Corrections

ECOLOGYCorrection for “Recovery of a top predator mediates negativeeutrophic effects on seagrass,” by Brent B. Hughes, Ron Eby, EricVan Dyke, M. Tim Tinker, Corina I. Marks, Kenneth S. Johnson,and Kerstin Wasson, which appeared in issue 38, September 17,2013, of Proc Natl Acad Sci USA (110:15313–15318; first publishedAugust 27, 2013; 10.1073/pnas.1302805110).

The authors note that Fig. 2 appeared incorrectly. Theauthors note that they “unintentionally labeled Fig. 2C ‘Grazerbiomass (g DW) *shoot (cm)−1’ instead of ‘Grazer biomass(mg DW) *shoot (cm)−1.’ ” The corrected figure and its leg-end appear below. This error does not affect the conclusionsof the article.

www.pnas.org/cgi/doi/10.1073/pnas.1401578111

Fig. 2. (A) Interaction web of top-down and bottom-up effects in the eelgrass study system. The top predator is the sea otter (E. lutris), the mesopredators arecrabs (Cancer spp. and Pugettia producta), the epiphyte mesograzers are primarily an isopod (I. resecata) and a sea slug (P. taylori), and algal epiphyte competitorsof eelgrass primarily consist of chain-forming diatoms, and the red alga Smithora naiadum. Solid arrows indicate direct effects, dashed arrows indicate indirecteffects, and the plus and minus symbols indicate positive and/or negative effects on trophic guilds and eelgrass condition. C, competitive interaction; T, trophicinteraction. (Original artwork by A. C. Hughes.) (B–E) Survey results testing for the effects of sea otter density on eelgrass bed community properties (Tables S2and S3). Elkhorn Slough (sea otters present and high nutrients) eelgrass beds (n = 4) are coded in red, and the Tomales Bay reference site (no sea otters, lownutrients) beds (n = 4) are coded in blue. (B) Crab biomass and size structure of two species of Cancer crabs; (C) grazer biomass per shoot and large grazer density;(D) algal epiphyte loading; and (E) aboveground and belowground eelgrass biomass. DW, dry weight; FW, fresh weight.

3644–3645 | PNAS | March 4, 2014 | vol. 111 | no. 9 www.pnas.org

Dow

nloa

ded

by g

uest

on

Dec

embe

r 12

, 202

0 D

ownl

oade

d by

gue

st o

n D

ecem

ber

12, 2

020

Dow

nloa

ded

by g

uest

on

Dec

embe

r 12

, 202

0 D

ownl

oade

d by

gue

st o

n D

ecem

ber

12, 2

020

Dow

nloa

ded

by g

uest

on

Dec

embe

r 12

, 202

0 D

ownl

oade

d by

gue

st o

n D

ecem

ber

12, 2

020

Dow

nloa

ded

by g

uest

on

Dec

embe

r 12

, 202

0 D

ownl

oade

d by

gue

st o

n D

ecem

ber

12, 2

020

MICROBIOLOGYCorrection for “Programmed Allee effect in bacteria causes atradeoff between population spread and survival,” by Robert Smith,Cheemeng Tan, Jaydeep K. Srimani, Anand Pai, Katherine A.Riccione, Hao Song, and Lingchong You, which appeared inissue 5, February 4, 2014, of Proc Natl Acad Sci USA (111:1969–1974; first published January 21, 2014; 10.1073/pnas.1315954111).The authors note that the following statement should be added

to the Acknowledgments: “This work was partially supported bya Society in Science–Branco Weiss Fellowship (to C.T.).”

www.pnas.org/cgi/doi/10.1073/pnas.1401930111

SYSTEMS BIOLOGYCorrection for “Analysis of proteome dynamics in the mouse brain,”by John C. Price, Shenheng Guan, Alma Burlingame, Stanley B.Prusiner, and Sina Ghaemmaghami, which appeared in issue 32,August 10, 2010, of Proc Natl Acad Sci USA (107:14508–14513; firstpublished August 10, 2010; 10.1073/pnas.1006551107).The authors note that the following grant should be added to

the Acknowledgments: “NIH Grant AG002132.”

www.pnas.org/cgi/doi/10.1073/pnas.1401576111

NEUROSCIENCECorrection for “Mapping the receptor site for α-scorpion toxinson a Na+ channel voltage sensor,” by Jinti Wang, Vladimir Yarov-Yarovoy, Roy Kahn, Dalia Gordon, Michael Gurevitz, ToddScheuer, and William A. Catterall, which appeared in issue 37,September 13, 2011, of Proc Natl Acad Sci USA (108:15426–15431;first published August 29, 2011; 10.1073/pnas.1112320108).The authors note that the following statement should be added

as a new Acknowledgments section: “This work was supported byNational Institutes of Health Research Grants R01 NS015751 (toW.A.C.) and U01 NS058039 (to W.A.C. and M.G.), by ResearchGrant IS-4313-10 from the US-Israel Binational AgriculturalResearch and Development Foundation (to M.G. andW.A.C.), andby Israeli Science Foundation Grant 107/08 (to M.G. and D.G.).”

www.pnas.org/cgi/doi/10.1073/pnas.1401985111

PNAS | March 4, 2014 | vol. 111 | no. 9 | 3645

CORR

ECTIONS

Dow

nloa

ded

by g

uest

on

Dec

embe

r 12

, 202

0

Analysis of proteome dynamics in the mouse brainJohn C. Pricea, Shenheng Guanb, Alma Burlingameb, Stanley B. Prusinera,c, and Sina Ghaemmaghamia,c,1

aInstitute for Neurodegenerative Diseases and Departments of bPharmaceutical Chemistry and cNeurology, University of California, San Francisco, CA 94143

Contributed by Stanley B. Prusiner, May 13, 2010 (sent for review April 20, 2010)

Advances in systems biology have allowed for global analyses ofmRNA and protein expression, but large-scale studies of proteindynamics and turnover have not been conducted in vivo. Proteinturnover is an important metabolic and regulatory mechanism inestablishing proteome homeostasis, impacting many physiologicaland pathological processes. Here, we have used organism-wideisotopic labeling to measure the turnover rates of ∼2,500 proteinsin multiple mouse tissues, spanning four orders of magnitude.Through comparison of the brain with the liver and blood, weshow that within the respective tissues, proteins performing similarfunctions often have similar turnover rates. Proteins in the brainhave significantly slower turnover (average lifetime of 9.0 d) com-pared with those of the liver (3.0 d) and blood (3.5 d). Within someorganelles (such as mitochondria), proteins have a narrow range oflifetimes, suggesting a synchronized turnover mechanism. Proteinsubunits within complexes of variable composition have a widerange of lifetimes, whereas those within well-defined complexesturn over in a coordinated manner. Together, the data representthe most comprehensive in vivo analysis of mammalian proteometurnover to date. The developed methodology can be adapted toassess in vivo proteome homeostasis in any model organism thatwill tolerate a labeled diet and may be particularly useful in theanalysis of neurodegenerative diseases in vivo.

isotopic labeling | protein | turnover | degradation | in vivo

Protein molecules are in dynamic equilibrium in vivo: they arecontinuously synthesized and degraded during the lifetime of

the organism (1, 2). The turnover rate of proteins can varyfrom minutes to years, often conforming to their biologicalfunctions (3, 4). The constant renewal of the protein population isan energy-intensive process, yet it allows the cell to rapidlymodulate protein levels in response to changes in the environ-ment (5, 6). Proper proteome dynamics are critical to normaldevelopment and maintenance of health (7, 8). For example, thedysregulation of protein turnover has been implicated in the agingprocess (9), increased degradation of the cystic fibrosis trans-membrane conductance regulator (CFTR) chloride channel isa primary cause of cystic fibrosis (10), and the inability to clearprotein aggregates leads to pathogenic accumulations in Alz-heimer’s, Parkinson’s, Creutzfeldt-Jakob, and other age-relateddiseases (11).The turnover rate of a protein is established by its relative rates

of synthesis and catabolism. Thus, the lifetime of a protein isinfluenced by a number of regulated processes at the level of thecell (transcription, translation, proteolysis, and autophagy) andtissue (development and regeneration) as well as its biochemicalproperties (structural stability, hydrophobicity, and sequencemotifs) (1, 12–15). The ability to measure turnover rates ona proteome-wide scale can help elucidate the interplay betweenthese regulated processes and identify novel variables that playa role in proteome homeostasis. It can also identify proteins whosedysregulation influences or results from pathological processes.Traditionally, protein turnover has been studied by measuring

the incorporation of radioactive, tracer amino acids into proteinsor bulk tissues (16–20). The advent of modern proteomics hasenabled scientists to use mass spectrometry to detect the in-corporation of stable isotopes into proteins (21, 22). In measuringturnover rates, the latter approach offers three potential advan-

tages. First, a large number of proteins can be simultaneouslyanalyzed within a biological sample. Second, labeling can be con-ducted with fully (100%) labeled amino acids. Therefore, turnoverrates can be assessed by the single-step kinetics of amino acid in-corporation without conducting complex pulse-chase analyses.Third, whereas tracer methods can only measure the total in-corporation of label, mass spectrometry can analyze the populationdistribution of partially labeled species for a given protein. Thus,turnover rates can be measured in instances in which upstreamprocesses, such as label uptake into tissue, are rate limiting (7, 8).Recent studies have shown that rats can be isotopically labeled

using a diet source supplemented with 15N-enriched, blue-greenalgae (Spirulina platensis) (23). Here, we have used this approachto measure the in vivo turnover kinetics of proteins in the brainsof wild-type, inbred mice (FVB) and provide a comparison ofthese dynamics to the blood and liver proteomes.

ResultsLarge-Scale Production of Ubiquitously 15N-Labeled Algae. To obtainthe algae necessary for long-term labeling studies, we constructeda closed-loop bioreactor based on a bubble-lift circulator (Fig.S1A and SI Text). Using 15N-enriched NaNO3 as the sole nitrogensource, we produced near-uniform 15N-labeled Spirulina at a yieldof 3 g of algae per liter of broth. Mass spectral analysis of theSpirulina indicated that the final isotopic enrichment was >99.5%15N (Dataset S1). The labeled feed required for the entirety of ourstudies was supplied by ∼60 L of Spirulina culture.

Maintaining Mice on Algae Diet. Mice fed a diet based on Spirulinahad no subchronic toxicities after 13 wk of continuous feeding(24). Our diet formulation was similar to published protocols(21, 23). Mice were examined daily for general health and at leasttwice a week for body mass. Body mass fluctuated daily (Fig.S1B), but no mice lost or gained a significant fraction (>20%)of their body mass subsequent to the introduction of the algaediet. At nine time points, mice were sacrificed and the labelingkinetics of proteins in the brain, liver, and blood were analyzed(Fig. 1, Methods).

Peptide Identification. The computational analysis of the data wasconducted using a novel suite of computer scripts (SI Text). Allprocessed data, including the peptides, proteins, and their mea-sured kinetic parameters, are included inDataset S2. All raw data,including LC elution profiles, survey scans, and MS/MS spectra,are available upon request. Only peptides assigned with a greaterthan 95% confidence in a Protein Prospector reverse databasesearch (25) were used in subsequent analysis. For samples col-lected after the initiation of labeling, 15N-labeled peptides wereidentified on the basis of (i) molecular weight region of the gel;(ii) the expected LC retention time, as observed in the 0-d sample;

Author contributions: J.C.P., A.B., S.B.P., and S. Ghaemmaghami designed research; J.C.P.and S. Guan performed research; S. Guan and A.B. contributed new reagents/analytictools; J.C.P., S. Guan, S.B.P., and S. Ghaemmaghami analyzed data; and J.C.P., S.B.P.,and S. Ghaemmaghami wrote the paper.

The authors declare no conflict of interest.1To whom correspondence should be addressed. E-mail: sina@ind.ucsf.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1006551107/-/DCSupplemental.

14508–14513 | PNAS | August 10, 2010 | vol. 107 | no. 32 www.pnas.org/cgi/doi/10.1073/pnas.1006551107

and (iii) expected mass distribution based on the elemental com-position of the peptide, natural isotopic distribution of C, O, S, andH atoms, and 14N/15N ratio ranging from 0.0037 to 0.995 (thenatural and algal abundance of 15N, respectively). For 4,619 brainpeptides, 7,226 liver peptides and 1,968 blood peptides, we wereable to identify and quantify the mass distribution at all nine timepoints with high statistical confidence (Table 1). Data obtainedfrom duplicate animals at any given time point showed minimalvariation (Fig. S2).As an example of a typical kinetic labeling pattern, the time-

dependent mass increase of the brain-derived Cofilin-1 peptide(NIILEEGKEILVGDVGQTVDDPYTTFVK) is shown in Fig.1C. Two independent parameters of 15N incorporation are evi-dent. First, there is a time-dependent increase in the fraction ofthe area of the peptide peaks that fall outside the expected un-labeled mass distribution (0 d). Second, the centroid mass of thelabeled population (Fig.1C, arrowheads) increases over time. Werefer to these two measurements, normalized with respect to themass difference between unlabeled and fully labeled samples,as “labeled population” and “mass shift,” respectively. For theCofilin-1 peptide, a 15N-labeled population is clearly visible insamples obtained after 1 d of labeling (Fig. 1C). The labeledprotein population is well resolved from the natural isotopic dis-tribution even at the earliest time points, negating the need for thedeconvolution of the natural and isotopic mass distributions.Plots of the labeled population and mass shift for Cofilin-1 and

several other peptides are shown in Fig. 2. Several kinetic trendsare evident in the data. First, the increase in mass shift is biphasic,with a rapid initial burst followed by an extended (slow) phase.Second, mass shift kinetics are similar among peptides extractedfrom the same tissue but different between the three analyzedtissues. Specifically, in the brain, the initial fast phase has a loweramplitude and the second phase has a slower rate in comparisonwith liver and blood. Third, labeled population has an initial lagphase of ∼1 d, followed by a single exponential increase. Andlastly, the kinetics of labeled population is highly variable amongthe analyzed peptides.

Calculating Protein Turnover. Historically, various simplified mod-els have been used to interpret the kinetics of protein turnover intracer labeling experiments (26–28). These models attempt toreconcile the observed kinetic trends of label uptake with reactionmechanisms consisting of kinetic influx and efflux of theoreticalpools of amino acids and proteins (17, 26). Here, we use a three-pool model (Fig. 1B) to explain the observed kinetic trends (ex-emplified by the Cofilin-1 peptide in Fig. 1C and the peptidesplotted in Fig. 2). The global (organism-wide) pool of amino acidscan be supplied by two sources: external diet and internal me-tabolism/catabolism. The global pool provides amino acids for

Table 1. Numbers of peptides and proteins analyzed in thisstudy

Brain Liver Blood

Detected peptides (0 d) 14,971 14,653 4,670Peptides used in protein analysis* 4,619 7,226 1,968Total proteins analyzed 1,010 1,122 3341 peptide† 379 313 1072–4 peptides† 353 343 1115–10 peptides† 165 244 71>10 peptides† 113 222 45Unique proteins (total)‡ 1,716

*Only peptides detected in all nine time points were used for proteinanalysis.

†Number of proteins for which the stated number of peptides were analyzedand averaged.

‡Number of unique proteins in the combined analysis of the three tissues.

Fig. 1. Protocol, model, and analysis of global turnover rates. (A) Exper-imental protocol. 15N inorganic salts are used to make broth for cultures ofSpirulina. Dried 15N-labeled algae is used to supply protein in mouse diet.At designated time points, samples were collected. Tissues were homog-enized and fractionated according to molecular weight in a 1D SDS/PAGEgel. In-gel digests liberate peptides from the gel, which are then analyzedby LC/MS/MS. The change in molecular weight and the relative populationsof labeled peptides are compared in proteome-wide bioinformatic analy-ses. (B) Three-pool kinetic model for the incorporation of 15N-labeled aminoacids into proteins. Amino acids in the global pool circulate between localtissue pools. (C ) Sample spectral data showing the incorporation of 15Nover time in a Cofilin-1 tryptic peptide. A shift in centroid mass (coloredarrowheads) as well as the change in the relative ratios of unlabeled tolabeled peptide populations (integrated peak areas) increase with time.The observed data fit with predictions in mass spectral changes as a func-tion of labeling time as calculated by a three-pool model (dotted lineoverlaying each spectrum.) The dotted red line overlaying the 0-d timepoint represents the predicted fully labeled spectrum. The correspondingschematics on the Right represent the theorized labeling of the protein,local amino acid, and global amino acid pools in accordance with thethree-pool model.

Price et al. PNAS | August 10, 2010 | vol. 107 | no. 32 | 14509

SYST

EMSBIOLO

GY

local (tissue-wide) pools used in protein synthesis. Amino acidscan exit the system (excreted) from the global pool. According tothis model, the labeled population in our studies represents thefraction of the protein pool that has turned over at a given time andmass shift of the labeled population reflects the enrichment of thelabile nitrogen in the local amino acid pool. The extent of massshift is a characteristic of the tissue of origin and not a feature ofany individual protein. Conversely, the kinetics of labeled pop-ulation represent the turnover rates of peptides and as such areexpected to be highly variable. After a lag period, the labeledpopulation increases exponentially. Fitting this phase to a singleexponential equation allows the measurement of the rate of pep-tide turnover (kturnover, Dataset S2) (29).Most proteins in our dataset were represented by more than

one peptide (Table 1). The variability in turnover kinetics amongpeptides encompassing a single protein was quite low, with thetypical coefficient of variation of ∼0.25 (Fig. S3). Peptides be-longing to a single protein and having similar kinetic profiles oflabeled population were averaged to obtain labeled populationcurves for each protein (Dataset S2). Outliers, defined as pep-tides with Pearson correlations less than 0.9 with respect to theprotein average, were excluded. The averaged protein curveswere fit to a single exponential equation and the turnover rate foreach protein was measured, as shown for blood-extracted serumalbumin (Fig. S3).The measured rates of turnover spanned four orders of mag-

nitude, from 0.002 d−1 to 10 d−1 (Dataset S2). In the brain, proteinshad longer turnover times whereas the distributions of the bloodand liver proteins were skewed toward faster turnover rates (Fig.3). The median turnover rate for the brain peptides was 0.075 d−1

compared with 0.23 and 0.20 d−1, respectively, for the blood andliver proteins. Thus, the average lifetimes of proteins in the brain,liver, and blood are 9.0, 3.0, and 3.5 d, respectively.We found that many of the proteins uniquely expressed in the

brain had slow rates of turnover. For example, myelin basic pro-tein, an abundant constituent of the myelin sheath, had a half-lifeof up to a year. Furthermore, proteins present in all three tissues

showed longer turnover times in the brain (Figs. 3–5). In partic-ular, mitochondrial proteins (Fig. 3B) tend to have a much slowerturnover rate in the brain in comparison with the liver and blood.

Correlation of Turnover Rates to Function. We next sought to un-cover statistically significant correlations between turnover ratesand other biological properties of the proteins kinetically ana-lyzed in our studies. We created a list of Gene Ontology (GO)terms (30) associated with the proteins in blood, liver, and brain,and then separated the proteins into bins according to turnoverrates, ranging from −3 to 2 log d−1 and overlapping by 0.5 log d−1

at intervals of 0.25 log d−1. The relative prevalence of GO terms ineach rate bin was calculated as the ratio between the number ofobserved proteins belonging to the GO term in that bin to thenumber expected by random chance. The statistical significance ofthe enrichment was determined by calculating the Fisher exact-test P value (31). We identified 330 GO terms that were enrichedin one or more rate bins with a statistical significance of P < 0.001,which included 108 terms for brain, 124 for liver, and 98 for blood(Fig. 4). Multiple GO terms can be related to one another ina hierarchical fashion (30). Thus, the same group of genes cancause the enrichment of multiple, related GO terms. To negatethis redundancy, GO terms that were represented by overlappinggroups of proteins in the data (overlap of 30% or more) weregrouped into empirically named clusters (Table S1). The bin

Fig. 2. The kinetics of peptide-labeled populations (A) and mass shifts (B).Measurements (symbols) were made for individual peptides from the desig-nated proteins extracted from brain, liver, and blood. UniProtKB/Swiss-Protaccession codes are indicated.

Fig. 3. Distribution and comparison of protein turnover rates in the brain,liver, and blood. (A) Distribution of turnover rates. In the brain, proteins hadlonger turnover times whereas the distributions of the blood and liverproteins were skewed toward faster turnover rates. The median turnoverrate for the brain peptides was 0.075 d−1 compared with 0.23 and 0.20 d−1,respectively, for the blood and liver proteins. Thus, the average lifetimes ofproteins in the brain, liver, and blood are 9.0, 3.0, and 3.5 d, respectively.Each bar represents the fraction of the total protein population within therate bin. The x axis represents the low limit of the bin at 0.25 log d−1

intervals. (B) Comparison of turnover rates of proteins shared between tis-sues. Gray dots represent mitochondrial proteins.

14510 | www.pnas.org/cgi/doi/10.1073/pnas.1006551107 Price et al.

enrichments for each cluster were determined by averaging theenrichments for each of the constituent GO terms.Secreted proteins (apolipoprotein, chylomicron, and comple-

ment factors) and proteins involved in signaling and proteinfolding (e.g., chaperones) have the fastest rates of turnover. Wemeasured half-lives of 2–10 h for these proteins. Proteins con-tained in the nucleosome (e.g., histones) and those involved in themaintenance of the myelin sheath showed the longest turnoverrates, withmeasured half-lives of up to 1 y. Proteins associatedwitha particular organelle often turn over at similar rates. For example,mitochondrial proteins generally showed long half-lives, and manyproteins residing in the ER had half-lives of 6–10 d (Fig. 4). Itshould be noted that Fig. 4 is not an exhaustive list of functionalcategories that are enriched for proteins with specific half-lives.The figure is limited to functional categories that were representedby a sufficient number of proteins in our dataset to enable themeasurement of enrichments with a high degree of statistical con-fidence (P < 0.001).

Turnover of Protein Complex Subunits.GOannotations were used toidentify proteins in the dataset belonging to multiprotein com-plexes, excluding highly heterogeneous protein complexes (e.g.,microtubule, nucleosome, etc.) We identified complexes that con-tained five or more protein subunits in our dataset and plotted thedistribution of their turnover rates (Fig. 5). The proteins containedwithin each complex and their respective turnover rates are listedin Table S2. Without exception, all protein complexes identified in

both brain and liver turned over more slowly in the brain than inthe liver. For example, 12 subunits of the proteasome were iden-tified in both the brain and the liver. The average half-life for theobserved subunits in the brain was 8 d, whereas the average half-life for subunits in the liver was 4 d. The half-lives among subunitsof the proteasomewere similar, with a SD of only 1.3 d in the brain.The subunits of many large, abundant complexes such as ATPsynthase and the ribosome have similarly narrow ranges of turn-over rates (Table S2).

DiscussionThe recent advent of global genomic and proteomic approacheshas enabled scientists to gain valuable insights into biologicalprocesses at a system-wide level. These techniques have mainlyfocused on analyzing steady-state levels of mRNA or proteinsunder varying biological conditions. However, it is clear thatadditional insights can be gained by furthering these studies toinclude the dynamics of protein expression and degradation (7,8). In the brain, for example, protein turnover is critical to syn-aptic plasticity (32, 33). Here, we report a large-scale analysis ofin vivo protein turnover in mice, a commonly used model or-ganism for the study of mammalian biology. We have shown thatthe rates of protein turnover span four orders of magnitude andcorrelate with a number of biological properties.Typically, in vivo labeling is achieved by the introduction of

food source in which 1 of 20 natural amino acids has been iso-topically labeled (SILAC) (22, 34, 35). Here, we have chosen toinstead use a ubiquitously labeled food source—an approach thatwas initially developed by Wu et al. (23). In measuring turnoverrates, the utilization of a ubiquitously labeled food source offersa number of advantages. First, most peptides produced by trypsindigestion lack the full complement of amino acids. For example,∼40% of the tryptic peptides analyzed in this study lacked lysines(the most commonly used SILAC probe) and could not have beenstudied by the introduction of labeled lysine alone (Fig. S4).Second, in cases where only a single amino acid in a peptide can

Fig. 4. Correlations between function and turnover rates. Functional cate-gories based on the Gene Ontology (GO) Database were clustered into thecategories listed along the y axis. The turnover rates for proteins belongingto the GO clusters were enriched (shown by grayscale shading) with highstatistical significance (P < 0.001) for the indicated rate bin. The constituentGO terms, proteins, and turnover rates for each cluster are listed in Table S1.

Fig. 5. Turnover rates of analyzed subunit proteins that comprise multi-protein complexes. Boxes show the interquartile range (IQR) of turnover ratesof protein complex subunits. The error bar represents the entire range ofrates and the dots represent outliers (1.5 IQR). Numbers in parentheses in-dicate the number of protein subunits analyzed and represented in the dis-tribution. Complexes observed in multiple tissues share the same box color;white boxes indicate that complex was detected in that tissue only.

Price et al. PNAS | August 10, 2010 | vol. 107 | no. 32 | 14511

SYST

EMSBIOLO

GY

be labeled, it is not possible to distinguish between the effects oflocal amino acid pool labeling and protein turnover. For example,analysis of a peptide with a single amino acid probe cannot dis-tinguish between a scenario where 50% of the proteins haveturned over within a fully labeled amino acid pool and one where100% of the proteins have turned over within a half-labeledamino acid pool (Fig. S4). Doherty and coworkers addressed thiscomplexity computationally through the careful analysis of pep-tides containing multiple valine residues (36). In our approach,the presence of multiple exchange probes enables the analysis ofpartially labeled protein populations, allowing the independentmeasurement of these two parameters. This ability enables themeasurement of turnover rates in lieu of the multiphasic andtissue-specific label uptake kinetics observed in vivo. Third, wehave shown that ubiquitously labeled algae can be produced ina typical laboratory setting and is significantly more cost effectivethan the utilization of commercially purified, labeled amino acids.Our data are consistent with a three-pool model of amino acid

incorporation. If the cellular amino acid pools of a tissue are infast equilibrium, the kinetics of mass shift are expected to becongruent among all of the peptides in a given tissue. This trendis clearly observed in our measurements (Fig. 2B). Our data showbursts in mass shifts, indicating rapid increases in 15N-labeledamino acids in the local pool from the dietary algae. The initialincrease was lower in brain than in liver and blood, reflecting thetrafficking of dietary amino acids. The 15N-labeled amino acidsenter the blood and traverse multiple tissues, including the gutand liver, before they reach the brain. The sequential fluxthrough multiple tissues enables the introduction of 14N-labeledamino acids (through local metabolic and catabolic processes)before flow into the local brain pool. Thus, in the brain the initialburst of 15N labeling of the local brain pool is reduced in com-parison with upstream tissues. The second, slower phase of themass shift corresponds to the “recycling” of amino acids throughcatabolic and metabolic processes. In other words, the break-down of internal proteins constantly dilutes the dietary supply of15N-labeled amino acids. Before complete labeling of the aminoacid pool can be achieved, the internal pool of 14N-containingproteins needs to be completely depleted. In the brain, theprolonged recycling phase is slower in comparison with the liverand blood. This result is expected given that brain proteins arerelatively long lived in comparison with liver and blood proteins,leading to an extended recycling phase. Future analysis of thefree amino acid enrichment kinetics in these various tissuescould be used to refine this model.It is important to note that steady-state protein levels, in

themselves, are not predictive of turnover rates.Whereas the staticlevel of a protein is established by the relative ratio of synthesis anddegradation rates, its lifetime is determined by the magnitudeof these rates. This idea is supported by our data. In proteomicanalyses of tryptic peptides, the relative steady-state level ofa protein can be crudely estimated by the ratio of observablepeptides to the theoretical number of peptides expected from thatprotein [Protein Abundance Index (PAI)] (37). Within our data,there is no significant correlation between turnover rates and PAI(Fig. S5A), suggesting that proteins with similar abundances canhave a wide range of turnover rates.Our data suggest that protein turnover is regulated at the level

of the tissue, organelle, and protein complex. The rate of turn-over is generally slower in the brain compared with the blood andthe liver. The relatively slow rate of bulk protein turnover in thebrain had been previously observed (18, 21). We show that this isnot only due the presence of stable proteins that are uniquelyexpressed in the brain, but also because proteins that are sharedbetween the three tissues have a longer half-life in the brain—bya factor of two to five. It is interesting to note that in rats, theorgan-specific metabolic rate per gram of liver has been esti-mated to be five times greater than the equivalent mass of brain

(38). The observation suggests that the difference in the meta-bolic rate of these two tissues may be largely due to differences inenergy commitment to proteome turnover.We observe statistically significant similarities within turnover

rates of proteins localized to specific organelles. Mitochondrialproteins, whether encoded by mitochondrial or nuclear DNA,have similar turnover rates (Fig. 4 andDataset S2).Mitochondrialand nuclear proteins tend to have longer half-lives than cytosolicproteins, which in turn, are more stable than proteins of the en-doplasmic reticulum. For some organelles, this coordinated turn-over may reflect autophagy as a primary route of degradation. Theturnover of mitochondria as a unit through autophagy (mitoph-agy) is known to be a primary method for mitochondrial regula-tion in the cell (39, 40). Mitochondrial protein lifetimes varybetween the liver and the brain, suggestive of different tissue-specific mitophagy rates.For many protein complexes, turnover rates of constituent

subunits fall within a small range. The 20S proteasome corecomplex in the brain and liver has a narrow range of turnoverrates. Although it has been suggested that multiple 20S subtypesare present in cells and tissues (41), our data suggest that alter-native proteasome compositions are either rare or have the samelifetime as the canonical core complex. Synthesis of abundantmultiprotein complexes, such as the ribosome and the protea-some, represent a considerable energy investment for the cell.The coordinated turnover of these complexes may represent anenergy-conservation strategy by the cell to avoid the presence oforphan subunits. For example, the regulation of turnover amongprotein and RNA ribosomal subunits had been previously estab-lished (20). For a few complexes, such as the Cop9 signalosomecomplex (CSN), we observed a broad range of turnover rates.CSN is a regulated component of the ubiquitine–proteasomedegradation pathway associated with specific developmentalprocesses (42). Distinct CSN populations with varying subunitcompositions and activities have been identified (43). Consistentwith these observations, our data suggest the presence of multipleCSN populations with distinct half-lives.A recent analysis of culturedHeLa cells succeeded inmeasuring

the turnover rate of ∼600 proteins (22). Of these, we found ∼150homologous mouse proteins in our in vivo dataset for at least onetissue. A comparison of the data reveals little correlation betweenturnover rates in culture with turnover rates in mice (Fig. S5B).Indeed, the turnover rates measured in culture were significantlyfaster than the in vivomeasurements. This variabilitymay be due tothe continuously proliferating nature of transformed cell lines.Unlike differentiated cells, a dividing cell line is in continuous needof protein production to supply newly generated daughter cells.The regeneration of liver mass that occurs through the pro-liferation of hepatic cells was shown to reduce apparent proteinhalf-life (20). Additionally, the range of half-lives in HeLa cellsappears to bemuch broader than the correspondingmeasurementsin vivo, perhaps because some of the mechanisms that regulateprotein turnover in vivo (e.g., autophagy and tissue regeneration)are absent in culture. These results highlight the limitations ofcultured cell lines as models of in vivo proteome homeostasis.Future studies that combine protein turnover measurements

with quantitative proteomic strategies will allow the absoluteprotein degradation rates to be established on a proteomewidescale. This work provides the methodology and theoretical frame-work necessary to conduct proteome-wide analyses of in vivoprotein turnover in any model organism and environmental con-dition where a labeled diet can be incorporated into the experi-mental design. The approach will be generally useful in analyzingrelationships between proteome homeostasis and biological phe-notypes of interest, particularly in the brain, where protein turn-over is critical to normal function (32, 33) and accumulation ofmisfolded protein aggregates is a primary characteristic of neu-rodegenerative disease (44, 45).

14512 | www.pnas.org/cgi/doi/10.1073/pnas.1006551107 Price et al.

MethodsAdditional methods are provided in SI Text.

Isotopic Labeling and Sample Collection. Labeling experiments were per-formedon9-wk-oldmalemice. At this age,mice are fully developed and tissuegrowth and differentiation are expected to play minimal roles in proteinturnover (20). Following a 7-d habituation period using unlabeled Spirulina,15N-labeled Spirulina was introduced as the sole dietary protein source for32 d (Fig. 1A). After the initiation of labeling, tissues were collected from twomice at 0, 0.4, 1, 2, 4, 8, 16, 24, and 32 d. Liver and brain were flash-frozen onsolid CO2 immediately after collection. Blood (1 mL) was obtained by cardiacpuncture and mixed thoroughly with anticoagulant buffer. The bufferedblood was fractionated by centrifugation, and a cell population depleted ofred blood cells was used for analysis of the blood proteome. Alkylated totalproteinwas resolved according tomolecular weight (Mr) with SDS/PAGE. Eachgel lane (representing a single time point) was divided into 9-Mr sectionsspanning 10–250 kDa. Cut gel fragments were trypsinized, and the extractedpeptides were analyzed by LC/MS/MS.

Gene Ontology Analysis. We created a list of gene ontology (GO) terms (30)associated with the analyzed proteins. The proteins were binned accordingto turnover rates, ranging from −3 to 2 log d−1. The relative prevalence of GO

terms in each rate bin was calculated as the ratio between the number ofobserved proteins belonging to the GO term in that bin to the numberexpected by chance. The statistical significance of the enrichment wasdetermined by calculating the Fisher exact-test P value (31). To negate re-dundancy, GO terms that were represented by overlapping groups of pro-teins in the data were grouped into empirically named clusters (Table S1). Forthe analysis of multiprotein complexes, we identified all proteins within ourdataset that mapped to “GO:0043234 protein complex.” In cases where over-lapping, hierarchical protein-complex GO terms were identified, the higher-order GO term was culled. Highly heterogeneous protein-complex GO terms(e.g., microtubule, nucleosome, etc.) were not included for analysis becausethey do not represent single discrete complexes.

ACKNOWLEDGMENTS. We thank the staff at the Hunters Point animalfacility for assistance in animal maintenance and tissue collection. We alsothank Jonathan Weissman for providing valuable comments and HangNguyen for editing the manuscript. This work was supported by grants fromthe National Institutes of Health (NIH) (AG10770, NCRR P41RR001614, andNCRR RR019934) and gifts from the G. Harold and Leila Y. Mathers CharitableFoundation and Sherman Fairchild Foundation. J.C.P. was supported by theLarry L. Hillblom Foundation. S. Ghaemmaghami was supported by the JohnDouglas French Alzheimer’s Foundation.

1. Goldberg AL, St John AC (1976) Intracellular protein degradation in mammalian andbacterial cells: Part 2. Annu Rev Biochem 45:747–803.

2. Mortimore GE, Pösö AR, Lardeux BR (1989) Mechanism and regulation of proteindegradation in liver. Diabetes Metab Rev 5:49–70.

3. Rousset S, et al. (2007) UCP2 is a mitochondrial transporter with an unusual very shorthalf-life. FEBS Lett 581:479–482.

4. Verzijl N, et al. (2000) Effect of collagen turnover on the accumulation of advancedglycation end products. J Biol Chem 275:39027–39031.

5. Ivan M, et al. (2001) HIFalpha targeted for VHL-mediated destruction by prolinehydroxylation: Implications for O2 sensing. Science 292:464–468.

6. Jaakkola P, et al. (2001) Targeting of HIF-alpha to the von Hippel-Lindau ubiquitylationcomplex by O2-regulated prolyl hydroxylation. Science 292:468–472.

7. Pratt JM, et al. (2002) Dynamics of protein turnover, a missing dimension in proteomics.

Mol Cell Proteomics 1:579–591.8. Hellerstein MK (2004) New stable isotope-mass spectrometric techniques for measuring

fluxes through intact metabolic pathways in mammalian systems: Introduction ofmoving pictures into functional genomics and biochemical phenotyping. Metab Eng 6:85–100.

9. van Remmen H, Ward WF, Sabia RV, Richardson A (1995) Gene expression and proteindegradation [in aging]. Handbook of Physiology Section 11: Aging, ed Masoro EJ(Oxford Univ Press, New York), pp 171–234.

10. Sun F, et al. (2006) Derlin-1 promotes the efficient degradation of the cystic fibrosistransmembrane conductance regulator (CFTR) and CFTR folding mutants. J Biol Chem281:36856–36863.

11. Martinez-Vicente M, Sovak G, Cuervo AM (2005) Protein degradation and aging. ExpGerontol 40:622–633.

12. Bachmair A, Finley D, Varshavsky A (1986) In vivo half-life of a protein is a function ofits amino-terminal residue. Science 234:179–186.

13. Dice JF, Goldberg AL (1975) A statistical analysis of the relationship betweendegradative rates and molecular weights of proteins. Arch Biochem Biophys 170:213–219.

14. Dice JF, Goldberg AL (1975) Relationship between in vivo degradative rates andisoelectric points of proteins. Proc Natl Acad Sci USA 72:3893–3897.

15. Tompa P, Prilusky J, Silman I, Sussman JL (2008) Structural disorder serves as a weaksignal for intracellular protein degradation. Proteins 71:903–909.

16. Buchanan DL (1961) Total carbon turnover measured by feeding a uniformly labeleddiet. Arch Biochem Biophys 94:500–511.

17. Garfinkel D, Lajtha A (1963) A metabolic inhomogeneity of glycine in vivo. I.

Experimental determination. J Biol Chem 238:2429–2434.18. Lajtha A (1959) Amino acid and protein metabolism of the brain. V. Turnover of

leucine in mouse tissues. J Neurochem 3:358–365.19. Lajtha A, Berl S, Waelsch H (1959) Amino acid and protein metabolism of the brain.

IV. The metabolism of glutamic acid. J Neurochem 3:322–332.20. Nikolov EN, Dineva BB, Dabeva MD, Nikolov TK (1987) Turnover of ribosomal proteins

in regenerating rat liver after partial hepatectomy. Int J Biochem 19:159–163.21. McClatchy DB, Dong MQ, Wu CC, Venable JD, Yates JR, 3rd (2007) 15N metabolic

labeling of mammalian tissue with slow protein turnover. J Proteome Res 6:2005–2010.

22. Doherty MK, Hammond DE, Clague MJ, Gaskell SJ, Beynon RJ (2009) Turnover of thehuman proteome: Determination of protein intracellular stability by dynamic SILAC. J

Proteome Res 8:104–112.

23. Wu CC, MacCoss MJ, Howell KE, Matthews DE, Yates JR, 3rd (2004) Metabolic labelingof mammalian organisms with stable isotopes for quantitative proteomic analysis.Anal Chem 76:4951–4959.

24. Salazar M, Martínez E, Madrigal E, Ruiz LE, Chamorro GA (1998) Subchronic toxicitystudy in mice fed Spirulina maxima. J Ethnopharmacol 62:235–241.

25. Chalkley RJ, et al. (2005) Comprehensive analysis of a multidimensional liquidchromatography mass spectrometry dataset acquired on a quadrupole selecting,quadrupole collision cell, time-of-flight mass spectrometer: II. New developments inProtein Prospector allow for reliable and comprehensive automatic analysis of largedatasets. Mol Cell Proteomics 4:1194–1204.

26. Garfinkel D (1963) A metabolic inhomogenity of glycine in vivo. II. Computersimulation. J Biol Chem 238:2435–2439.

27. Schimke RT (1970) Regulation of protein degradation in mammalian tissues.Mammalian Protein Metabolism, eds Munro HN, Allison JB (Academic, New York),Vol 4, pp 177–228.

28. Waterlow JC, Garlick PJ, Millward DJ (1978) Protein Turnover in Mammalian Tissuesand in the Whole Body (North-Holland Publishing, New York).

29. Taylor KB (2002) Enzyme Kinetics and Mechanisms (Kluwer Academic, Boston), p 227.30. Gene Ontology Consortium (2008) The Gene Ontology project in 2008. Nucleic Acids

Res 36(Database issue):D440–D444.31. Fisher RA (1922) On the interpretation of χ2 from contingency tables, and the

calculation of P. J Roy Statist. Soc. 85:87–94.32. Mao LM, et al. (2009) Stability of surface NMDA receptors controls synaptic and

behavioral adaptations to amphetamine. Nat Neurosci 12:602–610.33. Schwartz JH (2003) Ubiquitination, protein turnover, and long-term synaptic

plasticity. Sci STKE 2003:pe26.34. Krüger M, et al. (2008) SILAC mouse for quantitative proteomics uncovers kindlin-3 as

an essential factor for red blood cell function. Cell 134:353–364.35. Ong S-E, Mann M (2006) A practical recipe for stable isotope labeling by amino acids

in cell culture (SILAC). Nat Protoc 1:2650–2660.36. Doherty MK, Whitehead C, McCormack H, Gaskell SJ, Beynon RJ (2005) Proteome

dynamics in complex organisms: Using stable isotopes to monitor individual proteinturnover rates. Proteomics 5:522–533.

37. Ishihama Y, et al. (2005) Exponentially modified protein abundance index (emPAI) forestimation of absolute protein amount in proteomics by the number of sequencedpeptides per protein. Mol Cell Proteomics 4:1265–1272.

38. Wang Z, O’Connor TP, Heshka S, Heymsfield SB (2001) The reconstruction of Kleiber’slaw at the organ-tissue level. J Nutr 131:2967–2970.

39. Tal R, Winter G, Ecker N, Klionsky DJ, Abeliovich H (2007) Aup1p, a yeastmitochondrial protein phosphatase homolog, is required for efficient stationaryphase mitophagy and cell survival. J Biol Chem 282:5617–5624.

40. Journo D, Mor A, Abeliovich H (2009) Aup1-mediated regulation of Rtg3 duringmitophagy. J Biol Chem 284:35885–35895.

41. Drews O, et al. (2007) Mammalian proteasome subpopulations with distinct molecularcompositions and proteolytic activities. Mol Cell Proteomics 6:2021–2031.

42. Wei N, Deng XW (2003) The COP9 signalosome. Annu Rev Cell Dev Biol 19:261–286.43. Tomoda K, et al. (2002) The cytoplasmic shuttling and subsequent degradation of

p27Kip1 mediated by Jab1/CSN5 and the COP9 signalosome complex. J Biol Chem 277:2302–2310.

44. Prusiner SB (2001) Prions. Fields Virology, eds Knipe DM, Howley PM (LippincottWilliams and Wilkins, Philadelphia), 4th Ed, pp 3063–3087.

45. Dobson CM (1999) Protein misfolding, evolution and disease. Trends Biochem Sci 24:329–332.

Price et al. PNAS | August 10, 2010 | vol. 107 | no. 32 | 14513

SYST

EMSBIOLO

GY