Prioritization of Protein
Pedro Beltrao,1,3,* Ve ´ronique Albane `se,4Lillian R. Kenner,1,3Danielle L. Swaney,5Alma Burlingame,2,3Judit Ville ´n,5
Wendell A. Lim,1,3,6James S. Fraser,1,3Judith Frydman,4and Nevan J. Krogan1,3,7,*
1Department of Cellular and Molecular Pharmacology
2Department of Pharmaceutical Chemistry
University of California, San Francisco, San Francisco, CA 94107, USA
3California Institute for Quantitative Biosciences, QB3, San Francisco, CA 94107, USA
4BioX Program, Biology Department, Clark Center, Stanford, CA 94305, USA
5Department of Genome Sciences, University of Washington, Seattle, WA 98195, USA
6Howard Hughes Medical Institute, University of California, San Francisco, San Francisco, CA 94158, USA
7J. David Gladstone Institute, San Francisco, CA 94158, USA
*Correspondence: email@example.com (P.B.), firstname.lastname@example.org (N.J.K.)
Protein function is often regulated by posttransla-
tional modifications (PTMs), and recent advances in
mass spectrometry have resulted in an exponential
increase in PTM identification. However, the func-
tional significance of the vast majority of these modi-
ficationsremainsunknown. Toaddressthis problem,
we compiled nearly 200,000 phosphorylation, acety-
lation, and ubiquitination sites from 11 eukaryotic
species, including 2,500 newly identified ubiquityla-
tion sites for Saccharomyces cerevisiae. We devel-
oped methods to prioritize the functional relevance
of these PTMs by predicting those that likely partici-
pate in cross-regulatory events, regulate domain
activity, or mediate protein-protein interactions.
PTM conservation within domain families identifies
regulatory ‘‘hot spots’’ that overlap with functionally
important regions, a concept that we experimentally
validated on the HSP70 domain family. Finally, our
analysis of the evolution of PTM regulation highlights
potential routes for neutral drift in regulatory interac-
sites are likely to have a significant biological role.
The activity and localization of proteins inside the cell can be
regulated by reversible posttranslational modifications (PTMs),
including protein phosphorylation, acetylation, and ubiquityla-
tion. How these modifications regulate protein function and
how this regulation diverges during evolution is crucial for under-
standing signaling systems. Recent advances in mass spec-
trometry (MS) have increased the ability to identify PTMs with
thousands of sites now routinely discovered per study (Choudh-
aryand Mann,2010). However, the functional characterization of
these modifications is now rate limiting, a fact further compli-
cated by the recent findings that they can be highly divergent
across species (Beltrao et al., 2009; Holt et al., 2009; Landry
et al., 2009; Tan et al., 2009). Despite the poor conservation
within single proteins, the overall number of phosphosites per
protein within different functional modules (i.e., protein complex
or pathways) is conserved (Beltrao et al., 2009). This phenom-
enon could be explained by compensatory turnover of phos-
phorylation sites, similar to documented cases of compensatory
turnover of transcription factor binding sites in promoter regions
(Ludwig et al., 2000). The similarities in the evolutionary proper-
ties of transcriptional and posttranslational regulatory networks
(Moses and Landry, 2010) lend credence to the idea that pheno-
typic diversity is primarily driven by changes in regulatory
networks (Carroll, 2005).
Although phosphosites observed in high-throughput studies
are, on average, poorly conserved, sites with a known function
are more significantly constrained (Ba and Moses, 2010; Landry
et al., 2009). These trends have led some to speculate that there
is a substantial fraction of phosphorylation sites that are
nonfunctional (Landryetal.,2009;Lienhard, 2008).Conservation
of modification sites or regulatory interactions can be used to
prioritize experimental validation (Tan et al., 2009) but do not
provide a putative functional consequence for the modification.
Therefore, developing approaches to dissect the functional
importance of PTMs is currently the most significant bottleneck
in proteomic studies of posttranslational regulation.
In this study, we experimentally determined 2,500 ubiquityla-
tion sites for S. cerevisiae and compiled a list of nearly 200,000
modification sites across 11 eukaryotic species in order to
develop predictors of PTM functional relevance. These data,
as well as structural information, were used to identify modifica-
tions that might regulate protein-protein interactions, mediate
domain activity or be part of cross-regulatory events between
Cell 150, 413–425, July 20, 2012 ª2012 Elsevier Inc. 413
different PTMs. We show that sites with predicted function are
more likely to be conserved and that conservation of PTMs
within domain families identifies important regulatory regions
(termed here regulatory ‘‘hot spots’’). We validate these ap-
proaches by experimentally characterizing regulatory hot spots
within the HSP70 chaperone domain family and characterizing
a phosphosite within Skp1 (part of the Skp1/Cullin/F-box E3
ligase) as likely regulating the interaction between Skp1 and
the Met30 F-box protein. In summary, the resource developed
in this study, which is accessible online (http://ptmfunc.com),
can provide mechanistic functional annotations to PTMs and
generate specific predictions for experimental validation. This
analysis also allows for a better understanding of the evolution
of posttranslational networks and suggests that only a fraction
of PTMs is likely to have a regulatory role.
A Resource of Eukaryotic PTMs for Functional
and Evolutionary Studies
In order to study the evolutionary properties and functional role
of protein posttranslational regulation, we compiled previously
published in vivo, MS-derived PTMs (Table S1 available online).
We compiled a total of 153,478 phosphorylation sites for 11
eukaryotic species, retaining only sites that have high site local-
ization probability (Experimental Procedures). The phosphoryla-
tion data set covers a broad evolutionary time scale with infor-
mation for three fungi (S. cerevisiae, Schizosaccharomyces
pombe, and Candida albicans), two plant species (Arabidopsis
thaliana and Oryza sativa), three mammals (Homo sapiens, Mus
musculus, and Rattus norvegicus) as well as Xenopus laevis,
Drosophila melanogaster, and Caenorhabditis elegans. We also
compiled 13,133 lysine acetylation sites (covering H. sapiens,
M. musculus, and Drosophila melanogaster) and 22,000 human
ubiquitylation sites (Emanuele etal., 2011; Kim etal., 2011; Wag-
ner et al., 2011). In addition, we used a MS approach to experi-
mentally determine 2,500 ubiquitylation sites in S. cerevisiae
to facilitate comparative studies. Using a set of 12 different
S. cerevisiae phosphoproteomics experiments, we estimate
that the curated data sets should have less than 4% of false-
positive sites (Table S2).
Previous studies have used sequence conservation to study
the evolution of phosphosites (Ba and Moses, 2010; Holt et al.,
2009; Landry et al., 2009). In this work, we used the compiled
data to directly compare the phosphoproteomes across these
11 species and to evaluate the impact of data quality on the
evolutionary observations. We selected one of the species with
the highest coverage, the human data set, as reference and
compared the data from all other species to it (Figure 1A). We
aligned 1-to-1 orthologs of each species to H. sapiens proteins,
and for each phosphosite, determined the conservation in the
human protein of both the phospho-acceptor residue (i.e.,
sequence conservation) and the phosphorylation site (i.e., phos-
phosite conservation). In order to account for potential errors in
MS phosphosite positional assignments, we considered a phos-
phosphorylated within a window of ±2 alignment positions. Both
residue and phosphosite conservation were found to be propor-
tional to the divergence age away from H. sapiens (Figure 1A).
Phosphosite conservation ranged from ?8%–18% for the
distantly related plants and fungi to ?40% for the closely related
mouse and rat. We then asked whether the observed value was
higher than expected by randomly re-assigning the same
number of phosphosites within each orthologous protein. As
previously described by Landry and colleagues, we observed
that the sequence conservation of the phospho-acceptor re-
sidue was not higher than expected by chance (Figure 1B)
ylation sites was approximately two to three times higher than
random (Figure 1B). This difference suggests that the con-
servation of the phosphorylation state is a better indicator of
functional importance than sequence conservation of the phos-
pho-acceptor residue. For all the following analysis, we used the
conservation of the PTM state as the comparative metric.
In order to evaluate the generality of these evolutionary obser-
vations across different PTMs, we studied the conservation over
random expectation of ubiquitylation and acetylation sites
(Figure 1C). We compared the conservation of S. cerevisiae
phosphorylation and ubiquitylation sites in H. sapiens over a
random expectation calculated based on random sampling of
a similar number of modification acceptor residues within the
same proteins. Similarly, we compared the conservation of
D. melanogaster phosphorylation and acetylation sites in human
orthologs. The three modifications show a low level of evolu-
tionaryconstraint, ranging from1.3to2.2 timeshigherconserva-
tion than expected based on an equivalent random sample of
PTM acceptor residues (Figure 1C). Protein acetylation shows
a higher value of conservation over random when compared
to phosphorylation, consistent with previous work (Weinert
et al., 2011), whereas ubiquitylation appears to have a lower
evolutionary constraint when compared to phosphorylation
Given that these data sets gathered so far are likely to
be incomplete, the conservation values presented here are
teomics experiments to evaluate the error in the conservation
for each S. cerevisiae phosphoproteomic data set (Figure 1D).
Extrapolating from the regression analysis, we estimate that,
when corrected for coverage, ?10% of H. sapiens phosphosites
would be conserved in S. cerevisiae. We also calculated the
corrected conservation value for each data set independently
and estimated the corrected median conservation value for
H. sapiens phosphosites in S. cerevisiae as ?13% (Figure 1D).
This value is higher but comparable to the observed 8% conser-
vation measured with the complete data set. This suggests
that, at least for the extensively studied phosphoproteomes of
S. cerevisiae and H. sapiens, additional data are unlikely to
dramatically change the conservation estimates.
Besides coverage (i.e., false negatives), data quality (i.e., false
positives), and low phosphosite abundance are also important
factors when estimating phosphoproteome conservation. We
compared the conservation of S. cerevisiae phosphosites in
H. sapiens across different data quality criteria, including the
number of spectral counts, match to known kinase recognition
motifs and information on dynamically regulated phosphosites
414 Cell 150, 413–425, July 20, 2012 ª2012 Elsevier Inc.
(Figure 1E). The conservation of different classes of phosphory-
lation sites (Figure 1E, blue bars) was compared to an equivalent
random sample (Figure 1E, red bars). To determine statistical
significance of the results, the ratios of conserved over expected
values for the different phosphosite groups were compared
using a Mann-Whitney ranked test. We assumed that spectra
and/or peptide count for each phosphosite is correlated with
data quality and/or phosphosite abundance and observed that
peptide count >1) are more likely to be conserved than those
observed only once (Figure 1E, peptide count = 1, p
value < 10?8). Additionally, sites that represent well-matched
kinase-recognition motifs (Figure 1E, ‘‘Kinase preference’’) or
are known to be regulated (Figure 1E, ‘‘Regulated’’), as
measured in quantitative MS studies (Holt et al., 2009; Huber
et al., 2009; Soufi et al., 2009), are moderately more conserved
than average sites and more highly conserved than expected
by chance (Figure 1E, ‘‘S.cer phosphosites,’’ p value < 10?9).
Finally, sites that are known to be functionally important (Ba
and Moses, 2010) or have described in vivo kinase regulators
(http://www.phosphogrid.org) (Stark et al., 2010) are more than
three times more conserved than average sites (Figure 1E,
‘‘With known kinases’’ versus ‘‘S.cer phosphosites,’’ p value <
10?16). These results imply that higher phosphosite functionality,
quality, and/or abundance are correlated with conservation and
support previous observations made with sequence analysis
(Landry et al., 2009). It is likely that low abundance and/or
nonfunctional phosphosites, with low conservation, dominate
0 500 10001500 2000
Divergence Time (My)
Conservation in H.sap (%)
Divergence Time (My)
Conservation in H.sap / Random
Conservation in H.sap / Random
S.cer PTMs D.mel PTMs
Conservation in H.sap
0 0.1 0.2 0.3
With known kinase
>1 Peptide count <10
Peptide count = 1
Conservation of H.sap
phosphosites in S.cer (%)
S.cer dataset coverage (%)
y = 0.104
R² = 0.8
0 2550 75 100
Figure 1. Evolutionary Properties of Eukaryotic Posttranslational Modification Sites
(A) We analyzed the conservation of ten phosphoproteomes against that of H. sapiens using protein alignments of 1-to-1 orthologs. For each species, we
compared the conservation of the phosphoacceptor residues (i.e., sequence conservation) with the conservation of phosphorylation site based on the MS
experimental evidence. A random expectation or null model for each case was determined based on the random shuffling of phosphosite positions within each
(B) Ratio of observed conservation over random expectation.
(C) Ratio of conservation over random expectation for different PTMs. Error bars represent 1 SD.
(D) Predicted coverage and conservation in H. sapiens for 12 different previously reported phosphoproteomics experiments for S. cerevisiae.
(E) To access the impact of data quality on the conservation values we made use of different criteria to define subsets of S. cerevisiae phosphosites. For each
subset we calculated the observed conservation in H. sapiens as well as the expected value based on random sampling. Error bars represent 1 SD.
Cell 150, 413–425, July 20, 2012 ª2012 Elsevier Inc. 415
the overall measured divergence. These results further under-
score the need to devise methods to assign functional roles to
Phosphorylation Coupled to Other Classes of
On average, ?75% of known phosphorylation sites, 40% of
acetylation sites and 45% of ubiquitylation sites occur outside
known PFAM globular domains (Table S1). It has become
increasingly apparent that phosphosites within unstructured
regions not only recruit phospho-binding domains but also often
regulate other PTMs or localization signals (Hunter, 2007). This
described in histone tails where they have been proposed to
form a code to be read by different effectors and control gene
chromatin states (Strahl and Allis, 2000). More recently, exam-
ples of cross-regulation between adjacent PTMs have been
observed in several other proteins suggesting this to be
a universal mode of protein regulation (Hunter, 2007). Examples
include the promotion of sumoylation by a priming phosphoryla-
tion in several transcription factors (Yang and Gre ´goire, 2006),
the cross-inhibition between adjacent phosphorylation and
methylation sites in DNMT1 (Este `ve et al., 2011) and the positive
role of lysine acetylation on the phosphorylation of Cdc6 (Paoli-
nelli et al., 2009).
We hypothesized that it is possible to assign a functional role
to PTM sites by searching for the co-occurrence of different
modifications within the same protein. For the human proteome,
using the information on lysine acetylation, ubiquitylation and
sumoylation, we observed a significant overlap between pro-
teins containing these lysine modifications and the phosphopro-
teome (Figure 2A). Though 36% of all proteins are phospho-
proteins, >69% of proteins containing any of these lysine
modifications are also phosphorylated (Figure 2A). This enrich-
ment is highly significant (p value < 1 3 10?70, with a Fisher’s
exacttest) andnotmerely dueto MSdetection bias forabundant
proteins (Table S3). We next asked whether these PTMs tend to
cluster within the protein sequence (Figure 2B). Given the small
number of currently characterized sumoylation sites, we
grouped these together with ubiquitylation for the analysis. We
binned phospho-acceptor residues (i.e., serine, threonine, and
tyrosine) according to their smallest distanceto a modifiedlysine
residue. In each distance bin, we then calculated the fraction of
acceptor residues that is phosphorylated and compared this
observed value with a random expectation by randomly reas-
signing the same number of phosphorylation sites within each
protein. We observed that, on average, phospho-acceptors
near modified lysines are preferentially phosphorylated when
compared to more distant residues or an equivalent random
sample of sites. These results show that the different PTMs
tend to cluster within protein sequences. This result is not merely
due to preferential accumulation of PTMs in unstructured
regions (Figure S1) and was also observed using phosphoryla-
tion and lysine acetylation data for mouse and phosphorylation
and ubiquitylation data for S. cerevisiae (Figure S1). If phosphor-
ylation sites near other PTMs are more likely to be functionally
relevant, then we assumed that these should also show higher
conservation. We tested this by comparing the conservation, in
21664 H.sapiens proteins
(excluding splicing isoforms)
Ubiquitylated proteins (5399)
Fraction of phospho-acceptors phosphorylated in bin
Binned phospho-acceptor residues as a
function of distance to modified Lysine
Conservation of human phosphosites in S. cer
Figure 2. Association of Protein Phosphorylation with Lysine Posttranslational Modifications
(A) A Venn diagram representing the overlap of the different lysine modified proteins with the human phosphoproteome. Whereas 33% of the human proteins are
phosphorylated, 71% of the acetylated proteins, 69% of the ubiquitylated proteins, and 89% of the sumoylated proteins are also phosphorylated.
(B) The fraction of the phosphoacceptor residues was plotted as function of the distance to modified lysine residues. Observed values were compared with
expected values based on random sampling. Error bars represent 1 SD.
(C) Human phosphorylation sites that are near an acetylated lysine residue (<15 amino acid distance) were more likely conserved in S. cerevisiae than average
sites or an equivalent random sample. Error bars represent 1 SD.
See also Figure S1 and Table S3.
416 Cell 150, 413–425, July 20, 2012 ª2012 Elsevier Inc.
S. cerevisiae, of all human phosphorylation sites with those
phorylation sites near modified lysines were higher than for
average phosphosites and higher than an equivalent random
sample (p value < 10?7) (Figure 2C).
Regulation of Protein-Protein Interactions
For the PTM sites that occur within structured regions, we can
make use of the growing structural knowledge deposited in the
PDB (http://www.pdb.org) to assign putative functional roles
for the protein modifications, especially those found within
protein interfaces, since they may be involved in the regulation
of protein-protein interactions. For human or S. cerevisiae
protein-pairs that are known to physically interact, we used
X-ray structures, homology models or docking solutions to
define the most likely interface regions (Experimental Proce-
dures). Using these models, we identified 870 phosphorylation,
632 ubiquitylation, and 263 acetylation sites at putative interface
residues that can potentially regulate protein-protein interac-
tions (available at http://ptmfunc.com). To expand the number
of putative interface residues, we made use of the 3DID
database of domain-domain interactions (Stein et al., 2011).
For each domain family, as annotated in PFAM (http://pfam.
sanger.ac.uk), 3DID contains annotations of what residues
have been shown to participate in physical interactions in
X-ray structures. We have used these annotations to assign
interaction residues for PFAM domains in the 11 proteomes
1,802 ubiquitylation, and 1,691 acetylation sites that potentially
either of these definitions, we observed that S. cerevisiae phos-
phosites at interface residues are approximately two to three
times more likely to be conserved in H. sapiens than average
phosphosites (Figure 3A, ‘‘Interface residue’’ or ‘‘PFAM interac-
tion residue’’ versus ‘‘All phosphosites,’’ p value < 10?14). It is
known that globular domain regions are easier to align than the
unstructured regions where mostphosphosites occur. However,
the higher conservation of phosphosites at interface residues is
not merely due to alignment issues since phosphosites that
occur within PFAM domains are not more conserved than
average sites (Figure 3A). We confirmed these evolutionary
trends using interface models for human protein-protein interac-
tions (Figure S2A).
In order to test the generality of some of these observations
across different posttranslational modifications, we compared
the conservation (over random expectation) of all acetylation,
ubiquitylation and phosphorylation sites with those occurring
at predicted interface residues (Figure 3B). In line with the
observations made for protein phosphorylation, lysine acetyla-
tion at interface residues is more likely to be conserved (Fig-
ure 3B, p value < 10?5), however, ubiquitylation at interface resi-
dues shows a similar level of constraint when compared to
average ubiquitylation sites. These results suggest that phos-
phorylation and acetylation but not ubiquitylation sites at inter-
face residues are more likely to be functionally important than
average sites, suggesting that these PTMs are commonly
used by the cell to reversibly regulate the binding affinity of
The analysis of protein-protein interfaces creates specific
predictions for the functional role of PTMs. For example,
several alpha subunits of the proteasome are phosphorylated
at interface regions (Figures S2B and S2C). Serine 13 and
tyrosine 5 of Pre8 (the S. cerevisiae alpha 2) are phosphory-
lated in yeast and human, respectively, and could potentially
regulate the interactions with Pre9 (alpha 3). The N terminus
of alpha 5 is also phosphorylated in 7 of the 11 species.
This N-terminal region has been shown to be important for
proteasome activity (Groll et al., 2000) indicating that these
N-terminal phosphorylations might regulate the interactions
between alpha subunits or the activity of the proteasome
(Figures S2B and S2C). Similarly, we predicted that a phospho-
site at position S162 in the S. cerevisiae Skp1 could regulate
the interaction with Met30 (Figure 3C). Skp1 is a highly
conserved protein that is part of the Skp1/Cullin/F-box (SCF)
multisubunit E3 ubiquitin ligase complex (Petroski and Deshaies,
2005). Skp1 interacts with different F-box domain containing
proteins that can modulate the ubiquitylation substrate speci-
ficity (Petroski and Deshaies, 2005). In S. cerevisiae, Skp1 can
interact with the Met30 F-box protein to regulate proteins
involved in sulfur metabolism (Jonkers and Rep, 2009) an inter-
action that is known to be regulated under different stress condi-
tions (Jonkers and Rep, 2009). We postulated that the highly
conserved phosphorylation site in Skp1 might regulate the inter-
action with Met30 and/or other F-box proteins (Figure 3C). We
note that given the position of the residue at the end of the helix,
can access the kinase active site.
To experimentally probe the dependency of the Skp1:Met30
interaction on the phosphorylation status of S162, we used
a protein complementation assay (Ear and Michnick, 2009;
Michnick et al., 2010) that reports on the strength of the
protein-protein interaction in vivo. We fused Skp1 and Met30
to two fragments of the yeast cytosine deaminase and trans-
formed the constructs into a strain that lacks the endogenous
enzyme. Skp1 and Met30 interact directly in vivo allowing the
two fragments to reconstitute cytosine deaminase activity.
Reconstitution of enzyme activity permits growth on media
lacking uracil (-Ura) and leads to death on media containing
5-fluorocytosine (+5-FC) (Figure 3D). To test the idea that phos-
phorylation reversibly regulates the assembly of this interaction,
we mutated S162 to alanine (S162A) or the phosphomimetic
aspartic acid (S162D). The S162A mutant, similar to wild-type,
supported growth on –Ura media, which selects positively for
interacting proteins, and grew poorly on +5-FC media, which
counter-selects for interacting proteins, indicating that the un-
phosphorylated state binds Met30 (Figure 3D). In contrast, the
S162D mutant grew better on +5-FC than on –Ura media, indi-
cating that the phosphorylated state binds Met30 weaker than
the unphosphorylated state (Figure 3D). To validate this result,
Flag-tagged Skp1-S162A and Skp1-S162D were immunopre-
cipitated in the presence of Met30-Myc, and the Met30:Skp1
interaction was monitored using an a-Myc antibody. We found
that the Skp1:Met30 interaction is impaired in the phosphomi-
metic mutant, but not in the alanine mutant (Figure 3F), suggest-
ing that the phosphorylation of S162 acts as a reversible switch
for Met30 affinity.
Cell 150, 413–425, July 20, 2012 ª2012 Elsevier Inc. 417
The Skp1:Met30 interaction is required to keep the Met4 tran-
scription factor inactivated via ubiquitylation (Kaiser et al.,
synthesis of sulfur-containing amino acids and glutathione
metabolism but it also results in cell cycle arrest (Aghajan
et al., 2010). During our interaction studies, we observed that
type that was not observed with the overexpression of Skp1 WT
or S162A mutant (Figure 3E). These results suggest that Skp1
S162D impairs the interaction with Met30 resulting in an activa-
tion of Met4 and cell cycle arrest. The Met4 inactivation by
Skp1:Met30 is known to be promoted by SAM (Kaiser et al.,
2006). Consistent with the hypothesis that Skp1 S162D overex-
pression results in Met4 activation and cell cycle arrest, growth
in the presence of SAM relieves the impaired growth (Fig-
ure 3E), presumably by further activating the available pool
of Skp1:Met30 and/or relieving independently a cell cycle
block. These collective results strongly suggest that the
Within PFAM domains
PFAM interaction residue
Conservation in H.sap
Phos. at interface
Ace. at interface
Phos. at interface
Ubi. at interface
Conservation in H.sap / Random expectation
Number of colonies
Cell cycle arrest
Figure 3. Regulation of Interface Residues by PTMs
(A) The conservation of all S. cerevisiae phosphosites in H. sapiens was compared with the conservation of phosphosites within PFAM domain interface residues
(from the 3DID database) and at putative interface residues (from interaction models). Error bars represent 1 SD.
(B) We compared the conservation over random expectation of different PTMs at PFAM interface residues (from the 3DID database). The ratios of conservation
over random expectation were compared using a Mann-Whitney rank test. *p value < 0.05.
(C) TheSkp1:Met30interactionmodel. TheS162 position of theS.cerevisiaeSkp1 wasfound tobephosphorylated inS.cerevisiae,H.sapiens,M. musculus,and
(D) A protein complementation assay of cytosine deaminase activity reports on the Skp1:Met30 interaction. The phosphomimetic mutation reversed the growth
pattern on selective media reporting on interaction strength by growing on +5-FC plates and without any observable growth on –Ura plates suggesting that the
in vivo affinity of the Skp1:Met30 complex is reduced by phosphorylation.
a phenotype that was relieved in the presence of SAM.
(F) Skp1 S162D mutation shows a significant decrease in bound Met30 when compared to the S162A mutant by coimmunoprecipitation.
See also Figure S2.
418 Cell 150, 413–425, July 20, 2012 ª2012 Elsevier Inc.
phosphorylation of Skp1 on residue S162 has the potential to
reversibly alter the binding affinity of Skp1 with Met30 and regu-
late the function of Skp1:Met30.
Next, we used the interface models to differentiate between
the conservation of phosphorylation sites at interface residues
from the conservation of the predicted function (i.e., regulation
ylation of an interface might be conserved despite the diver-
gence of the actual phosphosite position. We observed that
>50% of the interfaces that are phosphorylated in S. cerevisiae
are also phosphorylated in H. sapiens despite only ?18% of
theinterface phosphorylation sites showingpositional conserva-
tion (Figure 4A). However, given that the current phosphopro-
teomes are likely to be incomplete, we cannot rule out that
some of the observed positional divergence is not due to
a coverage issue. A similar trend is observed using human inter-
face models (Figure S2A). If the conservation of function with
divergence of phosphosite position is mostly the product of
a neutral variation, we might expect to observe a conservation
of the kinase recognition for the phosphosites at the same in-
terface. To study this issue, we devised a metric of phospho-
site similarity based on the models of binding preferences of
63 S. cerevisiae kinases and calculated the similarity of
S. cerevisiae interface phosphosites with human phosphosites
at the same interface (Experimental Procedures). We then
compared these scores with the similarity scores for random
pairs of phosphosites and sites known to be regulated by the
same kinases (Figure 4B). The median phosphosite similarity
for interface phosphosites is higher than for random pairs (p
value < 2 3 10?16with a Kolmogorov-Smirnov test) suggesting
that a significant fraction of phosphosites observed at the
same interface in different species are phosphorylated by
kinases of similar specificity.
An example of conserved phosphorylation of an interface at
different positions is shown in Figure 4C for the interaction
between the S. cerevisiae Rho family GTPase Cdc42p and the
Rho inhibitor Rdi1p. Rdi1p is phosphorylated at the S40 position
in S. cerevisiae. Although the S40 equivalent position is phos-
phorylated in the C. albicans ortholog, it is currently not know
to be phosphorylated in human. Instead, the Rdi1p Y20 position
is phosphorylated in the human ortholog (Figure 4C), but it is
currently not know to be regulated in fungi. Regulation of Rho-
inhibitor interactions by phosphorylation has been previously
tion of Rho proteins (DerMardirossian et al., 2004). Our analysis
suggests that the phospho-regulation of the Cdc42:Rdi1 might
be highly conserved but achieved by the phosphorylation of
different positions in different species.
Posttranslational Hot Spots within Domain Families
We show above that PTMs with putative functional annotations
are more likely to be conserved across species than average
sites. We hypothesized that we could use conservation to iden-
tify regions within domain families with high regulatory potential.
Ten domain families that are extensively phosphorylated across
the 11 species with available phosphorylation data were initially
selected for this analysis (Table S4). For each domain family,
we selected a representative sequence/structure from the PDB
(Table S4), then aligned each domain from the 11 species to
the representative sequences/structures and mapped to them
Figure 4. Conservation of Interface Phosphorylation Can Be Achieved by Regulation of Different Positions
(A) We compared the conservation of all S. cerevisiae interface phosphorylation sites in H. sapiens (Interface residues) with the conservation of the phosphor-
ylation of S. cerevisiae interfaces in H. sapiens without regard to the actual phosphosite position (Phosphorylated interfaces). Error bars represent 1 SD.
(B)A metric of phosphositesimilarity (ExperimentalProcedures) was used tocomparephosphosite pairs found at thesame interfaces inthe two different species
(S. cerevisiae and H. sapiens) with random phosphosite pairs and pairs known to be regulated by the same kinases. Open circles represent the top and bottom
(C) The model of the S. cerevisiae Cdc42p:Rdi1p interface was annotated with currently known phosphorylation data. The Rdi1p Y20 and S40 positions refer to
the S. cerevisiae protein sequence positions. The S40 position is currently known to be phosphorylated in S. cerevisiae and C. albicans but not in human.
Conversely the Y20 position is known to be phosphorylated in human but not in fungi.
Cell 150, 413–425, July 20, 2012 ª2012 Elsevier Inc. 419
all phosphorylation sites. Putative regulatory regions were iden-
hot spots, were determined based on random sampling with a p
value cutoff of 0.005 or less (Experimental Procedures). We
hypothesize that phosphorylation of residues within these regu-
latory hot spots are more likely to regulate domain function. A
similar analysis for lysine acetylation was performed for the
protein kinase domain.
In Figure 5, we show the sequence and structural mapping for
the enrichment of phosphorylation or acetylation sites for two
example domains (protein kinase and HSP90). As expected,
the most significantly phosphorylation enriched region for the
kinase family is the activation loop region (Nolen et al., 2004). A
second hot spot of phospho-regulation was observed within
the ‘‘glycine-rich’’ loop that contributes to ATP binding and has
been described to activate or inhibit the activity of kinases, in
particular, CDKs (Narayanan and Jacobson, 2009). Interestingly,
there is no significant enrichment of acetylation sites within the
activation loop of kinases but instead these are preferentially
observed within the N-terminal lobe region. This enrichment is
primarily due to a catalytic lysine residue that is often observed
to be acetylated, a modification that has been previously shown
to be important for the regulation of kinase activity (Choudhary
et al., 2009).
The HSP90 domain family is a highly conserved dimeric heat-
shock protein family that facilitates the folding of client proteins
involved in a multitude of biological functions (Taipale et al.,
2010). We identified 145 phosphorylation sites within members
of the HSP90 domain that were preferentially enriched in the
C-terminal region (Figure 5). The strongest enrichment segment
corresponds to the residues 600–610 of the yeast HSP90
(HSC82) sequence that projects from the C-terminal region
and forms contacts with the equivalent segment of the opposing
dimer (Ali et al., 2006). Phosphorylation of this region is therefore
0 200 400 600
Position within domain
Enrichment over random
0 100200 300
Enrichment over random
Position within domain
0100 200 300
Enrichment over random
Position within domain
Figure 5. Phosphorylation Enrichment Analysis
Identifies Regulatory Hot Spots
For each domain family under analysis, we selected
a representative structure from the PDB database
(Table S4). Phosphorylation and acetylation data was
mapped to the representative sequence/structure using
sequence alignments and random sampling was used to
calculate the enrichment over random. This value was
plotted along the domain sequence position of the repre-
sentative structure as a moving average and a cutoff of
p value < 0.005 was used to identify significant enrichment
(dotted line). See also Figure S3.
likely to regulate HSP90 function. It has been
shown that the Ppt1 phosphatase binds to the
HSP90 C-terminal region and that the disruption
of this interaction results in hyperphosphoryla-
tion and misregulation of HSP90 (Wandinger
et al., 2006). Consistent with these ideas,
Soroka and colleagues validated this prediction
by demonstrating that the 600–610 region of the
yeast HSP90 is in fact regulated by Ppt1 and
that phosphorylation of this region has the
potential to regulate HSP90 function (Soroka et al., 2012).
We believe that this enrichment approach can be used to
study the regulatory potential of different domain families and
we provide additional examples in Supplemental Information
Regulation of the HSP70 Domain Family by Protein
The results above strongly suggest that our statistical enrich-
ment analysis can highlight functionally important sites subject
to regulation by PTMs. In order to further validate this approach,
we studied in more detail the regulation of the heat shock 70 kDa
(HSP70) domain family. The HSP70 is a highly conserved chap-
erone that folds client proteins through an ATP-dependent cycle
of binding and release (Kampinga and Craig, 2010). HSP70
proteins are constituted of two domains, an N-terminal nucleo-
tide binding domain (NBD) and a C-terminal substrate binding
domain (SBD) (Figure 6A). Although the HSP70 family has been
extensively studied and is implicated in a myriad of cellular
functions (Kampinga and Craig, 2010), its regulation by protein
phosphorylation has not been previously explored.
We identified 313 phosphosites within HSP70 proteins across
the 11 species and our enrichment analysis highlighted two
significant hot spots (Figure 6A). Strikingly, both of these map-
ped to functionally and structurally important regions, one near
the nucleotide binding pocket (Region 1) and the second near
the entrance to the peptide binding groove (Region 2). The two
regions were then used to predict the corresponding regulatory
phosphosites in SSA1, an abundant cytosolic HSP70 in the
budding yeast. SSA1 has been involved in multiple cellular func-
tions, including binding to polysomes and nascent chains, and
assisting the refolding of newly made and stress-denatured
polypeptides, as well as prevention of protein aggregation, the
posttranslational translocation of newly synthesized secreted
proteins into the endoplasmid reticulum (ER) and mitochondria,
420 Cell 150, 413–425, July 20, 2012 ª2012 Elsevier Inc.
and degradation of misfolded proteins (Albane `se et al., 2006;
Horton et al., 2001). To test the functional relevance of the pre-
dicted sites, SSA1 constructs were designed with alanine or
phosphomimetic mutations of residues that were known to be
phosphorylated and within these hot spot regions. Two closely
spaced phosphorylated threonines were mutated in Region 1
(T36, T38) and three phosphosites were mutated in Region 2
(T492, S495, T499) (Figure 6A). We also mutated serine 326,
a position known to be phosphorylated but outside the hot
spot regions to serve as a control. Since the cytosol of yeast
contains four nearly identical SSA homologs (SSA1–4) the
different SSA1 mutants were studied in two yeast strains engi-
neered to lack cytosolic Hsp70 function: (1) a strain lacking
sensitive point mutation renders it inactive above 37?C (ssa1-45)
and (2) a strain lacking both SSA1 and SSA2, but containing
functional copies of the less abundant SSA3 and SSA4. Similar
results were obtained in both types of cells.
Growth of the wild-type and mutant strains were measured
in liquid culture (Figure 6B) or using serial spot dilution assays
like wild-type, none of the phosphorylation mutants in Regions 1
and 2 were able to fully complement the growth even under
nonstress conditions of 30?C, indicating that the regulatory hot
0 100 200 300 400 500
Position within domain
Enrichment over random
1-A T36A S38A
1-D T36D S38D
2.1-A T492A S495A
2.1-D T492D S495D
2.2-A T492A S495A T499A
2.2-D T492D S495D T499D
Luciferase recovery (%)
Average aggregates per cell
0 30 60 90 120
0 60 120
Figure 6. Phosphorylation Hot Spots within the HSP70 Domain Family
(A) A cutoff of p value < 0.005 (dotted line) was used to identify two regions that are significantly enriched for phosphosites in the HSP70 family.
(B) Phosphorylation hot spot mutants do not complement SSA1 functions as measured by growth in liquid media.
compared to WT Ssa1 by western blot.
(D) The association of the indicated SSA1 mutants with polysomes was examined by immunoblot analysis. Ribosomal profiles (top) were determined by OD
254 nm and confirmed by immunoblot analysis of the ribosomal proteins Rpl3p.
(E) Recovery of luciferase activity is expressed as a percentage of activity before heat treatment and is an average of 2 experiments.
(F) Percentage of cells with multiple and single ubc9-2-GFP aggregates after heat shocked at 37?C for 30 min. Errors bars quantify the standard deviations of, at
least, four technical replicates.
See also Figure S4.
Cell 150, 413–425, July 20, 2012 ª2012 Elsevier Inc. 421
spot phosphorylation sites are important for SSA1 function. In
addition, we performed serial spot dilution assays on the
displayed a growth defect under heat shock conditions that is
not observed when the WT Ssa1 is expressed or when a control
mutation T326A is introduced (Figure S4A). Importantly, the
protein abundance of the mutants was comparable to WT
Ssa1 (Figure 6C). However, no dramatic differences were
observed between the alanine or phosphomimetic mutants sug-
gesting that either the phosphorylation cycle is important for the
function of Ssa1 or alternatively, the phosphorylation state
of region 1 and region 2 could distinctly affect the function of
Ssa1 in the multiple distinct cellular tasks required for cell
growth. To obtain further insight we explored the effect of the
phosphorylation mutants on a small subset of assays reporting
on different Hsp70/SSA functions, namely: association with
polysomes (Figure 6D); refolding of firefly luciferase following
heat stress (Figure 6E) and prevention of misfolded protein
aggregation (Figure 6F). We examined the association of Ssa1
with polysomes by fractionation extracts on 7%–47% sucrose
gradients followed by western blot analysis for the presence of
WT or mutant Ssa1 as well as the ribosomal protein Rpl3 (Fig-
ure 6D). Both Ssa1 WT and the control mutant associated with
polysomes as previously reported (Albane `se et al., 2006).
However, Ssa1 mutated in Region 2 were defective in binding
to polysomes. The Ssa1 mutants of Region 1 show similar to
wild-type association with polysomes (Figure S4B). In addition
we observed that a single phosphomimetic mutation in region
2 (S495D) is sufficient to disrupt the association with polysomes
(Figure S4C), a phenotype not observed for the equivalent
alanine mutant (S495A) (Figure S4C). These data suggest that
the regulatory hot spot we identified in Region 2 may be involved
the translational cofactors Sis1 and Pab1 to the ribosome
(Horton et al., 2001).
Hsp70 also assists the refolding of heat-denatured polypep-
tides, a function that can be monitored by following the recovery
of luciferase enzymatic activity following heat stress (Experi-
mental Procedures). As expected, the cells containing wild-type
Ssa1 showedrobust recoveryof luciferaseactivity,whereas little
recovery was observed in the SSA-defective cells transformed
ylation incompetent variant in Region 1 (i.e., the nucleotide
binding site) was significantly impaired in assisting the recovery
of stress-denatured luciferase when compared to WT or a phos-
of the phosphorylation mutant to prevent the aggregation of
conditions (Figure 6F). Similar to what was observed for the
recovery of stress-denatured luciferase, the phosphorylation
incompetent Ssa1 mutant (i.e., alanine mutant) in Region 1 was
impaired in the ability to prevent aggregate formation (Figure 6F)
phomimetic mutant appears to have a similar to WT luciferase
recovery and aggregation prevention capacity.
Taken together, these results indicate that the two conserved
phosphorylation hot spots in Hsp70 are functionally relevant,
Because phosphorylation of Hsp70 had been previously ob-
served, our approach provides evidence that such phosphoryla-
tion is important for Hsp70 regulation. Given the many distinct
functions of Hsp70 during the life cycle of the cell our results
open the way to future studies dissecting the precise contribu-
tion of regulation of each region to overall Hsp70 function as
well as the modulation of its activity under various growth
We have compiled a resource of nearly 200,000 PTMs covering
11 eukaryotic species and developed approaches to annotate
PTMs that are more likely to cross-regulate each other or to
regulate protein-interfaces or domain activity. To make this
resource easily available to others, these data are available
through a website (http://ptmfunc.com) that contains known
PTMs, spectral counts, information on conditional regulation,
conservation and putative functional assignments. Using these
methods, we have identified a phosphorylation site within Skp1
that is likely to reversibly alter the binding affinity of Skp1 to
Met30. Given the position in the crystal structure, it is possible
that the phosphorylation of Skp1 at S162 acts by sterically or
electrostatically repulsing Met30. However, given that Skp1
interacts with other F-box proteins, it is also possible that the
phosphorylation increases the affinity for another protein and
titrates Skp1 away from Met30. Based on the assumption that
conserved sites are more likely to be functionally relevant, we
have identified regions within domain families that show signifi-
cant enrichment of PTM sites across the 11 species analyzed
here (regulatory hot spots). We have experimentally character-
ized two such regions within the HSP70 chaperone domain
family showing that they affect Hsp70 function and provide addi-
tional examples for future studies. Putative functional annota-
tions for 8,776 phosphosites from these 11 species are available
through our website.
Besides the functional prioritization, this resource can also be
used to study the evolution of posttranslational regulation. Past
work on the evolution of cellular interaction networks has shown
that, whereas protein complex membership diverges slowly and
mostly through subunit duplication (Pereira-Leal et al., 2007; van
Dam and Snel, 2008), cellular interactions of broad specificity
such as protein interactions mediated by small peptide ‘‘linear
motifs’’ diverge quickly (Beltrao and Serrano, 2007; Neduva
andRussell, 2005).Our analysisof 11 partial phosphoproteomes
further validates previous observations regarding the fast diver-
gence of kinase-substrate interactions (Landry et al., 2009). In
addition, we show that lysine modifications are equally poorly
constrained when compared to an equivalent random sample
of lysine acceptor residues.
Although regulatory interactions diverge quickly, it is possible
for these changes to be neutral in respect to phenotype, much in
the same way that mutations within open reading frames can be
neutral to the coding sequence. Examples include the conserva-
422 Cell 150, 413–425, July 20, 2012 ª2012 Elsevier Inc.
complex and ribosome transcriptional regulation, despite the
divergence of the underlying interactions in some fungal species
(Lavoie et al., 2010; Moses et al., 2007; Tsong et al., 2006). The
existence of equivalent or ‘‘neutral networks’’ as described by
Andreas Wagner, among others, may in fact be important for
the exploration of novel phenotypes (Wagner, 2008). We
observed that PTMs that are known to be regulated in vivo or
are predicted to have a function are more likely to be conserved
across species than average sites. One explanation for these
results would be that there is a significant fraction of PTMs that
serve no functional role but result simply from the constant
evolutionary turnover of regulatory interactions. We also ob-
served several examples of conservation of the phosphorylation
phosphosite position. Although we cannot rule out that this
observation is due to incomplete phosphoproteome coverage,
it provides specific examples of possible neutral variation of
PTM regulation for future studies.
the long term, have practical biomedical applications. This is
particularly relevant for the study of disease since it has been
recently shown that disease causing mutations are significantly
associated with PTM sites (Li et al., 2010). This resource can
aid in the understanding of how mutations associated with
PTMs result in disease.
Posttranslational Modification Sites and Computational Methods
All of the sites compiled are provided in a searchable website (http://ptmfunc.
com). Protein sequences, protein identifiers and ortholog assignments
were obtained from the Inparanoid database (http://inparanoid.sbc.su.se,
version 7). For the comparative analysis we considered only 1-to-1 orthologs
with Inparanoid confidence scores greater than 90%. The total number of
human to species ortholog pairs used in this studied are listed in Table S5.
Protein sequence alignments were done with MUSCLE version 3.6 (Edgar,
2004). Additional information on the computational methods is provided in
the Extended Experimental Procedures.
Immunoprecipitation, Mass Spectrometry, and Ubiquitylation Site
S. cerevisiae Sub592 (containing a HisTag modified ubiquitin) and Sub62 were
grown separately in YPD and harvested during mid-log phase (OD 600 ?1.0).
Protein extract from Sub592 cells (?40 mg) was enriched for ubiquitylated
proteins via HisTag. Sub62 proteins and half of the ubiquitin-enriched
Sub592 protein were digested overnight with trypsin, while the remaining
half was digested with ArgC. After enzymatic digestion, all three samples
were desalted and enriched for diGly containing peptides using a polyclonal
antibody as previously described (Cell Signaling, Technology, Danvers, MA)
(Kim et al., 2011) and analyzed in an Orbitrap Velos mass spectrometer.
Raw files were searched with Sequest against the target-decoy S. cerevisiae
protein sequence database. Peptide spectral matches were filtered to a 1%
false-discovery rate at the peptide and protein level and diGly sites were
localized using a version of the Ascore algorithm that can accept any post-
translational modification (Ascore > 13) (Beausoleil et al., 2006). See Extended
Experimental Procedures for more details.
Protein Complementation Assays and Co-IP
Skp1 and Met30 were fused to fragments F1 and F2, respectively, of a
split cytosine deaminase by gap-repair cloning. Point mutants were con-
structed using PCR with site directed oligonucleotides. Protein complementa-
tion was assayed as previously described (Ear and Michnick, 2009; Michnick
et al., 2010). For Co-IP experiments, yeast cells expressing endogenous Myc-
tagged Met30 were transformed with a plasmid expressing a Flag-tagged
SKP1 S162A or S162D under control of the Gal promoter and selected
for Leucine auxotrophy. Detailed information is available in Extended Experi-
DNA, Yeast Strains
S. cerevisiae strains used were as follows: the ssa1 temperature sensitive
strain (mat alpha leu2 trp1 ura3 ade2 his3 lys2, ssa1-45BKD, ssa2::LEU2,
ssa3::TRP1, ssa4::LYS2) and was a gift from Betty Craig. The Dssa1::KanMX4
Dssa2::NAT was generated by direct replacement of the SSA2 coding region
with the NatMX4 cassette in the single deletion strain Dssa1::KanMX4 and
confirmed by PCR.
Drop Test Assay
Cells were grown overnight in selective medium and then diluted to
OD 600 nm O.4. Cells were grown for another 3 hr and then diluted to
OD 600 nm of 0.1. This sample was then subjected to 10-fold serial dilutions.
Ten microliters of each dilution was then spotted onto –URA plates and
allowed to grow at 30?C, 33?C, and 37?C for 2 days.
Lysate Preparation and Ribosome Fractionation
Yeast (200 ml) in exponential growth was treated with 100 mg/ml of
cycloheximide, harvested, washed with cold water, resuspended, and
frozen as drops in liquid nitrogen. The cell lysate was loaded on a 12 ml
7%–47% sucrose gradient and centrifuged for 150 min at 39,000 rpm at
4?C. Fractions were collected using a UA/6 detector. The fractions were
TCA precipitated and separated by SDS-PAGE and subjected to immunoblot
analysis. The detailed protocol is available in Extended Experimental
Luciferase In Vivo Refolding Assay
The ssa1-45 ts cells were transformed with firefly luciferase and a
plasmid driving the expression of the wild-type SSA1 or the phos-
phomutants.After growth at30?C,thecells wereshifted to44?Cfor 1hr,which
causes the heat-induced denaturation of luciferase. Cycloheximide was
added to 10 mg/ml 15 min before the end of the heat shock to prevent further
expression of luciferase. Cells were then transferred to 30?C to recover. At
different time points during the recovery, aliquots were taken, centrifuged,
and frozen in liquid nitrogen. The luciferase activity was measured and
recovery is expressed as a percentage of the activity before heat shock
Microscopy and Aggregation Assay
The ssa1-45 ts strain was transformed with the different SSA1 mutants as well
as with the Gal-Ubc9-2-GFP construct. Cells were grown overnight at 30?C
and then diluted to OD 600 nm 0.3 and induced with 2% galactose for 6 hr.
Cells were then shifted to 37?C for 30 min to induce the misfolding of Ubc9-
2-GFP. The formation of Ubc9-2-GFP puncta was then examined by fluores-
Supplemental Information includes Extended Experimental Procedures,
four figures, and five tables and can be found with this article online at
We thank Stephen Michnick and Peter Kaiser for strains and plasmids, Betty
Craig for the ssa1-45 ts cells, and Ste ´phanie Escusa and Jonathan Warner
for reagents. This work was supported by grants from the National In-
stitutesofHealth (AI090935, GM082250,
GM081879 [N.J.K.]); GM55040, GM062583, GM081879, PN2 EY016546
[W.A.L.]; GM56433 [J.F.]; DP5 OD009180 [J.S.F.]; and RR01614 [A.B.]), the
Howard Hughes Medical Institute (W.A.L.), and the Packard Foundation
Cell 150, 413–425, July 20, 2012 ª2012 Elsevier Inc. 423
(W.A.L.). P.B. is supported by the Human Frontier Science Program, J.S.F. is
a QB3@UCSF Fellow, and N.J.K. is a Searle Scholar and a Keck Young
Received: August 19, 2011
Revised: March 21, 2012
Accepted: May 18, 2012
Published: July 19, 2012
Aghajan, M., Jonai, N., Flick, K., Fu, F., Luo, M., Cai, X., Ouni, I., Pierce, N.,
Tang, X., Lomenick, B., et al. (2010). Chemical genetics screen for enhancers
of rapamycin identifies a specific inhibitor of an SCF family E3 ubiquitin ligase.
Nat. Biotechnol. 28, 738–742.
Albane `se, V., Yam, A.Y., Baughman, J., Parnot, C., and Frydman, J. (2006).
Systems analyses reveal two chaperone networks with distinct functions in
eukaryotic cells. Cell 124, 75–88.
Ali, M.M., Roe, S.M., Vaughan, C.K., Meyer, P., Panaretou, B., Piper, P.W.,
Prodromou, C., and Pearl, L.H. (2006). Crystal structure of an Hsp90-
nucleotide-p23/Sba1 closed chaperone complex. Nature 440, 1013–1017.
Ba, A.N., and Moses, A.M. (2010). Evolution of characterized phosphorylation
sites in budding yeast. Mol. Biol. Evol. 27, 2027–2037.
Beausoleil, S.A., Ville ´n, J., Gerber, S.A., Rush, J., and Gygi, S.P. (2006). A
probability-based approach for high-throughput protein phosphorylation
analysis and site localization. Nat. Biotechnol. 24, 1285–1292.
Beltrao, P., and Serrano, L. (2007). Specificity and evolvability in eukaryotic
protein interaction networks. PLoS Comput. Biol. 3, e25.
Beltrao, P., Trinidad, J.C., Fiedler, D., Roguev, A., Lim, W.A., Shokat, K.M.,
Burlingame, A.L., and Krogan, N.J. (2009). Evolution of phosphoregulation:
comparison of phosphorylation patterns across yeast species. PLoS Biol. 7,
Carroll, S.B. (2005). Evolution at two levels: on genes and form. PLoS Biol. 3,
Choudhary, C., and Mann, M. (2010). Decoding signalling networks by mass
spectrometry-based proteomics. Nat. Rev. Mol. Cell Biol. 11, 427–439.
Choudhary, C., Kumar, C., Gnad, F., Nielsen, M.L., Rehman, M., Walther, T.C.,
and co-regulates major cellular functions. Science 325, 834–840.
DerMardirossian, C., Schnelzer, A., and Bokoch, G.M. (2004). Phosphoryla-
tion of RhoGDI by Pak1 mediates dissociation of Rac GTPase. Mol. Cell 15,
Ear, P.H., and Michnick,S.W. (2009). Agenerallife-death selection strategy for
dissecting protein functions. Nat. Methods 6, 813–816.
Edgar,R.C. (2004). MUSCLE: multiple sequence alignment with high accuracy
and high throughput. Nucleic Acids Res. 32, 1792–1797.
Emanuele, M.J., Elia, A.E., Xu, Q., Thoma, C.R., Izhar, L., Leng, Y., Guo, A.,
Chen, Y.N., Rush, J., Hsu, P.W., et al. (2011). Global identification of modular
cullin-RING ligase substrates. Cell 147, 459–474.
Este `ve, P.O., Chang, Y., Samaranayake, M., Upadhyay, A.K., Horton, J.R.,
Feehery, G.R., Cheng, X., and Pradhan, S. (2011). A methylation and phos-
phorylation switch between an adjacent lysine and serine determines human
DNMT1 stability. Nat. Struct. Mol. Biol. 18, 42–48.
Groll, M., Bajorek, M., Ko ¨hler, A., Moroder, L., Rubin, D.M., Huber, R., Glick-
man, M.H., and Finley, D. (2000). A gated channel into the proteasome core
particle. Nat. Struct. Biol. 7, 1062–1067.
Holt, L.J., Tuch, B.B., Ville ´n, J., Johnson, A.D., Gygi, S.P., and Morgan, D.O.
(2009). Global analysis of Cdk1 substrate phosphorylation sites provides
insights into evolution. Science 325, 1682–1686.
Horton,L.E., James, P.,Craig,E.A.,and Hensold, J.O.(2001). Theyeasthsp70
homologue Ssa is required for translation and interacts with Sis1 and Pab1 on
translating ribosomes. J. Biol. Chem. 276, 14426–14433.
Huber, A., Bodenmiller, B., Uotila, A., Stahl, M., Wanka, S., Gerrits, B., Aeber-
sold, R., and Loewith, R. (2009). Characterization of the rapamycin-sensitive
phosphoproteome reveals that Sch9 is a central coordinator of protein
synthesis. Genes Dev. 23, 1929–1943.
Hunter, T. (2007). The age of crosstalk: phosphorylation, ubiquitination, and
beyond. Mol. Cell 28, 730–738.
Jonkers, W., and Rep, M. (2009). Lessons from fungal F-box proteins. Eukar-
yot. Cell 8, 677–695.
Kaiser, P., Su, N.Y., Yen, J.L., Ouni, I., and Flick, K. (2006). The yeast ubiquitin
division. Cell Div. 1, 16.
Kampinga, H.H., and Craig, E.A. (2010). The HSP70 chaperone machinery: J
proteins as drivers of functional specificity. Nat. Rev. Mol. Cell Biol. 11,
Kim, W., Bennett, E.J., Huttlin, E.L., Guo, A.,Li, J., Possemato, A., Sowa, M.E.,
Rad, R., Rush, J., Comb, M.J., et al. (2011). Systematic and quantitative
assessment of the ubiquitin-modified proteome. Mol. Cell 44, 325–340.
Landry, C.R., Levy, E.D., and Michnick, S.W. (2009). Weak functional
constraints on phosphoproteomes. Trends Genet. 25, 193–197.
Lavoie, H., Hogues, H., Mallick, J., Sellam, A., Nantel, A., and Whiteway, M.
(2010). Evolutionary tinkering with conserved components of a transcriptional
regulatory network. PLoS Biol. 8, e1000329.
Li, S., Iakoucheva, L.M., Mooney, S.D., and Radivojac, P. (2010). Loss of post-
translational modification sites in disease. Pac. Symp. Biocomput., 337–347.
Ludwig, M.Z., Bergman, C., Patel, N.H., and Kreitman, M. (2000). Evidence for
stabilizing selection in a eukaryotic enhancer element. Nature 403, 564–567.
Michnick, S.W., Ear, P.H., Landry, C., Malleshaiah, M.K., and Messier, V.
(2010). A toolkit of protein-fragment complementation assays for studying
and dissecting large-scale and dynamic protein-protein interactions in living
cells. Methods Enzymol. 470, 335–368.
Moses, A.M., and Landry, C.R. (2010). Moving from transcriptional to phos-
pho-evolution: generalizing regulatory evolution? Trends Genet. 26, 462–467.
Moses, A.M., Liku, M.E., Li, J.J., and Durbin, R. (2007). Regulatory evolution in
proteins by turnover and lineage-specific changes of cyclin-dependent kinase
consensus sites. Proc. Natl. Acad. Sci. USA 104, 17713–17718.
Narayanan, A., and Jacobson, M.P. (2009). Computational studies of protein
regulation by post-translational phosphorylation. Curr. Opin. Struct. Biol. 19,
Neduva, V., and Russell, R.B. (2005). Linear motifs: evolutionary interaction
switches. FEBS Lett. 579, 3342–3345.
Nolen, B., Taylor, S., and Ghosh, G. (2004). Regulation of protein kinases;
controlling activity through activation segment conformation. Mol. Cell 15,
Paolinelli, R., Mendoza-Maldonado, R., Cereseto, A., and Giacca, M. (2009).
Acetylation by GCN5 regulates CDC6 phosphorylation in the S phase of the
cell cycle. Nat. Struct. Mol. Biol. 16, 412–420.
of protein complexes by duplication of homomeric interactions. Genome Biol.
Petroski, M.D., and Deshaies, R.J. (2005). Function and regulation of cullin-
RING ubiquitin ligases. Nat. Rev. Mol. Cell Biol. 6, 9–20.
Soroka, J., Wandinger, S.K., Ma ¨usbacher, N., Schreiber, T., Richter, K., Daub,
H., and Buchner, J. (2012). Conformational switching of the molecular chap-
erone Hsp90 via regulated phosphorylation. Mol. Cell 45, 517–528.
Soufi, B., Kelstrup, C.D., Stoehr, G., Fro ¨hlich, F., Walther, T.C., and Olsen, J.V.
(2009). Global analysis of the yeast osmotic stress response by quantitative
proteomics. Mol. Biosyst. 5, 1337–1346.
Stark, C., Su, T.C., Breitkreutz, A., Lourenco, P., Dahabieh, M., Breitkreutz,
B.J., Tyers, M., and Sadowski, I. (2010). PhosphoGRID: a database of
424 Cell 150, 413–425, July 20, 2012 ª2012 Elsevier Inc.
experimentally verified in vivo protein phosphorylation sites from the budding
yeast Saccharomyces cerevisiae. Database (Oxford) 2010, bap026.
Stein, A., Ce ´ol, A., and Aloy, P. (2011). 3did: identification and classification of
domain-based interactions of known three-dimensional structure. Nucleic
Acids Res. 39 (Database issue), D718–D723.
Strahl, B.D., and Allis, C.D. (2000). The language of covalent histone modifica-
tions. Nature 403, 41–45.
Taipale, M., Jarosz, D.F., and Lindquist, S. (2010). HSP90 at the hub of protein
homeostasis: emerging mechanistic insights. Nat. Rev. Mol. Cell Biol. 11,
Tan, C.S., Bodenmiller, B., Pasculescu, A., Jovanovic, M., Hengartner, M.O.,
Jørgensen, C., Bader, G.D., Aebersold, R., Pawson, T., and Linding, R.
(2009). Comparative analysis reveals conserved protein phosphorylation
networks implicated in multiple diseases. Sci. Signal. 2, ra39.
Tsong, A.E., Tuch, B.B., Li, H., and Johnson, A.D. (2006). Evolution of alterna-
tive transcriptional circuits with identical logic. Nature 443, 415–420.
van Dam, T.J., and Snel, B. (2008). Protein complex evolution does not involve
extensive network rewiring. PLoS Comput. Biol. 4, e1000132.
Wagner, A. (2008). Neutralism and selectionism: a network-based recon-
ciliation. Nat. Rev. Genet. 9, 965–974.
Wagner, S.A., Beli, P., Weinert, B.T., Nielsen, M.L., Cox, J., Mann, M., and
Choudhary, C. (2011). A proteome-wide, quantitative survey of in vivo ubiqui-
tylation sites reveals widespread regulatory roles. Mol. Cell. Proteomics 10,
Wandinger, S.K., Suhre, M.H., Wegele, H., and Buchner, J. (2006). The phos-
phatase Ppt1 is a dedicated regulator of the molecular chaperone Hsp90.
EMBO J. 25, 367–376.
Weinert, B.T., Wagner, S.A., Horn, H., Henriksen, P., Liu, W.R., Olsen, J.V.,
Jensen, L.J., and Choudhary, C. (2011). Proteome-wide mapping of the
Drosophila acetylome demonstrates a high degree of conservation of lysine
acetylation. Sci. Signal. 4, ra48.
Yang, X.J., and Gre ´goire, S. (2006). A recurrent phospho-sumoyl switch in
transcriptional repression and beyond. Mol. Cell 23, 779–786.
Cell 150, 413–425, July 20, 2012 ª2012 Elsevier Inc. 425