Genome-wide identification of F-box genes in eight genomes

The hidden Markov model (HMM) profile of the F-box domain (PF00646) was downloaded from Pfam [16]. All sequence data were downloaded from ENSEMBL (version 69, October 2012) [17]. The downloaded HMM profile was used to search the entire set of annotated proteins from eight species, namely, Callithrix jacchus (marmoset), Gorilla gorilla (gorilla), Homo sapiens (human), Macaca mulatta (macaque), Mus musculus (mouse), Pan troglodytes (chimpanzee), Pongo abelii (orangutan), and Rattus norvegicus (rat) using the hmmsearch program implemented in the HMMER package [18]. We used the default cut-off values to filter the results of these queries. HMMER-predicted proteins were then scanned for F-box and other domains using InterProScan, which is a tool that integrates multiple signature-recognition methods into one resource [19], [20].

For each F-box domain-containing protein identified by the hmmsearch and InterProsScan programs, additional PSI-BLAST [21] searches (with an e-value cut-off of 1e-20) were performed against the entire set of annotated proteins to identify additional F-box proteins that were not found using the HMM profile because of their diverged F-box domains. A second scan for the F-box domain was performed for PSI-BLAST hits.

In order to detect pseudogenes related to F-box genes, all of the F-box protein sequences retrieved from the hmmsearch and PSI-BLAST programs were used as queries to perform a TBLASTN search of the entire set of annotated pseudogenes of each species with an e-value cut-off of 1e-40. Finally, TBLASTN hits were translated and then scanned for presence of F-box domains.

The genomic distribution of identified F-box genes was evaluated by comparing the observed number of genes in each chromosome with its expected number under a Poisson distribution.

Phylogenetic analysis

Multiple sequence alignments of the F-box protein sequences, which are shown in File S1, were generated using MUSCLE [22] and then manually checked and trimmed with TRIMAL 1.2 (gt = 0.3) [23]. Subsequently, following the Akaike Information Criterion (AIC) computed with ProtTest 3.2 [24], the JTT+F model was chosen to construct a maximum likelihood (ML) tree with PhyML 3.0 [25]. Topological robustness of the phylogenetic tree was assessed by bootstrapping with 100 replicates.

Orthology assignments of F-box genes from the eight species were downloaded from the ENSEMBL database using Biomart. Orthologous groups (also known as orthogroups) were inferred based on phylogenetic relationships and confirmed by reciprocal BLAST. Subsequently, homology relationships of F-box genes in the same orthogroup were checked against the data obtained from ENSEMBL. We nominated human and mouse F-box genes according to the nomenclature proposed by Jin et al. [26]. F-box genes in the other six species considered in this study had the same names as their orthologs in humans.

Inference of gene-gain and gene-loss events and their underlying mechanisms

Gene-gain and gene-loss events were inferred using the species/gene tree reconciliation approach with NOTUNG software [27]. For these analyses, the reference species tree used was reconstructed according to TIMETREE [28]. The reconciled tree was manually adjusted by applying information on orthologs (from outgroups Danio rerio and Gallus gallus), because the gene tree may not always be reliable for species/gene tree reconciliation.

We investigated several potential underlying mechanisms responsible for gene-gain events inferred from previous estimations. First, we explored whether retrotransposition may have contributed to such duplications. The nucleotide sequences of duplicated genes were inspected for signatures of retrosequences, such as lack of introns, stretches of poly (A) at the 3′ end, and short direct repeats at both ends. Second, if the duplicated gene was not found to have been generated by retrotransposition, we considered two types of segmental duplications (>90% identity and >1 kb in length) [29]: (i) tandem duplication, where the two genes are located within the same chromosomal region (i.e., fewer than 20 genes apart from each other) [30], or (ii) interspersed duplication.

Next, we performed a more exhaustive search for genes that are absent in certain lineages. The orthologs present in closely related species were used as queries for BLASTN searches against the genomic sequences in the NCBI database. When a high-identity match (identity >60%) was produced, we examined whether it was annotated as a gene by NCBI; if it wasn't, we manually annotated it using FGENESH+ (www.softberry.com). Next, the proteins annotated were scanned for known functional domains by InterProScan. If an F-box domain was found, the protein was designated as an F-box protein; otherwise, the protein was noted as having lost its F-box domain. If a high-identity match was not obtained, the gene was assumed to have been removed from the genome by deletion.

Selective pressure analyses

For each orthogroup, orthologous amino-acid sequences with similar length were aligned with MUSCLE [22] and then manually checked and trimmed with TRIMAL 1.2 [23]. TRIMAL removed poorly aligned columns and incomplete sequences, considering the remaining sequences in the MSA using three specified thresholds (- resoverlap 0.75, -seqoverlap 80, -gt 0.7). Subsequently, the sequences that did not pass the sequence overlap threshold were replaced by another transcript of the associated genome. Finally, only columns and sequences that passed the thresholds were retained in the final alignments. These retained protein alignments were used to guide the alignments of the corresponding CDSs using TRIMAL. The multiple coding sequence alignments thus generated are shown in File S3.

Average codon-based evolutionary divergence over all sequence pairs within each orthologous group was measured in terms of Ka (the number of non-synonymous substitutions per non-synonymous site) and Ks (the number of synonymous substitutions per synonymous site) using MEGA [31]. Next, the codon-based Z-test was used to evaluate the significance of the Ka/Ks substitution rate.

The distribution of selective pressure across the gene was investigated using the sliding window method in which Ka/Ks (ω) ratios were calculated by DnaSP with a window length of 30 bp and a step size of 6 bp [32]. The statistic ω was calculated in each window, and its value was assigned to the nucleotide at the midpoint of the window. Values of ω from each window were plotted against the nucleotide position, thus enabling the visualization of distinct selective pressure acting across the gene.

To assess variations in selective pressure among sites, the site-specific models were tested comparatively using M0 (one ratio) and M3 (discrete), both of which are implemented in the PAML software package [33]. To detect positive selection affecting a few sites along particular lineages, we applied branch-site model A for analysis.

The ω value for each codon was calculated using the MEC model that is implemented in the Selecton software package [34]. The MEC model is a combination of empirical and mechanistic model, which accounts for differing empirical amino acid mutation probabilities. Compared with the more conservative M8 model, a smaller proportion of sites (with ω values >1) may be sufficient in the MEC model to indicate positive selection despite a low global ω value for the protein [35]. Program Selecton was run with the MEC model and M8a (a null model that only allows purifying and neutral selection), and their second-order Akaike Information Criterion (AICc) score was compared. To map the detected positive selection sites onto the protein's three-dimensional (3D) structure, we used the homology modeling method to construct a 3D structure of F-box proteins in the SWISS-MODEL workspace [36]. The PyMOL (http://www.pymol.org) graphical interface was used to manipulate and display the F-box protein 3D structure.

Single-nucleotide polymorphism (SNP) analysis of F-box genes in the human population

Variant calls from the 1000 Genomes project [37] Phase 1 release v3 were obtained from the NCBI ftp server (ftp://ftp-trace.ncbi.nih.gov/1000genomes/ftp/release/20110521/). Variants were annotated using the SNPEff (v3.4) software (http://snpeff.sourceforge.net/). The annotated variants were filtered to retrieve SNP variants in F-box protein coding regions. Statistical analyses were performed using distribution-free non-parametric tests in the R program. The significance of differences in SNP density distributions in different regions of F-box genes was assessed using the Kruskal-Wallis test or the Mann-Whitney U test. The Fisher's exact test was used to compare ratios of non-synonymous to synonymous substitution SNPs between different regions of F-box genes.

Identifying F-box proteins and their domain architectures

In order to comprehensively identify F-box genes, we used an integrated hmmsearch–BLAST–InterProScan approach. A total of 559 protein-coding F-box genes were identified in eight genomes (Table S1 A). The F-box gene number in all genomes was approximately 70 (except in the mouse genome), accounting for over 0.3% of the total protein-coding genes (Table S1 B). We also identified nine annotated pseudogenes duplicated from protein-coding F-box genes and 21 corresponding homologous DNA regions (Table S2).

The chromosomal distribution of F-box gene was uniform across all autosomes, except for several chromosomes (Table S3). In five primate genomes, F-box genes were significantly enriched in chromosomes 16 and 19, compared with other chromosomes. Chromosomal translocation may have contributed to the higher F-box gene density in chromosome 16. On the other hand, the high density of F-box genes in human chromosome 19 may be attributed simply to the fact that human chromosome 19 is the most gene-rich chromosome [38]. Interestingly, in all genomes analyzed, F-box genes were absent from chromosome X, and this absence was not random (Table S3). The absence of F-box genes on chromosome X of all the eight species examined is consistent with the fact that mammalian X chromosomes are highly conserved across species [39]. No F-box gene was found on chromosome Y, but the departure from uniformity was statistically insignificant. The mechanisms underlying the absence of F-box genes from sex chromosomes remain unclear.

Twenty-seven types of domains (excluding the F-box domain) were identified within F-box protein sets, and all but nine were located at the C-terminus (Figure 1). Among them, the proportion of the LRR domain was the largest, followed by the WD40 domain. The remaining domains were present in small subsets of F-box proteins. Except the F-box domain, no other recognizable domain was found in 30% of the 559 F-box proteins identified in this study.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Number and domain structure of F-box proteins from marmoset, gorilla, human, macaque, mouse, chimpanzee, orangutan, and rat.

https://doi.org/10.1371/journal.pone.0094899.g001

Clustering orthogroups

The 559 F-box genes identified may be clustered into 71 distinct orthogroups (Figure S1). The deep branches were poorly supported by bootstrapping, because F-box genes from different subgroups are of great divergence. However, the vast majority of the single orthogroups reached well-resolved topology with strong bootstrap support. Furthermore, the members in each identified orthogroup were the best blast hits of each other (File S2), and their orthology relationships were confirmed by data obtained from ENSEMBL. These orthogroups include stable and unstable clades based on evolutionary stability. Stable clades are those that contain at least one member of each species (but species-specific duplications could as well result in more copies). Unstable clades in turn do not contain genes of each species. Fifty-seven orthogroups were evolutionarily stable, and the remaining 14 orthogroups were unstable but varied only slightly. The gene orthology relationships among the eight organisms were one-to-one, one-to-many/one-to-one and one-to-zero/one-to-one for 54, 3, and 14 orthogroups, respectively. The one-to-many and one-to-zero orthology relationships are the results of lineage-specific gene duplications and losses, respectively. Although proteins in the majority of the orthogroups contained the same C-terminal domain, species-specific domain accretion or reduction also took place (Figures 1 and S1). For example, the primate FBXL22 lost the LRR domain, and the PRY domain was accreted in the FBXW10 of both gorilla and orangutan. Moreover, orthogroups with the same or similar domain architecture were not necessarily the closest neighbors. For instance, several LRR-domain-containing orthogroups were scattered across the phylogenetic tree.

All F-box genes were directly clustered into their corresponding orthogroups, with the exception of Fbxo6 and Fbxo44. Gene conversion may be a potential source of conflict between a gene tree and a species tree [40]. Hence, gene conversion tests were performed using Geneconv (http://.math.wustl.edu/~sawyer). Statistically significant evidence of a gene conversion event was found between Fbxo6 and Fbxo44 of human, chimpanzee, gorilla, orangutan, and macaque at nucleotide 1477 and nucleotide 1707 (site numbering refers to human Fbxo44 with Ensembl Gene ID ENSG00000132879) (Figure S2). These results suggest that a gene conversion event occurred in the ancestor of these five primates (Fbxo44 was not found in marmoset, so this species was excluded from our analysis). These gene conversion regions contained sequences of F-box functional domains and partial F-box associated (FBA) region.

Evolutionary change of F-box gene number and underlying mechanisms

As mentioned above, lineage-specific gene gain and loss events appeared to have given rise to one-to-many or one-to-zero orthologous relationships among the eight organisms. Therefore, we estimated the number of F-box genes in all ancestral organisms and their change at different stages of the evolution of Euarchontoglire animals. Evolutionary changes in the number of the F-box genes are shown in Figure 2. Although the number of F-box genes was found to be conserved, as a general trend, during the evolution of Euarchontoglires, lineage-specific gene gain and loss events still took place.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Estimated numbers of ancestral, gained, and lost genes during the evolution of Euarchontoglire animals.

Names of extant and ancestral species are on the right-hand side of each external and internal node. Numbers within circles and boxes indicate the numbers of genes in each extant and ancestral species, respectively. Branches are not drawn in proportion to their lengths. The numbers above and below each branch are the numbers of genes gained and lost, respectively.

https://doi.org/10.1371/journal.pone.0094899.g002

Multiple mechanisms contributed to F-box gene number variation among these organisms. It should be noticed that a closely linked gene cluster includes 12 neighboring genes in mouse chromosome 9 with similar gene structures and a high level of sequence identity, ranging from 63.7% to 97.9%, in their coding regions (Figure 3). Therefore, this gene cluster likely arose from a series of tandem duplication events. However, these genes appear to have diverged in terms of their intron length, which we identified to be caused by variations in internal intron sequence repeats, such as the 510 bp repeats in intron 6 of Fbxw22 (Figure 3).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. Chromosomal location, phylogenetic relationships and gene structure of the mouse Fbxw12 gene cluster.

These genes diverged at intron length, caused in part by internal sequence repeats.

https://doi.org/10.1371/journal.pone.0094899.g003

Both primates and mice contain a single Fbxl18 gene, while rats contain two distinct Fbxl18 genes. A schematic illustration of the gene structures of Fbxl18I and its paralogue Fbxl18II are shown in Figure 4. Evidences that Fbxl18II was formed by retrotransposition of Fbxl18I are as follows: (i) Fbxl18I and Fbxl18II are located on chromosomes 5 and 12, respectively; (ii) the two sequences are identical over 111 bp in their 5′ UTRs and their 3′ UTR sequences have over 80% similarity; (iii) Fbxl18II is flanked by a short sequence repeat of ‘agaagaagggaga’; (iv) stretches of 12 adenine (A) occur in Fbxl18II as relics of poly (A) structure; and (v) Fbxl18II contains a 114-bp intron in its 5′ UTR. In addition, a 91-bp intron whose counterpart in Fbxl18I is flanked by an intron sequence with a GT-AG boundary is also observed (Figure 4). Based on these facts, rat Fbxl18II likely represents a semi-processed retrogene. Although functional experimental evidence for rat Fbxl18II is lacking, an expressed sequence tag (EST) (GenBank: CB765629.1) with sequence identity with Fbxl18II, rather than Fbxl18I, was found in the NCBI database. Fbxl18II sequence analysis showed that (i) the DNA sequence of Fbxl18II has a perfect open reading frame (ORF); (ii) using RepeatMasker (http://www.repeatmasker.org), a 937 bp sequence upstream of the start codon was predicted as a long interspersed nuclear element (LINE) that provides original evolutionary materials for promoter formation, and (iii) a highly likely transcription start site was predicted at about 200 bp upstream of the start codon using FirstEF [41] (p = 0.422) and Promoter 2.0 [42] (prediction score  = 1.224) (Figure S3). These findings suggest that the retrogene Fbxl18II may indeed be a functional gene.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 4. Gene structures of two paralogs Fbxl18I (Ensembl Gene ID: ENSRNOG00000001117) and Fbxl18II (Ensembl Gene ID: ENSRNOG00000033326).

Dotted lines between the two paralogs indicate identical sequence fragments.

https://doi.org/10.1371/journal.pone.0094899.g004

Absence of a gene in some genomes does not necessarily indicate removal by deletion. In fact, F-box gene sequences without typical F-box domains often remain in the genome, as supported by our findings that they exist as diverged orthologs of their counterparts with intact F-box domains across multiple species. For instance, the rat Fbxo45 does not contain the typical F-box domain, while its orthologs in mouse and primate do.

Sequence divergence of orthologs and region- and lineage-specific positive selection

To explore the selection forces driving the evolution of the F-box gene family, we first calculated sequence divergence in the complete coding sequences (CDSs) for each orthogroup. The values of Ka, Ks, and ω showed extensive discrepancy among orthogroups (Figure S4). Ks was positively correlated with Ka, as assessed using the Kendall test (tau  = 0.419, p<0.001). Codon-based Z-test indicated that the ω ratios of all orthogroups were significantly lower than 1 (p<0.001), except for Lrrc29 (p = 0.107).

To determine the presence of specific regions within F-box genes under positive selection, sliding window analyses were performed. As expected, ω ratios greater than 1 were observed in regions such as WD40, LRR, and Cyclin_C domains (Figure 5), as well as other uncharacterized domains (Figure S5). Contrary to this pattern, the ω ratios of F-box domains were mostly lower than those of other regions within F-box genes. Such functional selective constraints on the F-box domain may be a consequence of co-evolution of the F-box domain and the SKP1 protein since SKP1 residues are highly conserved, particularly in the core portion of the SKP1–F-box protein interface [43]. For many orthogroups, all regions were under selective constraint with ω <1 although the exact ω value varied among regions (Figure S5).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 5. Sliding window analysis of sequence divergence across the protein-coding regions of six orthogroups using a window length of 30 bp and a step size of 6 bp.

https://doi.org/10.1371/journal.pone.0094899.g005

To verify the statistical significance of variation in selective pressure among sites, we applied a pair of models, namely M0 and M3, to perform likelihood ratio tests. Model M0 assigns a homogeneous ω among sites, whereas M3 assumes several ω site classes. The likelihood ratio test indicated that model M3 exhibited significantly better fit with the data than model M0 for all orthogroups, with the exception of Fbxw7, Fbxl7, Fbxl14, Fbxl15, Fbxl18, Fbxl20, Kdm2A, Lrrc29, Fbxo2, Fbxo4, Fbxo11 Fbxo25, Fbxo33, and Fbxo44, the p values of which were greater than 0.05 (Table S5).

Next, we investigated whether some sites within the F-box gene were under positive selection along specific lineages using branch-site models. The orthogroups that included eight orthologs from the eight species were selected for this analysis. Results showed that some sites were under positive selection along specific lineages. A total of 22 out of 34 orthogroups exhibited site-specific positive selection along a specific branch (Table S6). Furthermore, certain F-box genes, such as Fbxw12, underwent adaptive evolution independently in different lineages.

Molecular adaptation at individual sites and their effects on function

Specific residues may be selected individually, regardless of the pattern of selection governing the global sequence. We calculated ω values at each codon position. Fourteen out of 71 orthogroups underwent adaptive evolution, as indicated by the presence of some positively selected residues in them (Akaike information criterion score < M8a, and posterior mean of ω >1.5) (Table 1). A large number of individual positively selected sites were observed in Fbxw12 and Fbxl13. Although positive selection on these sites were not statistically significant (the lower bound of the confidence interval ω was not >1), the results were in agreement with the interpretation that a large number of sites in the two genes had diverged in Euarchontoglires (File S3). Some of these positive selection residues were located in identified substrate-interacting domains.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Detection of positive selection at individual sites by Bayesian method with MEC model using Selecton.

https://doi.org/10.1371/journal.pone.0094899.t001

The relationship between positively selected amino acid sites and their effects on function was preliminarily obtained by mapping the sites onto the 3D structure of the functional domain. A 3D structure was built for FBXW12 in automated mode (Figure 6). Currently known WD40 domain-peptide interaction sites were located on all three major surfaces: top, bottom and circumference [44]. The α-helix region was populated with residues with low ω values, whereby the pivotal corrected 3D structure was safeguarded through purifying selection. By contrast, the majority of residues under positive selection of FBXW12 were located in the substrate-binding channel of the WD40 domain.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 6. Projection of positive-selection sites onto WD40 domains.

The Ka/Ks (ω) value for each residue is color coded on a seven-step scale from violet (<<1) to dark yellow (posterior mean of ω >1.5). Based on the template 3ow8, a 3D structure was constructed for a peptide fragment (residues 89—408) of FBXW12. Most of the residues under strong positive selection (orange-colored stick surface) were mapped to the substrate binding channel.

https://doi.org/10.1371/journal.pone.0094899.g006

Stronger negative selection on domains than on non-domains of F-box genes in the human population

The 1000 Genomes data provided an opportunity to compare the mutational burden within different regions of the F-box genes. Here, we analyzed SNP density distribution and the difference of the ratios of non-synonymous to synonymous SNP numbers in the domains and non-domains of F-box genes. We mapped 1,254 SNPs genotyped in the 1000 Genomes project to coding regions of human F-box genes, of which 391 mapped to predicted domain regions (File S4). The average SNP density in the F-box genes was 1.080 per 100 bp, indicating that the F-box genes in humans had a high occurrence of single nucleotide substitutions. This result is similar to that found by Clark et al. in Arabidopsis thaliana [45]. In the current study, 15 SNPs were found in the coding regions of Fbxo32 from 1,092 persons. However, only one SNP was found among the coding regions of the same gene from 1,313 cattle [46].

The average SNP density in the F-box domain, other domain and non-domain regions was 1.042, 1.040, and 1.100 SNPs per 100 bp, respectively (Figure 7). The Kruskal-Wallis test for differences in SNP density across regions of F-box genes indicated a significantly uneven distribution (Kruskal-Wallis test p = 0.023, Table S7A). Non-synonymous SNP density in domain regions was significantly lower than that in non-domain regions (Mann-Whitney U test p = 0.038) (Table S7B). Consistent with this, the ratio of non-synonymous to synonymous SNP numbers in the domain regions (0.769) was significantly lower than that in the non-domain regions (1.391) (Fisher's exact test p = 1.403e-06, Table S7C), reflecting stronger purifying selection on domain regions. Contrary to what might have been expected, the F-box domain region was not found to be more conservative than other domain regions based on both SNP density distribution and the ratio of non-synonymous to synonymous SNP number (Figure 7, Table S7C). However, F-box domain regions were the most conservative according to sequence diversity analyses conducted based on the orthogroups (Figures 5 and S5). The differences in results were likely caused by the following factors. First, after the species diverged from Euarchontoglires, certain lineages maintained advantageous mutations in the C-terminal functional domains of F-box genes that facilitated adaptation to changing environments by altering their substrate specificities. Second, in humans, C-terminal functional domains of F-box genes have evolved under strong purifying selection to accurately recognize specific substrates.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 7. SNP distributions across different regions of F-box genes suggest stronger negative selection on domain regions compared with non-domain regions.

A. The total number of synonymous and non-synonymous substitution SNPs in F-box domain, other domain (non F-box domain), domain (including F-box and other domains) and non-domain regions. B. The distributions of average SNP density (SNP numbers per 100 bp) across different regions of F-box genes.

https://doi.org/10.1371/journal.pone.0094899.g007

Evolutionary conservation and variation in F-box gene number

Conflicting reports exist in the literature regarding the number of F-box genes in humans [26], [47]. We performed a comprehensive search of F-box genes using an integrated method. Approximately 70 F-box genes were identified in each Euarchontoglire species, which is much less than the hundreds of F-box genes known to exist in plants. This might be due to the sessile life-style of plants, which prevents them from escaping unfavorable environmental conditions, and necessitates more extensive molecular signaling machineries. Compared with the F-box genes predicted in humans by Jin et al. [26], only Lmo7 was not detected in our study. A published report indicates that an isoform of LMO7 can bind to SKP1 [48]. The F-box domain of LMO7 contains significant changes compared with the consensus sequence. However, LMO7 can still bind to SKP1 because the mutations occur in non-key residues. In such a case, the F-box domain cannot be predicted by a profile search. The method of identifying functional F-box proteins solely based on the consensus sequence inevitably leads to limitations such as filtering out of diverged F-box proteins like Fbox45 (Table S4). We also assigned the genes Ect2l, Fbxo47, and Fbxo48, all of which contain F-box domains, to the F-box gene family. In addition, Jin et al. [26] had noted that Fbxw12-related sequences are expanded over a cluster of six genes on chromosome 9 in mouse. Surprisingly, we found six other paralogs, namely Fbxw20, Fbxw21, Fbxw22, Fbxw24, Fbxw26, and Fbxw28, in this gene cluster (Figure 3). Unequal crossover at meiosis is one of the possible reasons for tandem gene duplication, which occurs more readily in the presence of tandem-repeating genes [49]. Therefore, we inferred that the Fbxw12 gene cluster may have been formed by tandem duplication through a series of unequal crossover events. Genes included in a gene cluster often differentiate from each other with respect to expression patterns, such as the mammalian Hox gene cluster [50]. Therefore, future experimental investigations should provide insight into the significance of expansion of the Fbxw12 gene cluster in mouse.

In general, F-box gene expansion or contraction events did not appear to occur massively during the course of Euarchontoglires evolution. Hence, F-box gene numbers are conserved among the eight organisms investigated in the current study. Orthologous F-box proteins generally contain conserved domain architectures in these organisms. As such, they are likely to mediate essential biological pathways by interacting with similar substrates. In eukaryotes, although orthologous proteins typically have the same domain architectures and functions, significant exceptions and complications to this generalization may be observed [51]. This assumption can also be applied to F-box genes, since evidence of both lineage-specific domain accretion and reduction was found in this study. Differences in homologous protein domain architectures among species may play a role in the functional diversification of orthologs.

Generally speaking, gene duplication can eventually lead to an imbalance in gene quality, and most of the extra gene copies degenerate via accumulation of mutations and become pseudogenes [49]. For instance, several CYP2D genes and olfactory receptor gene cluster have been reported to be pseudogenes in humans [52], [53]. In Euarchontoglires, nine F-box related pseudogenes and their corresponding orthologous DNA regions were found. Certain homologous F-box genes were absent in specific lineages, which may be due to bias caused by incomplete genome sequence, variations in genome assembly quality, or loss of the homologous gene from the genome. Although some homologous genes were still present, their F-box domains could hardly be detected because of long-term divergence. However, it should be noted that loss of the F-box domain in an F-box protein homologue does not necessarily imply loss of function of the ubiquitin ligase. For example, NIPA binds SKP1 and CUL1 despite not having a clear F-box domain, and appears to function as an F-box-like protein [54]. Taken together, our results indicate that although F-box gene gain and loss events do not occur as frequently in Euarchontoglires as they do in plants, the evolutionary pattern of the F-box gene family in these species is consistent with the birth-and-death evolution model [30], [55].

C-terminal domains of F-box proteins have undergone adaptive evolution

C-terminal regions of F-box proteins may have evolved under different selective pressures, such as strong purifying, neutral or positive selection. The protein structure may only be conserved in some parts of an active site responsible for catalysis, while the remaining peripheral regions may have changed considerably, causing change in substrate specificity [56]. Indeed, the analysis of selective pressure at individual sites showed that the vast majority of functional domains are rich in residues with low ω, whereas the residues in the core portion of the protein–protein interface underwent excess amino-acid fixation during the course of Euarchontoglires evolution. These results are consistent with previous findings in other organisms. In plants, while the F-box domain appears rather stable, some C-terminal protein-protein interaction domains such as Kelch and FBA show strong signatures of positive selection [12], [13]. Our results are also in agreement with previous findings in nematodes [14]. Co-evolution of the substrate and F-box protein interface may explain the apparent fast evolution of the substrate-binding domain. Alternatively, mutation of residues that are solvent-exposed to a structural fold may be more tolerated than those located at the highly structured core. Therefore, further extensive research is required before the cause of this positive selection can be definitively determined.

Seventy-one F-box gene orthogroups were found to have been under significantly different selective pressure over the course of Euarchontoglire evolutionary history. We propose that the orthogroups under strongly purifying selection pressure across the genes still recognize the same or similar targets in the eight organisms, whereas others under positive selection pressure conferring adaptive evolution have evolved to recognize different targets. These findings are in contradiction with previous reports that samples of F-box genes from several specific mammalian families show no evidence of positive selection. This discrepancy is likely caused by sample bias [14]. We suggest that mammalian members of the F-box gene family may be involved in both endogenous and exogenous protein degradation.

Comparison of F-box genes between animals and plants

F-box genes are small in number and quite conserved in Euarchontoglires. Much like in Euarchontoglires, no evidence of drastic changes in the total number of F-box genes (42–47) was found in the 12 extant Drosophila species considered in a previous study [57]. However, F-box gene family is one of the largest and fastest evolving gene families in plants [12], [13]. For instance, the number of F-box Kelch genes (FBKs) varies dramatically among Arabidopsis thaliana, Oryza sativa, Populus trichocarpa and Vitis vinifera[12]. The large number of F-box proteins in plants might be required by their species-specific physiology, such as responses to various hormones [58], [59], the circadian clock and photomorphogenesis [60], [61], flower development [62], and defense responses [63].

For both plants and animals, F-box proteins often carry one of a variety of protein-protein interaction domains in the C-terminal regions in addition to the loosely conserved N-terminal F-box domain [7], [64]. The differences in domain distributions among Euarchontoglires, nematodes and plants were striking. Among 27 identified C-terminal domains in Euarchontoglires, LRRs, one of the most abundant C-terminal domains in plants [30], was the most common. However, LRRs was rare in nematodes (unpublished data from our laboratory). In C. elegans, most F-box proteins contain either the FTH or FBA2 domain [14], both of which are absent in plants and Euarchontoglires. In addition to FTH and FBA2, many other distinct domains were also found to be lineage-specific. By contrast, only a single member of the F-box protein family contained the Kelch domain in each species of Euarchontoglires as well as in C. elegans. However, it has been reported that Kelch-containing F-box proteins expanded dramatically among terrestrial plants [12]. Thus, the very distinctive domain distribution of F-box proteins may reflect their divergent functional roles in plants and animals.