Evolutionary Dynamics of Recently Duplicated Genes: Selective Constraints on Diverging Paralogs in the Drosophila pseudoobscura Genome
- First Online:
- Cite this article as:
- Meisel, R.P. J Mol Evol (2009) 69: 81. doi:10.1007/s00239-009-9254-1
- 113 Views
Duplicated genes produce genetic variation that can influence the evolution of genomes and phenotypes. In most cases, for a duplicated gene to contribute to evolutionary novelty it must survive the early stages of divergence from its paralog without becoming a pseudogene. I examined the evolutionary dynamics of recently duplicated genes in the Drosophila pseudoobscura genome to understand the factors affecting these early stages of evolution. Paralogs located in closer proximity have higher sequence identity. This suggests that gene conversion occurs more often between duplications in close proximity or that there is more genetic independence between distant paralogs. Partially duplicated genes have a higher likelihood of pseudogenization than completely duplicated genes, but no single factor significantly contributes to the selective constraints on a completely duplicated gene. However, DNA-based duplications and duplications within chromosome arms tend to produce longer duplication tracts than retroposed and inter-arm duplications, and longer duplication tracts are more likely to contain a completely duplicated gene. Therefore, the relative position of paralogs and the mechanism of duplication indirectly affect whether a duplicated gene is retained or pseudogenized.
KeywordsDrosophilaGene duplicationPseudogeneCopy number polymorphism
Gene duplication is an important evolutionary process (Ohno 1970) that allows for the expansion of gene families (e.g., Nozawa and Nei 2007), for the evolution of genes involved in the regulation of development (Sidow 1996), and for the evolution of cellular biochemical pathways (Wang and Gu 2001). If a newly duplicated gene is to contribute to evolutionary novelty as a protein coding sequence, it must survive the early stages of divergence from its paralog without becoming a pseudogene. I examined the early evolution of duplicated genes in the Drosophila pseudoobscura genome. Recently duplicated genes were interrogated to understand the factors influencing the nucleotide divergence between paralogs and the forces governing the retention of their open reading frames.
Immediately following the duplication of a gene, the derived copy will be found on a single chromosome. If it survives transmission to the next generation, the duplicated gene will segregate as a copy number polymorphism (CNP) until it is either lost or fixed in the population (Dopman and Hartl 2007; Emerson et al. 2008; Sebat et al. 2004; Turner et al. 2008). As paralogs diverge, they may experience one of a number of possible evolutionary trajectories (Lynch and Conery 2000; Moore and Purugganan 2005), including pseudogenization, subfunctionalization (Force et al. 1999; Hughes 1994), or neofunctionalization (Byrne and Wolfe 2007; He and Zhang 2005). The evolutionary fate of a duplicated gene may be influenced by the molecular mechanism responsible for the duplication. For example, retroposition will duplicate an open reading frame, but not untranscribed cis regulatory regions responsible for transcriptional control. DNA-based mechanisms, on the other hand, may duplicate cis regulatory regions along with the coding region. In this case, the mechanism of duplication could affect the expression profile of a duplicated gene, which may influence whether the derived copy becomes a pseudogene. Additionally, the functional properties of a coding sequence can influence the likelihood that a duplication of the coding sequence will be retained (Kondrashov et al. 2002; Papp et al. 2003; Seoighe and Gehring 2004). Furthermore, the relative location of the two paralogs has also been shown to affect the selective constraints on duplicated genes (Cusack and Wolfe 2007; Zhou et al. 2008).
Comparative studies of genome wide patterns of duplication are now possible with the availability of multiple sequenced genomes from closely related species (e.g., Demuth et al. 2006; Hahn et al. 2007; Zhou et al. 2008). Previous work on the sequence divergence between paralogs within the Drosophila melanogaster genome suggests that the majority of newly arising duplicated genes are pseudogenized (Lynch and Conery 2000), although this conclusion is not unanimous (Long and Thornton 2001). With multiple sequenced Drosophila genomes (Adams et al. 2000; Drosophila 12 Genomes Consortium 2007; Richards et al. 2005) we can now infer duplication events along individual lineages (Hahn et al. 2007; Heger and Ponting 2007; Zhou et al. 2008). The ancestral karyotype in the genus consists of a telocentric X-chromosome, four telocentric major autosomes, and a small dot chromosome. The six chromosome arms are referred to as Muller elements A-F (Muller 1940). Throughout the evolution of the genus, chromosome arms have fused and gene order has been shuffled within arms, but very little gene movement occurs between arms (Bhutkar et al. 2007; Drosophila 12 Genomes Consortium 2007; Heger and Ponting 2007). This facilitates the mapping of gene duplications in these genomes. Additionally, many species, such as D. pseudoobscura, harbor inversion polymorphisms (Krimbas and Powell 2000), which may prevent the spread of duplicated genes between different chromosomal arrangements within a single species (Popadic et al. 1995).
I present an analysis of recently duplicated genes in the D. pseudoobscura genome using a dataset containing the duplications of complete open reading frames and those that appear to have been pseudogenized. I also have information regarding the structure of gene duplications, as well as high confidence assignments of ancestral versus derived copies. Paralogs with low divergence between copies were interrogated for CNP. My results indicate that nucleotide divergence between paralogs is not an adequate proxy for the age of very recent duplications because of gene conversion between paralogs. I am able to classify duplication events by their mechanism, and I consider the role that the duplication mechanisms play in the selective constraints on duplicated genes. Additionally, I examine how the relative positions of paralogs, size of duplications, nucleotide divergence between paralogs, and functional properties of the genes influence the likelihood of pseudogenization.
Identifying, Annotating, and Aligning Duplicated Genes
Duplicated protein coding genes were identified in the D. pseudoobscura genome by comparisons with the D. melanogaster genome (Meisel 2009). In the preliminary annotation of the D. pseudoobscura genome, one to one best hit orthologs to D. melanogaster protein coding genes were identified (Richards et al. 2005). The sequence covered by the orthologs (both exonic and intronic) was classified as “genic” sequence, and the sequence between the identified genes was classified as “intergenic”. If a gene was duplicated in the D. pseudoobscura genome after the split with D. melanogaster, there should be a second homolog located in one of the sequences initially categorized as intergenic. Genes may also be duplicated into the introns of other genes, and my approach will miss these duplications.
To identify genes duplicated along the D. pseudoobscura lineage, the D. pseudoobscura genic regions were searched against the intergenic regions using MegaBLAST (Zhang et al. 2000), after masking for all known Drosophila transposable elements and a common repetitive sequence found throughout the D. pseudoobscura genome (Richards et al. 2005) (Genbank accessions AY693425 and AY693426) using RepeatMasker (Smit et al. 2004). The D. pseudoobscura intergenic regions were also searched against all D. melanogaster proteins using BLASTX (Altschul et al. 1997) to mimic the initial search for one to one orthologs. (The parameters used in these BLAST searches are available in the Supplementary Methods.) All intergenic sequences that matched orthologous D. pseudoobscura and D. melanogaster genes were retained for further analysis. While this approach will not identify duplications of D. pseudoobscura genes that were missed by the initial search for one to one orthologs, these types of genes make up a minority of the genic content of the D. pseudoobscura genome (Drosophila 12 Genomes Consortium 2007). However, duplicated genes with degenerated open reading frames will be identified with this method because it relies solely on sequence identity; this differs from other collections of duplicated genes in Drosophila genomes which rely on gene prediction algorithms (e.g., Hahn et al. 2007).
The duplication endpoints were identified using the nucleotide BLAST alignments of the intergenic and genic regions. The genome was partitioned into genic and intergenic regions prior to the BLAST search, and alignments would often reach the end of the partitioned regions. To remedy this, flanking partitions were added until the duplication endpoints no longer reached the end of the outermost partitions. As each partition was added, the endpoints were re-identified using BLAST.
The protein coding sequences of the paralogs were annotated using the following approaches. First, the initial published annotation often contained a predicted open reading frame for the one to one best hit ortholog in the D. pseudoobscura genome (Richards et al. 2005). Additionally, BLASTX (Altschul et al. 1997) searches of the duplicated sequence against the D. melanogaster proteome were used to identify protein coding sequences with homology to the D. melanogaster ortholog. Finally, the duplicated sequences were used as queries in Genscan (Burge and Karlin 1997), and the predicted open reading frames were also used in the annotation. The open reading frames identified using these three approaches were used to create a protein coding sequence for each copy with maximal overlap with the D. melanogaster gene. Predicted exons were included if they had a homologous exon in the D. melanogaster gene. Once the protein coding sequences were inferred, the amino acid sequences of the two D. pseudoobscura paralogs were aligned with the D. melanogaster ortholog using the CLUSTALW (Thompson et al. 1994) implementation in MEGA 3.1 (Kumar et al. 2004). Finally, the nucleotide sequences of the D. pseudoobscura genes were overlaid on the amino acid alignment, and the intergenic and intronic nucleotide sequences were aligned using CLUSTALW. (The alignments between the D. pseudoobscura paralogs and the D. melanogaster orthologous coding sequence are available as Supplementary Material.)
Draft genome assemblies have many assembly gaps. If an assembly gap was found in a duplicated region, that region was also extracted from the reconciled assembly of the D. pseudoobscura genome (http://rana.lbl.gov/drosophila/caf1.html). In some cases, the reconciled assembly also contained an assembly gap or was missing the duplication entirely; in these cases the duplicated gene was removed from the dataset. If the reconciled assembly contained the duplication without assembly gaps, the sequence from the reconciled assembly was annotated as described earlier. The duplications were further confirmed to not be assembly artifacts by MegaBLAST (Zhang et al. 2000) against the trace sequences from the whole genome sequencing project. Duplications with multiple traces spanning the endpoints were consider real, while those that did not have multiple traces spanning their endpoints were excluded. Finally, the ten paralogs with the highest sequence identity were reconstructed as follows to confirm that they are not assembly artifacts. Trace sequences from within each copy were obtained by using the mate-pairs located outside the duplicated region. These traces were assembled and aligned to the genomic sequence used in the analysis.
Estimating Nucleotide Divergence Between Paralogs and the Lengths of Duplicated Regions
Nucleotide divergence between D. pseudoobscura paralogs was calculated for all sites aligned between paralogs, non-coding sites, synonymous sites within coding exons (Nei and Gojobori 1986), and non-coding plus synonymous sites. Duplicated genes with frameshift mutations were shifted to be in the proper reading frame for these calculations. Paralogs were excluded from the analysis if they had less than 80% sequence identity at all sites, non-coding sites, synonymous sites, or non-coding plus synonymous sites and if there were at least 100 nucleotides in that particular class of sites (the 100 site cutoff was chosen to prevent spurious results due to a small sample size of nucleotides). The 80% sequence identity cutoff was chosen for four reasons: (1) it ensures that the duplications arose after the split between D. pseudoobscura and D. melanogaster; because divergence between these species is saturated at synonymous and non-coding sites (Richards et al. 2005), 80% sequence identity is a conservative cutoff. (2) It allows for reliable alignment of non-coding sequences (introns and intergenic regions). (3) Recent duplications are more likely to currently be located at the same genomic location in which they arose, which decreases the confounding effect of secondary relocation in the analysis of divergence between paralogs. (4) The BLAST settings should be able to identify a large majority of paralogs with >80% sequence identity (Gotea et al. 2003). The length of each duplication event was estimated using the number of nucleotide sites in each copy, the average of the lengths of the two copies, and the number of nucleotide sites in the alignment of the two copies (excluding gaps in the alignment). Finally, I excluded genes that had been recently duplicated multiple times to ensure the phylogenetic independence of all genes in the dataset. The excluded genes had more than one paralog with at least 80% sequence identity.
Completeness of Genes and Degeneration of Open Reading Frames
Amino acid alignments between paralogs were used to classify duplicated genes as “complete” or “partial”. If both copies of a duplicated gene contain the same beginning- and end-points of their annotated protein coding sequence, the duplicated gene was considered complete—cis regulatory regions were not considered because they are poorly annotated for D. pseudoobscura genes (Richards et al. 2005). Conversely, partially duplicated genes are missing either the 5′ end, the 3′ end, or both ends of the annotated coding sequence. Coding sequences containing a frameshift mutation, premature stop codon, or mutation at a 5′ or 3′ intron splice site were categorized as “degenerated”. Degenerating mutations were confirmed with comparisons to the original trace files. All duplicated genes fall into one of four mutually exclusive categories: (1) complete non-degenerated, (2) complete degenerated, (3) partial non-degenerated, and (4) partial degenerated.
Mechanisms of Duplication and Inferring Ancestral and Derived Copies
The intron-exon structure of the duplicated genes, the relative position of the D. pseudoobscura paralogs, and the location of the D. melanogaster ortholog were used to classify each copy as ancestral or derived. First, duplications of single genes were classified as retroposed if one copy is missing all introns present in the other copy (over the region of the gene that was duplicated). The copy missing introns was considered derived. Intron loss via recombination between processed transcripts and the genes that encode them may also give rise to intronless genes (Coulombe-Huntington and Majewski 2007; Fink 1987). However, independent evidence suggests that the intronless genes in this dataset were generated via retroposition (Meisel 2009). For example, the ancestral and derived copies of all putatively retroposed duplications are found on different chromosome arms, and the ancestral copies are all found on the same arm as their D. melanogaster orthologs. Also, there are no ancestral copies that have lost introns, suggesting that intron-loss is a rare event in Drosophila. Duplications of multiple genes were classified as DNA-based duplications. Additionally, duplications of single genes were considered DNA duplications if both copies contain the same intron-exon structure and at least one intron is found in the duplicated region. Duplications of single exon genes and single exons were classified as ambiguous. A single duplicated gene (the orthologs of CG7730) contains two of the three introns present in the ancestral copy; this gene was classified as ambiguously duplicated.
Ancestral and derived copies were inferred for DNA duplications as follows. The location of the D. melanogaster copy was inferred to be the ancestral location of the gene prior to duplication. D. pseudoobscura paralogs were first classified based on their relative position; they were divided into those in which both paralogs are located on the same chromosome arm (intra-arm) and those in which the paralogs are located on different chromosome arms (inter-arm). A portion of Muller element A (chromosome arm XL) was relocated to the proximal region of chromosome XR (Muller element D) along the D. pseudoobscura lineage after the split with the D. melanogaster lineage (Schaeffer et al. 2008). My analysis is robust to whether this region is considered part of element A or element D. In the case of inter-arm duplications, the ancestral copy was inferred to be the one located on the same homologous chromosome arm as the D. melanogaster ortholog. Paralogs located on the same chromosome arm with no protein coding genes between them were classified as adjacent (regardless of their orientation), and intra-arm duplications with at least one gene between them were classified as non-adjacent. For non-adjacent intra-arm duplications, the ancestral copy was inferred to be the one with the same flanking genes as the D. melanogaster ortholog. If neither copy has the same flanking genes as the D. melanogaster ortholog, I was able infer ancestral and derived copies only if one copy has a partial coding sequence—the copy with a partial coding sequence was inferred to be the derived copy.
Adjacent duplications were assigned as ancestral and derived based on the completeness of the coding sequence and the conserved orientation with the D. melanogaster orthologs. Partial coding sequences may arise via either the partial duplication of a coding sequence or the complete duplication of a coding sequence followed by the subsequent loss of the 5′ or 3′ end of the coding sequence. An adjacent duplication containing the partial coding sequence of a gene was inferred to be the derived copy because that copy is missing a portion of the coding sequence found in the other copy and the D. melanogaster ortholog. If adjacent paralogs are in opposite orientations, the copy in the same orientation (relative to flanking genes) as the D. melanogaster ortholog was inferred to be the ancestral copy. In the case of adjacent duplications with complete coding sequences and both paralogs in the same orientation, neither copy was assigned as ancestral or derived. This approach is expected to give reliable assignments of ancestral and derived copies because it is independent of most of the evolutionary events that occur subsequent to gene duplication. However, there may be errors in the assignment of ancestral and derived copies for adjacent duplications because either one of the paralogs or the D. melanogaster ortholog may have changed its orientation (via a microinversion). Additionally, the assumption that partial duplications are derived may not be appropriate. The error associated with these potential flaws appears to be negligible—treating the ancestral and derived copies as unknown for adjacent duplications results in the same conclusions as assigning copies as ancestral and derived.
Expression Profiles of D. melanogaster Orthologs
Expression data for the D. melanogaster orthologs of the D. pseudoobscura duplicated genes were downloaded from the FlyAtlas (Chintapalli et al. 2007), which contains expression data from 13 different body parts. Genes with an mRNA signal >100 in a particular body part were said to be expressed in that body part. G tests were used to assess whether expression in a particular body part is independent of whether completely duplicated genes are degenerated.
Copy Number Polymorphism
Duplicated genes with derived copies on one of two autosomes—Muller elements C and E (chromosomes 3 and 2, respectively)—were tested to see if the derived copy has fixed in a sample of chromosomes from natural populations. Duplicated genes with the highest sequence identity between paralogs are most likely to be segregating as CNPs. I selected paralogs with high sequence identity for which PCR primers could be designed such that they only amplify a region if the derived copy of a duplicated gene is present (primer sequences are available in Supplementary Table 1). No explicit sequence identity cutoff was used, and the sampled paralogs differ at 0.44–6.13% of non-coding and synonymous sites. Primers were placed in sequence that is conserved between D. pseudoobscura and D. melanogaster to minimize the possibility of false negatives. For two of the duplications (the orthologs of CG2412 and CG11552) two sets of primers were used to test for CNP.
A sample of 63 lines that had been made isochromosomal for the third chromosomes (Schaeffer et al. 2003) was assayed for CNP on the third chromosome. Each line carries one of six different arrangements differentiated by chromosomal inversions: Arrowhead (AR), Standard (ST), Pikes Peak (PP), Chiricahua (CH), Santa Cruz (SC), and Tree Line (TL). Duplicated genes were identified on the AR background because the strain sequenced in the D. pseudoobscura genome project was homozygous for an AR third chromosome (Richards et al. 2005). Only one chromosome was assayed for SC and TL, each, and at least 12 chromosomes were assayed for the other four arrangements. Twelve lines that had been inbred for ten generations using single-pair sib-matings were assayed for duplicated genes on chromosome 2, which harbors no inversion polymorphism. Each of these inbred lines was started from a single wild-caught female. DNA sequencing from loci not discussed here reveals that these lines have very little heterozygosity (data not shown).
Results and Discussion
I identified 88 duplications that occurred in the D. pseudoobscura genome after the split with the D. melanogaster lineage, containing a total of 101 genes (Supplementary Table 2). Nine duplications contain two genes, two duplications contain three genes, and the rest are duplications of single genes. I analyzed duplications of single genes separately and found similar results as when I analyzed all duplications; only the analysis of all duplications is presented. In the analysis below, nucleotide divergence between paralogs is measured at non-coding and synonymous sites and duplication length by the number of sites in the alignment of the paralogs (excluding gaps), but the results are robust to all measures of nucleotide divergence and duplication length.
On the Path to Fixation of a New Duplication and Gene Conversion Between Paralogs
The path to fixation of a duplicated gene will be influenced by a combination of stochastic and deterministic processes (Lynch et al. 2001). Recently duplicated genes may be segregating as CNPs if the derived copy has yet to fix in the population. Nucleotide divergence between paralogs can be used a proxy for the age a duplication. However, gene conversion between paralogs (Drouin 2002; Lazzaro and Clark 2001; Osada and Innan 2008; Petes and Fink 1982; Semple and Wolfe 1999; Slightom et al. 1980; Thornton and Long 2005) may slow the rate of divergence between the two copies, resulting in an underestimate of the age of the duplication (Teshima and Innan 2004). The data on divergence between paralogs and CNP can be used to indirectly infer the effect of gene conversion on the sequence divergence between paralogs.
Copy Number Polymorphism
Duplicated genes with minimal divergence between copies have the highest likelihood of segregating as CNPs. I used PCR to determine if the derived copy is present for duplicated genes with low divergence between paralogs. Derived copies on chromosomes 2 and 3 (Muller elements E and C, respectively) were interrogated. While there are technical limitations with using PCR to infer CNP, these are more likely to lead to false negatives, rather than false positives. Therefore, in the analysis of CNP below, I focus primarily on what the presence of derived copies in the sampled chromosomes reveals about the evolutionary dynamics of the duplicated genes.
Copy number polymorphism frequency of duplicated genes on two chromosomes
Chromosome 3—Muller element C (n = 63)
Chromosome 2—Muller element E (n = 12)
Five duplicated genes were assayed on the second chromosome, which harbors no inversion polymorphism. One duplicated gene on chromosome 2 was not found in the sampled chromosomes despite the ability of the PCR primers to amplify the derived copy from the genome strain. This paralog had the lowest nucleotide divergence between copies of all the duplications sampled for CNP, suggesting it may be the most recent duplication. The other four derived copies are fixed in the sampled chromosomes, despite very little nucleotide divergence between paralogs (Table 1). This suggests that the fixation of recently duplicated genes occurs rapidly. Alternatively, gene conversion between paralogs may slow the rate of divergence between paralogs, causing nucleotide divergence to be a poor indicator of the age of the duplications. The evidence for gene conversion between paralogs is examined below.
The Effect of Relative Position on the Divergence Between Paralogs
Another possible explanation for the low nucleotide sequence divergence between proximally located paralogs is that the more distantly located the two copies are, the greater the genetic independence between the paralogs. That is, linkage disequilibrium between paralogs located on different chromosome arms can be decreased via independent assortment, and associations between intra-arm paralogs can be broken via crossing over. This will make paralogs with less genetic linkage appear more diverged under the following scenario. The derived copy of a duplicated gene originates from a single allele of the ancestral locus, which had segregating polymorphisms at the time of the duplication event. Initially, the derived copy will have more sequence identity with that allele at the ancestral locus than to the other alleles at the ancestral locus. More tightly linked paralogs have a higher probability of the derived copy being sampled along with a descendant of the ancestral allele from which it arose, causing measures of divergence between genetically linked paralogs to be artificially lower than unlinked paralogs. Unfortunately, it is not possible to determine whether differences in rates of gene conversion or differences in the genetic independence of paralogs cause the relationship between divergence and relative positions.
The other explanations for the relationship between relative position and nucleotide divergence between paralogs are not as convincing. For example, derived copies may be located near the ancestral copy when they arise and disperse throughout the genome over time—a hypothesis that has been previously considered, and rejected, in Caenorhabditis elegans (Katju and Lynch 2003). This is also unlikely for the duplicated genes in the D. pseudoobscura genome because most distantly located paralogs contain no hallmarks of relocation; the majority of the non-adjacent and inter-arm duplications lie within regions of conserved gene order between D. pseudoobscura and D. melanogaster, so they could not have moved away via a simple rearrangement event. It is more parsimonious to assume that the current location of the derived copy is an adequate approximation of its location when it was generated. Codon usage bias has also been shown to slow the rate of divergence between paralogs (Lin et al. 2006). However, the effect of relative position on the divergence between paralogs is also observed when only intergenic and intronic sequences are examined. Therefore, the result is unlikely to be affected by codon bias. Furthermore, it has been suggested that proximal duplications in Drosophila are usually pseudogenized or lost because they are unlikely to gain a new function, while dispersed duplications are more likely to be retained because they can evolve a beneficial function (Zhou et al. 2008). Therefore, one expects adjacent duplications to be of a more recent origin than non-adjacent duplications. However, this model also predicts a higher rate of pseudogenization for adjacent duplications, which is not observed in the D. pseudoobscura genome (see below).
On the Likelihood of Pseudogenization
The Effect of Complete/Partial Coding Sequences
Although pseudogenes have been extensively studied in Drosophila genomes (e.g., Harrison et al. 2003; Petrov and Hartl 2000), the processes governing why particular duplicated genes become pseudogenes while others are retained are still unclear (Lynch et al. 2001; Zhang 2003). One hallmark of a pseudogenized gene is a nonsense, frameshift, or intron-splice-site mutation (Harrison and Gerstein 2002), and I refer to genes harboring at least one of these mutations as “degenerated”. Two different ancestral copies of duplicated genes have accumulated mutations that disrupt their open reading frames (Supplementary Table 2). These degenerating mutations presumably occurred after the duplication event because the mutations are not shared by the derived copy. There is no difference in the conclusions of the analysis presented below if those two duplicated genes are treated as degenerated or not degenerated.
Counts of degenerated derived copies of duplicated genes for various classes of duplicated genes
Derived copy degenerated?
All duplicated genes
Completely duplicated genes
Amb & retra
It is possible that partially duplicated genes arose as complete genes and subsequently lost the 5′ or 3′ end of the duplicated region. If the loss of one end of a duplicated gene occurs via the fixation of neutral mutations, we would expect the probability that a complete duplication becomes a partial duplicate to increase with time. However, there is not a significant difference in nucleotide divergence between paralogs when one compares complete and partial duplications (Fig. 3b). To control for the selective constraints on non-degenerated genes, I looked at degenerated genes alone. There is not a significant difference in nucleotide divergence between partial and complete degenerated genes (P = 0.33, Wilcoxon test). While a small number of duplications that were classified as partial may have originally arisen as complete duplications (and subsequently lost their 5′ or 3′ end), this does not appear to be a common occurrence.
I further examined the selective constraints on completely duplicated genes. First, I compared the rates of amino acid evolution between completely duplicated genes in which the derived copied had not degenerated and those with degenerated derived copies. The two copies of completely duplicated genes for which one copy has degenerated evolve at more asymmetrical rates than duplicated genes that had not degenerated (Fig. 3a). The increased rate asymmetry is the result of faster amino acid evolution along the lineages leading to degenerated copies (Supplementary Fig. S1). This indicates that non-degenerated completely duplicated genes are under more selective constraints than degenerated complete genes—another unsurprising result. However, degenerated partially duplicated genes also have more asymmetrical rates of amino acid evolution than non-degenerated partial duplicates (Fig. 3a). Furthermore, no significant differences in the rates of amino acid evolution are detected between the two copies of non-degenerated partially duplicated genes (Supplementary Fig. S1). This suggests that there are more selective constraints on non-degenerated partially duplicated genes than on degenerated partial genes. But there is also less nucleotide divergence between paralogs for non-degenerated partially duplicated genes than degenerated partial genes (Fig. 3b). Therefore, the failure to detect asymmetrical rates of amino acid evolution between the two copies of non-degenerated partially duplicated genes is a by-product of their recent origin—not enough time has passed for them to accumulate amino acid differences. The same phenomenon cannot explain the differences in relative rates of amino acid evolution between degenerated and non-degenerated completely duplicated genes; degenerated and non-degenerated completely duplicated genes do not have significantly different amounts of nucleotide divergence between paralogs (Fig. 3b).
The Role Other Factors Play in Pseudogenization
The effects of the length of a duplication event, the ancestral expression profile, the duplication mechanism, and the relative position of paralogs on the pseudogenization of duplicated genes were examined. I find that none of these factors significantly affect the likelihood of pseudogenization for completely duplicated genes. The size of a duplication event may influence whether or not the complete coding sequence of a gene gets duplicated (Katju and Lynch 2003). Indeed, completely duplicated genes are contained in significantly longer duplication blocks than partially duplicated genes (Fig. 3c). Additionally, non-degenerated genes are also contained in longer duplications than degenerated genes (Fig. 3c). This suggests that the length of a duplication event influences the likelihood that the gene contained within it will degenerate. Furthermore, adjacent duplications are longer than non-adjacent duplications, and intra-arm duplications are longer than inter-arm duplications (H = 21.67, P < 0.001) (Fig. 1b). If duplication length influences the selective constraints on completely duplicated genes, then the relative position of a duplication may influence the selective constraints on the coding sequence as well. However, there is not a significant difference in duplication length between non-degenerated and degenerated duplicated genes when one looks at completely or partially duplicated genes separately (Fig. 3c). Non-degenerated derived copies are contained in longer duplications than degenerated genes because longer duplications have a greater likelihood of containing completely duplicated genes, and complete genes are less likely to be degenerated.
It has previously been observed that different functional classes of genes are preferentially retained following duplication, while other classes are underrepresented as multigene families (Kondrashov et al. 2002; Papp et al. 2003; Seoighe and Gehring 2004). I was unable to examine the prevalence of different gene ontology classes among degenerated and non-degenerated completely duplicated genes because of small sample sizes. Additionally, the only whole genome expression data available for D. pseudoobscura are from whole males and females (Zhang et al. 2007), and there is no evidence that sex-biased expression predicts degeneration (data not shown). As a proxy for the expression profile of the ancestral copy, data from 13 different body parts were obtained for the D. melanogaster orthologs of each completely duplicated gene (Chintapalli et al. 2007). Completely duplicated genes with orthologs expressed in the hindgut, midgut, and brain have a higher frequency of degeneration than completely duplicated genes not expressed in those tissues (Supplementary Fig. S2). A sequential Bonferroni correction for multiple tests (Sokal and Rohlf 1995), however, causes one to fail to reject the null hypothesis of independence between degeneration and tissue expression. It is also possible that the presence of cis regulatory sequences at the derived locus (either arriving with a DNA-based duplicated gene or acquired from sequences flanking a duplicated gene) influences whether a duplicated gene becomes a pseudogene. This could be studied by examining the expression profiles of recently duplicated genes in future experiments, but these experiments would be beyond the scope of the results presented here.
The process giving rise to a completely duplicated gene does not influence whether the derived copy has degenerated either. There is an insignificant excess of non-degenerated DNA-based completely duplicated genes compared to ambiguous and retroposed complete duplications (P = 0.32, F.E.T.) (Table 2). Additionally, the relative position of the paralogs does not significantly affect the probability of pseudogenization of completely duplicated genes; intra-arm duplications have an insignificantly lower frequency of degenerated derived copies than inter-arm duplications (P = 0.10, F.E.T.) (Table 2). In rodents, duplications in which the ancestral and derived copies are distantly located from each other evolve at more asymmetrical rates than duplications located in close proximity (Cusack and Wolfe 2007); this may be the result of relaxed constraints on relocated duplications. In D. pseudoobscura, however, there is not a significant difference in relative rates of amino acid evolution between complete non-degenerated duplications in different relative positions (H = 2.34, P = 0.310), nor is there a significant difference in the ratio of non-synonymous to synonymous substitutions between paralogs for intra- and inter-arm completely duplicated genes.
While none of the aforementioned factors alone significantly affect the probability that a completely duplicated gene will become a pseudogene, their interactions may. DNA duplications and intra-arm duplications produce longer duplication tracts, and longer duplications are more likely to contain a complete gene. Therefore, the duplication mechanism and relative position of paralogs seem to indirectly affect the selective retention of duplicated genes. However, DNA-based duplications are more likely to be on the same chromosome arm as the ancestral copy (Meisel 2009), so the mechanism of duplication and relative position are not independent.
The Neutrality of Pseudogenization
If none of the aforementioned factors contribute significantly to the likelihood that a completely duplicated gene will degenerate, it is possible that the degeneration of completely duplicated genes is primarily a neutral process. Using data from D. melanogaster, it was previously estimated that half of all duplicated genes are lost by the time the paralogs have diverged at 8.5% of their synonymous sites (Lynch and Conery 2000). This approach toward estimating a half-life for duplicated genes assumes that the probability of degeneration for duplicated genes that arrive with functional potential (i.e., completely duplicated genes) will increase with time. Contrary to this assumption, degenerated completely duplicated genes do not have significantly more nucleotide divergence between paralogs than non-degenerated completely duplicated genes (Fig. 3b). This suggests that a large fraction of completely duplicated genes are under enough selective constraints to prevent degeneration by mutational pressure alone. Therefore, there may not be an appropriate method for estimating the half-life of duplicated genes in Drosophila; the probability and frequency of pseudogenization may depend on specific properties of individual duplicated genes rather than on the nucleotide divergence between paralogs. Unfortunately, the attempts to identify those properties for the duplicated genes in this dataset were not fruitful.
The Possibility of Chimeric Genes
While partially duplicated genes are most likely pseudogenes, it is also possible that they make up part of a functional chimeric gene (Arguello et al. 2006, 2007; Long et al. 2003). It is unlikely that a considerable fraction of the partial duplications in the D. pseudoobscura genome are part of functional chimeric genes for three reasons. First, previously described chimeric genes have intact open reading frames over the portion of the gene that was duplicated (Arguello et al. 2006; Jones and Begun 2005; Long and Langley 1993; Nozawa et al. 2005; Yang et al. 2008), whereas the partially duplicated genes in this dataset tend to have degenerated open reading frames (Table 2). Second, even though unequal rates of evolution between ancestral and derived copies is also a hallmark of chimeric genes (Jones and Begun 2005), the partially duplicated genes presented here with unequal rates of amino acid evolution also have degenerated open reading frames (Supplementary Fig. S1). Third, partially duplicated genes with degenerated coding sequences are more diverged from their ancestral paralogs at the nucleotide sequence level than those with intact open reading frames (Fig. 3b). This indicates that partially duplicated genes with degenerated open reading frames are older than those with non-degenerated open reading frames. Therefore, partially duplicated genes probably degenerate via a neutral process because the probability of degenerating increases with time—in contrast to completely duplicated genes for which there is no relationship between degeneration and nucleotide divergence (Fig. 3b). That is not to say that chimeric genes are not evolutionarily important; however, it is unlikely that many of the partially duplicated genes in this dataset make up part of functional chimeric genes.
N. Hasan, B. B. Hilldorfer, R. LeGros, and R. L. Zindren helped with sorting the BLAST hits, and N. Hasan and B. B. Hilldorfer assisted in testing the recently duplicated genes for CNP. S. W. Schaeffer and J. R. Arguello provided useful discussion and comments on the manuscript. V. Gotea and W. Makalowski provided assistance with RepeatMasker and MegaBLAST, and V. Gotea also commented on the manuscript. This material is partially based on work supported by the National Science Foundation under Grant No. 0608186, awarded to RPM. Any opinions, findings and conclusions or recommendations expressed in this material are those of the author and do not necessarily reflect the views of the National Science Foundation.