
evolution. These structural changes in the neighborhoods sug-
gest a molecular network model of evolution (Fig. 3B): For
orthologous proteins in two species, as the evolutionary distance
between the species increases, the proteins’ local network neigh-
borhoods become increasingly different and the neighborhood
becomes more interconnected in the species that has undergone
more genetic change.
Network Rewiring of Protein–Protein Interactions. To study evolu-
tionary mechanisms of structural changes in the interactomes, we
investigate network motifs (34, 35). We first identify orthologous
protein pairs from evolutionarily close species (SI Appendix, sec-
tion S3), resulting in 2,485,564 protein pairs, which we then use
to calculate interaction rewiring rates (IRRs) for selected net-
work motifs (Fig. 4A). We calculate the number of times each
motif appears in each protein neighborhood and derive the IRR
by comparing the motif occurrences between the interactomes
of the older and the younger species of each protein pair (SI
Appendix, section S7.1). We find strong statistical evidence that
network motifs rewire during evolution (P < 10
−33
for all net-
work motifs; Fig. 4B), suggesting that rewiring of interactions is
an important mechanism for the evolution of interactomes. For
example, proteins in evolutionarily older species on average par-
ticipate in a factor of 0.861 fewer protein–protein interactions
compared with proteins in evolutionarily younger species (IRR =
−0.215; Fig. 4B). This significant negative correlation between a
protein’s number of interactions and the protein’s evolutionary
age confirms earlier studies of Saccharomyces cerevisiae (36). We
also find that square motifs of interactions become more com-
mon in protein neighborhoods during evolution (IRR = 0.016;
Fig. 4B). A range of biological evidence (18, 37, 38) supports
this positive rate of change in the number of square motifs:
From a structural perspective (38), protein–protein interactions
often require complementary interfaces; hence two proteins with
similar interfaces share many of their neighbors. However, they
might not interact directly with each other, which manifests in the
interactome as a square motif of interactions (see SI Appendix,
Fig. S6 for an illustration of interaction interfaces recognizing
the binding sites in proteins). Evolutionary arguments follow-
ing gene duplication (18) reach the same conclusion; proteins
with multiple shared interaction partners are likely to share
even more partners and thereby produce new square motifs of
interactions. To test the predictive power of our motif-based
model of structural network changes, we estimate the size of the
whole human interactome by extrapolating the S. cerevisiae inter-
actome, using IRRs from Fig. 4B (SI Appendix, section S7.3).
Assuming one splice isoform per gene, we predict the number
of interactions in humans to be ∼160,000. This prediction is
in surprisingly good agreement with three previous estimates of
the size of the human interactome, which range from 150,000
to 370,000 interactions (15, 17, 39) and have proved crucial in
establishing the complexity of the human interactome (19).
Discussion
Our analyses reveal how protein–protein interaction networks
change through evolution and how changes in these networks
affect phenotypes and organismal response to environmen-
tal complexity. This systematic investigation of protein–protein
interaction networks from an evolutionary perspective was
enabled by a dataset of interactomes, consisting of protein–
protein interaction networks from 1,840 species. To date, most
evolutionary analyses of biological networks have focused on a
small number of organisms with high-coverage protein–protein
interaction data, such as S. cerevisiae, Mus musculus, and humans.
This is because interactomes mapped by unbiased tests of all
possible pairwise combinations of proteins on the same plat-
form remain scarce, an important limitation of the present study.
Furthermore, experimentally documented protein interactions
are currently subject to a high number of false positives and
negatives. As more protein interaction data are collected, and
more genomes become available, the generalizability of our
findings can be further evaluated. However, our results are con-
sistent across both different subsets of protein interaction data
(SI Appendix, Table S2) and different phylogenetic lineages (SI
Appendix, Fig. S10) and are not explained by many possible
genomic and network confounders (SI Appendix, section S8,
Fig. S8, and Table S1), thus providing confidence that our key
findings cannot be attributed to biases in the datasets.
Interactome resilience is an important aspect of our study. The
resilience measures fragmentation of the interactome into iso-
lated components and thus represents a global measure of the
interactome’s topological stability. Beyond fragmentation, there
are other possible modifications of the interactome that could
alter the network’s biological function without necessarily dis-
connecting the network (40–42). As more detailed information
about functions of individual proteins in the interactome (43), as
well as dynamic protein-expression data (44), becomes available,
our measure of interactome resilience could be adapted to give
a more complex definition of resiliency, which might yield more
detailed evolutionary predictions. Additionally, information on
how protein–protein interactions change dynamically both in
time and space (45–47) might reveal how topological stability of
the interactome depends on large-scale interactome connectivity
as well as on the interactome’s dynamic properties (40).
Our study presents an additional paradigm for evolutionary
studies by demonstrating that interactomes reveal fundamental
structural principles of molecular networks. Our findings high-
light evolution as an important predictor of structural network
change and show that evolution of a species predicts resilience
of the species’ interactome to protein failures. The findings
offer quantitative evidence for the biological proposition that
an organism that has undergone more genetic change has a
more resilient interactome, which, in turn, is associated with
the greater ability of the organism to survive in a more com-
plex, variable, or competitive environment. Our findings can also
help clarify the mechanisms of how interactomes change dur-
ing evolution, why currently observed network structures exist,
and how they may change in the future and facilitate the extrap-
olation of functional information from experimentally charac-
terized proteins to their orthologous proteins in poorly studied
organisms.
Materials and Methods
Detailed description of data, statistical methodology, and additional analy-
ses are provided in SI Appendix.
Code and Data Availability. Software implementation of statistical method-
ology is publicly available at snap.stanford.edu/tree-of-life. All data used
in this paper, including the processed interactomes, are shared with the
community and available from snap.stanford.edu/tree-of-life.
ACKNOWLEDGMENTS. For helpful discussion and comments on the manu-
script, we thank Aaron Goodman, Hunter B. Fraser, Alice Y. Ting, Emma Pier-
son, Michael An, and Pratyaksh Sharma. M.Z., R.S., and J.L. were supported
in part by NSF, NIH, Defense Advanced Research Projects Agency, Boeing,
Stanford Data Science Initiative, and Chan Zuckerberg Biohub. M.W.F. was
supported in part by the Stanford Center for Computational, Evolutionary,
and Human Genomics and a grant from the John Templeton Foundation.
1. Castelle CJ, Banfield JF (2018) Major new microbial groups expand diversity and alter
our understanding of the tree of life. Cell 172:1181–1197.
2. Barab
´
asi AL, Oltvai ZN (2004) Network biology: Understanding the cell’s functional
organization. Nat Rev Genet 5:101–113.
3. Hu JX, Thomas CE, Brunak S (2016) Network biology concepts in complex disease
comorbidities. Nat Rev Genet 17:615–629.
4. Huttlin EL, et al. (2017) Architecture of the human interactome defines protein
communities and disease networks. Nature 545:505–509.
4432
|
www.pnas.org/cgi/doi/10.1073/pnas.1818013116 Zitnik et al.