is located at the locus or origin, it likely retained the ancestral expression pattern.This gene has also been maintained as a single copy gene in almost all mammals examined, suggesting that the function of this gene has not changed.In contrast, Retnl transposed to a new genomic location, displays evidence of accelerated sequence evolution after duplication, and has been retained in variable copy number on different mammalian lineages.The changes in genomic location, rate of sequence evolution, and copy number suggest that Retnl has evolved new, and possibly lineage-specific, functions.Our analyses also suggest that rodent Retn evolved more rapidly than Retn genes in other mammals, thus the difference in function of Retn in diabetes between mouse and humans likely represents the gain of a new function in rodents, and that rodents will not be a good model for understanding the function of the human RETN and RETNL genes.

Fig 1 .
Fig 1.Alignment of Retn and Retnl protein sequences.Alignment of predicted resistin and resistin-like protein sequences from human, mouse, opossum, Anole lizard, and coelacanth (lobe-finned fish).The human Retn sequence is shown at the top in single letter code.Dots in the alignments represent identity to the human Retn sequence, with differences indicated in single letter code.Asterisks below the alignment identify residues that are perfectly conserved among all of the selected sequences.doi:10.1371/journal.pone.0130188.g001

Fig 2 .
Fig 2. Phylogeny of Retn and Retnl genes.Phylogeny inferred by the Bayesian method for 136 Retn and Retnl sequences from diverse vertebrates.The phylogeny was rooted with the Coelacanth Retn sequence.Similar phylogenies were obtained if sequences from Sauria (coelacanth sequence not used) were used as the outgroup or if other methods were used (e.g., see S6 Fig).Numbers at the nodes indicate posterior probabilities, with those for the nodes in early mammalian evolution shown in bold.Branch lengths are proportional to the inferred amount of change, with the scale bar at the bottom.Diamonds indicate inferred gene duplication events.Retnl genes are shown in the upper part of the tree while Retn genes are below.doi:10.1371/journal.pone.0130188.g002 Fig 2 and S6 Fig being found in conserved genomic neighborhoods (Fig 4).These results indicate that the gene duplication that generated Retn and Retnl occurred prior to the divergence of placental and marsupial mammals.

Fig 3 .
Fig 3.  Genomic organization of genes near Retn genes of representative vertebrate species.Retn genes are labeled in red.Genes that share genomic location with human genes are labeled in yellow, while genes labeled in green are either lineage-specific genes or are found at a different genomic location in the human genome (genes without name do not have a human ortholog).The painted turtle genomic neighborhood is composed of two scaffolds that are likely adjacent.Chromosome, genomic scaffold, or sequence accession numbers, with approximate coordinates and size, of the displayed fragment are shown.See S1 Table for details on genomic locations of Retn genes.Gene sizes and distances between genes are not to scale.Arrowheads indicate direction of transcription.Gene symbols are: Retn, resistin; Xab2, XPA binding protein 2; Pet100, PET100 homolog; Pcp2, Purkinje cell protein 2; Stxbp2, Syntaxin binding protein 2; Mcemp1, Mast cell-expressed membrane protein 1; Trappc5, Trafficking protein particle complex 5; Fcer2, Fc fragment of IgE, low affinity II, receptor for (CD23); Clec4g, C-type lectin domain family 4, member G; Kiss1r, KISS1 receptor; Jrk, Jerky; Pglyrp6, Peptidoglycan recognition protein 6; Rasal3, RAS protein activator like 3; Emr-like, Egf-like module containing, mucin-like, hormone receptor-like; Wiz, Widely interspaced zinc finger motifs; and Fam102b, Family with sequence similarity 102, member B.

Fig 4 .
Fig 4.  Genomic organization of genes near Retnl genes of representative vertebrate species.Retnl genes are labeled in red.Genes that share genomic location with human genes are labeled in black, while genes labeled in green are either lineage-specific genes or are found at a different genomic location in the human genome (genes without names do not have a human ortholog).Chromosome, genomic scaffold, or sequence accession numbers, with approximate coordinates and size, of the displayed fragment is shown.See S1 Table for details on genomic locations of Retnl genes.Gene sizes and distances between genes are not to scale.Arrowheads indicate direction of transcription.Gene symbols are: Retnl, resistin-like; Morc1, MORC family CWtype zinc finger 1; Guca1c, Guanylate cyclase activator 1C; Trat1, T cell receptor associated transmembrane adaptor 1; Dzip3, DAZ interacting zinc finger protein 3; Kiaa1524, KIAA1524; Myh15, Myosin, heavy chain 15; Hhla2, HERV-H LTR-associating 2; Hjurp, Holliday junction recognition protein; Sh2d1b, SH2 domain containing 1B; Ift57, Intraflagellar transport 57; and Mrps23, Mitochondrial ribosomal protein S23.

doi: 10 .
Fig), and thus could be in the locus of origin.The platypus Retn gene is located on a short genomic contig (24kb) that also contains the Mcemp1 gene, suggesting that the genomic neighborhood is similar to that of Retn in other mammals (Fig 3).A platypus genomic neighborhood homologous to the Retnl genomic neighborhood was identified that lacks a Retnl gene (Fig 4).These results suggest that the locus of origin was the Retn genomic neighborhood and that during duplication of this gene yielded a daughter Retnl gene that transposed to a new genomic neighborhood.However, since both posterior probabilities (Fig 2) and bootstrapping (S6 Fig) provide good, but not excellent, support for the outgroup placement of the platypus Retn gene sequence, we cannot exclude the possibility that the gene duplication occurred earlier in mammalian evolution, with the platypus subsequently losing the Retnl gene.To exclude this possibility, we examined additional outgroup species also have a Retn gene, and have a divergence prior to the Retn-Retnl duplication that is very strongly supported (Fig 2 and S6 Fig).Analysis of the genomic neighborhoods surrounding Retn genes in species from clade Sauria shows that many genes are shared with mammalian Retn gene neighborhoods (Fig 3), and that the genomic neighborhoods homologous to the mammalian Retnl gene neighborhood lack Retn-like genes (Fig 4).The Xenopus Retn gene was on a short genomic contig (<11 kb) that did not contain any additional genes thus provided no information on the ancestral gene order.Gene order near the Retn gene in coelacanth differed greatly from that of other species with Retn genes, with only the Trappc5 being in the same genomic neighborhood (Fig3).Similarly, a conserved genomic neighborhood could not be found in other fish, as the closely linked genes tended to be dispersed among different chromosomes (e.g., in zebrafish: Stxbp2, Chromosome 3; Trappc1, Chromosome 1; Kiss1r, Chromosome 5).Our analyses indicates that a conserved genomic neighborhood is shared between mammals and species of clade Sauria, with the neighborhood surrounding the non-mammalian Retn-like genes supporting the conclusion that the mammalian Retn gene is located at locus of origin and that Retnl was transposed to a new genomic location (see Fig5for evolutionary model).

Fig 2 ,
S6 and S7 Figs) and were found as a cluster of duplicated genes (see Fig 4, S8 Fig and S1 Table

Fig 5 .
Fig 5.  Model for the evolution of the resistin and resistin-like genes.Retn and Retnl genes are indicated by the red arrows, with the arrow pointing in the direction of transcription.Other genes are shown in yellow (locus of origin) or blue (location of inserted Retnl gene).Curved arrows indicated gene duplications that were either a transposition (to generate Retnl) or tandem, on the rodent and primate lineages.X's indicate inactivating mutations in the primate and artiodactyl genes that generate pseudogenes.(A).In the ancestor to mammals, Retn was located in the locus of origin.(B).On an early mammalian lineage, prior to the placental mammal-marsupial divergence, a copy of Retn was transposed to a new genomic location to generate the Retnl gene.The transposition likely allowed Retnl to acquire a novel expression pattern.(C).Retn and Retnl genes have different fates on divergent mammalian lineages.While Retn remained as a single copy intact gene on different mammalian lineages, Retnl had different fates, raising the possibility that it acquired lineage-specific functions.Retnl remained as a single copy gene in perissodactyls, duplicated intact (potentially functional) copies in rodents, duplicated and generated a pseudogene in primates, or was inactivated in artiodactyls.
doi:10.1371/journal.pone.0130188.g005functions.Rodent Retnl gene sequences also have longer branch lengths in the phylogenies (Fig 2 and S6 Fig), implying that they have accumulated a larger number of changes.This would also suggest that the rodent Retnl genes have experienced relaxed selection or adaptive evolution, both of which would be consistent with lineage-specific changes in gene function.Intriguingly, among the Retn sequences identified in this study, the rodent Retn sequences show the greatest amount of change (see Fig 2 and S6 Fig).Previous studies have shown that human and rodent Retn are expressed in different places and appear to have differing roles in diabetes

S8
Fig. Genomic organization of genes near rodent Retnl genes.(PDF) S1 Table.Genomic locations of Retn and Retnl genes in vertebrate genomes.(PDF)

Table 1 .
Table and S1 Fig).As illustrated in Fig 1, all of the coding regions predict proteins of similar length (approximately 110 Numbers of Retn and Retnl genes found in groups of vertebrate species.
1Number of species examined / number of species with a Retn or Retnl gene / number of species with at least one intact Retn or Retnl coding sequence for each group.2Genomicsequencesthatfail to predict part of a Retn or Retnl gene sequence, possibly due to a gap in the genome assembly.3Sequencecontains mutations that cause a frame shift or introduce premature stop codons in the Retn or Retnl coding sequence.doi:10.1371/journal.pone.0130188.t001 Table and S5 Fig).It is possible that an intact Retnl gene exits in a gap in the gibbon genome.