Regulatory evolution across the protein interaction network.

Abstract

Protein-protein interactions may impose constraints on both structural and regulatory evolution. Here we show that protein-protein interactions are negatively associated with evolutionary variation in gene expression. Moreover, interacting proteins have similar levels of variation in expression, and their expression levels are positively correlated across strains. Our results suggest that interacting proteins undergo similar evolutionary dynamics, and that their expression levels are evolutionarily coupled. These patterns hold for organisms as diverse as budding yeast and fruit flies.

Full text

Page 1

Regulatory evolution across the
protein interaction network
Bernardo Lemos, Colin D Meiklejohn & Daniel L Hartl
Protein-protein interactions may impose constraints on both
structural and regulatory evolution. Here we show that
protein-protein interactions are negatively associated with
evolutionary variation in gene expression. Moreover,
interacting proteins have similar levels of variation in
expression, and their expression levels are positively correlated
across strains. Our results suggest that interacting proteins
undergo similar evolutionary dynamics, and that their
expression levels are evolutionarily coupled. These patterns
hold for organisms as diverse as budding yeast and fruit flies.
Recent studies on variation in gene expression in natural populations
indicate that a number of factors influence regulatory evolution,
including membership in a particular functional class or biological
process1and the pattern of sex-biased expression2,3. Protein-protein
interactions may also be relevant to regulatory evolution if they mean
that the interacting partners of a given protein impose stricter
stoichiometric requirements. For proteins that participate in com-
plexes, such requirements may result in increased selection pressure to
maintain evolutionarily stable expression levels and might be mani-
fested in a negative association between evolutionary variation in gene
expression and the number of protein interactions.
We analyzed variation in expression of 5,978 yeast (Saccharomyces
cerevisiae) genes among four natural isolates grown under identical
laboratory conditions4, of 4,439 fly (Drosophila melanogaster) genes in
adult males from eight wild-type strains of geographically diverse
origin2and of a similar number of genes assayed in both sexes of two
species of Drosophila3. We matched these data to protein-protein
interactions that could be assigned with high confidence (confidence
score 4 0.55): 8,728 in yeast5and 3,964 in flies6. Choosing interac-
tions with more stringent selection criteria produced similar results
(Supplementary Note online).
The breadth of variation of expression for a given gene in both flies
and yeast (gene expression polymorphism) was negatively correlated
with the number of protein-protein interactions found for the product
of that gene (yeast: r ¼ –0.11, P o 0.0001, n ¼ 3,536; flies: r ¼ –0.15,
P ¼ 0.022, n ¼ 225; Fig. 1). Gene expression divergence between two
Drosophila species was also negatively associated with the number of
protein-protein interactions (males: r ¼ –0.18, P ¼ 0.012, n ¼ 205;
females: r ¼ –0.19, P ¼ 0.007, n ¼ 205). Furthermore, the con-
nectivity of a protein was an excellent predictor of the mean variation
in expression in classes of genes with a similar number of interactions
(Supplementary Fig. 1 online). Taken together, these results suggest
that protein interactions might be a constraint that reduces evolu-
tionary variation in gene expression both within and between species.
Among yeast proteins, the rate of accumulation of replacement
substitutions decreases as the number of protein-protein interactions
increases7,8. The importance of this conclusion has been disputed on
the grounds that the effect may be confounded with gene expression
level9(absolute transcript abundance) and is driven disproportio-
nately by highly connected proteins10. In our data, both of these
hypotheses can be ruled out (Supplementary Note online). First, the
bivariate associations between evolutionary variation in gene expres-
sion, number of protein-protein interactions and gene expression level
indicate that gene expression level may obscure the negative associa-
tion between the number of protein-protein interactions and gene
expression polymorphism and divergence. Accordingly, when we
include gene expression level as a covariate, the partial Spearman
rank correlation between evolutionary variation in gene expression
and number of protein-protein interactions not only remains statis-
tically significant but tends to become stronger. Second, the correla-
tion between gene expressionvariation and number of protein-protein
interactions remains significant in both yeast and flies when removing
proteins with more than four interactions, whereas it is lessened by
removing minimally connected proteins or proteins that do not
participate in complexes. This is in agreement with the observation
that the disruption of genes participating in protein complexes has
a stronger effect on fitness than the disruption of genes not involved
in complexes11.
Quantitative genetic models identify a number of forces that may
influence the breadth of variation in quantitative traits in natural
populations, including population size, the mutation rate for the trait
and the strength of stabilizing selection. In particular, mutation-
selection models12and the neutral theory of evolution13indicate
that the strength of stabilizing selection is an important determinant
of standing levels of variation. All else being equal, weaker stabilizing
selection is expected to maintain higher levels of variation than
0.4
0.8
Number of protein-protein interactions
1020
0.1
0.2
Gene expression polymorphism
Number of protein-protein interactions
204060
0
S. cerevisiaeD. melanogaster
0
Gene expression polymorphism
Figure 1 Relationship between the population genetic variation in gene
expression (gene expression polymorphism) in S. cerevisiae and
D. melanogaster and the number of protein-protein interactions.
Published online 19 September 2004; doi:10.1038/ng1427
Department of Organismic and Evolutionary Biology, Harvard University, 16 Divinity Avenue, Hartl Lab, Room 2105, Cambridge, Massachusetts, 02138, USA.
Correspondence should be addressed to B.L. (blemos@oeb.harvard.edu).
NATURE GENETICS VOLUME 36 [ NUMBER 10 [ OCTOBER 20041059
BRIEF COMMUNICATIONS
© 2004 Nature Publishing Group http://www.nature.com/naturegenetics

Page 2

stronger stabilizing selection. Therefore, differences in the breadth of
population genetic variation in expression levels (gene expression
polymorphism) may reflect differences in the strength of stabilizing
selection across genes. In yeast, deletion of individual interacting
proteins has similar fitness consequences8. This suggests that proteins
that interact with each other may have similar effects on overall fitness
and may, therefore, be under similar strengths of stabilizing selection.
Data on gene expression polymorphism and absolute transcript
abundance were available for both members of 8,607 (yeast) and
550 (flies) interacting protein pairs. In these data sets, we calculated
the normalized difference in gene expression polymorphism between
interacting proteins averaged across all pairs (Supplementary
Methods online). We calculated null distributions from 10,000
samples generated by shuffling the list of interacting partners. The
amount of expression polymorphism among genes that encode
interacting proteins was more similar than that among random
pairs of genes (yeast, P o 10–4; flies, P ¼ 0.0992 in genes with
more than one allele in both interacting proteins), and these results
hold true when gene expression polymorphism is standardized by
absolute expression level (yeast, P o 10–4; flies, P ¼ 0.0715; Fig. 2a).
These results confirm our prediction that interacting proteins have
similar levels of gene expression polymorphism. This observation
suggests that the two proteins in an interacting pair undergo similar
evolutionary dynamics and, more specifically, may be subject to
similar strengths of stabilizing selection.
Expression levels of interacting proteins may covary across geno-
types if both proteins are regulated by the same segregating genetic
variants or if there is coevolution to minimize the deleterious effects of
imbalance in protein concentration between members of an interact-
ing protein pair7,14. In either case, we would expect the gene expres-
sion levels of interacting proteins to be positively correlated across
strains. To test this prediction, we calculated the Spearman rank
correlation in gene expression between interacting proteins averaged
across all pairs (yeast, r ¼ 0.12; flies, r ¼ 0.05). Expression of genes
that encode interacting proteins was significantly more positively
correlated across strains than that of randomly paired proteins
(yeast, P o 10–4; flies, P ¼ 0.0580; Fig. 2b).
In yeast, genes whose products are involved in complexes have
larger fitness consequences when deleted11, are precisely coregulated
with their interacting partners11and have transcription and transla-
tion properties that minimize noise in the final protein concentra-
tion15, compared with genes whose products do not participate in
protein complexes. Our results on variation in gene expression within
and between species indicate that protein-protein interactions may
constrain regulatory evolution, with such constraints presumably
serving to maintain the protein’s stoichiometric balance with its
interacting partners11,14. These results indicate that the protein inter-
action network may provide a common ground for understanding
regulatory evolution in disparate organisms.
Note: Supplementary information is available on the Nature Genetics website.
ACKNOWLEDGMENTS
We thank D. Hewitt, R. Frankham, J. Blumenstiel, J. Parsch and members of the
laboratory of D.L.H. for suggestions. This work was supported by grants from the
US National Institutes of Health to D.L.H.
COMPETING INTERESTS STATEMENT
The authors declare that they have no competing financial interests.
Received 11 May; accepted 18 August 2004
Published online at http://www.nature.com/naturegenetics/
1. Rifkin, S.A., Kim, J. & White, K.P. Nat. Genet. 33, 138–144 (2003).
2. Meiklejohn, C.D., Parsch, J., Ranz, J.M. & Hartl, D.L. Proc. Natl. Acad. Sci. USA 100,
9894–9899 (2003).
3. Ranz, J.M., Castillo-Davis, C.I., Meiklejohn, C.D. & Hartl, D.L. Science 300, 1742–
1745 (2003).
4. Townsend, J.P., Cavalieri, D. & Hartl, D.L. Mol. Biol. Evol. 20, 955–963 (2003).
5. Bader, J.S., Chaudhuri, A., Rothberg, J.M. & Chant, J. Nat. Biotechnol. 22, 78–85
(2004).
6. Giot, L. et al. Science 302, 1727–1736 (2003).
7. Fraser, H.B., Hirsh, A.E., Steinmetz, L.M., Scharfe, C. & Feldman, M.W. Science 296,
750–752 (2002).
8. Fraser, H.B., Wall, D.P. & Hirsh, A.E. BMC Evol. Biol. 3, 11 (2003).
9. Pal, C., Papp, B. & Hurst, L.D. Nature 421, 496–497 (2003).
10. Bloom, J.D. & Adami, C. BMC Evol. Biol. 3, 21 (2003).
11. Papp, B., Pal, C. & Hurst, L.D. Nature 424, 194–197 (2003).
12. Lande, R. Genet. Res. 26, 221–235 (1975).
13. Kimura, M. The Neutral Theory of Molecular Evolution (Cambridge University Press,
Cambridge, Massachusetts, 1983).
14. Veitia, R.A. BioEssays 24, 175–184 (2002).
15. Fraser, H.B., Hirsh, A., Giaever, G., Kumm, J. & Eisen, M.B. PloS Biology 2, 834–838
(2004).
Average normalized difference in gene expression polymorphism
S. cerevisiaeD. melanogaster
a
1,000
100
200
300
400
–0.002 0.0270.0560.0850.114
100
200
300
400
–0.0100.0050.0200.0360.051
Average correlation in gene expression level across strains
S. cerevisiaeD. melanogaster
b
0.534 0.542 0.550 0.557 0.565 0.573 0.581 0.589
200
400
600
800
100
200
300
0.39 0.41 0.43 0.45 0.47 0.49 0.51 0.53 0.55 0.57
Figure 2 Interacting proteins have similar levels of population genetic
variation in expression (gene expression polymorphism) and expression levels
that are positively correlated across strains. (a) Distributions of the average
normalized difference in gene expression polymorphism in sets of proteins
randomly paired in yeast and fruit flies. The average normalized difference in
gene expression polymorphism between proteins that interact is indicated by
arrows. (b) Distributions of the average Spearman rank correlation of
proteins randomly paired in yeast and fruit flies. The average Spearman rank
correlation between proteins that interact is indicated by arrows.
1060VOLUME 36 [ NUMBER 10 [ OCTOBER 2004 NATURE GENETICS
BRIEF COMMUNICATIONS
© 2004 Nature Publishing Group http://www.nature.com/naturegenetics