Molecular evolution under increasing transposable element burden in Drosophila: a speed limit on the evolutionary arms race.

Abstract

Genome architecture is profoundly influenced by transposable elements (TEs), and natural selection against their harmful effects is a critical factor limiting their spread. Genome defense by the piRNA silencing pathway also plays a crucial role in limiting TE proliferation. How these two forces jointly determine TE abundance is not well understood. To shed light on the nature of factors that predict TE success, we test three distinct hypotheses in the Drosophila genus. First, we determine whether TE abundance and relaxed genome-wide purifying selection on protein sequences are positively correlated. This serves to test the hypothesis that variation in TE abundance in the Drosophila genus can be explained by the strength of natural selection, relative to drift, acting in parallel against mildly deleterious non-synonymous mutations. Second, we test whether increasing TE abundance is correlated with an increased rate of amino-acid evolution in genes encoding the piRNA machinery, as might be predicted by an evolutionary arms race model. Third, we test whether increasing TE abundance is correlated with greater codon bias in genes of the piRNA machinery. This is predicted if increasing TE abundance selects for increased efficiency in the machinery of genome defense.
Surprisingly, we find neither of the first two hypotheses to be true. Specifically, we found that genome-wide levels of purifying selection, measured by the ratio of non-synonymous to synonymous substitution rates (ω), were greater in species with greater TE abundance. In addition, species with greater TE abundance have greater levels of purifying selection in the piRNA machinery. In contrast, it appears that increasing TE abundance has primarily driven adaptation in the piRNA machinery by increasing codon bias.
These results indicate that within the Drosophila genus, a historically reduced strength of selection relative to drift is unlikely to explain patterns of increased TE success across species. Other factors, such as ecological exposure, are likely to contribute to variation in TE abundances within species. Furthermore, constraints on the piRNA machinery may temper the evolutionary arms race that would drive increasing rates of evolution at the amino acid level. In the face of these constraints, selection may act primarily by improving the translational efficiency of the machinery of genome defense through efficient codon usage.

Full text

Page 1

RESEARCH ARTICLEOpen Access
Molecular evolution under increasing
transposable element burden in Drosophila:
A speed limit on the evolutionary arms race
Dean M Castillo1,2, Joshua Chang Mell3, Kimberly S Box1and Justin P Blumenstiel1*
Abstract
Background: Genome architecture is profoundly influenced by transposable elements (TEs), and natural selection
against their harmful effects is a critical factor limiting their spread. Genome defense by the piRNA silencing
pathway also plays a crucial role in limiting TE proliferation. How these two forces jointly determine TE abundance
is not well understood. To shed light on the nature of factors that predict TE success, we test three distinct
hypotheses in the Drosophila genus. First, we determine whether TE abundance and relaxed genome-wide
purifying selection on protein sequences are positively correlated. This serves to test the hypothesis that variation
in TE abundance in the Drosophila genus can be explained by the strength of natural selection, relative to drift,
acting in parallel against mildly deleterious non-synonymous mutations. Second, we test whether increasing TE
abundance is correlated with an increased rate of amino-acid evolution in genes encoding the piRNA machinery,
as might be predicted by an evolutionary arms race model. Third, we test whether increasing TE abundance is
correlated with greater codon bias in genes of the piRNA machinery. This is predicted if increasing TE abundance
selects for increased efficiency in the machinery of genome defense.
Results: Surprisingly, we find neither of the first two hypotheses to be true. Specifically, we found that genome-
wide levels of purifying selection, measured by the ratio of non-synonymous to synonymous substitution rates (ω),
were greater in species with greater TE abundance. In addition, species with greater TE abundance have greater
levels of purifying selection in the piRNA machinery. In contrast, it appears that increasing TE abundance has
primarily driven adaptation in the piRNA machinery by increasing codon bias.
Conclusions: These results indicate that within the Drosophila genus, a historically reduced strength of selection
relative to drift is unlikely to explain patterns of increased TE success across species. Other factors, such as
ecological exposure, are likely to contribute to variation in TE abundances within species. Furthermore, constraints
on the piRNA machinery may temper the evolutionary arms race that would drive increasing rates of evolution at
the amino acid level. In the face of these constraints, selection may act primarily by improving the translational
efficiency of the machinery of genome defense through efficient codon usage.
Background
In sexual species, genetic parasites such as transposable
elements (TEs) can proliferate to the detriment of the
host [1]. Natural selection is widely considered the domi-
nant force limiting TE proliferation [2-4]. The selective
forces limiting TE abundance are thought to act against
three primary consequences of TE proliferation- gene
mutation by TE insertion, chromosomal rearrangement
caused by ectopic recombination among dispersed
copies, and the energetic burden imposed on the host
arising from the costs of replication, transcription and
translation of TE copies. TE insertion alleles often segre-
gate at low frequencies in populations, consistent with
natural selection limiting their increase. However, studies
in recent years have shown that different modes of RNA
silencing also play an important role in constraining TE
proliferation [5-7]. In particular, within the germline of
animals, the piRNA machinery functions as an immune
* Correspondence: jblumens@ku.edu
1Department of Ecology and Evolutionary Biology, University of Kansas,
Lawrence Kansas 66045, USA
Full list of author information is available at the end of the article
Castillo et al. BMC Evolutionary Biology 2011, 11:258
http://www.biomedcentral.com/1471-2148/11/258
© 2011 Castillo et al; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons
Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in
any medium, provided the original work is properly cited.

Page 2

system to protect the genome against TE proliferation
[8-10]. TE copies that have inserted into distinct chromo-
somal regions known as piRNA clusters are recognized as
aberrant and their transcripts are directed to the piRNA
biogenesis machinery. In complex with Argonaute pro-
teins, 26 - 31 nt piRNAs use sequence identity to target
transcripts of dispersed TE copies for degradation. In
turn, this generates secondary piRNAs that feed into the
cycle of piRNA biogenesis and TE silencing.
These two forces - natural selection and genome
defense by small RNAs - directly limit TE proliferation,
but in distinct ways. In particular, natural selection will
primarily act to limit the increase of harmful TE insertion
alleles. Conversely, the piRNA machinery will act by
directly limiting the transposition rate, and thus the rate
of production of new insertion alleles. How these forces
jointly determine the rate of TE proliferation is poorly
understood [11]. In fact, the distinction between natural
selection and genome defense as separable forces may be
artificial, since the genome defense machinery is itself the
product of natural selection.
To what degree does the strength of natural selection
explain variation in TE abundance across species? Popula-
tion genetic theory predicts that the strength of selection
relative to genetic drift will be greater in larger populations
[12]. Thus, as indicated by Ohta, mildly deleterious substi-
tutions will fix at a greater rate in smaller populations.
This prediction has been confirmed in a variety of systems
where purifying selection was estimated using the rate
ratio of non-synonymous to synonymous substitution (ω).
Consistent with natural selection acting more strongly
against deleterious non-synonymous substitutions, larger
populations tend to have smaller ω values. For example, in
mammals there is a strong negative correlation between
population size and ω [13]. A large study comparing
branch-specific ω between closely related mainland and
island species found that ω estimates were higher for
island species, likely due to their smaller population size
[14]. This was observed in both vertebrates and inverte-
brates. If variation in TE abundance across species is also
influenced by the strength of selection relative to drift, TE
abundance and genome-wide ω values for protein-coding
genes should be positively correlated. This is because dif-
ferences in population size will modulate the efficacy of
selection against both types of mutation.
This hypothesis, however, relies on several assumptions.
First, it assumes that variation in ω is more strongly
explained by the influence of mildly deleterious substitu-
tions rather than adaptive substitutions. If beneficial muta-
tions are common and much more likely to fix in larger
populations, the negative relationship between ω and
population size will be ameliorated. Studies in Drosophila,
however, have indicated that population size has little
effect on the rate of adaptive fixation [15,16]. Second, it
assumes that the rates of both forms of mutation are inde-
pendent of population size. Among related species, this is
likely for mutations at the nucleotide level, but may be vio-
lated for TE insertions if exposure to TE invasion is
greater in larger populations. Finally, it assumes that the
distribution of fitness effects is similar between TE inser-
tions and non-synonymous mutations and doesn’t con-
sider the non-linear scaling of the probability of fixation of
mildly deleterious alleles with effective population size
[17]. When the product of the effective population size
and deleterious selection coefficient (Nes) is substantially
greater than one, the chance that such a deleterious allele
fixes becomes vanishingly small. Only nearly neutral dele-
terious mutations for which Nes is less than one are
expected to fix. Thus, the degree to which variation in
population size governs the accumulation of deleterious
alleles, such as TE insertions, itself depends on population
size. In organisms with very large population sizes drift
becomes extremely weak, and modest variation in popula-
tion size among related species may only impact the rate
of accumulation of very mildly deleterious alleles with
selection coefficients very close to zero [18]. In this case,
the distribution of fitness effects of mildly deleterious
mutations will be an important factor in determining the
relationship between population size and ω. Setting aside
the effects of beneficial mutations, a negative correlation
will be observed across all population sizes only if there
are a sufficient number of nearly neutral mutations at all
population sizes. If the distribution of fitness effects for
deleterious mutations does not display this characteristic,
modest increases in population size in very large popula-
tions may not always lead to decreased ω. For similar rea-
sons, one might not expect a simple relationship between
TE accumulation and population size when population
sizes are large.
This latter point is highlighted by studies of TE
dynamics yielding contrasting results for different spe-
cies. In Drosophila melanogaster, studies have suggested
that the deterministic forces of selection against TEs
may greatly outweigh genetic drift as a factor [19,20].
D. melanogaster also has a large effective population
size. Thus, modest variation in population size among
species across the genus Drosophila may not be an
important factor contributing to variation in TE abun-
dance. In contrast, studies in vertebrates with much
smaller effective population sizes have shown that
genetic drift can be an important factor contributing to
TE accumulation [21]. For example, the frequency dis-
tribution of TE insertion alleles in the pufferfish is con-
sistent with neutrality [22]. This indicates that variation
in effective population size at these low population sizes
may have a greater impact on the fate of mildly deleter-
ious TE insertions, relative to species with larger popu-
lation sizes such as Drosophila. It also indicates that in
Castillo et al. BMC Evolutionary Biology 2011, 11:258
http://www.biomedcentral.com/1471-2148/11/258
Page 2 of 16

Page 3

species with larger population sizes, there may only be a
weak correlation between the rate of non-synonymous
substitution and TE abundance. This may explain why,
in one study, after correcting for phylogenetic signal, no
apparent relationship between genomic TE number and
population size was found [23]. Considering these issues,
we aimed to test the simple hypothesis that TE abun-
dance is positively correlated with genome-wide ω
across the Drosophila genus. Rejection of this hypothesis
would support alternative models and provide further
testable hypotheses in the study of TE dynamics in large
populations.
An added level of complexity is the evolutionary
dynamic between TEs and the piRNA machinery. In sev-
eral species of Drosophila, many, but not all, components
of the piRNA machinery show a high rate of adaptive
evolution [24-26]. This has been proposed to arise from
an evolutionary arms race between the host and TEs.
Evolutionary arms races between hosts and parasites
drive cycles of adaptation and counter-adaptation, lead-
ing to increased rates of adaptive evolution in host
immune systems. In the case of TEs, there is likely strong
selection to avoid silencing by the piRNA machinery.
This may select for functions analogous to those
observed in viruses, which have mechanisms that directly
antagonize the machinery of RNA silencing [27,28]. Reci-
procally, natural selection acting on the host likely selects
for changes that counteract these strategies, driving a
high rate of adaptive evolution in the proteins of the
piRNA machinery. But importantly, the classic evolution-
ary arms race is not sufficient to explain all evolutionary
dynamics between host and parasite [29-31]. For exam-
ple, trench warfare may occur when a diversity of para-
sites selects for the maintenance of multiple defense
strategies, thus favoring a mode of balancing selection
[32]. While an evolutionary arms race with TEs is sug-
gested by the high rate of evolution in the piRNA
machinery in some Drosophila species, it is not clear that
increasing TE abundance could drive ever-increasing
levels of amino-acid evolution. Constraint on core func-
tion will eventually pose some limit to rates of amino-
acid evolution. Additionally, with increasing TE abun-
dance, the balance of forces may begin to favor purifying
selection over adaptation. To explore these issues, we test
the simple hypothesis that in the Drosophila genus,
increasing TE abundance drives a higher rate of amino-
acid substitution in the piRNA machinery as measured
by ω.
In this study, we found that genome-wide levels of puri-
fying selection are greater (smaller ω) in Drosophila spe-
cies with higher TE abundance, inconsistent with a model
in which increasing TE abundance and ω are jointly
explained by weaker selection relative to drift. Compared
to control genes, we also find that this observation is more
evident in the piRNA machinery. Strikingly, we find that
TE abundance and levels of codon bias are positively cor-
related in the piRNA machinery but not in control genes
or in the rest of the genome. Rather than an increasing
rate of amino-acid evolution, the primary response of the
piRNA machinery to increasing TE abundance appears to
be through improved codon usage for increased transla-
tional efficiency.
Results
Genome-wide levels of purifying selection and TE
abundance are positively correlated
Using the twelve sequenced Drosophila genomes
(Figure 1a), we determined the relationship between
TE abundance and previous estimates of genome-wide
average ω on terminal branches [33]. TE abundance
was quantified using the total amount of assembled
euchromatin comprised of repeats as determined from
the 12 Drosophila genomes consortium [34]. This measure
captures each of the components of selection considered
important against TEs - gene mutation, ectopic recombi-
nation and metabolic cost. This measure does ignore the
influence of heterochromatic TEs and some species with
low euchromatic content may in fact have many TEs
within the large domains of pericentric heterochromatin.
The converse may also be true. However, it is clear that
the gene rich and highly recombining environment of the
euchromatin makes insertions in this portion of the gen-
ome substantially more harmful. Thus, euchromatic TE
content is likely to serve as the best metric for the selective
burden of TE content on a species. It is also correlated
with the genomic number of TE families (Figure 1b).
Using computationally predicted PILER-DF libraries [35]
as a measure of TE family number, we find that the num-
ber of different TE families within each species’ genome
is a good predictor of total TE abundance (Figure 1b,
p = 0.001).
We found a significant negative correlation between
genomic TE abundance and tip ω estimates (Figure 2a,
p = 0.025). When assessing correlations among species, it
is essential to account for phylogenetic non-independence
[36,37]. This was achieved using the implementation of
Continuous in the BayesTraits package which models evo-
lution of continuous traits on a phylogeny using a general-
ized least-squares approach under Brownian motion
model of evolution. First, we tested whether phylogenetic
correction was needed by testing whether the l scaling
parameter significantly deviated from 0 (hereafter desig-
nated the l test). A l parameter equal to zero indicates
trait evolution can be effectively modeled as if it were on a
star phylogeny. In this case, a phylogenetic correction was
not mandated using a likelihood ratio test (l test, p =
0.99). Nonetheless, we tested whether the correlation was
significant with phylogenetic non-independence taken into
Castillo et al. BMC Evolutionary Biology 2011, 11:258
http://www.biomedcentral.com/1471-2148/11/258
Page 3 of 16

Page 4

account. This was performed by testing whether the covar-
iance between the two traits was significantly greater than
zero, assuming no scaling transformation of the phylogeny
(hereafter designated the BayesTraits test). Nonetheless,
after accounting for phylogenetic non-independence, the
negative correlation between genomic TE abundance and
genomic ω estimates remained significant using a likeli-
hood ratio test (BayesTraits test, p = 0.0078). To account
for potential problems with estimating ω over long term-
inal branches, we also evaluated the relationship between
TE abundance and ω estimates only across the D. melano-
gaster subgroup. Considering only D. simulans, D. melano-
gaster, D. yakuba and D. erecta, there is still a negative
relationship between TE abundance and genome-wide ω
(rpearson= -0.60, b = -0.0022), albeit non-significant with
only four data points. Overall, these results indicate that
genomes more encumbered by TEs have had a historically
greater magnitude of purifying selection acting on protein
coding sequences. This is difficult to reconcile with a
model in which TE abundance is largely determined by
the degree to which mildly deleterious alleles are removed
by natural selection within populations.
Increased purifying selection on the piRNA machinery in
species with high TE abundance
Considering the background genome-wide correlation
between TE abundance and levels of purifying selection
on coding sequences, we next determined how the rate
of amino-acid evolution of the piRNA machinery was
shaped by TE abundance (Figure 2b and 2c). The rate of
molecular evolution for eleven known components of the
piRNA machinery and eleven control genes was deter-
mined by estimating ω [38] on the evolutionary paths
leading to all twelve Drosophila species with sequenced
genomes. Path estimates of ω were obtained, rather than
terminal branch estimates, since terminal branch lengths
vary widely between species. Since piRNA genes are
known to evolve quickly, a set of control genes was
selected, matched for a similar rate of evolution as the
piRNA genes. If increasing TE abundance drives a faster
0
0.02
0.04
0.06
0.08
0.1
0.12
0 10 20 30 40 50
0
0.05
0.1
0.15
0.2
0.25
0.3
0 10 20 30 40 50
0
0.05
0.1
0.15
0.2
0.25
0.3
0 10 20 30 40 50
TE Content (Mb) TE Content (Mb)TE Content (Mb)
?
A)
B)
C)
Genome AverageControl GenespiRNA Genes
rp = -0.73
??= -0.9 x 10-3
rp = -0.65
??= -1.9 x 10-3
rp = -0.85
??= -2.5 x 10-3
Figure 2 Genomic TE abundance vs. ω from: A) Genome wide estimates on terminal branches obtained from Heger and Ponting [33].
Excludes D. sechellia, D, willistoni and D. persimilis (p = 0.026). B) Path ω estimates for control genes. (p = 0.022). C) Path ω estimates for piRNA
genes (p < 0.001). rp: Pearson’s correlation coefficient. b: regression coefficient (slope).
D. melanogaster 6.3
D. simulans 3.0
D. sechellia 4.2
D. yakuba 15.3
D. erecta 9.3
D. ananassae 43.9
D. pseudoobscura 3.5
D. persimilis 11.7
D. willistoni 29.1
D. mojavensis 14.6
D. virilis 24.0
D. grimshawi 3.9
0.09
TE Content (Mb)
0
50
100
150
200
250
0 10 20 30 40 50
rp =0.84, ?=3.74
PILER Centroids
A)B)
Figure 1 A) Phylogenetic tree of the Drosophila species used in this study, with TE abundance indicated in Mb. Dots indicate nodes for
which ω was estimated on descendant foreground branches. B) TE abundance is correlated with TE family number. Unique, species-specific
PILER-DF centroids vs. genomic TE abundance (p = 0.001). rp: Pearson’s correlation coefficient. b: regression coefficient (slope).
Castillo et al. BMC Evolutionary Biology 2011, 11:258
http://www.biomedcentral.com/1471-2148/11/258
Page 4 of 16

Page 5

arms race, we might expect elevated ω on the piRNA
machinery along lineages with greater TE abundance.
However, this was not the case. Without phylogenetic
correction, control genes (p = 0.022) and piRNA genes
(p = 0.0005) both show a negative relationship between
TE abundance and average ω, consistent with the
observed genome-wide negative correlation between TE
abundance and ω. However, both the rpearsonand b values
were more negative for the piRNA genes compared to
control genes. Furthermore, for control genes, the l test
indicated that phylogenetic correction was needed (l
test, p = 0.024) and this correlation was not significant
after the correction was made (BayesTraits, p = 0.243). In
contrast, even though the l test indicated no phyloge-
netic correction was needed for the piRNA genes (l test,
p = 0.439), the result was still significant after phylogeny
was taken into account (BayesTraits, p = 0.029)
We then sought to determine if this effect was more evi-
dent in the piRNA machinery due to artifacts of estimating
ω on long branches. We did this by testing whether this
trend was also observed in the more closely related
D. melanogaster group (D. melanogaster, D. simulans,
D. sechellia, D. yakuba and D. erecta). Focusing only on
this set of species, path ω was re-estimated over these
shorter divergence times (Figure 1a). With only five spe-
cies, there is limited power in a single correlation using
average values, so we examined the distribution of correla-
tion coefficients between TE abundance and ω in piRNA
genes relative to the control gene set (Figures 3, 4 and 5).
This served to test whether TE abundance better
explained variation in ω for the piRNA genes, compared
to control genes, across different levels of divergence. The
distribution of correlation coefficients is significantly more
negative for piRNA genes than control genes. This is the
case in the five species within the D. melanogaster group
(Bootstrap test for difference in mean, p < 0.001) and also
all twelve species (Bootstrap test for difference in mean,
p < 0.006). In addition, to compare effect size, we also
tested whether the distribution of b regression coefficients
(measures of slope) were different between piRNA genes
and control genes. b values were more negative for both
the D. melanogaster and 12 species comparison. In the
case of the D. melanogaster comparison, this was signifi-
cant (Mean b: Control genes, 0.007, piRNA genes, -0.016,
p = 0.016, t-test). In the case of the 12 species comparison,
b was 30% steeper for the piRNA machinery, but this
wasn’t significant (Mean b: Control genes, -0.0019, piRNA
genes, -0.0025, p = 0.464, t-test). Thus, we can conclude
that in comparison to the control genes, TE abundance
better explains variation in ω in the piRNA machinery
across different levels of divergence. Moreover, the
strength of the effect of TE abundance on ω is especially
strong when shorter time scales of divergence are
examined.
The above analysis does not account for phylogenetic
non-independence. Given the excess of negative correla-
tion coefficients for piRNA genes compared to control
genes, we next sought to determine whether this excess
would be maintained after accounting for phylogenetic
non-independence. With the small sample size of five spe-
cies in the melanogaster group, phylogenetic correction is
precluded due to low power [39] and we instead focused
solely on all twelve species. After controlling for phyloge-
netic non-independence where mandated, six of the eleven
piRNA genes (Table 1), but none of the control genes
(Table 2), show a significant relationship between ω and
TE abundance, and this enrichment is significant (Fisher’s
exact test, 2-tail, p = 0.012). Thus, even after accounting
for phylogeny, the effect of TE abundance on ω is more
evident in the piRNA machinery than for other quickly
evolving genes.
Observed correlation between TE abundance and
purifying selection in the piRNA machinery is
independent of effects of dS
Variation in ω can be explained by a variety of factors not
directly related to the forces of divergence on protein-
coding function. For example, lower values of ω in the
piRNA machinery of species with greater TE abundance
could be explained by an increased rate of silent substitu-
tion rather than a decreased rate of amino-acid substitu-
tions. To test this, we determined whether either dN or
dS were separately correlated with TE abundance after
accounting for phylogenetic non-independence. Consid-
ering dN alone, seven piRNA genes show a significant
relationship with TE abundance (Table 3). In contrast,
only two piRNA genes show a significant relationship
between dS and TE abundance (Table 4). Of eleven
piRNA genes, only krimper shows a significant relation-
ship between TE abundance and dS, but not dN. Exclud-
ing krimper, there is still a significant enrichment in the
piRNA genes for a negative relationship between TE
p < 0.001p < 0.006
12 Speciesmelanogaster group
piRNA Genes Control Genes
-1
1
piRNA Genes Control Genes
Corr. Coeff.
Figure 3 Distribution of raw Pearson correlation coefficients
for the correlation between TE abundance and ω. Boxes are
95% confidence intervals for the mean.
Castillo et al. BMC Evolutionary Biology 2011, 11:258
http://www.biomedcentral.com/1471-2148/11/258
Page 5 of 16