The evolutionary trajectory of duplicated genes has been an active area of investigation since gene duplication was first recognized as an important force in species evolution 1. The availability of new sequence data and analyses has challenged the hypothesis suggesting most duplicated genes are destined to lose their function and become pseudogenes, with a few exceptions establishing new biological roles (reviewed in 2). It is believed that the occurrence of gene duplication would result in relaxed selection of redundant copies permitting genes to evolve specialized sub-functions 34. Moreover, nearly identical genomic regions provide important substrates for chromosomal rearrangements that permit rapid evolutionary changes to occur in a short period of time 5.

An estimated 3.5-5% of the human genome has undergone recent duplication 6789, and these segmental duplications (also termed duplicons or low-copy repeats) are found to be hot spots, or predisposition sites, for the occurrence of nonallelic homologous recombination. This recombination can lead to genomic mutations such as deletion, duplication, inversion, or translocation, resulting in human disease 10. Many mouse strains with chromosomal aberrations are known 11, and it remains to be seen whether segmental duplications have a role in any of these genomic mutations.

A high-quality draft genome sequence not only makes it possible to expand the known set of duplicate genes, but also reveals their genomic context. This genomic context contains the regulatory and structural elements responsible for gene expression that need to be interrogated for a better understanding of the mechanisms and consequences of gene duplication. Furthermore, an accurate and well-annotated mouse genome is an essential resource for many in the biomedical research community, especially those who use the sequence to design and interpret transgenic, mutagenesis, microarray, and proteomic studies.

Several lines of evidence show that the whole-genome shotgun (WGS) approach yielded a high-quality draft sequence that covers roughly 96% of the euchromatic genome excluding chromosome Y (a female C57BL/6J mouse was used in the sequencing project) 12. The WGS sequence reads were assembled into sequence contigs using sequence-assembly programs to produce the February 2002 MGSCv3 working draft 1213. The newly released February 2003 assembly was a hybrid assembly comprising 705 megabases (Mb) of finished bacterial artificial chromosome (BAC) sequences incorporated into the MGSCv3 assembly.

We previously analyzed several versions of the human genome draft assemblies (NCBI Builds 28, 29, and 30) 9, and found substantial potential genome assembly errors in all builds, including approximately 40 Mb of sequence in Build 30. These assembly errors probably arose from difficulties in merging finished sequence or from incorrectly assigning sequence contigs into the genome assembly. In such cases, completely identical or nearly identical sequences (due to allelic differences or sequencing errors) would be present at distinct regions in the genome sequence. These sequence misassignment errors would yield near-perfect duplication artifacts, detected as having extremely high sequence identities (exceeding 99.5% and over 5 kilobase (kb) in length), in genome assemblies. However, a small subset of such results could represent duplications that arose from very recent evolutionary events and will require further experimental analysis.

A number of web-based resources, specifically those provided by the National Center for Biotechnology Information (NCBI), Ensembl (at the European Bioinformatics Institute and Sanger Centre), and the University of California Santa Cruz (UCSC), make the genome sequence and associated annotations readily accessible. Because of the success of the mouse genome sequencing consortium (MSGC), investigators worldwide are utilizing the draft 'as is' in both medical and evolutionary studies. In this paper we show that even though the genome assembly is still in draft form, an initial analysis of the sequence can reveal novel genomic duplications and demarcate regions of the genome that require additional examination.

We performed a search for all recent segmental duplications that were larger than 5 kb in size and showed greater than 90% sequence identity from both the February 2002 (numerical results for February 2002 assembly are presented at our web site 14) and the February 2003 mouse genome sequence assemblies 15. Our method was based on pairwise (mega-) BLAST2 16 sequence comparisons between entire chromosome sequences. From our analysis of the February 2003 assembly, a total of 33.6 Mb (1.2%) of the genome sequence (2,695 Mb) was found to be involved in recent segmental duplications (Table 1) and 8.9 Mb of this sequence was unmapped data (found in the unmapped chromosome sequence). On the basis of the 20 mapped chromosomes, more than 712 distinct intrachromosomal segmental duplications, comprising 19.9 Mb of sequence (Figure 1), and 475 distinct interchromosomal duplications, comprising 7.1 Mb of sequence, were identified. We also found that 57% of the duplications were in tandem, which we defined as two related intrachromosomal duplicons located within 200 kb of one another.

Duplications can be found in all chromosomes analyzed, with chromosomes 6, 7, 17, and X having the highest, and chromosome 18 having the least, duplicated content (Table 1, Figure 1). Substantial amounts (8.9 Mb) of the duplicated content are found in the unmapped chromosome (ChrUn) sequence, suggesting that the correct chromosomal assignment of these segments remains a major assembly challenge. It is possible that small subsets of these duplications are due to chimeric reads and other sequencing artifacts and thus should not be part of the finished genome sequence. On the other hand, these unmapped duplicated sequences represent true duplications that have been excluded from the assembly. One example of this occurs with a member of the mouse Bcl2 family of apoptosis regulators, Bcl2a1. Bcl2a1 contains four highly similar genes (> 97% identical at the nucleotide level) that have been mapped together on chromosome 9 of the C57BL/6 and 129SV genomes 1718. Currently, the Bcl2a1 genes are not assembled on the mapped chromosome and are found in three distinct unmapped contigs. In the human genome only one copy of BCL2A1 is found, although a recent, independent 8.5 kb tandem duplication containing the last exon of BCL2A1 has occurred, forming a novel BCL2A1-related transcript (AF249277). An example of a region that has changed between assemblies is the Amy2 locus. Amy2 is known to vary in copy number between inbred strains of mice 19. In the February 2002 assembly, only one copy of the Amy2 gene resided on chromosome 3 in addition to a second copy found on a large 10 kb unmapped contig. In addition, partial high identity matches (> 95%) to four distinct unmapped contigs were found (note that these partial copies were not detected in our analysis as they are less than 5 kb long). In the February 2003 assembly, six Amy2 genes exist, which is close to the five Amy2-like genes that were detected in the genome of strain A/J mice using quantitative densitometry of Southern blots 20. It is, however, important to note that a gap, not bridged by a clone, still exists between the Amy2 locus and the Amy1 gene, and so the copy number in the C57BL/6J genome assembly may still vary.

We analyzed the distribution of segmental duplication content by sorting the duplications into six different sequence-similarity categories: 90-92%, 92-94%, 94-96%, 96-98%, 98-99.5%, and 99.5-100%, for both the February 2002 and 2003 assembly builds (Table 2). The amount of duplication content appears to be unevenly distributed across these categories, with a distinct rise in the 94-96% category. This might suggest recent duplicative events in the mouse genome have not occurred at a steady rate. However, it is unclear at this point how these results were affected by the draft status of the genome assemblies. Between the 2002 and the 2003 assembly builds we found that the amount of duplication content is nearly the same within each percent category except for the 99.5-100% category, which contained 4.8 Mb of sequence in 2002 and 18.5 Mb in 2003 (Table 2). Furthermore, we determined that the majority (88%) of the duplicated sequence in the 99.5-100% category occurred intrachromosomally, within 200 kb of each other. Using the assembly component tables (provided by UCSC 21), which contain information about the underlying makeup of the February 2003 genome assembly (shotgun-assembled scaffolds and BAC sequences), we found that 215/216 (99.5%) of these duplications involved a BAC sequence. Hence, we suspect that the large increase in near-identical duplications could be the result of sequence misassignment errors arising from the inherent difficulty of merging finished BAC sequence with shotgun sequence contigs.

We previously observed that the human genome sequence assembled by Celera's WGS method 22 showed poor quality in regions with near-identical segmental duplications 23. To assess the finishing status of duplicated regions in the WGS mouse genome assembly (February 2002 MGSCv3 assembly), we calculated the amount of unfinished sequence (regions with gaps or Ns) within the immediate neighborhood (20 kb) of each duplicon (the unmapped chromosome sequence was excluded from this analysis). We observed substantially higher amounts of unfinished sequence (number of Ns) in these regions. Whereas 8.0% of the assembly is comprised of Ns, regions harboring duplications contain an average of 12.2%. This average rises to 16.6% for duplications with more than 98% sequence identity (statistics can be obtained from our website 14). This suggests that the WGS assembler had difficulty assembling regions containing recent sequence duplication and that these regions are good candidates for finishing using clone resources.

Using the NCBI Refseq and Ensembl mouse gene annotation, we identified 675 genes that mapped to duplicated regions of the mouse genome (a full list of genes can be obtained from our website 14); 414 of these genes were found to be fully contained within a segmental duplication, thus representing the best candidates for whole-gene duplication. While it is likely that some of these duplicate copies have become pseudogenes, others may have evolved specialized functions 3. Moreover, we sought to use the identified gene sequences, which were expressed sequence tags (ESTs) and/or cDNAs, as experimentally derived resources to help validate the genomic duplication content presented in this study. We aligned duplicated gene sequences to each genomic region using UCSC BLAT 17 and determined their percent identity matches. Unambiguous gene-to-genomic identity matches were established for all 128 gene pairs we examined. Each gene sequence was mapped to their respective genomic region with at least 99.1% identity (examples are shown in Table 3; a full table is available at 14). We also examined the identified duplicated genes using their InterPro protein-domain classification present in 608 Ensembl genes to see whether specific kinds of genes or protein domains have been preferentially duplicated. We found that genes containing protein domains related to signal transduction (rhodopsin-like G-protein-coupled receptor superfamily), olfaction (olfactory receptors, vomeronasal receptors) immunity (immunoglobulin/MHC, serine protease), and drug metabolism (cytochrome P450) are significantly enriched (by at least threefold) (Table 4).

From this list of genes, we performed a detailed analysis of Mater, a maternal-effect gene of potential medical importance. Mater encodes an autoantigen in a mouse model for human autoimmune premature ovarian failure 24. Knockout studies have shown that it is essential for early embryonic development in mice 25. Mater encodes a protein of 1,111 amino acids from a 3.5 kb transcript that spans 57 kb on mouse chromosome 7. A 42 kb segmental duplication involving two duplicons (DUP1, where Mater is located; DUP2, where a novel Mater2 is located) are situated about 5 Mb apart and in an inverted orientation (Figure 2). DUP1 and DUP2 are on average 91.1% identical over the entire 42 kb genomic region, with a 96.6% average in the exonic regions. Furthermore, we identified an intron-less Mater pseudogene (MaterP), which shares 87% DNA sequence identity to Mater, at a location 10 Mb proximal to Mater (Figure 2; see Additional data files for a detailed comparative genomic analysis of the Mater locus). The mapping locations of these duplications have been confirmed by fluorescence in situ hybridization (FISH) (Figure 3). Thus, Mater serves as one example of a gene that has been knocked out in mice but for which there is a second, highly similar transcript whose biological role is not yet known.

In addition, we were interested in determining whether any of the 675 genes have undergone recent (≥ 90% sequence identity over ≥ 5 kb) and independent duplication in the human genome. Some of these genes could be recently evolving via the 'birth and death model of evolution' which has been used to describe the evolution of the major histocompatibility complex (MHC) and immunoglobulin multigene families 26. This model describes genes that are repeatedly created through duplication, with some genes becoming fixed while others are rendered nonfunctional by deleterious mutations 26.

We examined the 675 duplicated mouse genes using best reciprocal BLAST hits to identify their putative human orthologs. We subsequently analyzed regions containing these putative orthologs for recent sequence duplication in the human genome. Sixteen of the 675 genes were found to be involved in recent, independent gene duplication in mouse and human (see Table 5). Some of these regions containing whole-gene duplications are part of multigene families known to be evolving via duplication and are found in tandem duplicated arrays in both species (that is, the Amy2, H2-Q1, Gsta1, and Olfr54 genes). An interesting example of a recent and apparently independent whole-gene duplication that occurred in mouse and human involves Bmp8a and a second intronic transcript Oxct2. Of the partial gene duplications, the recent duplication within the Tnxb gene and its human ortholog TNXB (found at the MHC III locus of mouse chromosome 17 and human 6p21) is particularly intriguing. In humans, this locus consists of a tandem array of genes (RP, C4, CYP21, and TNXB (RCCX)), which through gene duplication, can exist as mono-, di- and tri-modular forms in the caucasian population 27. Recent studies have also shown the presence of a deletion haplotype in one individual, leading to a fusion of the TNXA/TNXB gene on one chromosome and a duplication of CYP21 on the other chromosome 28. Furthermore, complex haplotypes of the complement genes (C4A and C4B) residing in the RCCX module have been characterized and postulated to have a role in individual susceptibility to infection and autoimmune disease 29. A closer inspection of the genomic region surrounding this recent duplication in the mouse reveals that the C57BL/6J duplication encompasses homologous genes (Tnxb, Slp (a C4 paralog), Cyp21a1, and C4). Similarly, in humans, this orthologous region of the mouse genome has been shown to undergo multiple recombination events, giving rise to a variety of haplotypes 30. Overall, many of the genes that have recently experienced duplications in the mouse and human genomes are of biomedical and evolutionary interest. The complexity and polymorphic nature of these recent duplications underscores the need for, and the difficulty of, performing the detailed structural and functional analyses that will help discern their true genomic organization, evolutionary history, and biological implications.

Our current analysis of the presence and organization of recent segmental duplications in the mouse genome has identified recent gene-duplication events and potentially problematic regions of the mouse genome assembly. At a practical level, identifying regions with segmental duplication will be useful in highlighting the most dynamic regions of any mammalian genome assembly. For the genome-sequencing community, these potential misassemblies/putative duplications can become initial targets for clone-based finishing; and for the biologist, they can serve as sentinels for regions of the genome most likely to change in subsequent assemblies. Additional hierarchical shotgun sequencing effort 12 will undoubtedly be critical to finish the mouse genome sequence and reveal additional duplicated regions that are incomplete at the moment.

Many of the duplicated genes are of evolutionary and medical importance (that is, genes involved in immune defense, olfaction, and drug metabolism). Knowledge of these duplicated regions could be important for accurately mapping mutants derived from ethylnitrosourea (ENU) mutagenesis, designing targeting vectors for embryonic stem cell alterations, and validating putative single-nucleotide polymrophisms (SNPs) that may have arisen from recently duplicated sequences rather than allelic variants 31. The ability to create large, sophisticated targeting vectors, by engineering BACs using homologous recombination in Escherichia coli 32, should prove very useful for designing in vivo experiments aimed at dissecting the function of recently duplicated genes. Knowledge of all recent duplications in mouse may also highlight regions subject to chromosomal rearrangement and polymorphism within and between species, and provide an opportunity to model the stability of such genomic architecture in a mammalian genome.

Further analysis of Mater duplication, including a figure showing a multiple percent identity plot of Mater versus Mater2, MATER (human), and MaterP, is available (additional data file 1).