Divergence in Expression between Duplicated Genes in Arabidopsis
Eric W. Ganko,* Blake C. Meyers, and Todd J. Vision*
*Department of Biology, University of North Carolina at Chapel Hill and Department of Plant and Soil Science, University of
Delaware
New genes may arise through tandem duplication, dispersed small-scale duplication, and polyploidy, and patterns of
divergence between duplicated genes may vary among these classes. We have examined patterns of gene expression and
coding sequence divergence between duplicated genes in Arabidopsis thaliana. Due to the simultaneous origin of
polyploidy-derived gene pairs, we can compare covariation in the rates of expression divergence and sequence
divergence within this group. Among tandem and dispersed duplicates, much of the divergence in expression profile
appears to occur at or shortly after duplication. Contrary to findings from other eukaryotic systems, there is little
relationship between expression divergence and synonymous substitutions, whereas there is a strong positive relationship
between expression divergence and nonsynonymous substitutions. Because this pattern is pronounced among the
polyploidy-derived pairs, we infer that the strength of purifying selection acting on protein sequence and expression
pattern is correlated. The polyploidy-derived pairs are somewhat atypical in that they have broader expression patterns
and are expressed at higher levels, suggesting differences among polyploidy- and nonpolyploidy-derived duplicates in
the types of genes that revert to single copy. Finally, within many of the duplicated pairs, 1 gene is expressed at a higher
level across all assayed conditions, which suggests that the subfunctionalization model for duplicate gene preservation
provides, at best, only a partial explanation for the patterns of expression divergence between duplicated genes.
Introduction
Functional divergence among duplicated genes is
among the most important sources of evolutionary innovation in complex organisms. Population genetic theory suggests that most duplicate gene pairs revert to single copy on
a short evolutionary timescale and that those pairs that are
retained are likely to have at least partially diverged in function (Lynch and Conery 2000). Two indicators of functional divergence that are amenable to systematic study
are changes in protein sequence and alterations in the timing, location, and relative number of gene transcripts.
A number of studies in recent years have measured the
divergence between duplicate genes in their spatiotemporal
profiles of gene expression and compared this with proxies
of duplication age, such as the number of synonymous substitutions between the 2 aligned protein coding sequences.
Using spotted cDNA data from yeast, Gu et al. (2002)
found a significant positive relationship between the divergence in expression and the estimated number of synonymous (KS) and nonsynonymous (KA) substitutions per site,
though the relationship held only for KA 0.3. This contradicted an earlier, more limited, study where the weak relationship between correlation in expression and protein
sequence divergence was not significant (Wagner 2000).
In agreement with Gu et al. (2002), Makova and Li
(2003) showed that KS and KA are positively related to
the divergence in expression among different tissues in human duplicate gene pairs for KA , 0.2. Considerable divergence in expression had occurred in about 50% of the pairs
even when there was very little sequence divergence (KS ,
0.1). This suggests that gene pairs diverge rapidly in their
spatiotemporal pattern of expression at or near the time of
inception and continue to diverge steadily thereafter.
Studies such as those cited above depend on the assumption that KS, and to a lesser degree KA, can be treated
Key words: duplicate gene, subfunctionalization, linkage, MPSS,
AtGenExpress, Arabidopsis thaliana.
E-mail: [email protected].
Mol. Biol. Evol. 24(10):2298–2309. 2007
doi:10.1093/molbev/msm158
Advance Access publication August 1, 2007
as proxies for time. Although it is well known that this is not
strictly true (e.g., Zhang et al. 2002), it is generally assumed
that variation in substitution rates will only obscure whatever trends may exist between expression divergence and
time but would not be positively misleading. Furthermore,
it is assumed that variation in the substitution rate among
gene pairs is not related to variation in the rate of expression
divergence. Contrary to this assumption, recent studies of
multiple pairs of orthologous genes between closely related
Drosophila species (Nuzhdin et al. 2004; Lemos et al. 2005)
and between distantly related Caenorhabditis elegans and
Caenorhabditis briggsae (Castillo-Davis et al. 2004) suggest that divergence in protein sequence and gene expression may be coupled. These findings raise the question of
whether divergence in protein sequence and gene expression are also coupled in paralogous gene pairs.
Arabidopsis thaliana is an ancient polyploid and
thus has large numbers of ‘‘segmental duplicates’’ of
near-simultaneous age (Vision et al. 2000; Zhang et al.
2002; Blanc et al. 2003). For this reason, and because of
the broad and deep expression data available (Meyers,
Tej, et al. 2004; Schmid et al. 2005), Arabidopsis is an attractive model for studying the relationship between substitution rate and expression divergence.
Previously, Haberer et al. (2004) estimated that twothirds of duplicate gene pairs had divergent expression in
Arabidopsis. Casneuf et al. (2006) concluded that nonsegmentally duplicated pairs diverge more rapidly than segmentally duplicated pairs and also that the rate of
expression divergence is somewhat dependent upon inferred gene function.
In addition to segmental duplicates, we recognize
2 other categories of duplicated gene pairs in this study.
‘‘Tandem duplicates’’ are closely adjacent to one another
in the genome and most likely arise through a mechanism
of unequal crossing over. ‘‘Dispersed duplicates’’ are found
neither adjacent to each other in the genome nor within
homeologous chromosome segments. A number of different mechanisms may directly create dispersed pairs of duplicated genes in eukaryotes (Katju and Lynch 2003). Some
dispersed pairs may also be due to relocation of one copy of
a tandem or segmental duplication through a secondary
Ó 2007 The Authors.
This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/
by-nc/2.0/uk/) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
Divergent Expression in Duplicated Genes 2299
chromosomal rearrangement. Recent work suggests that the
evolutionary dynamics of duplicate gene retention and
functional divergence may differ substantially among these
categories (reviewed in Vision 2005; Freeling and Thomas
2006).
Gene expression divergence may play an important
role in the preservation of duplicated genes. The probability
of preserving both duplicates is quite small if preservation
depends solely on ‘‘neofunctionalization,’’ the appearance
of a mutation that modifies one copy to perform a new,
adaptively favored, function. More frequently, a deleterious
mutation would be expected to appear first that would lead
to nonfunctionalization or loss of one of the duplicate copies (Walsh 1995). An alternative mechanism by which both
copies of a pair can be prevented from undergoing nonfunctionalization is via ‘‘subfunctionalization’’ or the acquisition of complementary degenerative mutations in the 2
copies that partition some of the functionality of the ancestral gene into its 2 descendents (Force et al. 1999). One way
in which subfunctionalization can occur in organisms with
complex gene regulation is through complementary
changes in transcription factor–binding sites that result in
differential expression in the 2 copies. In the simplest formulation of the model, there would be at least 1 condition or
cell type in which each copy is expressed and the other is
not.
Lynch et al. (2001) have shown that the probability
that a duplicate pair is retained through subfunctionalization versus neofunctionalization depends on several factors,
including population size and the frequency of recombination between the pair. Because subfunctionalization does
not depend on the occurrence of a favorable mutation, it
is more likely than neofunctionalization to explain the retention of duplicate genes in organisms with small population sizes, such as multicellular plants. Additionally, tight
linkage facilitates subfunctionalization because of population genetic processes acting during fixation of the new duplication. Because tandem duplicates are in close linkage,
this leads to the prediction that subfunctionalization should
be a more frequent mechanism for the retention of tandem
duplicates than for other kinds of duplicated pairs. We can
test this prediction by comparing young tandem pairs with
young dispersed pairs.
Two Arabidopsis gene expression data sets are used in
this study. One derives from Massively Parallel Signature
Sequencing (MPSS, Brenner et al. 2000) described more
fully in Meyers, Tej, et al. (2004). MPSS provides a digital
signature of mRNA abundance for each gene, similar to that
obtained using Serial Analysis of Gene Expression, and has
2 attractive features for this study. 1) mRNA abundance can
be reliably quantified even at low levels because there is
essentially no noise. 2) The expression measurements for
closely related duplicated genes are not subject to crosshybridization artifacts, which is a major impediment to
studying divergence among gene family members using
hybridization-based gene expression assays (Xu et al.
2001).
The second data set is the Arabidopsis Development
Atlas (ADA, Schmid et al. 2005), which contains triplicate
expression estimates for ;80% of known Arabidopsis
genes across 79 different tissues and developmental time
points using the Affymetrix ATH1 microarray. In contrast
to the MPSS data set, which consists of 5 coarsely sampled
mRNA libraries, the ADA data set allows us to detect relatively fine anatomical and developmental differences between expression profiles. However, in order to minimize
the effects of cross-hybridization in the ADA data set, it was
necessary to exclude array probes that matched more than 1
gene; as a result, there is relatively poor sampling of duplicates with small amounts of sequence divergence.
Materials and Methods
Sequence Analysis of Duplicate Pairs
For this study, we focus on gene families containing
only 2 members. Because divergence between such pairs is
(relatively) free of influence by functional overlap with
other gene family members, the divergence in expression
pattern between them is relatively straightforward to interpret. However, we cannot exclude the possibility that extinct gene family members may have influenced the
divergence of the pairs we study.
Pairs of duplicated genes were selected as follows. An
all-by-all BlastP (Altschul et al. 1997) with default parameters was performed with all annotated Arabidopsis protein
sequences from TIGR version 3.0. The Blast output was
processed by TribeMCL (Enright et al. 2002), with an expansion parameter of 5, to cluster each gene uniquely into
one gene family. Each pair was classified as either tandem
(T), segmental (S), or dispersed (D). If the 2 members of
a pair were within 20 annotated genes of each other on
the same chromosome, they were assigned to class T. If
they were identified as anchors of a segmental homology
using fluorescent in situ hybridization (Calabrese et al.
2003), they were assigned to class S. The remaining pairs
were assigned to class D.
Each pair of protein sequences was aligned using
ClustalW (Thompson et al. 1994) using default parameters.
The DNA sequences were aligned codon-by-codon, using
the protein sequences as guides, and per-site synonymous
(KS) and nonsynonymous (KA) substitution rates were calculated using PAML (Yang 1997). Pairs with KS 1 were
discarded. Codon bias, measured as the effective number of
codons (ENC), was calculated for each gene pair using CodonW (Wright 1990) to test whether highly biased genes
were present in the data set that would lead to anomalously
low estimates of the number of synonymous substitutions
between duplicate genes (Friedman and Hughes 2001). Because we wished to restrict our study to functional genes,
annotated pseudogenes were ignored. In addition, we examined each gene pair for evidence of premature truncation.
Less than 10% of gene pairs differ in size by more than 25%
(relative to the smaller of the pair). These pairs are evenly
distributed among the duplication categories.
Measuring Gene Expression Using MPSS Data
We analyzed transcript abundance in 5 cDNA libraries
derived from callus, inflorescence, leaf, root, and silique
tissues harvested under standard growth conditions from
A. thaliana accession Col-0. Details of plant growth, library
2300 Ganko et al.
construction, and the analysis of the MPSS sequence signatures obtained are reported in Meyers, Tej, et al.
(2004) and Meyers, Vu, et al. (2004). Signatures were obtained using the ‘‘classic’’ MPSS method. Although the total number of signatures obtained from each library ranged
from approximately 1.7 to 3.6 million, all signatures are
normalized to transcripts per million (TPM) (Meyers,
Tej, et al. 2004). These data are available from http://
mpss.udel.edu/at/.
In order to compare the observed statistics with those
expected for gene pairs that had diverged to the point of
randomness, the same statistics were calculated for 100
MPSS and ADA data sets in which gene identities and expression profiles had been randomly paired.
Each measure of divergence was compared with KA
and KS by linear regression for both observed and shuffled
data sets. A Bonferroni-corrected P 5 5.2 104 (corresponding to an alpha value of 0.05) was used as a threshold
of significance to account for multiple testing.
Measuring Gene Expression Using Affymetrix Data
We also analyzed expression data from the ADA (ExpressionSet_ME00319 at ftp://ftp.arabidopsis.org/), which
contains triplicate expression estimates for each of 79 tissues and developmental time points using the Affymetrix
ATH1 chip (Schmid et al. 2005). The ATH1 chip has
22,746 probes representing .80% of known Arabidopsis
genes. We matched each Affymetrix probe to the TAIR
Arabidopsis genome annotation (ftp://ftp.arabidopsis.org/
home/tair/Microarrays/Affymetrix/old/affy8k_array_elements2004-06-01.txt) and excluded from analysis any probes that
matched multiple TAIR genes. The mean signal value of each
triplicate was calculated for each probe, under each condition,
only when all 3 replicates had a ‘‘present’’ call. All other instances were treated as absent with a signal value of 0.
Analysis of Expression Divergence between Duplicates
We employed a number of different measures of the
divergence between 2 expression profiles, including AE, average level of expression per library; IL, proportion of
shared libraries; and NL, number of libraries where at least
1 copy of a pair is expressed (see Results). The normalized
Manhattan distance dN calculates the distance between the
expression vectors of genes i and j,
5 xj;k
1X
xi;k
P5
dN 5
P5
;
2 k51 xi;k
xj;k k51
k51
where xi,k is the expression of gene i in P
library k and each
element of the vector is standardized by 5k51 xi;k , the total
expression of that gene over all libraries. It is more sensitive
to the proportion of libraries that differ among the genes
than an alternative measure based on Euclidean distance.
Another measure of the similarity in the expression
profiles of 2 genes is the Pearson product-moment correlation, rP; this measure is more useful for the ADA data set
because there are an insufficient number of libraries in the
MPSS data set to obtain a reliable correlation estimate. Note
that both measures are only sensitive to differences in the
relative abundance of transcripts among libraries and not to
absolute differences in abundance between 2 pairs of genes.
Following Gu et al. (2002) and Makova and Li (2003), we
used a log[(1 þ rP)/(1 rP)] transformation for the ADA
pffiffiffiffi
data set. However, we found an arcsin rp transformation
to yield much more normally distributed residuals for the
MPSS data set, and so we report these results using this
transformation. The qualitative conclusions are not dependent on the transformation chosen.
Measuring Symmetry of Expression Divergence
We used different tests to determine whether duplicated genes differed in expression level within a library.
For the MPSS data set, we compared the higher signal with
a 95% binomial confidence interval around the lower signal
value. For the ADA data set, the higher signal was compared with the 95% prediction limit (Sokal and Rohlf
1995) for a sample size of 3 fitted to the lower signal value.
Results
Divergence between gene pairs from 2-member gene
families is relatively free of influence by functional overlap
with other gene family members, simplifying the interpretation of expression pattern divergence between them. Clustering of the pairwise matches obtained from an all-by-all
sequence similarity search of the predicted Arabidopsis proteome yielded 984 gene families of size 2. Of these, 503
pairs were estimated to have an average level of synonymous divergence among sites KS , 1 and were retained
for analysis (supplementary table 1, Supplementary Material online). All genes in this set had an ENC greater than
35, indicating that synonymous substitutions were not constrained by codon usage. A comparison of the chromosomal
positions of the members of each pair relative to one another and to duplicated genomic segments within Arabidopsis allowed us to subdivide all the duplicated pairs into 65
tandem, 152 segmental, and 286 dispersed pairs (supplementary table 1, Supplementary Material online). Of the
152 segmental pairs, all but 5 were present in the data
set of segmentally duplicated pairs of Blanc et al.
(2003), and all those that were present were classified as
having occurred during the most recent large-scale duplication event 40–80 MYA (Lynch and Conery 2000; Vision
et al. 2000; Simillion et al. 2002; Blanc et al. 2003; Bowers
et al. 2003). We treat these duplicates as having arisen simultaneously and interpret the variability in KS and KA
among these pairs to be due to variability in the rate of substitution as in Zhang et al. (2002).
Transcript Abundance in the Data Set Relative to Other
Genes
Levels of expression were measured by MPSS for
each gene in 5 different libraries (callus, flower, leaf, root,
and silique). Of the 503 gene pairs, 11 had a total expression
level of less than 5 TPM summed across all 5 libraries in
Divergent Expression in Duplicated Genes 2301
both members of the pair and were discarded. In all, 492
gene pairs remained for study in the MPSS data set and
were subdivided into 63 tandem, 149 segmental, and 280
dispersed pairs. The 984 genes represented by the 492 pairs
had median expression levels ranging from 11.5 to 15 TPM
and maximum expression levels ranging from 3,000 to
5,000 TPM among the libraries. Using a threshold of at least
5 TPM to declare that a gene is expressed in a library, there
is an unusual bimodality in the pattern of gene pair expression observed. The majority of genes are expressed in either
less than 2 libraries or constitutively in all 5. The distribution is similar among all Arabidopsis genes, though there
are somewhat more constitutively expressed genes among
the 984 in the duplication data set (table 1A). In over twothirds of the pairs (345), each copy is expressed in at least 1
library. Interestingly, these jointly expressed pairs are more
frequent in the segmental class (76.5%) than in the dispersed (67.5%) and tandem (66.6%) classes, respectively
(Fisher’s exact test, two-tailed: P , 2 1016).
We performed a similar analysis on the ADA data set.
To account for potential cross-hybridization problems in
these data, gene pairs that could not be unambiguously
matched to probes on the ATH1 chip were discarded.
The 325 remaining pairs there were subdivided into 185
dispersed, 110 segmental, and 30 tandem duplicate pairs.
Using the present call as a threshold for minimal expression
within the ADA data set, we observed a median signal of
231, with levels ranging from 5 to over 37,000, across all 79
libraries. Unlike the MPSS data set, we did not observe a bimodal distribution of expression among duplicate pairs but
rather an abundance of pairs with broad expression in many
libraries (table 1B). Approximately one-third (32%) of duplicate genes were expressed in all 79 libraries and over
two-thirds were expressed in .70 tissue and developmental
libraries. Only 11 duplicate gene pairs (2%) lacked a significant signal value in all 79 libraries, much lower than the
12% of all genes lacking a single significant value (table 1B).
Transcript Abundance and Sequence Conservation
We then compared the total expression of a pair, TE,
with the rate of synonymous (KS) and nonsynoymous substitution (KA) between pairs. As with all the subsequent
analyses relating sequence divergence to some measure
of expression, the data are quite noisy and the linear trends
we identify only explain a portion of the variance. Nonetheless, a number of striking patterns are evident.
Summing across all 5 libraries in the MPSS data, we
found that TE declined precipitously with KA (log(TE), P ,
6.72 1013; seen in fig. 1A). The decline was approximately twice as steep among segmental pairs as in dispersed
and tandem pairs (interaction between KA and duplication
type: P 5 0.024), and the negative relationship was significantly different from 0 for the segmental and dispersed duplication types when considered separately. The median KA
for pairs with a total expression level greater than 1,000
TPM was 0.05 versus 0.13 for pairs with expression levels
of 1,000 TPM or less. In contrast, the total expression level
is weakly related to KS only within segmental duplicates
and not significant after correcting for multiple tests. A significant negative relationship was also observed between TE
Table 1
The Number of Libraries in Which Each Gene Is Expressed
A. MPSS
Libraries
0
1
2
3
4
5
Genes in pairsa (%)
153 (16)
124 (13)
73 (7)
93 (9)
109 (11)
432 (44)
All genesb (%)
4160 (23)
2643 (15)
1727 (10)
1777 (10)
1930 (11)
5612 (31)
B. ADA
Libraries
0
1–10
11–20
21–30
31–40
41–50
51–60
61–70
71–78
79
Genes in pairsc (%)
11 (2)
38 (6)
26 (4)
11 (2)
18 (3)
23 (3)
34 (5)
41 (6)
240 (37)
208 (32)
All genesd (%)
2693 (12)
3063 (14)
1403 (6)
993 (4)
893 (4)
846 (4)
957 (4)
1647 (7)
6646 (29)
3605 (16)
a
The 984 genes analyzed in the MPSS data set.
All 17,849 Arabidopsis genes in the MPSS data set.
c
The 650 genes analyzed in the ADA data set.
d
All 22,746 Arabidopsis genes in the ADA data set.
b
and KA (log(TE), P., 2.2 1016) for all gene pairs in the
ADA data (fig. 1B) and also within the segmental and dispersed duplication types separately. In agreement with the
MPSS data, no significant relationship was observed between TE and KS. TE was, on average, higher than that found
in the randomized data only for gene pairs with KA less than
;0.1 (fig. 1).
TE conflates the breadth of expression and the average
level of expression within those libraries in which the genes
are expressed, so we separately quantified the breadth of
expression by counting the number of libraries (NL) in
which at least one of the copies is expressed. We found that
NL declined with increasing KA independent of duplication
class in both data sets (fig. 2) and showed no significant
relationship to KS (results not shown).
We examined abundance independently of breadth by
measuring the average level of expression (AE) among those
libraries in which expression was substantially different
from 0 (5 TPM in the MPSS data set and a present call
for all 3 replicates in the ADA data set). With MPSS, average expression showed a significant interaction between
KA and duplication type even when NL was included in the
model (P 5 2.3 105). The interaction was due to a significant negative relationship between AE and KA among
segmental and dispersed, but not tandem, pairs. The slope
was approximately 3 times greater for segmental than dispersed pairs (fig. 3A). We also observed a significant interaction between KA and AE in segmental and dispersed pairs
in the ADA data, where the slope of segmental pairs was
approximately 2 times greater than dispersed pairs (fig. 3B).
In both data sets, no significant interaction was detected between KS and AE. Thus, considering the combined results
for TE, NL, and AE in segmental pairs, it appears that both
the breadth of expression and the absolute level of expression decline most strongly with the rate of nonsynonymous
substitution.
2302 Ganko et al.
FIG. 1.—Log10 transformed total expression (TE) as a function of KA
for all duplication types combined. A solid black line shows the
regression through the data, whereas the dotted line shows the mean value
among randomly shuffled data sets. The boxplot shows the median
(central dot), interquartile range (box), and maximum and minimum
scores (whiskers) for each data set. Outliers are shown as small dots
outside of the whiskers. (A) MPSS; (B) ADA.
Divergence in Expression as a Function of Sequence
Divergence
In the analyses reported above, the 2 members of
a gene pair were treated as 1 unit. However, we are particularly interested in divergence between members of a duplicated pair.
The normalized Manhattan distance, dN, is one measure of divergence that compares differences in the relative
abundance of 2 genes across a set of libraries (even when
2 genes have very different baseline expression levels). In
both data sets, we found no relationship between dN and KS
either for the combined data or for each duplication type
separately (results not shown) but did find a highly significant positive relationship between dN and KA independent
of duplication type (fig. 4). For MPSS, the linear regression
FIG. 2.—Breadth of expression (NL) as a function of KA in the data
sets for all duplication types combined: (A) MPSS, (B) ADA. Figure
conventions as in figure 1.
equation was dN 5 0.34 þ 0.67 KA (P , 2.7 106),
whereas for ADA the equation was dN 5 0.19 þ 0.90
KA (P , 7.2 108). It is noteworthy that the intercepts
of both these equations were significantly greater than 0
(P , 2 1016 for MPSS, P , 1 1012 for ADA), indicating that even pairs identical in amino acid sequence
would be predicted to have quite distinct expression patterns, especially in the MPSS data set. No significant interaction with duplication type was observed in the MPSS data
but there was a significant interaction in the ADA data. Dispersed pairs showed a positive relationship (dN 5 0.19 þ
0.93 KA, P , 2 1005) with a significantly positive intercept (P , 2 106) but segmental and tandem duplicates did not.
An alternative measure is the Pearson productmoment correlation (rP), which summarizes the similarity
in relative levels of expression among libraries. As with
dN, we found neither significant overall relationship between rP and KS nor any interaction with duplication type
Divergent Expression in Duplicated Genes 2303
FIG. 4.—Normalized expression distance (dN) as a function of
coding sequence divergence for all duplication types combined: (A)
MPSS, (B) ADA. Figure conventions as in figure 1.
(log[(1 þ rP)/(1 rP)] 5 1.9–6.5 KA, P 5 4.7 104,
fig. 5).
Overlap in Specificity Declines with KA
FIG. 3.—Log10 transformed average expression (AE) as a function of
KA for all duplication types combined: (A) MPSS, (B) ADA. Figure
conventions as in figure 1.
in either data set. For KA, the slope of the relationship for all
duplication types combined and the interaction among duplication types were not significantly different from 0 in the
MPSS data set, nor was the overall slope significantly nonzero in the ADA data set. However, there was a significant
interaction and a strong negative relationship between rP
and KA among segmental pairs in the ADA data set
An alternative way to measure expression divergence
is to quantify the overlap in expression by counting the
number of libraries in which both duplicates are expressed
relative to all the libraries in which either are expressed.
This measure, which we refer to as IL for ‘‘library identity,’’
varies from 0 for duplicates with completely nonoverlapping expression patterns to 1 for duplicates expressed in
all the same libraries.
Taking 5 TPM as the threshold for expression of a gene
in a given MPSS library, the distribution of IL was strongly
bimodal. Approximately 32% of the pairs shared no libraries and 27% shared all libraries. There were considerably
more true pairs with IL 5 1 than among shuffled pairs
(19.7%). We were unable to find an optimal transformation
2304 Ganko et al.
FIG. 5.—Transformed correlation (rP) as a function of KA in the
ADA data set for segmental pairs. Figure conventions as in figure 1.
for IL that yielded normally distributed residuals, so statistical tests must be treated with caution. Nonetheless, we
found no significant overall relationship between IL and
KS and found a significant overall negative relationship
with KA (fig. 6A, IL 5 0.63 1.12 KA, P 5 7.85 1008); there were no interactions with duplication type.
In the ADA data set, the majority of pairs were expressed
in nearly the same set of libraries (55% have IL . 0.9 and
23% have IL 5 1). Only 11 pairs (3%) have IL 5 0. There is
no significant relationship between IL and KS. However,
a significant negative relationship between IL and KA is observed (fig. 6B, IL 5 0.92 1.22 KA, P , 4.0 1009),
particularly among dispersed duplications (IL 5 0.91 1.26 KA, P , 5.9 1006). The regression equation for
the ADA data set predicts that IL would be approximately
equal to 1 for KA 5 0, indicating that pairs identical in
amino acid sequence would be expressed in a nearly identical set of libraries.
Asymmetry of Expression between Duplicates
The model of duplicate gene preservation by subfunctionalization by Force et al. (1999) predicts complementary
expression patterns following duplication, especially in
young duplications. Although the presence of one copy
and the absence of another from a particular library (as measured above) may reflect an extreme form of specialization
within a gene pair, a more subtle specialization may be detected as a complementary expression pattern in which each
copy has higher expression in at least 1 library. Alternatively, the copies may not differ significantly among the libraries or may show an asymmetric pattern in which one
copy predominates in all libraries that differ between the
copies.
FIG. 6.—The proportion of shared libraries (IL) as a function of
coding sequence divergence in the data sets for all duplication types
combined: (A) MPSS, (B) ADA. Figure conventions as in figure 1.
Surprisingly, in the MPSS data set, we found that most
pairs had an asymmetric pattern of expression across the
libraries. In over 70% of pairs, more than 2 libraries differed
between the gene copies and one of the copies had higher
expression in all of them (table 2A). The distribution varied
somewhat between duplication types. Young (KS 0.5)
dispersed and segmental duplicates contained the most such
asymmetric pairs.
One concern with the MPSS data set is the small number of libraries and the coarse sampling of tissues for each.
It is conceivable that more complementary pairs would
emerge if more, or more finely sampled, libraries were examined. However, in the ADA data set, which contains 79
different libraries, we find evidence of a similar asymmetry
of expression between copies (table 2B). Nearly 70% of duplicate pairs demonstrated asymmetric expression across 10
or more libraries. In contrast, a complementary expression
pattern was observed in only ;10% of the pairs. Older dispersed pairs were more likely than young ones to show such
complementary expression, whereas segmental pairs had
the lowest level of complementarity.
Divergent Expression in Duplicated Genes 2305
Table 2
The Number of Gene Pairs with Complementary Versus Asymmetric Expression Divergencea
A. MPSS
Dispersed
Youngf
Oldg
Tandem
Youngf
Oldg
Segmental
B. ADA
Dispersed
Youngf
Oldg
Tandem
Youngf
Oldg
Segmental
Complementaryb
Asymmetricc
Ambiguousd
Equivalente
2
4.0%
4
1.8%
1
1.9%
0
0.0%
5
3.4%
35
70.0%
161
68.8%
31
59.6%
6
60.0%
107
73.2%
13
26.0%
58
24.8%
18
34.6%
4
40.0%
34
23.2%
0
0.0%
5
2.1%
2
3.8%
0
0.0%
0
0.0%
0
0.0%
20
12.3%
4
18.2%
4
50.0%
7
6.3%
15
68.2%
106
65.0%
15
68.2%
3
37.5%
77
70%
1
4.5%
14
8.6%
1
4.5%
0
0.0%
16
14.5%
5
22.7%
23
14.1%
2
9.1%
1
12.5%
10
9.1%
a
The percentage among all pairs in a given class and age is shown below each count.
Each copy had higher expression in at least 1 library.
c
One duplicate had higher expression in all libraries that differed and at least 2 libraries differed.
d
Only 1 library differed between duplicates.
e
No libraries differed between duplicates.
f
KS 0.5.
g
KS . 0.5.
b
The loss of functionality in 1 gene of the pair could
lead to rapid asymmetric-like changes in expression level
with little selective constraint. Treating size as a proxy
for function, we found that ,10% of our gene pairs differ
in size by more than 25% (relative to the smaller of the pair).
Further, these pairs show no distribution bias among duplication categories and are more likely to have complementary or equivalent expression patterns rather than
asymmetric expression patterns.
Discussion
Rapid Expression Divergence among Duplicate Pairs
Our study highlights a number of striking patterns in
the divergence of expression profiles between duplicated
gene pairs in Arabidopsis. First, it is clear that many pairs
have substantially diverged expression patterns shortly after
‘‘birth’’ as evidenced by the dissimilarity between pairs that
are nearly identical in sequence. Previous studies in yeast,
humans, and Arabidopsis have also suggested a rapid phase
of initial divergence between duplicates (Gu et al. 2002;
Makova and Li 2003; Blanc and Wolfe 2004; Gu et al.
2005; Yang et al. 2005; Casneuf et al. 2006). At the same
time, however, the variability in the extent of expression
divergence among very young pairs is quite high, as has
also been seen in these other systems.
One way in which rapid divergence could occur is if,
as a result of incomplete duplication, the cis-regulatory sequences of the 2 genes are initially structurally dissimilar
(Katju and Lynch 2003). This would only apply to tandem
and dispersed duplicates; segmental pairs are structurally
identical, at least initially. But the relationship between
cis-regulatory change and expression divergence is highly
idiosyncratic. Haberer et al. (2004) noted that tandem and
segmental duplicate gene pairs had divergent expression in
Arabidopsis even when they shared many similar cisregulatory sequences and suggested that changes to a small
fraction of cis-elements could be sufficient for neofunctionalization or subfunctionalization.
In addition, epigenetic differences between duplicates
that are not apparent in our data set may contribute to rapid
expression differentiation. For example, organ-specific differentiation of duplicate gene expression immediately following polyploidy has been observed in cotton (Adams
et al. 2003), as has silencing of polyploid-derived duplicates due to hypermethylation in Arabidopsis polyploids
(Wang et al. 2004). In fact, a number of epigenetic mechanisms are known that could potentially differentiate newly
arisen, structurally identical, duplicates (Matzke et al. 2003;
Rapp and Wendel 2005; Chen and Ni 2006). Such epigenetic effects on expression divergence could apply to all 3
classes of duplicates and lead to rapid subfunctionalization
(Rodin and Riggs 2003) or nonfunctionalization.
Despite the overall pattern of rapid expression divergence, there is strong similarity in the expression profiles of
some duplicate pairs that, on the basis of their sequence
similarity, appear to be rather old. Some theoretical models,
such as those of Lynch et al. (2001), predict that duplicate
genes with redundant functions should resolve to become
2306 Ganko et al.
nonredundant within several million generations. Although
expression in the same library does not constitute strong
evidence for functional redundancy, a number of functional
studies have raised the possibility that some duplicate pairs
generated during the most recent polyploidization event
still have not fully lost the overlap in their functionality
(e.g., Pinyopich et al. 2003; Azevedo et al. 2006). Nowak
et al. (1997) and Wagner (1999) have proposed models to
explain the remarkable evolutionary persistence of redundant functionality between some duplicates; the driving
force behind these models is selection for the production
of offspring that do not suffer deleterious mutations in nonredundant genes.
with covariation among duplicated pairs in the strength
of positive selection acting on protein sequence and expression pattern (Shiu et al. 2006).
Our results also suggest that there may be stronger
sequence conservation, both in protein sequence and in
expression profile, for genes with greater expression
breadth and greater average expression level. We found that
both the breadth and the average level of expression declined with KA, even among simultaneously duplicated
pairs. This is consistent with a number of studies from
mammals suggesting greater protein sequence conservation
and selective constraint for genes with broader expression
patterns (Zhang and Li 2004; Jordan et al. 2005).
Covariation of Expression Divergence and
Nonsynonymous Substitution Rate
Frequent Asymmetric Divergence between Duplicates
Another major finding is that we did not see a positive
relationship between divergence in expression profile and
the number of synonymous substitutions per site (KS),
but we did observe a positive relationship with the number
of nonsynonymous substitutions per site (KA). Most strikingly, rP was negatively correlated with KA among segmental pairs in the ADA data set (supplementary table 2,
Supplementary Material online). Because the segmentally
duplicated pairs in this study diverged nearly simultaneously and the sequence divergence since the most recent
polyploidy would dwarf any pre-existing allelic variation
between the future paralogs, the relationship does not appear to be due to covariation with the age of the paralogs.
Rather, there appears to be a coupling between the rate of
expression divergence and the rate of nonsynonymous substitution for a gene pair, presumably through covariation in
the strength of functional constraint (or positive selection)
among gene pairs. Zhang et al. (2002) previously reported
a wide range of KA:KS ratios among segmentally duplicated
Arabidopsis genes (from less than 0.01 to over 0.6) showing that levels of constraint clearly vary among gene pairs.
Previous suggestions for a coupling between protein
and regulatory divergence have come primarily from studies of orthologous genes. Duret and Mouchiroud (2000)
identified a negative correlation between KA and broad tissue expression for both protein coding and untranslated regions of human–mouse orthologs with no correlation to KS.
Nuzhdin et al. (2004), studying orthologs between Drosophila melanogaster and Drosophila simulans, also found
a correlation between KA and expression divergence but no
correlation to KS. In nematodes, Castillo-Davis et al. (2004)
reported a correlation between protein and cis-regulatory
sequence evolution in C. elegans and C. briggsae orthologs
but, contrary to our findings, found no such relationship
among paralogs.
Khaitovich et al. (2004) suggested, from an analysis of
mammalian expression data, that expression patterns diverged as if they were, on the whole, neutral traits. Because
neutrality does not preclude variable-purifying selection,
this idea is not irreconcilable with the above findings on
principle. But such a neutral model would also predict a positive relationship between expression divergence and KS, as
a proxy for time, among tandem and dispersed pairs, which
we do not observe. Thus, our results are more consistent
In both the MPSS data and the ADA data, we found
pervasive expression asymmetry, where one copy had
a higher expression level than the other in all the libraries
in which they differed. Approximately 70% of the gene
pairs in both data sets showed asymmetric divergence compared with less than 10% of gene pairs that showed complementary divergence. In a subset of these cases, the
asymmetry is extreme; in approximately 15% of the MPSS
gene pairs and 6% of the ADA gene pairs, one copy was not
expressed in any of the libraries.
A slightly lower proportion of asymmetry is seen in
tandem pairs within the MPSS data, which has a larger tandem pair sample. One possible explanation for the lower
proportion may be gene conversion, which is thought to occur preferentially, though not exclusively, within tandem
arrays when it occurs between genic sequences in plants
(Benovoy et al. 2005; Mondragon-Palomino and Gaut
2005).
A number of gene family studies in Arabidopsis have
also reported asymmetric expression divergence. For example, Casneuf et al. (2006) found asymmetric expression divergence between duplicate genes that were translocated or
part of small-scale duplications. In another case, Duarte
et al. (2006) identified paralogous pairs of Type II
MADS-box genes in Arabidopsis, where the expression
of one paralog was significantly lower than the expression
of its sister paralog in virtually all tissues, yet still exhibited
selective constraint. Earlier, Zhao et al. (2003) reported
consistently higher expression for some genes within the
SKP-1–like gene family in Arabidopsis across multiple tissues when compared with their sister genes.
It is not possible to exclude the option that some of the
apparently hypofunctional copies may, in fact, be functional at very low concentrations or expressed at higher levels under conditions or in tissues that have yet to be sampled
(e.g., see Adams et al. 2003). For instance, if the 2 copies
were expressed in different cells within an organ, or even in
different compartments within a cell, it would not be apparent in a microarray experiment using whole-organ RNA
extracts. The duplicated gene pair GMD1 and GMD2
(AT5G66280 and AT3G51160) provides an example of
such a case. These genes produce variants of GDP-d-mannose 4,6-dehydratase in the cell wall. GMD2 is expressed
nearly ubiquitously in Arabidopsis, excluding the root tip,
whereas GMD1 is expressed almost exclusively in the root
Divergent Expression in Duplicated Genes 2307
tip (Bonin et al. 2003). Presumably, RNA sampled from the
root tip, rather than the whole root, would demonstrate that
GMD1 is more highly expressed in this specific tissue and,
thus, the pair would be seen to have a complementary,
rather than asymmetric, pattern of expression divergence.
Despite this caveat, asymmetric functional divergence
does appear to be a common feature of duplicated genes
(e.g., Ferris and Whitt 1979). Wagner (2002) reported that
duplicated pairs in yeast show asymmetry not only in the
number of stress conditions in which each copy is expressed
but also in the number of interacting partners in experimental protein-interaction data and in the number of differentially expressed genes in knockout mutants of the 2
copies. Papp et al. (2003b) found that the number of cisregulatory sequences shared between duplicate genes decreased over time, even though the number of regulatory
sequences remained relatively constant and that the change
in the cis-regulatory structure leads to functional novelty.
Conant and Wagner (2002) found that the KA:KS ratio
was related to the degree of asymmetry between duplicated
pairs in yeast, indicating that asymmetry may be driven by
relaxed functional constraint or positive selection acting on
one copy. Alternatively, purifying selection may be at work
on the slower evolving copy as compared with the ancestral
gene (Kim and Yi 2006). Recent studies of yeast and
humans have identified a pattern of network asymmetry
among duplicate genes, where 1 gene in the pair acquires
more coexpressed gene partners than the other (Chung et al.
2006; Conant and Wolfe 2006).
The frequency of asymmetric, and the rarity of complementary, expression divergence suggests that the
subfunctionalization model (Force et al. 1999) is not adequate to describe overall patterns of divergence. Nonetheless, the footprint of subfunctionalization may still be
present if it is responsible for initial preservation of duplicates (Rastogi and Liberles 2005). The results (table 2) do
not bear out the prediction of a higher proportion of young
pairs with complementary expression patterns, though we
do observe a nonsignificant trend of greater complementarity among tandem than dispersed duplicates, as is weakly
predicted by the model (Lynch et al. 2001). Furthermore,
despite its overall rarity, there are some striking examples
of complementarity among the gene pairs examined. One
such example is the duplicate pair AtMST1 and AtMST2
(AT1G79230 and AT1G16460), 2 mercaptopyruvate sulfurtransferases that not only show a complementary expression pattern but also have been found to differ in their
subcellular localization to the mitochondria and cytoplasm,
respectively (Nakamura et al. 2000).
Differences among Modes of Duplication
To what extent does the mode of duplication affect
patterns of expression divergence? Because only a small
portion of dispersed duplicates have KS values that match
the relatively young tandem duplicates, comparing the nonpolyploid, and given the problems with KS as a proxy for
time (e.g., Zhang et al. 2002), comparing the nonpolyploid
tandem and dispersed duplicates is difficult. However, by
comparing segmental duplicates with nonsegmental dupli-
cates, we found that gene pairs in which both members were
expressed were somewhat more frequent among segmental
duplicates. A number of differences have also been observed in the literature. In general, segmental pairs tend
to have longer retention times (Otto and Yong 2002). In
yeast, the gene pairs retained from a genome duplication
event are enriched in members of multisubunit protein complexes (Papp et al. 2003a) and have higher than average
levels of transcriptional activity (Seoighe and Wolfe
1999). A number of recent studies have reported that the
genes retained from the most recent polyploidy event in
Arabidopsis (40–80 MYA) are enriched in signal transduction genes and transcription factors (Blanc and Wolfe 2004;
Maere et al. 2005). Schranz and Mitchell-Olds (2006) report
a correlation in the retention probability of genes following the
most recent, and independent, polyploidy events in Arabidopsis and Cleome. Thosegenes thatare not lostmay thenbeunder
stronger purifying selection. Chapman et al. (2006) report that
polymorphisms in the retained segmentally duplicated genes
of Arabidopsis and rice tend to have fewer radical amino acid
substitutions. The retained genes in Arabidopsis may also
have a tendency to be those that, on average, are more dosage
sensitive (Freeling and Thomas 2006), consistent with the
balance hypothesis (e.g., Birchler et al. 2003), which predicts
that genes retained from large-scale duplication events are
more likely to be those belonging to stoichiometrically sensitive pathways.
Assessment of MPSS and Affymetrix Data
We found that the 2 different expression platforms,
MPSS and Affymetrix, had complementary strengths and
weaknesses for the study of expression divergence in
duplicated genes. The chief advantage of MPSS and other
signature sequencing–based platforms is the ability to
distinguish the signatures of duplicate copies that would
cross-hybridize on a microarray. The potential for crosshybridization resulted in the loss of one-third of the duplicated gene pairs from the ADA data set. Nonetheless, the
breadth of tissues and conditions assayed in the ADA data
set provides us greater assurance that the asymmetric patterns of divergence within gene pairs were not an artifact of
limited tissue sampling in the MPSS data. In MPSS, the
coarseness of tissue sampling is necessitated by the expense
of the technology, though this problem may be mitigated by
more recent technologies developed for short-read sequencing. Recently developed methods, recently developed
methods offer even more fine-grained analysis of expression levels among cell types (Birnbaum et al. 2003). The
results presented here motivate further phylogenetic study
of gene families using experimental data that can provide
greater resolution for characterizing patterns of functional
divergence between duplicated genes (e.g., Nakhamchik
et al. 2004).
Supplementary Material
Supplementary tables 1 and 2 are available at Molecular Biology and Evolution.online (http://www.mbe.
oxfordjournals.org/).
2308 Ganko et al.
Supplementary table 1 lists all the gene pairs used in
this study along with all expression measurements (where
applicable). Supplementary table 2 lists all P values for all
statistically significant tests.
Acknowledgments
E.W.G. is supported by a Seeding Postdoctoral Innovators in Science and Education fellowship from the National
Institute of General Medical Sciences (GM 00678). This
work was also supported by National Science Foundation
grants DBI-0027314 to T.J.V. and DBI-0110528 to
B.C.M. T.J.V. acknowledges R. Moore, M. Rausher, and
J. Willis for useful suggestions.
Literature Cited
Adams KL, Cronn R, Percifield R, Wendel JF. 2003. Genes
duplicated by polyploidy show unequal contributions to the
transcriptome and organ-specific reciprocal silencing. Proc
Natl Acad Sci USA. 100:4649–4654.
Altschul SF, Madden TL, Schaffer AA, Zhang J, Zhang Z,
Miller W, Lipman DJ. 1997. Gapped BLAST and PSIBLAST: a new generation of protein database search
programs. Nucleic Acids Res. 25:3389–3402.
Azevedo C, Betsuyaku S, Peart J, Takahashi A, Noel L,
Sadanandom A, Casais C, Parker J, Shirasu K. 2006. Role
of SGT1 in resistance protein accumulation in plant
immunity. EMBO J. 25:2007–2016.
Benovoy D, Morris RT, Morin A, Drouin G. 2005. Ectopic gene
conversions increase the G þ C content of duplicated yeast
and Arabidopsis genes. Mol Biol Evol. 22:1865–1868.
Birchler JA, Auger DL, Riddle NC. 2003. In search of the
molecular basis of heterosis. Plant Cell. 15:2236–2239.
Birnbaum K, Shasha DE, Wang JY, Jung JW, Lambert GM,
Galbraith DW, Benfey PN. 2003. A gene expression map of
the Arabidopsis root. Science. 302:1956–1960.
Blanc G, Hokamp K, Wolfe KH. 2003. A recent polyploidy
superimposed on older large-scale duplications in the
Arabidopsis genome. Genome Res. 13:137–144.
Blanc G, Wolfe KH. 2004. Functional Divergence of Duplicated
Genes Formed by Polyploidy during Arabidopsis Evolution.
Plant Cell. 16:1679–1691.
Bonin CP, Freshour G, Hahn MG, Vanzin GF, Reiter WD. 2003.
The GMD1 and GMD2 genes of Arabidopsis encode isoforms
of GDP-D-mannose 4,6-dehydratase with cell type-specific
expression patterns. Plant Physiol. 132:883–892.
Bowers JE, Chapman BA, Rong J, Paterson AH. 2003. Unravelling
angiosperm genome evolution by phylogenetic analysis of
chromosomal duplication events. Nature. 422:433–438.
Brenner S, Johnson M, Bridgham J, et al. (24 co-authors). 2000.
Gene expression analysis by massively parallel signature
sequencing (MPSS) on microbead arrays. Nat Biotechnol. 18:
630–634.
Calabrese PP, Chakravarty S, Vision TJ. 2003. Fast identification
and statistical evaluation of segmental homologies in
comparative maps. Bioinformatics. 19(Suppl 1):i74–i80.
Casneuf T, De Bodt S, Raes J, Maere S, Van de Peer Y. 2006.
Nonrandom divergence of gene expression following gene
and genome duplications in the flowering plant Arabidopsis
thaliana. Genome Biol. 7:R13.
Castillo-Davis CI, Hartl DL, Achaz G. 2004. cis-Regulatory and
protein evolution in orthologous and duplicate genes. Genome
Res. 14:1530–1536.
Chapman BA, Bowers JE, Feltus FA, Paterson AH. 2006.
Buffering of crucial functions by paleologous duplicated
genes may contribute cyclicality to angiosperm genome
duplication. Proc Natl Acad Sci USA. 103:2730–2735.
Chen ZJ, Ni Z. 2006. Mechanisms of genomic rearrangements
and gene expression changes in plant polyploids. Bioessays.
28:240–252.
Chung WY, Albert R, Albert I, Nekrutenko A, Makova KD.
2006. Rapid and asymmetric divergence of duplicate genes in
the human gene coexpression network. BMC Bioinformatics.
7:46.
Conant GC, Wagner A. 2002. GenomeHistory: a software tool
and its application to fully sequenced genomes. Nucleic Acids
Res. 30:3378–3386.
Conant GC, Wolfe KH. 2006. Functional partitioning of yeast coexpression networks after genome duplication. PLoS Biol.
4:e109.
Duarte JM, Cui L, Wall PK, Zhang Q, Zhang X, LeebensMack J, Ma H, Altman N, dePamphilis CW. 2006. Expression
pattern shifts following duplication indicative of subfunctionalization and neofunctionalization in regulatory genes of
Arabidopsis. Mol Biol Evol. 23:469–478.
Duret L, Mouchiroud D. 2000. Determinants of substitution rates
in mammalian genes: expression pattern affects selection
intensity but not mutation rate. Mol Biol Evol. 17:68–74.
Enright AJ, Van Dongen S, Ouzounis CA. 2002. An efficient
algorithm for large-scale detection of protein families. Nucleic
Acids Res. 30:1575–1584.
Ferris SD, Whitt GS. 1979. Evolution of the differential
regulation of duplicate genes after polyploidization. J Mol
Evol. 12:267–317.
Force A, Lynch M, Pickett FB, Amores A, Yan YL,
Postlethwait J. 1999. Preservation of duplicate genes
by complementary, degenerative mutations. Genetics. 151:
1531–1545.
Freeling M, Thomas BC. 2006. Gene-balanced duplications, like
tetraploidy, provide predictable drive to increase morphological complexity. Genome Res. 16:805–814.
Friedman R, Hughes AL. 2001. Gene duplication and the
structure of eukaryotic genomes. Genome Res. 11:373–381.
Gu X, Zhang Z, Huang W. 2005. Rapid evolution of expression
and regulatory divergences after yeast gene duplication. Proc
Natl Acad Sci USA. 102:707–712.
Gu Z, Nicolae D, Lu HH, Li WH. 2002. Rapid divergence in
expression between duplicate genes inferred from microarray
data. Trends Genet. 18:609–613.
Haberer G, Hindemitt T, Meyers BC, Mayer KF. 2004.
Transcriptional similarities, dissimilarities, and conservation
of cis-elements in duplicated genes of Arabidopsis. Plant
Physiol. 136:3009–3022.
Jordan IK, Marino-Ramirez L, Koonin EV. 2005. Evolutionary
significance of gene expression divergence. Gene.
345:119–126.
Katju V, Lynch M. 2003. The structure and early evolution of
recently arisen gene duplicates in the Caenorhabditis elegans
genome. Genetics. 165:1793–1803.
Khaitovich P, Weiss G, Lachmann M, Hellmann I, Enard W,
Muetzel B, Wirkner U, Ansorge W, Paabo S. 2004. A neutral
model of transcriptome evolution. PLoS Biol. 2:E132.
Kim SH, Yi SV. 2006. Correlated asymmetry of sequence and
functional divergence between duplicate proteins of Saccharomyces cerevisiae. Mol Biol Evol. 23:1068–1075.
Lemos B, Bettencourt BR, Meiklejohn CD, Hartl DL. 2005.
Evolution of proteins and gene expression levels are coupled
in Drosophila and are independently associated with mRNA
abundance, protein length, and number of protein-protein
interactions. Mol Biol Evol. 22:1345–1354.
Divergent Expression in Duplicated Genes 2309
Lynch M, Conery J. 2001. The evolutionary fate and consequences of duplicate genes. Science. 290:1151–1155.
Lynch M, Conery JS. 2000. The evolutionary fate and
consequences of duplicate genes. Science. 290:1151–1155.
Lynch M, O’Hely M, Walsh B, Force A. 2001. The probability of
preservation of a newly arisen gene duplicate. Genetics.
159:1789–1804.
Maere S, De Bodt S, Raes J, Casneuf T, Van Montagu M,
Kuiper M, Van de Peer Y. 2005. Modeling gene and genome
duplications in eukaryotes. Proc Natl Acad Sci USA.
102:5454–5459.
Makova KD, Li WH. 2003. Divergence in the spatial pattern of
gene expression between human duplicate genes. Genome
Res. 13:1638–1645.
Matzke MA, Mette MF, Kanno T, Matzke AJ. 2003. Does the
intrinsic instability of aneuploid genomes have a causal role in
cancer? Trends Genet. 19:253–256.
Meyers BC, Tej SS, Vu TH, Haudenschild CD, Agrawal V,
Edberg SB, Ghazal H, Decola S. 2004. The use of MPSS for
whole-genome transcriptional analysis in Arabidopsis. Genome Res. 14:1641–1653.
Meyers BC, Vu TH, Tej SS, Ghazal H, Matvienko M,
Agrawal V, Ning J, Haudenschild CD. 2004. Analysis of
the transcriptional complexity of Arabidopsis thaliana by
massively parallel signature sequencing. Nat Biotechnol.
22:1006–1011.
Mondragon-Palomino M, Gaut BS. 2005. Gene conversion and
the evolution of three leucine-rich repeat gene families in
Arabidopsis thaliana. Mol Biol Evol. 22:2444–2456.
Nakamura T, Yamaguchi Y, Sano H. 2000. Plant mercaptopyruvate sulfurtransferases: molecular cloning, subcellular localization and enzymatic activities. Eur J Biochem. 267:
5621–5630.
Nakhamchik A, Zhao Z, Provart NJ, Shiu SH, Keatley SK,
Cameron RK, Goring DR. 2004. A comprehensive expression
analysis of the Arabidopsis proline-rich extensin-like receptor
kinase gene family using bioinformatic and experimental
approaches. Plant Cell Physiol. 45:1875–1881.
Nowak MA, Boerlijst MC, Cooke J, Smith JM. 1997. Evolution
of genetic redundancy. Nature. 388:167–171.
Nuzhdin SV, Wayne ML, Harmon KL, McIntyre LM. 2004.
Common pattern of evolution of gene expression level and
protein sequence in Drosophila. Mol Biol Evol. 21:1308–1317.
Otto SP, Yong P. 2002. The evolution of gene duplicates. Adv
Genet. 46:451–483.
Papp B, Pal C, Hurst LD. 2003a. Dosage sensitivity and the
evolution of gene families in yeast. Nature. 424:194–197.
Papp B, Pal C, Hurst LD. 2003b. Evolution of cis-regulatory
elements in duplicated genes of yeast. Trends Genet. 19:
417–422.
Pinyopich A, Ditta GS, Savidge B, Liljegren SJ, Baumann E,
Wisman E, Yanofsky MF. 2003. Assessing the redundancy of
MADS-box genes during carpel and ovule development.
Nature. 424:85–88.
Rapp RA, Wendel JF. 2005. Epigenetics and plant evolution.
New Phytol. 168:81–91.
Rastogi S, Liberles DA. 2005. Subfunctionalization of duplicated
genes as a transition state to neofunctionalization. BMC Evol
Biol. 5:28.
Rodin SN, Riggs AD. 2003. Epigenetic silencing may aid
evolution by gene duplication. J Mol Evol. 56:718–729.
Schmid M, Davison TS, Henz SR, Pape UJ, Demar M,
Vingron M, Scholkopf B, Weigel D, Lohmann JU. 2005. A
gene expression map of Arabidopsis thaliana development.
Nat Genet. 37:501–506.
Schranz ME, Mitchell-Olds T. 2006. Independent ancient
polyploidy events in the sister families Brassicaceae and
Cleomaceae. Plant Cell. 18:1152–1165.
Seoighe C, Wolfe KH. 1999. Yeast genome evolution in the postgenome era. Curr Opin Microbiol. 2:548–554.
Shiu SH, Byrnes JK, Pan R, Zhang P, Li WH. 2006. Role of
positive selection in the retention of duplicate genes in
mammalian genomes. Proc Natl Acad Sci USA. 103:
2232–2236.
Simillion C, Vandepoele K, Van Montagu MC, Zabeau M, Van
de Peer Y. 2002. The hidden duplication past of Arabidopsis
thaliana. Proc Natl Acad Sci USA. 99:13627–13632.
Sokal RR, Rohlf FJ. 1995. Biometry. New York: Freeman and
Company.
Thompson JD, Higgins DG, Gibson TJ. 1994. CLUSTAL W:
improving the sensitivity of progressive multiple sequence
alignment through sequence weighting, position-specific gap
penalties and weight matrix choice. Nucleic Acids Res. 22:
4673–4680.
Vision TJ. 2005. Gene order in plants: a slow but sure shuffle.
New Phytol. 168:51–60.
Vision TJ, Brown DG, Tanksley SD. 2000. The origins of
genomic duplications in Arabidopsis. Science. 290:
2114–2117.
Wagner A. 1999. Redundant gene functions and natural
selection. J Evol Biol. 12:1–16.
Wagner A. 2000. Decoupled evolution of coding region and
mRNA expression patterns after gene duplication: implications for the neutralist-selectionist debate. Proc Natl Acad Sci
USA. 97:6579–6584.
Wagner A. 2002. Asymmetric functional divergence of duplicate
genes in yeast. Mol Biol Evol. 19:1760–1768.
Walsh JB. 1995. How often do duplicated genes evolve new
functions? Genetics. 139:421–428.
Wang J, Tian L, Madlung A, et al. (10 co-authors). 2004.
Stochastic and epigenetic changes of gene expression in
Arabidopsis polyploids. Genetics. 167:1961–1973.
Wright F. 1990. The ‘effective number of codons’ used in a gene.
Gene. 87:23–29.
Xu W, Bak S, Decker A, Paquette SM, Feyereisen R,
Galbraith DW. 2001. Microarray-based analysis of gene
expression in very large gene families: the cytochrome
P450 gene superfamily of Arabidopsis thaliana. Gene. 272:
61–74.
Yang J, Su AI, Li WH. 2005. Gene expression evolves faster in
narrowly than in broadly expressed mammalian genes. Mol
Biol Evol. 22:2113–2118.
Yang Z. 1997. PAML: a program package for phylogenetic
analysis by maximum likelihood. Comput Appl Biosci.
13:555–556.
Zhang L, Li WH. 2004. Mammalian housekeeping genes evolve
more slowly than tissue-specific genes. Mol Biol Evol. 21:
236–239.
Zhang L, Vision TJ, Gaut BS. 2002. Patterns of nucleotide
substitution among simultaneously duplicated gene pairs in
Arabidopsis thaliana. Mol Biol Evol. 19:1464–1473.
Zhao D, Ni W, Feng B, Han T, Petrasek MG, Ma H. 2003.
Members of the Arabidopsis-SKP1-like gene family exhibit
a variety of expression patterns and may play diverse roles in
Arabidopsis. Plant Physiol. 133:203–217.
Neelima Sinha, Associate Editor
Accepted July 28, 2007