Expression of conjoined genes: another mechanism for gene regulation in eukaryotes.

Abstract

From the ENCODE project, it is realized that almost every base of the entire human genome is transcribed. One class of transcripts resulting from this arises from the conjoined gene, which is formed by combining the exons of two or more distinct (parent) genes lying on the same strand of a chromosome. Only a very limited number of such genes are known, and the definition and terminologies used for them are highly variable in the public databases. In this work, we have computationally identified and manually curated 751 conjoined genes (CGs) in the human genome that are supported by at least one mRNA or EST sequence available in the NCBI database. 353 representative CGs, of which 291 (82%) could be confirmed, were subjected to experimental validation using RT-PCR and sequencing methods. We speculate that these genes are arising out of novel functional requirements and are not merely artifacts of transcription, since more than 70% of them are conserved in other vertebrate genomes. The unique splicing patterns exhibited by CGs reveal their possible roles in protein evolution or gene regulation. Novel CGs, for which no transcript is available, could be identified in 80% of randomly selected potential CG forming regions, indicating that their formation is a routine process. Formation of CGs is not only limited to human, as we have also identified 270 CGs in mouse and 227 in drosophila using our approach. Additionally, we propose a novel mechanism for the formation of CGs. Finally, we developed a database, ConjoinG, which contains detailed information about all the CGs (800 in total) identified in the human genome. In summary, our findings reveal new insights about the functionality of CGs in terms of another possible mechanism for gene regulation and genomic evolution and the mechanism leading to their formation.

Full text

Page 1

Expression of Conjoined Genes: Another Mechanism for
Gene Regulation in Eukaryotes
Tulika Prakash, Vineet K. Sharma, Naoki Adati, Ritsuko Ozawa, Naveen Kumar, Yuichiro Nishida,
Takayoshi Fujikake, Tadayuki Takeda, Todd D. Taylor*
MetaSystems Research Team, Computational Systems Biology Research Group, Advanced Computational Sciences Department, RIKEN Advanced Science Institute (ASI),
Yokohama, Japan
Abstract
From the ENCODE project, it is realized that almost every base of the entire human genome is transcribed. One class of
transcripts resulting from this arises from the conjoined gene, which is formed by combining the exons of two or more
distinct (parent) genes lying on the same strand of a chromosome. Only a very limited number of such genes are known,
and the definition and terminologies used for them are highly variable in the public databases. In this work, we have
computationally identified and manually curated 751 conjoined genes (CGs) in the human genome that are supported by at
least one mRNA or EST sequence available in the NCBI database. 353 representative CGs, of which 291 (82%) could be
confirmed, were subjected to experimental validation using RT-PCR and sequencing methods. We speculate that these
genes are arising out of novel functional requirements and are not merely artifacts of transcription, since more than 70% of
them are conserved in other vertebrate genomes. The unique splicing patterns exhibited by CGs reveal their possible roles
in protein evolution or gene regulation. Novel CGs, for which no transcript is available, could be identified in 80% of
randomly selected potential CG forming regions, indicating that their formation is a routine process. Formation of CGs is not
only limited to human, as we have also identified 270 CGs in mouse and 227 in drosophila using our approach. Additionally,
we propose a novel mechanism for the formation of CGs. Finally, we developed a database, ConjoinG, which contains
detailed information about all the CGs (800 in total) identified in the human genome. In summary, our findings reveal new
insights about the functionality of CGs in terms of another possible mechanism for gene regulation and genomic evolution
and the mechanism leading to their formation.
Citation: Prakash T, Sharma VK, Adati N, Ozawa R, Kumar N, et al. (2010) Expression of Conjoined Genes: Another Mechanism for Gene Regulation in
Eukaryotes. PLoS ONE 5(10): e13284. doi:10.1371/journal.pone.0013284
Editor: Pawel Michalak, University of Texas Arlington, United States of America
Received April 30, 2010; Accepted September 14, 2010; Published October 12, 2010
Copyright: ? 2010 Prakash et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits
unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: Funding support from operational expenditure fund of RIKEN is duly acknowledged. The funders had no role in study design, data collection and
analysis, decision to publish, or preparation of the manuscript.
Competing Interests: The authors have declared that no competing interests exist.
* E-mail: taylor@riken.jp
Introduction
Eukaryotic transcription is a highly complex process typically
accomplished by interaction of several proteins and regulatory
sequences at different levels to generate a variety of gene products.
The ENCODE project recently uncovered complex patterns of
dispersed regulation and pervasive transcription for at least 1% of
the human genome [1]. Subsequently, the long-standing conven-
tional definition of a gene is fading and it is now realized that the
genome is full of overlapping and other complex transcripts [2].
One such intriguing example is the read-through transcript or
conjoined or co-transcribed gene (see Table S1 for a list of alternative
and proposed names). A ‘‘conjoined gene’’ (CG) is defined as a gene,
which gives rise to transcripts by combining at least part of one
exon from each of two or more distinct known (parent) genes
which lie on the same chromosome, are in the same orientation,
and often (95%) translate independently into different proteins. In
some cases, the transcripts formed by CGs are translated to form
chimeric or completely novel proteins. Currently, only 34 CGs are
described in the NCBI Entrez Gene database, including well-
known examples such as TRIM6-TRIM34 and NME1-NME2 (see
http://metasystems.riken.jp/conjoing/faqs#ques2 for a complete
list). This ‘‘lack of annotation’’ indicates that this is either a rare
phenomenon or that this type of gene has not yet been well
characterized in the human genome due to the lack of consensus
within the genome annotation community. Also, the use of
different gene names to address such transcripts compounds the
problem of their identification.
The most widely used resources for accessing human genome
annotation information include NCBI Entrez Gene (http://www.
ncbi.nlm.nih.gov/sites/entrez?db=gene), the UCSC Genome
Browser (http://genome.ucsc.edu/),
(http://uswest.ensembl.org/index.html), and the Vertebrate Ge-
nome Annotation (Vega) database (http://vega.sanger.ac.uk/
Homo_sapiens/index.html). Even now, seven years after the
completion of the human genome, there exist many discrepancies
for annotating human genes (including CGs) among these
resources, thus somewhat limiting the identification of CGs. For
example, CG FPGT-TNNI3K is not reported in either NCBI or
UCSC although both parent genes, FPGT and TNNI3K, are
present, while in Vega and Ensembl this CG is reported as
TNNI3K and the parent gene FPGT is not present at all. To add to
the confusion, the locus representing the FPGT gene in NCBI and
UCSC is represented as a variant of TNNI3K in Vega and
theEnsembldatabase
PLoS ONE | www.plosone.org1October 2010 | Volume 5 | Issue 10 | e13284

Page 2

Ensembl (see Figure S1). Furthermore, the descriptive contents
and the terms used to address CGs across the various resources are
not uniform and could be misleading, making it difficult to find
and to search for detailed information about them. For example,
in plants, read-through transcription commonly refers to genes
having multiple polyadenylation sites leading to transcripts that
extend for variable distances into the 39 flanking region of the
genes [3]. Confusingly, many known CGs are also termed ‘‘read-
through transcripts’’ in these databases. Similarly, ‘‘fusion gene’’ is
another misnomer for CGs because the commonly understood
definition of a fusion gene is a hybrid gene formed from two
previously separate genes that occurs as the result of a
translocation, interstitial deletion, or chromosomal inversion.
Thus, we propose the term ‘conjoined gene’ (CG) for these specialized
genes that give rise to transcripts by joining the exons of two or
more distinct parent genes during transcription. For clarity, it is
important to assign these genes with unique and meaningful
names, as is done by the HUGO Gene Nomenclature Committee
(http://www.genenames.org/) for all other genes in the human
genome.
During the gene annotation of human chromosome 11, eleven
CGs were identified by our group [4]. Concurrently, three other
research groups independently analyzed the human genome and
identified unique CGs using mRNA and EST information
available in the public databases [5,6,7]. Recently, in two cancer
cell lines, MCF7 (breast) and HCT116 (colon), 70 putative CGs
were identified using Paired-End diTags [8]. Confusingly, the
definition of ‘conjoined genes’ in these analyses is variable,
including transcripts formed by combining the exons of two genes
in opposite orientations and fused transcripts formed by
chromosomal translocations or other rearrangements such as
insertions or deletions. Similarly, Li and Zhao et al. [9] recently
identified nearly 31,000 ‘‘chimeric RNAs’’ in the human genome,
however, transcripts formed by CGs were not included in their
analysis. According to Denoeud et al. [10], more than 150 CGs
were found in the 1% of the human genome analyzed for the
ENCODE project; thus, there ought to be many more instances of
CGs than have yet been found in the entire genome. Therefore,
there is still a need for systematic scanning of the human genome
for the identification of more true instances of ‘‘conjoined genes’’.
In the present study, we report the findings of our ‘‘Conjoin’’
algorithm for the identification of CGs in any genome given the
availability of its mRNA or EST sequences, along with its
reference gene annotation in the NCBI database. Detailed
examination of these genes provides useful insights about their
roles in the human genome. Furthermore, we have developed a
comprehensive database with more uniform and descriptive
annotation for the CGs to provide the research community a
specialized resource to visualize, examine, and study the
characteristics of these genes.
Results
Computational identification of CGs
To estimate the total number of CGs in the human genome,
alignments of the known genes, mRNAs, and ESTs to the human
genome generated by UCSC (Human assembly (hg18) March,
2006) were used. The algorithm ‘‘Conjoin’’ was developed and
applied to the entire human genome (see Text S1 for the details).
This resulted in 623 and 942 CG candidates from the mRNA and
EST data, respectively. In a manual curation step, false positives
arising out of misalignments of genes from the same gene family,
poor quality sequences, and short EST sequences were removed.
We also removed the false positive cases arising from splice
variants of single gene loci with multiple gene names (see Figure S2
for an example). As a result we obtained 317 and 434 CGs
supported by at least one mRNA or EST sequence, respectively.
These 751 CGs connected a total of 1,451 known, separate parent
genes, with at least one case found on each chromosome.
Interestingly, more than one third of the CGs are formed by
parent genes belonging to the functional class of ‘‘Cellular
Processes and Signaling’’ as defined in the clusters of orthologous
groups for eukaryotes (KOGs) [11] (see Text S2). In addition, 11%
(83/751) of the CGs involved parent genes with related functions,
such as TRIM6-TRIM34, CCL15-CCL14, TNFSF12-TNFSF13, etc.
Figure 1 shows an example of a known CG, NME1-NME2, in the
NCBI Entrez Gene database on chromosome 17 formed by
connecting parent genes NME1 and NME2. Only 12% (88/751) of
CGs are supported by more than one transcript, suggesting that
these genes are either relatively rare, or that they are not
ubiquitously expressed and are limited to only certain tissue types.
Interestingly, when multiple transcripts were detected for a CG,
alternative splicing was observed in 61% of the cases. Out of the
34 CGs listed in the NCBI Entrez Gene database, 30 were
identified by our approach. For the remaining four cases, no
aligned mRNA or EST sequences connecting the parent genes
were found in the version of data used from the UCSC Genome
Browser at the time of this study.
A vast majority (98%) of CGs are formed by only two parent
genes, with only a few cases where three (13) or four (4) genes are
involved. As expected, the number of CGs per chromosome is
significantly correlated with the average gene density and amount
of transcriptome data available for each chromosome. However,
an inverse correlation was observed for the average distance
between genes on each chromosome (see Text S3), implying that
genes lying closer on the genome have a higher tendency for
forming CGs. Nevertheless, these observations ruled out the
possibility of any bias towards certain chromosomes over others.
Splicing patterns of exons
For the 317 CGs supported by at least one mRNA sequence, we
evaluated the exon splicing patterns (see Methods). An interesting
pattern was observed in 42% of conjoined mRNAs, where a new
intron was created which spans the terminal exon, the 39-UTR of
the upstream (59-) gene, followed by the intergenic region between
the two parent genes, and then the 59-UTR and initial exon of the
downstream (39-) gene (http://metasystems.riken.jp/conjoing/
faqs#ques5). Similar observations were also reported by Akiva
et al. [5]. In addition, in 46% of the CGs, novel exons were
included from the intergenic or intronic regions, or from the
flanking regions of the upstream (59-) or the downstream (39-)
parent genes. New splice sites have been used in 58% of the CGs;
however, in the majority of these cases (85%) splicing occurred at
canonical sites (GT-AG). In almost all the CGs (99%), in which
splice sites remain conserved from the parent genes, splicing
occurred at canonical sites.
Experimental validation of conjoined genes
We attempted to experimentally validate 353 out of 751 CGs
using RT-PCR and sequencing methods in 16 human tissues (see
Methods). The CGs in 291 out of 353 (82%) cases were confirmed
by sequencing the expected CG transcript in at least one tissue.
One representative example is shown in Figure 2. This CG,
ZC3H10-ESYT1, is supported by one EST sequence (DB062879)
in the NCBI GenBank database and was confirmed by sequencing
of the RT-PCR product. PCR products of kidney, skeletal muscle,
spleen, and testis were selected for sequencing because they
contained representative bands of obtained PCR products.
Conjoined Genes in Eukaryotes
PLoS ONE | www.plosone.org2 October 2010 | Volume 5 | Issue 10 | e13284

Page 3

Obtained sequences revealed alternative splicing and novel exons,
as well as tissue-specific expression. The expected mRNAs could
not be sequenced for the remaining 62 CGs. This could be due to
non-expression of these genes in the tissues used for validation or
because they are expressed at very low and undetectable levels.
Interestingly, alternative splicing was observed in 63% (184/291)
of the CGs, many of which also harbor novel exons. For the
remaining 37% of the CGs, either only a single sequence was
obtained by our experiments (for example, CG PHOSPHO2-
KLHL23), or all of the sequences exhibited the same splicing
pattern (for example, CG TGIF2-C20orf24), so alternative splicing
could not be detected.
These experiments also revealed important information about
CG expression patterns. Among the tissues examined, brain and
reproductive tissues, including testis, prostate, and uterus, are
particularly enriched in CG sequences (http://metasystems.riken.
jp/conjoing/faqs#ques11). The reproductive tissues exhibit dis-
tinct polyadenylation signal biases [12], which possibly confer them
RNA diversity. Mammalian brain has been identified as the tissue
that expresses the greatest number of alternative mRNA isoforms
[13]. In addition, a large variety of non-coding RNAs have been
associated with the central nervous system and impart high
complexity to the brain by acting as gene expression regulators
[14]. This is likely to be related to the fact that the brain is
populated by thousands of highly specialized unique cell types that
undergo dynamic changes. Therefore, finding a large number of
transcripts encoded by CGs which can increase both protein-
coding and non-coding RNA diversity in these tissues is not so
surprising, and a similar role of imparting functional diversity and
gene expression regulation by CGs cannot be ruled out.
Since 69% (202/291) of CGs were found to be widely expressed
(http://metasystems.riken.jp/conjoing/faqs#ques11),formation
of CGs is expected to be a more widespread phenomenon
occurring in most tissues. However, 31% (89/291) of CGs showed
tissue-specific expression, suggesting selective expression for many
of them. Certainly, these CGs could be expressed in tissues other
than the ones we examined. Using the tissue expression data from
the NCBI UniGene database (http://www.ncbi.nlm.nih.gov/
unigene), about 76% of the conjoined and participating parent
genes were found to be expressed in tissues from various tumors.
Although several fusion genes, particularly those produced by
chromosomal rearrangements, have been found to be involved in
carcinogenesis [15], only a few CGs have been shown to be
expressed in cancer tissues, including RBM6-RBM5 [16], HHLA1-
OC90 [17], and LY75-CD302 [18]. Therefore, studying the role of
these CGs in both normal and cancerous cells in more detail may
provide important insights about the diseased conditions.
Comparison of our findings with those of others
In early 2006, Akiva et al. [5] and Parra et al. [6] independently
analyzed the human genome and identified 212 and 127 CGs,
respectively, using mRNA and EST information available in the
public databases. At the same time, another analysis done by Kim
et al. resulted in the identification of 258 unique CGs in the
human genome [7]. On close examination, we found that some of
these genes have now become obsolete due to the lack of
appropriate alignments of the CG mRNAs with the parent genes,
revision of the coordinates of the participating parent genes in the
current version of the human genome databases, or different
chromosomal location or opposite orientation of the parent genes.
As a result, only 193, 123, and 83 CGs remain valid in the Akiva,
Kim, and Parra datasets, respectively. To estimate the total
number of unique CGs, we compared the 751 CGs identified in
this study with these three datasets (Figure 3). 232 out of 751 CGs
Figure 1. CG formed by parent genes NME1 and NME2 on chromosome 17. This CG is found in the NCBI Entrez Gene database (NME1-NME2).
Two mRNA sequences (DQ109675, and BC107894) and 37 EST sequences (two are shown here, BI603164 and BQ277261) support this CG. The
upstream (59-) and downstream (39-) parent genes are shown in red and blue colors, respectively. The CG transcripts and CDSs are shown in magenta
and grey colors, respectively. Exons and introns are shown as boxes and lines, respectively. Arrows indicate the strand of the transcript.
doi:10.1371/journal.pone.0013284.g001
Conjoined Genes in Eukaryotes
PLoS ONE | www.plosone.org3 October 2010 | Volume 5 | Issue 10 | e13284

Page 4

were included in at least one other dataset, with 106 being
confirmed in at least one human tissue by our experiments. The
remaining 519 (69%) CGs were uniquely identified by our method
only, 185 of which were experimentally validated. Forty-three
CGs were uniquely identified by at least one of the Akiva, Kim, or
Parra analyses, but not by us.
We also compared the 13 chimeric transcripts obtained by RT-
PCR by Denoeud et al. [10] in the ENCODE region to our
dataset. Only two were found in common, and five were doubtful
by the criteria mentioned above. Six CGs in the ENCODE study
were uniquely identified by them. The sequences used to confirm
these CGs were not found in the version of the human genome
data from UCSC that was used for our analysis. Thus, adding the
unique CGs identified by Akiva, Kim, Parra and the ENCODE
analysis to ours results in 800 unique CGs identified in the entire
human genome to date. This clearly indicates that there may still
be many more instances of CGs in the genome and that the
existing genomic information is not yet sufficient to identify all of
them.
Functional roles of conjoined genes
Protein evolution by chimeric proteins.
of gene fusion in eukaryotic evolution has been previously
The significance
Figure 2. Example of experimentally verified CG, ZC3H10-ESYT1, on chromosome 12. (A) PCR products amplified from human tissues. This
CG (based on EST accession DB062879) was confirmed by RT-PCR in 16 human tissues. It showed various patterns in each tissue. Arrowheads indicate
the bands which were cloned and sequenced. Marker: OneSTEP Ladder 100 (0.1–2 kbp) (NIPPON GENE CO., LTD., Toyama, Japan). (B) Confirmed CG
regions displayed on the UCSC Genome Browser (hg18). The upper tracks (black) represent the seed EST and the identified CG sequences, and the
lower tracks (blue) show the RefSeq genes annotated in this region.
doi:10.1371/journal.pone.0013284.g002
Figure 3. Overlap of the CGs identified by our approach versus
those identified by others. Twelve CGs are common among all four
datasets. 519 CGs were uniquely identified by our approach. (Akiva et al
[5], Parra et al [6], and Kim et al [7] (ChimerDB)).
doi:10.1371/journal.pone.0013284.g003
Conjoined Genes in Eukaryotes
PLoS ONE | www.plosone.org4October 2010 | Volume 5 | Issue 10 | e13284

Page 5

demonstrated [19]. The specialized splicing patterns observed in
CGs can result in novel proteins, which may play a role in the
additional complexity of the human genome as demonstrated in
case of chimeric protein Kua-UBE2V1 [20]. For open reading
frame (ORF) prediction, only those CGs were selected for which at
least one mRNA sequence was found and a reasonable ORF could
be obtained (297 CGs, 409 mRNAs, see Methods). Transcripts
arising from 16% of the selected CGs used conserved reading
frames of translation from all the parent genes, thereby forming
chimeric proteins by joining the domains of their respective parent
genes (Table 1). It is well known in prokaryotes that the vast
majority of gene pairs whose orthologs are fused are either part of
the same complex, or function in the same pathway [21]. Thus
CG formation can result in co-regulation of gene expression of
functionally related proteins such as for TNFSF12-TNFSF13 [22]
and TRIM6-TRIM34 [23].
Gene regulation by CG expression.
perform other regulatory roles, such as altering the expression of
parent genes. They can achieve this by using altogether different
frames of translation in the conjoined transcript with respect to all
the parent genes, thus preventing their expression by forming a
novel protein or non- protein-coding transcript (5% of CG
mRNAs). Alternatively, the CG transcript may be translated using
the conserved reading frame from only one parent gene. The
sequence corresponding to the other parent gene is then either
translated in a different reading frame or remains as an
untranslated region (79% of CG mRNAs) (Table 1). In either of
these scenarios, expression of one of the parent genes is affected. In
some mouse tissues, CG Ankhd1-Eif4ebp3 shows a similar
expression pattern as that of Ankhd1, whereas the expression of
Eif4ebp3 is significantly lower in these tissues [24]. A similar role of
gene regulation is expected for some other CGs such as INS-IGF2,
ZFP91-CNTF, MUTED-TXNDC5, etc., in which the CG ORF is
similar to only one of the parent gene’s ORF. Interestingly, in a
slightly larger number of CGs (58%), the predicted ORF was more
like that of the downstream (39-) parent gene as compared to the
upstream (59-) parent gene (42%).
Protein expression, however, is highly controlled by the
Nonsense Mediated Decay (NMD) mechanism [25]. Thus, it is
not surprising that 18% of CGs are expected to undergo NMD
due to the appearance of a premature stop codon as a result of
frame-shift caused by altered exon-intron splicing with respect to
the parent genes, or by the inclusion of novel exons. Even in these
cases, expression of the parent genes may be temporally regulated
by the formation of a conjoined transcript.
Alternatively, CGs can
Occurrence of conjoined genes in other genomes.
functional role of CGs can be further strengthened if they are
shown to survive through selective evolutionary pressure. Hence,
we examined the conservation of CG ‘junction exons’ across 23
other vertebrate genomes using BLAT (e#1026). The ‘junction
exons’ can be defined as the exons which contain DNA sequence
from both participating parent genes, that is to say, the last exon of
the upstream (59-) gene and the first exon of the downstream (39-)
gene, in the transcripts formed by the CGs. In cases where the last
and first exons of the two parent genes, respectively, form separate
exons in the CG transcripts, with or without any novel exons
between them, the entire region was used.
More than 70% of the human CG junction exons were found to
be conserved across eight other vertebrate genomes (http://
metasystems.riken.jp/conjoing/faqs#ques4). No significant con-
servation was observed in the lower-order vertebrates including
zebrafish, X. tropicalis, medaka, stickleback, lamprey, tetraodon,
and fugu although a large number of mRNA and EST sequences
were available for all these organisms. Among the higher-order
vertebrates, maximum conservation of CG junction exons was
observed in the chimpanzee genome. We observed a large
decrease in the number of conserved human CG junction exons
from chimpanzee to macaque and orangutan. This could be due
to the poor quality of the transcriptome annotation for these
genomes. With the advancement in high-throughput transcrip-
tome sequencing technologies, such as RNAseq, more RNA
sequence data is expected to be available in the near future,
leading to the detection of additional conserved human CGs in
these and other genomes. Nevertheless, all other genomes showed
much less conservation of human CG junction exons. Therefore, it
is evident from our analysis that CG conservation does not depend
on the amount of sequence data available for any given genome;
instead, it is correlated with the order of complexity of the
vertebrate genomes.
Conservation of CGs across several different vertebrate
genomes implies that their formation is not only limited to
human. To further explore this possibility, we applied the
‘Conjoin’ algorithm on three eukaryotic genomes namely, mouse,
fruit fly, and dog. Interestingly, 270 and 227 CGs were identified
in mouse and fruit fly respectively, whereas no CGs could be
detected in the dog genome using the currently available mRNA
and EST data from UCSC. The junction exons from the mouse
and fruit fly CGs were extracted as described above for human
CGs, and were searched in the human mRNA and EST datasets
using BLAT (e#1026). Only 0.03% of the mouse CG junction
The
Table 1. Reading frames used for the formation of the CG ORFs.
Reading frames used by CG Protein Product% CG (## of CGs/Total Selected)Examples
Predicted NMD
candidates
Predicted Non-NMD
candidates
Same as all parent genes
(conserved)
Chimeric16% (48/297)NME1-NME2
5 43
Same as one of the parent
genes (partially conserved)
Similar to one of the
parent genes
79% (234/297) ARID4B-RBM34, OMA1-DAB1
47 187
Different than all parent
genes (different)
Novel or non-coding 5% (15/297)DCDC1-DCDC5
0 15
doi:10.1371/journal.pone.0013284.t001
Conjoined Genes in Eukaryotes
PLoS ONE | www.plosone.org5 October 2010 | Volume 5 | Issue 10 | e13284