Articles

Global analysis of protein structural changes in complex proteomes

© 2014 Nature America, Inc. All rights reserved.

Yuehan Feng1,5, Giorgia De Franceschi1,2,5, Abdullah Kahraman1,5, Martin Soste1, Andre Melnik1, Paul J Boersema1, Patrizia Polverino de Laureto2, Yaroslav Nikolaev3, Ana Paula Oliveira4 & Paola Picotti1 Changes in protein conformation can affect protein function, but methods to probe these structural changes on a global scale in cells have been lacking. To enable large-scale analyses of protein conformational changes directly in their biological matrices, we present a method that couples limited proteolysis with a targeted proteomics workflow. Using our method, we assessed the structural features of more than 1,000 yeast proteins simultaneously and detected altered conformations for ~300 proteins upon a change of nutrients. We find that some branches of carbon metabolism are transcriptionally regulated whereas others are regulated by enzyme conformational changes. We detect structural changes in aggregation-prone proteins and show the functional relevance of one of these proteins to the metabolic switch. This approach enables probing of both subtle and pronounced structural changes of proteins on a large scale. Protein function is modulated by regulation of expression levels, protein-protein interactions, chemical modifications and structural changes. Mass spectrometry (MS)-based proteomic techniques are routinely used to measure changes in protein abundance, posttranslational modification and protein interactors 1–3, but much less is known about protein conformational changes4–6, owing to the lack of suitable approaches available to study global changes in protein folds in cells. Proteins change structure upon ligand binding, interaction with other proteins, chemical derivatization, mutations or environmental changes. Structural transitions range from local fluctuations, through larger domain motions, to drastic switches between folded and unfolded conformations, or between monomeric and polymeric states7–9. For example, allosteric enzyme conformational changes modulate enzyme activity in response to changes in the concentration of specific metabolites10. In amyloidoses and some neurodegenerative diseases, specific proteins undergo a structural change that increases their propensity to form insoluble deposits, resulting in pathological changes including cell and tissue damage11. The capability to capture protein structural transitions both in complex biological samples and on a large scale would have a profound impact on our understanding of protein structure-function relationships. X-ray crystallography, nuclear magnetic resonance (NMR) and various spectroscopic techniques have been used to analyze simple protein systems reconstructed in vitro, but studying protein structural changes in complex biological systems is not feasible with these methods12–14. Förster resonance energy transfer (FRET)15 and in-cell NMR16 enable monitoring of conformational changes of specific proteins in their cellular environment, but both methods require labeling the protein that is being analyzed and are therefore not suitable for

proteome-wide applications. Similarly, approaches based on chemical cross-linking and MS are not efficient in capturing large-scale structural changes in unfractionated cell extracts17. To address this problem, we present a method that enables probing of structural transitions of proteins in complex biological environments on a large scale. It couples limited (or native) proteolysis (LiP)18 with targeted proteomic tools. LiP involves the use of broad-specificity proteases under controlled conditions such that initial cleavage sites are dictated by the structural features of the protein18. To date, LiP has typically been applied to purified proteins19–22, owing to the challenge of identifying LiP sites in complex backgrounds. To enable the proteomewide application of LiP, our approach exploits: (i) a double-digestion step that generates peptides amenable to bottom-up proteomic analysis and (ii) the sensitivity and background filtering capabilities of selected reaction monitoring (SRM) MS23 to reproducibly probe LiP patterns in complex matrices. LiP-coupled SRM (LiP-SRM) allows probing of both subtle and pronounced structural changes of proteins and can be used in both targeted and discovery experiments, as demonstrated here by systematic identification of structural transitions of proteins from yeast cultures subjected to a metabolic transition. RESULTS The LiP-SRM workflow In the first step, a proteome extracted from a biological sample under nondenaturing conditions is subjected to double digestion (Fig. 1). The first digestion is conducted using a broad-specificity protease (e.g., proteinase K, thermolysin, subtilisin, papain, chymotrypsin or elastase) at a low enzyme to substrate ratio (E/S) and for a short time. Under these conditions, the sites of initial proteolysis are dictated by the structural properties of the substrate, generating (on average)

1Institute

of Biochemistry, Department of Biology, ETH Zurich, Zurich, Switzerland. 2CRIBI Biotechnology Centre, University of Padua, Padua, Italy. 3Institute of Molecular Biology and Biophysics, Department of Biology, ETH Zurich, Zurich, Switzerland. 4Institute of Molecular Systems Biology, Department of Biology, ETH Zurich, Zurich, Switzerland. 5These authors contributed equally to this work. Correspondence should be addressed to P.P. ([email protected]). Received 9 December 2013; accepted 25 July 2014; published online 14 September 2014; doi:10.1038/nbt.2999

nature biotechnology  advance online publication



Articles LC-SRM

Discovery LC-MS/MS LiP (native)

Trypsin (denaturing)

VLEHAA FDPVAK

(optional)

RT

Condition 1

Discovery LC-MS/MS Trypsin (denaturing)

VLEHAAFDPVAK

(optional)

RT

LC-SRM VLEHAAFDPVAK Trypsin (denaturing)

© 2014 Nature America, Inc. All rights reserved.

LiP (native)

Discovery LC-MS/MS (optional)

RT

Condition 2

Trypsin (denaturing)

Discovery LC-MS/MS (optional)

VLEHAAFDPVAK

RT

Figure 1  LiP-SRM workflow. In the first step of the LiP-SRM workflow a proteome is extracted from cells under nondenaturing, native conditions and limited proteolysis is conducted with a broad-specificity protease, such as proteinase K, for a short time. Red arrows indicate limited proteolysis sites. Protease activity is quenched by shifting the proteome to denaturing conditions. A complete trypsin digestion is then performed. A fraction of the same sample is only subjected to the trypsin step under denaturing conditions. The samples are quantitatively analyzed by MS and levels of the resulting and peptides are compared. A fully tryptic peptide containing a LiP cleavage site will be detected in the trypsin control and replaced by two half-tryptic halves in the sample subjected to LiP. Each proteome (biological triplicates) undergoes the double-digestion step to detect protein structural differences in differently treated proteome extracts and the resulting digests are compared. Discovery-based shotgun LC-MS/MS analyses, followed by SRM validation or hypothesis-based SRM alone, are applied to the analysis, in case of discovery or targeted applications, respectively.

large protein fragments. The proteome is then shifted to denaturing conditions, quenching the activity of the LiP protease. Next, a complete trypsin digestion generates peptides suitable for bottom-up MS analysis. As a control, an aliquot of the same proteome taken before LiP is subjected to trypsinization only. As a result, fully tryptic peptides embedding LiP sites are less abundant in the doubly digested sample than in the control, and half-tryptic peptides generated by intra-tryptic peptide cleavage are present only in the doubly-digested sample. Peptides are detected and quantified using SRM assays for all fully tryptic peptides from the protein of interest. Fully tryptic peptides with considerably lower abundance in the doubly digested sample compared with the trypsin-only sample embed LiP cleavage sites and are followed up by SRM measurement of all possible halftryptic peptides that can be generated by their internal cleavage. These steps allow the identification of peptides that are specific to a given protein conformation. We name these ‘conformotypic peptides’. To detect structural transitions induced by a perturbation, we compared proteolytic patterns obtained under two conditions after normalization with trypsin-only controls to correct for protein abundance changes across samples, incomplete trypsin specificity and endogenous protease cleavages. Using LiP-SRM to analyze an amyloid-forming protein To validate the LiP-SRM strategy, we analyzed the human protein α-Syn. α-Syn is predominantly unfolded under physiological conditions24. In Parkinson’s disease patients, α-Syn switches to a β-sheetrich fold and polymerizes into fibrillar, amyloid aggregates24. We prepared monomeric, predominantly unfolded α-Syn (M-α-Syn) by 

dissolving purified recombinant α-Syn in a physiological buffer25; we prepared β-sheet-rich α-Syn (F-α-Syn) by prolonged incubation under conditions that favor fibrillization25 and confirmed the formation of amyloid fibrils by electron microscopy (Fig. 2a), thioflavin T fluorescence, circular dichroism and Fourier transform infrared spectroscopy (Supplementary Fig. 1). To mimic a complex biological background, we spiked M- or F-α-Syn into Saccharomyces cerevisiae proteome extracts. The two proteome samples differed only in the conformation of the added protein, as yeast is devoid of α-Syn homologs. We then analyzed the samples using the LiP workflow with proteinase K as the LiP protease and identified conformotypic peptides specific to M- and F-α-Syn conformations (Fig. 2b,c and Supplementary Table 1). The fully tryptic peptides EGVVHGVATVAEKTK, EQVTNVGGAVVTGVTAVAQK and TVEGAGSIAAATGFVK map to the amyloid β-sheet-rich core of F-α-Syn26; in M-α-Syn these peptide regions are disordered. These three peptides were consistently protected from proteolytic attack and more abundant (between 26- and 37-fold) in the sample containing F-α-Syn. The half-tryptic peptides GVTAVAQK and AATGFVK, derived from internal cleavage of peptides EQVTNVGGAVVT GVTAVAQK and TVEGAGSIAAATGFVK, respectively, were observed only for M-α-Syn. These data indicate that the regions undergoing the most pronounced conformational changes were the N terminus and the amyloid core; the region associated with the amyloid core transitioned from a disordered state in M-α-Syn to a β-sheet-rich conformation in F-α-Syn26. This experiment shows that LiP-SRM can be used to probe conformational transitions of a protein in a complex biological background. advance online publication  nature biotechnology

a

b M-α-Syn

41.1 ± 2.5

F-α-Syn C terminus

N terminus

25.9 ± 3.3

0.5 µm

0.08 ± 0.02

0.19 ± 0.04

2.4 × 104

11.6 12.3 13.0

RT

RT

8.4

9.1

9.8

Intensity, cps 11.5 12.2 12.9

RT

RT

Apomyoglobin

Holomyoglobin

e

Heme

Intensity, cps 9.2

9.9

4.2

8.6

1.6 ± 0.4

1.4 ± 0.1

5.6

4.0 × 104

9.3 10.0

4.3

RT

1.6 ± 0.1

4.9

RT

6,200

12.0 12.7 13.4

RT

8.5

RT

9.2 × 104

2.2 × 104

Intensity, cps

F-α-Syn in yeast

Intensity, cps

Intensity, cps

12.0 12.7 13.4

Intensity, cps

9.3 10.0

RT

Intensity, cps

8.6

4

4.0 × 10

6,200

9.2 × 104

Intensity, cps

2.2 × 104

Intensity, cps

2.4 × 104

M-α-Syn in yeast

d

28.4 ± 5.0

Fibril core 36.9 ± 4.4

c

4.2 ± 0.1

10.4 ± 0.5

15 d

Intensity, cps

5.0

5.7

RT

2.3 ± 0.2

0.11 ± 0.02

1.3 ± 0.2

8.0 ± 1.4 >32.8 ± 6.1

1.5 ± 0.3 Helix F

Loop

6.6 8.1

9.6

9.5 11.0 12.5

RT

Using LiP-SRM to detect a subtle structural transition During amyloid conversion, α-Syn undergoes a global conformational change. To assess whether LiP-SRM can also detect subtle conformational changes, we used a second model system, myoglobin. Holomyoglobin is globular with an α-helical fold (Fig. 2d)27. Upon heme dissociation, the apo conformational form of myoglobin is produced28. Based on NMR data, seven of the eight helices of holomyoglobin remain structured in apomyoglobin, but helix F (13 amino acids) becomes disordered (Fig. 2d)28. We prepared apomyoglobin and holomyoglobin from purified myoglobin, spiked them separately into yeast proteome extracts, and analyzed the samples by LiP-SRM. The fully tryptic peptide GHHEAELKPLAQSHATK (Fig. 2e,f), encompassing helix F, was detected only in the ­holomyoglobin-containing sample at an intensity comparable to that of the trypsin control. In contrast, in the apomyoglobin-containing sample the fully tryptic peptide corresponding to helix F was almost undetectable (Fig. 2f, ~tenfold change). Two half-tryptic peptides derived from cleavage within peptide GHHEAELKPLAQSHATK, which is located in helix F in holomyoglobin28, were detected only in the apomyoglobin sample (Fig. 2e). This indicates that apomyoglobin nature biotechnology  advance online publication

9.4

Intensity, cps

2.0 × 10

Intensity, cps 7.9

RT

17.0 18.5 20

RT 4

3.0 × 10

6.4

9.1 10.6 12.1

RT 4

3,500 Apomyoglobin in yeast

Intensity, cps

Intensity, cps

Holomyoglobin in yeast

1.8 × 105

2.0 × 104

3.0 × 104

RT 5

1.8 × 10

Intensity, cps

3,500

Intensity, cps

f

Intensity, cps

Figure 2  LiP-SRM of α-Syn and myoglobin spiked into complex cell extracts. (a) Two conformational states of α-Syn. Left, illustration of the monomeric (M), disordered conformation adopted by α-Syn in a physiological buffer. Right, amyloid fibrils (F) formed upon incubation of α-Syn at 37 °C for 15 d. (b) Amino acid sequence of α-Syn. The fibril core region of F-α-Syn26 is marked in orange. Two replicates of the same yeast total proteome extract were spiked with α-Syn in either M or F conformations at a concentration that mimics a protein of medium-high abundance and subjected to LiP-SRM. (c) SRM peaks obtained for conformotypic peptides in the cell extract. The different SRM transitions measured for each peptide (Supplementary Table 1) are shown in different colors. RT, retention time; cps, counts per second. (d) Left, the holomyoglobin fold, comprised of eight helical segments, based on the crystallographic structure of the protein27. Right, structure of apomyoglobin, the conformation adopted after removal of the heme group from holomyoglobin. The image was obtained based on PDB file 1YMB and the structural details are from 28. Images were prepared using the software WebLab (Accelrys). (e) Amino acid sequence of myoglobin. Two replicates of the same yeast total proteome extract were spiked with myoglobin in either holo or apo conformations at a concentration that mimics a protein of medium-high abundance and samples were subjected to LiP-SRM. (f) SRM peaks obtained for conformotypic peptides in the cell extract. The different SRM transitions measured for each peptide (Supplementary Table 1) are shown in different colors. In b and e abundance changes of fully tryptic (red) and half-tryptic (blue) peptides across the two conditions are mapped to the sequence of the proteins. Average fold changes ± s.d. are reported next to each peptide. Peptides that did not change significantly are shown in gray boxes in e.

Intensity, cps

© 2014 Nature America, Inc. All rights reserved.

Articles

9.3 10.8 12.3

8.9 10.4 11.9

17.2 18.7 20.2

RT

RT

RT

is selectively cleaved within the loop generated upon heme removal, whereas holomyoglobin is resistant to this proteolytic event. Most of the apomyoglobin/holomyoglobin abundance ratios for fully tryptic peptides mapping to the other helical segments (Fig. 2f) were close to one. This suggests that the structure of apomyoglobin is similar to that of holomyoglobin with the exception of the F-region. This shows that LiP-SRM can capture a subtle protein conformational change in a cell extract. Analysis of the myoglobin system demonstrated that the LiP step was highly reproducible (Supplementary Table 2 and Supplementary Figs. 2 and 3). LiP patterns were also qualitatively and quantitatively independent of the abundance of target protein substrates in complex proteome extracts (Supplementary Table 3). The myoglobin system was also used to assess the influence of the E/S ratio, protease incubation time and type of protease (Supplementary Figs. 4 and 5, and Supplementary Note). Last, we used conformotypic peptides to calculate the amount of each conformation in a mixture of two conformational states by analysis of mixtures of holomyoglobin and apomyoglobin, containing 90%, 50% and 10% apomyoglobin, in a yeast proteome background (Supplementary Tables 4 and 5). 

Articles a Probability

Figure 3  Global analysis of protein conformational changes. (a,b) Analysis of LiP cleavage sites in the proteome of yeast grown in glucose. (a) WebLogo diagram of the proteinase K cleavage site and the neighboring five amino acids (AA). Amino acids were plotted based on their frequency and colored according to their physicochemical properties. (b) Pie-chart showing the type of secondary structure element where LiP cleavage sites occurred. (c) Proteins that changed LiP pattern, and thus structural properties, in the transition of yeast from glucose- to ethanolbased metabolism. Protein identifiers were mapped to known protein functions using the Saccharomyces Genome Database and manually grouped based on function using keywords associated to that function. Only functional clusters including at least three proteins were reported. The size of each box is proportional to the number of proteins from the respective group. Drawings inside boxes show a schematic representation of the most represented biological process in the respective class. Enrg., energy metabolism; Polys., polysaccharide metabolism; Nucleot., nucleotide metabolism; TIEFs, translation initiation or elongation factors; Eis., eisosomes; 14-3-3, isoforms of 14-3-3 proteins, acidic proteins that act as regulators of signal transduction. Actin/tubulin, actin- and tubulinrelated proteins; p-bodies, proteins involved in cytosolic processing bodies; RNA pol., proteins involved in RNA polymerase complexes.

b

1.0

β-Sheet Loops or unstructured

0.5 0

–5

0 5 AA distance from cleavage

c Actin/tubul.

ER-Golgi

α-Helix

Prions

Eis. 14-3-3

Translation

Protein

TIEFs

Structural changes

Ribosome

Degradation

60S 40S

Folding

p-bodies RNA pol.

Metabolism

Core carbon metabolism

Based on the calculations described in the Supplementary Note, the LiP data indicated that the samples contained 82.8%, 56.2% and 14.8% of apomyoglobin, respectively (Supplementary Table 6). Agreement with actual amounts of myoglobin conformations in the mixtures demonstrates that conformotypic markers, when available, can be used to calculate the conformational composition of a sample containing unknown amounts of the different conformations. Analyzing protein conformational changes on a large scale Next, we assessed whether a LiP-based approach would enable unbiased identification of proteins that undergo structural transitions upon a given stimulation in a proteome. Yeast metabolism has been used in multiple studies to evaluate the performance of various approaches because of its well-characterized topology and the availability of high-throughput data sets (e.g., refs. 23,29,30). We focused on the metabolic transition from glucose- to ethanol-based growth, a transition known to induce metabolic remodeling in yeast30. Samples were cultured in triplicate to exponential phase in either glucose- or ethanol-based medium and subjected to LiP. The resulting 12 samples (two conditions, three biological replicates, each sample subjected to proteinase K and trypsin or trypsin-only treatment) were first analyzed by liquid chromatography (LC)-MS/MS on a high-resolution MS followed by SRM-based validation (Supplementary Note). The shotgun analysis identified 21,899 peptides mapping to 1,622 unique proteins (protein false discovery rate (FDR) ~1%). By comparing proteinase K–treated and control samples, we detected 4,267 LiP sites, mapping to 1,001 proteins (Supplementary Table 7). First, we analyzed the locations and properties of LiP sites in the proteome of yeast grown on glucose. We focused on a subset of LiP peptides that map to unique proteins and assessed their precise abundance changes by SRM (Supplementary Tables 1 and 8). The associated proteins spanned a range of over a million to below 2,000 copies/cell31 and had a variety of functions (Supplementary Table 8). No obvious consensus motif for proteinase K cleavage was identified. Only a slight preference for cleavage at alanine, an abundant amino acid in the yeast proteome, was observed (Fig. 3a). This confirms that proteinase K LiP sites are dictated by the structure of the substrate. We mapped half-tryptic peptides to the available structures of the corresponding proteins retrieved from the Protein Data Bank or homology models. Most LiP sites occurred at loops or regions devoid 

Amino acid

Enrg. Nucleot.

Lipid

© 2014 Nature America, Inc. All rights reserved.

Carbon

Polys. Detox

ATP synthase

of secondary structure (Fig. 3b and Supplementary Table 8). Helical or β-sheet segments were also cleaved, though at a lower frequency. Next, we analyzed proteins with differences in LiP pattern in glucose- versus ethanol-based metabolism. We focused on unique halftryptic peptides with an abundance difference of at least twofold, after corrections for changes in protein abundance and intracellular proteolytic activity. The 586 half-tryptic peptides that met this criterion mapped to 283 proteins (Fig. 3c and Supplementary Table 9). Of these half-tryptic peptides, 8% were located at known or predicted protein-protein interaction sites32 and 7% embedded the site of a known post-translational modification33. For the 283 proteins, the most common Gene Ontology terms were “Catalytic Activity” and “Metabolic Process” (Supplementary Fig. 6). Functional enrichment analysis using proteins detected in our experiment as a background proteome (Supplementary Table 10) revealed that the top-scoring functional module was “Precursor Metabolites and Energy Generation” (FDR-adjusted q value, 1.45 × 105), specifically the glycolysis pathway (q value, 4.7 × 105, Supplementary Fig. 7). Thus, proteins with structural properties that change upon the metabolic transition are significantly enriched for glycolytic and carbon metabolism enzymes. Other proteins that changed their LiP pattern were involved in endoplasmic reticulum–to-Golgi trafficking, cytoskeleton dynamics, protein folding and degradation, translation and various metabolic branches. Conformational changes in core carbon metabolism proteins To assess the functional meaning of the conformational changes in proteins involved in core carbon metabolism, we asked whether we could identify a relationship between protein structural changes and enzyme activity and how structure-based regulatory mechanisms were interfaced with transcriptional regulation. We focused on metabolic enzymes and known isoenzymes in this metabolic subnetwork advance online publication  nature biotechnology

Articles Glucose

© 2014 Nature America, Inc. All rights reserved.

Figure 4  LiP-SRM analysis of core carbon metabolism upon a shift from glycolytic to gluconeogenic growth. Schematic representation of core carbon metabolism in S. cerevisiae, comprising the glycolytic pathway, the TCA and glyoxylate cycles, and the ethanol production branch. Proteins are colored according to the results of LiP and protein abundance measurements by SRM. Abundance changes are differences between levels in ethanol relative to glucose growth. Only significant abundance changes are reported (peptide fold change > 2, q value < 0.02, multiple SRM transitions per peptide, three biological replicates, for LiP patterns; and protein fold change > 2, q value < 0.02, multiple SRM transitions, multiple peptides per protein, three-biological replicates, for protein abundances). Protein abundance changes were determined based on measurements conducted on the trypsin control (Supplementary Table 11). Changes in the LiP pattern (Supplementary Table 11) were calculated from SRM data obtained on the doubly digested sample and after normalization for protein abundance changes and endogenous protease activity. The information on reversed and inactivated reactions in the transition from glucose to ethanol grown cells is from reference 34.

(56 proteins). To evaluate transcriptional regulation in the network, we precisely measured protein abundance differences between the two growth conditions using SRM (Fig. 4 and Supplementary Table 1). Most enzymes in the tricarboxylic acid cycle (TCA) and glyoxylate cycles significantly increased abundance in ethanol-grown relative to glucose-grown cells (q value < 0.02, fold change > 2; average, 11- and 112-fold, respectively, Supplementary Table 11). Four (Tdh1, Hxk1, Glk1 and Gpm3) and three (Cdc19, Gpm1 and Eno2) of the 18 glycolytic enzymes increased and decreased slightly in abundance, respectively (average, 2.5-fold). Three of the ten ethanol-metabolism enzymes increased (Adh2, Adh5 and Ald6, average 75-fold) and one enzyme decreased slightly in abundance (Pdc1, approximately threefold) in ethanol compared to glucose. These data suggest that TCA and glyoxylate cycle enzymes undergo coordinated transcriptional regulation; the same is not apparent for glycolytic and ethanol production enzymes, as previously observed23. Next, we used SRM to validate the structural transitions in the metabolic system. Levels of 19 half-tryptic LiP peptides, corresponding to ten proteins (Cdc19, Gpm1, Hxk2, Pgk1, Tdh1, Tdh2, Tdh3, Pfk1, Pdc1 and Adh1), changed significantly (q value < 0.01) between the two conditions (Supplementary Table 11 and Supplementary Fig. 8). These enzymes function in the glycolysis and ethanol­production branches of the network. None of the enzymes we tested changed subcellular distribution upon the considered metabolic transition (Supplementary Fig. 9). Interestingly, structural changes mostly mapped to enzymes that underwent no or minor abundance changes (Fig. 4). No significant changes were found in the LiP patterns of either TCA or glyoxylate cycle enzymes. All 17 TCA cycle and some glyoxylate cycle enzymes were readily detectable in both conditions. One explanation for this observation is that the TCA cycle is not regulated by enzyme structural changes. To exclude the possibility that this is an artifact due to undersampling, we quantified by SRM all possible fully tryptic peptides (319 peptides, Supplementary Table 1) for the 17 TCA cycle enzymes. We reasoned that if any of these enzymes underwent local structural transitions, this would be reflected in a change in abundance of the fully tryptic peptide mapping to the region associated with the transition, as we showed for myoglobin and α-Syn. No peptide abundances changed significantly (Supplementary Table 12), suggesting that no TCA cycle enzyme changed structural properties upon the metabolic transition. Next, we mapped metabolic flux data, previously resolved by flux balance analysis of yeast grown under the same two conditions 34, to the metabolic network (Fig. 4) and asked whether there was a nature biotechnology  advance online publication

Hxk1 Hxk2

Detected

Glk1

Conformational change

G6P

Abundance increase

Pgi1

Abundance decrease

F6P Pfk1 Pfk2

Reversed

Fbp1

Inactivated

FBP Fba1 GA3P Tdh1 Tdh3 Tdh2 13BPG Pgk1 Gpm1 Gpm2 Eno1 Eno2

Pck1

3PG Gpm3 2PG

Adh6 Adh4 Adh5 Adh3 Adh2 Adh1

PEP Cdc19 Pyk2 PYR

Pdc1 Pdc6 Pdc5

Ethanol

Ald6

AcA

AcATE Acs1 Acs2

Cytosol Pyc1 Pyc2

Pda1 Pdb1

Mitochondria

Mdh1

Cit1 Cit3

MAL

Sdh1 Sdh2 Sdh3 Sdh4

Mls1

ICIT

CIT Aco1 Aco2

GLX

Dal7

Icl1

ICIT Idp1

AKG

Osm1

SUC

MAL

Aco1

Idh1 Idh2

Fum1

FUM

Mdh2 Mdh3

Cit2

CIT

AcCoA OAc

OAc

AcCoA

PYR

Lsc1

SUCCoA

Kgd1 Kgd2

relationship between the structural transition of an enzyme and the predicted change in metabolic activity. Strikingly, of the eight reactions catalyzed by enzymes that changed structure based on our data, four (those catalyzed by Gpm1, Tdh1-3, Adh1 and Pgk1) underwent reversion of the associated flux and three (Cdc19, Pfk1 and Hxk2) were inactivated or no longer used when yeast was grown on ethanol rather than glucose. No flux data were available for the reaction catalyzed by Pdc1, but its activity is much higher in glucose than in ethanol35. Thus, the reactions catalyzed by each of the enzymes identified by LiP to change structural features were either reversed or inactivated upon the metabolic shift. Reactions in the TCA cycle generally did not change direction, but rather flux increased in magnitude in ethanol compared to glucose, suggesting that increased TCA cycle activity upon growth in ethanol is realized by increased expression of the associated enzymes. Regulation of reactions that need to be reversed or inactivated in glycolysis and ethanol production branches involved instead enzyme structural changes and little or no alteration in expression levels. Structural transitions dependent on FBP concentration To determine whether structural transitions of glycolytic enzymes are due to allosteric regulation, we focused on pyruvate kinase, Cdc19. The reaction catalyzed by Cdc19 needs to be inactivated upon the switch from glucose to ethanol36. Close-to-zero flux is observed in ethanol34,36; however, the levels of this enzyme decrease only 2.4-fold compared to in glucose (Fig. 4). Thus, we asked whether the observed structural change reflected an allosteric inactivation. Based on SRM measurements (Supplementary Tables 1 and 13), the peptides with 

Articles c

Glu 1.2 × 10

IENQQGVNNFDEILK

Eth 1.2 × 105

5

1.2 × 10

5

2.1 ± 0.27 GDTYVSIQGFK

0.16 ± 0.03

0.39 ± 0.01

EPVSDWTDDVEAR

0.25 ± 0.008 0.33 ± 0.03

0.20 ± 0.03

0.20 ± 0.008

0.44 ± 0.04

2.0 × 10 VSDWTDDVEAR

Intensity, cps

Intensity, cps

19.2 20.3 21.4 19.2 20.3 21.4 19.2 20.3 21.4 RT RT RT 4 4 2.0 × 10 2.0 × 10

4

Intensity, cps

© 2014 Nature America, Inc. All rights reserved.

0.09 ± 0.004 0.29 ± 0.03

20.4 21.1 21.8 20.5 21.2 21.9 20.5 21.2 21.9 RT RT RT 3.2 × 104 3.2 × 104

3.2 × 104

Intensity, cps

0.10 ± 0.006

e FBP binding site

0.19 ± 0.01

0.27 ± 0.009

11.0 12.1 RT

22.6 23.7 24.8 RT

Intensity, cps

0.45 ± 0.01

Intensity, cps

0.41 ± 0.06 0.49 ± 0.06

10.2 11.3 12.4 9.9 10.9 12.0 RT RT 5 5 4.6 × 10 4.6 × 10

Intensity, cps

0.40 ± 0.03

5

Intensity, cps

0.48 ± 0.04

4.6 × 10

Intensity, cps

0.49 ± 0.02

Intensity, cps

9.8

b

Intensity, cps

VTDGVMVAR

Intensity, cps

FBP

Cdc19 active site

22.9 24.2 25.5 23.0 24.3 25.6 23.0 24.3 25.6 RT RT RT 5 5 8.4 × 10 8.4 × 10 8.4 × 105

PYR

d

Eth + FBP Intensity, cps

n.s. 2–5 >5

Intensity, cps

Fold-change

Intensity, cps

a

22.7 23.8 24.9 RT

22.5 23.6 24.7 RT

Figure 5  Structural changes of Cdc19 upon a switch from glucose- to ethanol-based metabolism. (a) Structure of yeast pyruvate kinase, Cdc19. One subunit of the homotetrameric crystal structure of Cdc19 (PDB file 1A3W) is shown in a cartoon representation. The product analog of (2-phosphoglycolic acid, PGA, multicolor) and an associated potassium cation (purple) are emphasized in the active site of Cdc19. The allosteric regulator FBP (multicolor) is depicted in its allosteric site. The conformational change of Cdc19 upon transition from glycolytic to gluconeogenic conditions is mapped to the protein structure as derived from changes in the proteolytic pattern of the protein. Abundance changes of fully tryptic peptides in ethanol growth conditions relative to glucose growth conditions are mapped to the structure of the protein with the following color code: lavender, no significant change (n.s.); black, abundance change between two- and fivefold; green, abundance change larger than fivefold. (b) Amino acid sequence of Cdc19. Differences in the proteolytic pattern of Cdc19 in yeast grown in glucose versus ethanol are mapped to the enzyme sequence. Abundance changes of fully tryptic peptides are indicated with the color code used in a. Dashed line, nonproteotypic peptide. Significant average fold changes ± s.d. (fold change > 2, q value < 0.02, multiple SRM transitions per peptide, three biological replicates) are reported above each peptide sequence. (c) SRM peaks obtained for five LiP peptides of Cdc19 quantified from glucose- or ethanol-growth conditions or upon addition of FBP in excess to the extract from cells grown in ethanol. The five peptides displayed map to the active site or to the FBP-binding site of Cdc19. The different SRM transitions measured for each peptide (Supplementary Table 1) are shown in different colors. Signal intensities were normalized by the abundance change of the protein upon the considered metabolic transition. Further data on the structural reversion of Cdc19 by FBP are in Supplementary Table 14. (d,e) Crystallographic, single-subunit structures of the yeast pyruvate kinase in the absence (d, PDB entry 1A3X) and presence (e, PDB entry 1A3W) of FBP (multicolored small molecule), bound to the allosteric site. Peptides highlighted in green are those displaying the largest fold change in our LiP experiments and were found to be part of the allosteric site. Interestingly, each peptide covers a protein region that is partially disordered and becomes ordered upon binding of FBP. The C-terminal peptide is in a random coil conformation and becomes part of a β-sheet upon FBP binding.

the largest fold changes mapped to the active site and the binding site of an allosteric regulator of Cdc19, fructose-1,6-bisphosphate (FBP) (Fig. 5)36. The FBP concentration was 100-fold higher in glucose than in ethanol; therefore, we repeated the LiP experiment after addition of an excess of FBP to the lysate of yeast grown in ethanol. Addition of FBP specifically reversed the proteolytic pattern of Cdc19 to that observed in glucose (Fig. 5 and Supplementary Table 14) and substantially restored its activity (Supplementary Fig. 10). Addition of excess FBP to the lysate of yeast grown in glucose caused only minor alterations in the Cdc19 proteolytic pattern. Thus, FBP does not affect the activity of the proteases used in our workflow. The LiP pattern obtained from a pure Cdc19 preparation in the presence of FBP matched the LiP pattern of Cdc19 from glucose-grown yeast. The structural rearrangement we predicted based on LiP data closely matched that reported from X-ray analysis of purified Cdc19 in the presence and absence of FBP (Fig. 5d,e). These results confirm that the structural transition of Cdc19 and the resulting activation of the 

enzyme in cells grown in glucose are triggered by changes in the concentration of FBP and that the metabolite directly binds Cdc19. Given its massive change in intracellular concentration, we asked whether FBP was responsible for structural rearrangements of other proteins. We compared LiP patterns of lysates of cells grown in ethanol in the absence and upon addition of FBP. Addition of FBP significantly (q value < 0.01) changed LiP patterns of various proteins (Supplementary Table 15). Among the top ten candidate FBP targets, based on the extent of their structural rearrangement (Supplementary Table 15), were Cdc19, Fba1 (of which FBP is a natural substrate) and four other enzymes that were known to be directly regulated by FBP in other species (Fas1, Fas2, Acc1 and Icl1)37. Other known FBP interactors (Mdh1, Mdh2, Pfk1, Pfk2 and Pgi1) were also detected, though with less pronounced structural changes37. The top candidate FBP interactor was the fatty acid synthase subunit 1, Fas1, with 23 LiP peptides changing abundance more than fivefold upon FBP addition (Supplementary Table 16). Most changes in the LiP pattern of Fas1 advance online publication  nature biotechnology

Articles

© 2014 Nature America, Inc. All rights reserved.

were replicated with a pure preparation of the native fatty acid synthase complex with and without FBP (Supplementary Table 16), thus confirming that FBP directly interacts with the yeast FAS complex. Thus, the LiP workflow enabled the detection of known (Cdc19) and novel (yeast Fas1) FBP-protein interactions. Structural transitions of 14-3-3 and prion-like proteins Nine proteins with structural properties dependent on growth conditions were rich in asparagine, glutamine or glycine and underwent pronounced structural transitions at regions enriched for repeats of these amino acids. These sequence motifs are typically found in proteins that form amyloid structures, and seven had previously been classified as yeast prions38. Two proteins with polyglutamine (polyQ) domains were isoforms of the yeast 14-3-3 protein, Bmh1 and Bmh2; the Q-rich, C-terminal regions of these proteins have unknown function39. Based on SRM measurements, the abundances of these two proteins did not change significantly, but seven LiP peptides changed significantly (q value < 0.01) between the two conditions, with the largest changes observed for four half-tryptic peptides from the C terminus (Fig. 6a and Supplementary Table 17). No substantial differences were observed in the subcellular distributions (Fig. 6b). To evaluate whether these proteins were important in the adaptation to the considered metabolic transition, we compared growth of Bmh1-deletion strains to that of wild-type strains in glucose- and ethanol-rich media. No significant differences were observed during glucose-based growth, but the ∆bmh1 strain clearly showed a growth defect upon switch to ethanol (Fig. 6c). This demonstrates that Bmh1 is important for the capability of yeast to grow in ethanol-rich media, thus providing further functional validation of the LiP results. DISCUSSION A complete description of a proteome and how it responds to perturbations cannot be contemplated without a global analysis of the structural features of proteins. Here we provide the first demonstration, to our knowledge, of the feasibility of analyzing protein structural changes on a large scale. Structural changes were measured directly in a complex proteome matrix without the need for sample enrichment, purification or protein labeling. The LiP-SRM approach can be applied in a targeted manner to analyze structural rearrangements of specific proteins or protein networks. This requires measuring all possible fully tryptic peptides of the target protein using SRM and comparing their intensities in different conditions. Putative half-tryptic nature biotechnology  advance online publication

a

*

2.9 ± 1.9

*

*

*

*

*

*

2.0 ± 0.9

*

*

* *

n.s.

0.3 ± 0.2

2–3

0.3 ± 0.2

>3

0.2 ± 0.1

c 280

*

2.0 ± 1.9

b

0.3 ± 0.1

2% Glu 2% Eth

Bmh1–GFP Bmh2–GFP

240 Scattered light (a.u.)

Scattered light (a.u.)

Figure 6  Structural transition of the yeast 14-3-3 protein Bmh1 upon a switch from glucose- to ethanol-based metabolism. (a) Amino acid sequence of yeast Bmh1. Bold indicates the sequence covered in SRM measurements. Abundance changes of Bmh1 peptides in ethanol relative to glucose growth conditions are indicated with the following color code: purple, no significant change; black, abundance change between two- and threefold; green, abundance change larger than or equal to threefold. Significant average fold changes ± s.d. are reported above each peptide sequence (fold change > 2, q value < 0.02, multiple SRM transitions per peptide, three biological replicates). Red asterisks (*), indicate shared peptides between Bmh1 and Bmh2. See also Supplementary Table 17. (b) Fluorescence microscopy images of yeast strains expressing GFP-tagged Bmh1 and Bmh2, grown in glucose and ethanol. (c) Growth curves of wild-type (wt) yeast (strain BY4711) and the ∆bmh1 strains grown in the glucose- or ethanol-based media. Upon glucose growth, a diauxic shift occurs after about 16 h and results in subsequent ethanol-based growth, reflecting the metabolic transition analyzed in our discovery phase. s.d. was obtained from biological triplicates and are depicted as faint shades around the data points (average of the triplicates) a.u., arbitrary unit.

240 200 2% Glucose wt ∆bmh1

160 0

10

20 30 Time (h)

40

200 2% Ethanol wt ∆bmh1

160 50

0

10

20 30 Time (h)

40

50

peptides generated by LiP can be measured for confirmation. Alternatively, the method can be applied in an unbiased manner on a large scale. This is achieved by a discovery phase using shotgun proteomics, followed by SRM validation. The approach is directly applicable to biological samples, though with a lower resolution than that of NMR or X-ray crystallography. It enables analysis of multiple proteins simultaneously and is compatible with the analysis of human specimens. The approach was validated by analysis of α-Syn, myoglobin and Cdc19 and comparison to orthogonal data from NMR, crystallography, circular dichroism, fluorescence and Fourier transform infrared spectroscopy. Genetic, phenotypic and in vitro experiments validated the functional relevance of structural transitions of Cdc19, Bmh1 and Fas1/Fas2 detected by the unbiased application of LiP. In yeast, our approach detected structural changes of proteins spanning a broad, but not complete, range of abundances. Using long separation gradients and chromatographic columns40 will likely expand the sensitivity to lower-abundance proteins. For more complex mammalian proteomes, identification of LiP peptides from low-abundance proteins will require the addition of appropriate sample-enrichment steps, such as a dendrimer-based enrichment to isolate peptides produced by LiP41. Based on preliminary experiments using human cell lines, the coverage of a mammalian proteome in the absence of enrichment is ~3,500 LiP cleavage sites and ~1,200 proteins. If a protein in a sample is represented by a population of different conformers, one limitation of LiP-SRM is that it produces an average description of the conformational properties of the protein and the features of the different conformers are not distinguished. This limitation also applies to classical proteomic approaches that measure the ensemble of post-translationally modified versions of a protein or its multiple protein complexes. If conformotypic peptides for the different protein conformations are derived from homogeneous protein samples, however, quantification of such markers in a heterogeneous sample enables an estimation of the conformational composition, as shown for myoglobin. Some proteins may not be susceptible to LiP cleavage in any of their conformations and will be refractory to analysis. This in turn implies that certain proteins for which we did not detect a change in the LiP pattern are not necessarily conformationally invariant. 

© 2014 Nature America, Inc. All rights reserved.

Articles Use of multiple digestion times may mitigate this issue. The method does require extraction and handling of a proteome under nondenaturing conditions, which restricts its use to soluble proteins. Application of LiP to membrane preparations to characterize membrane-embedded proteins will require further protocol optimization. In the discovery phase, our approach detects protein structural alterations based on their effect on LiP patterns. These include pure protein conformational changes, differential post-translational modifications and binding events. Post-translational modifications and binding of interactors often drastically affect the conformational properties of a protein and thus these three types of structural events are tightly interconnected. A recent analysis of pairs of protein crystal structures in the PDB, with and without a bound small molecule, revealed that binding of small molecules mostly resulted in significant structural motions in the protein substrate 42. Similar results were obtained for protein-protein interactions43,44. As demonstrated in our analysis of Cdc19, the detected structural alterations may be further evaluated in a targeted phase, where the cause of the structural change is unraveled. In principle, post-translational modifications may affect proteolytic patterns due to steric hindrance without necessarily affecting the structural properties of the substrate. In our workflow this is taken into account by analysis of the trypsin-only control sample, in which substantial changes in the post-translational modification status of a protein can be deduced based on abundance changes of some of the associated tryptic peptides. Our application of LiP to the metabolic transition is to our knowledge the first systematic analysis in which the structural features of >1,000 proteins were probed simultaneously. It allowed unraveling of a system property of carbon metabolism and identification of interesting proteins for functional follow-up (e.g., Cdc19, Fas1/2 and 14-3-3 proteins). Our analysis revealed that glycolytic enzymes are controlled by structural transitions associated with reversion or inactivation of the corresponding reaction fluxes. The increase in flux through the TCA cycle during ethanol growth was instead achieved by a coordinated increase in the abundance of TCA cycle enzymes. Regulation of central carbon metabolism is therefore highly modular, with branches predominantly regulated at the transcriptional level and others characterized by a more complex regulation including enzyme structural rearrangements. The conformational changes of glycolytic enzymes could be driven by differential post-translational modifications, binding of small molecules, or formation of macromolecular complexes and inclusions45. Our microscopy data excluded the dynamic formation of cytosolic assemblies. Previous phosphoproteomic measurements, acquired under the same conditions, detected changes in the phosphorylation pattern of two (Pfk1 and Gpm1) of these glycolytic enzymes46. A few of the observed conformational changes in the yeast proteome were dependent on the FBP concentration. Our analysis of Cdc19, which was susceptible to LiP when inactive and resistant to LiP when active, demonstrated that conformotypic peptides, once identified, can be used to assess the activation state of an enzyme in cell extracts. Further, the unbiased application of LiP to proteomes treated with FBP enabled the identification of novel, candidate protein–FBP interactions, such as that of Fas1/2, validated in vitro using the purified enzymatic complex. Thus, LiP patterns can be used to identify proteins that bind a given small molecule, as ­previously proposed for the DARTS method47, and can pinpoint the binding pocket. In the absence of orthogonal validation, caveats of these analyses are the detection of unspecific or artifactual interactions. For example, loss of subcellular compartmentalization during cell lysis will release proteins that are normally noncytosolic and thus normally not accessible to cytosolic metabolites. Other limitations 

include the lack of sufficient sensitivity to access low-abundance interactors or the risk of monitoring the effect of secondary metabolites of the administered molecule produced by enzymatic activities in the cell extract. 14-3-3 proteins underwent structural changes upon the metabolic switch. Their Q-rich C termini underwent a selective rearrangement to a more rigid and protease-resistant structure upon the transition from glycolytic to gluconeogenic conditions and Bmh1 was revealed to be important for yeast to efficiently switch to gluconeogenic metabolism in ethanol. Further studies will be required to establish whether this transition involves the formation of oligomers or binding of 14-3-3 interactors. No changes in abundance of 14-3-3 proteins or transcripts30 or protein phosphorylation46 were detected upon this metabolic transition. Therefore, LiP-SRM identified proteins involved in the response to the metabolic transition that were not detected by other proteomic techniques. Our analysis yielded conformotypic peptides for the soluble and disease-associated conformations of α-Syn. If analytical sensitivity allows, applying these markers to biological fluids from Parkinson’s disease patients might reveal whether the ratio between fibrillar versus monomeric α-Syn is modified in Parkinson’s disease and allow the exploration of the novel concept of ‘conformational biomarkers’ as, for example, opposed to classical, concentration-based disease biomarkers. In summary, the LiP-SRM approach couples proteolytic probes and a targeted proteomic workflow to probe both pronounced and subtle structural transitions of many proteins simultaneously. Protein structural changes can be analyzed directly in complex biological samples, such as cell or blood extracts on a large scale without the need for protein purification or enrichment. Methods Methods and any associated references are available in the online version of the paper. Accession codes. PeptideAtlas: PASS00538, PASS00539, PASS00540. Note: Any Supplementary Information and Source Data files are available in the online version of the paper. Acknowledgments P.P. is supported by a ‘Foerderungsprofessur’ grant from the Swiss National Science Foundation (grant PP00P3_133670), by an EU Seventh Framework Program Reintegration grant (FP7-PEOPLE-2010-RG-277147) and by a Promedica Stiftung (grant 2-70669-11). Y.F. is supported by an ETH Research Grant (grant 4412-1); M.S. is supported by a Natural Sciences and Engineering Research Council of Canada Postgraduate Scholarship D award. G.D.F. is supported by a post-doctoral fellowship of the University of Padua. A.P.O. is supported by the SystemsX.ch project YeastX. We thank R. Costenoble, K. Kochanowski and U. Sauer (ETH Zurich) for insightful discussions and for the measurements of the intracellular concentrations of FBP and M. Peter for access to plasmid and strain collections. We are grateful to O. Vitek and M. Choi (Purdue University), R. Riek, C. Chi and P. Navarro (ETH Zurich) for helpful discussions. We also thank P. Nanni and R. Schlapbach from the Functional Genomics Centre Zurich for access to mass spectrometry instrumentation, F. Allain for access to the D-BIOL Biomolecular NMR Spectroscopy Platform at the ETH Zurich, N. Ban and M.A. Leibundgut for providing a sample of purified yeast fatty acid synthase. AUTHOR CONTRIBUTIONS P.P. conceived and supervised the project. Y.F., G.D.F. and P.P. designed and performed the experiments. Y.F., G.D.F., M.S. and A.M. performed experiments and analyzed the data. P.B. contributed to mass spectrometry measurements. P.P.d.L. supervised parts of the project. A.K. analyzed the data. A.P.O. and Y.N. contributed to the analysis and validation of the metabolic data. P.P., Y.F., G.D.F. and A.K. wrote the manuscript.

advance online publication  nature biotechnology

Articles COMPETING FINANCIAL INTERESTS The authors declare competing financial interests: details are available in the online version of the paper.

© 2014 Nature America, Inc. All rights reserved.

Reprints and permissions information is available online at http://www.nature.com/ reprints/index.html.

1. Aebersold, R. & Mann, M. Mass spectrometry-based proteomics. Nature 422, 198–207 (2003). 2. Mischerikow, N. & Heck, A.J. Targeted large-scale analysis of protein acetylation. Proteomics 11, 571–589 (2011). 3. Olsen, J.V. et al. Global, in vivo, and site-specific phosphorylation dynamics in signaling networks. Cell 127, 635–648 (2006). 4. Goh, C.S., Milburn, D. & Gerstein, M. Conformational changes associated with protein-protein interactions. Curr. Opin. Struct. Biol. 14, 104–109 (2004). 5. Wrabl, J.O. et al. The role of protein conformational fluctuations in allostery, function, and evolution. Biophys. Chem. 159, 129–141 (2011). 6. Rabiller, M. et al. Proteus in the world of proteins: conformational changes in protein kinases. Arch. Pharm. (Weinheim) 343, 193–206 (2010). 7. Fink, A.L. Natively unfolded proteins. Curr. Opin. Struct. Biol. 15, 35–41 (2005). 8. Korn, E.D., Carlier, M.F. & Pantaloni, D. Actin polymerization and ATP hydrolysis. Science 238, 638–644 (1987). 9. Cox, S., Radzio-Andzelm, E. & Taylor, S.S. Domain movements in protein kinases. Curr. Opin. Struct. Biol. 4, 893–901 (1994). 10. Bu, Z. & Callaway, D.J. Proteins move! Protein dynamics and long-range allostery in cell signaling. Adv. Protein Chem. Struct. Biol. 83, 163–221 (2011). 11. Chiti, F. & Dobson, C.M. Protein misfolding, functional amyloid, and human disease. Annu. Rev. Biochem. 75, 333–366 (2006). 12. Eftink, M.R. Fluorescence techniques for studying protein structure. Methods Biochem. Anal. 35, 127–205 (1991). 13. Ilari, A. & Savino, C. Protein structure determination by x-ray crystallography. Methods Mol. Biol. 452, 63–87 (2008). 14. Pelton, J.T. & McLean, L.R. Spectroscopic methods for analysis of protein secondary structure. Anal. Biochem. 277, 167–176 (2000). 15. Heyduk, T. Measuring protein conformational changes by FRET/LRET. Curr. Opin. Biotechnol. 13, 292–296 (2002). 16. Sakakibara, D. et al. Protein structure determination in living cells by in-cell NMR spectroscopy. Nature 458, 102–105 (2009). 17. Herzog, F. et al. Structural probing of a protein phosphatase 2A network by chemical cross-linking and mass spectrometry. Science 337, 1348–1352 (2012). 18. Fontana, A. et al. Probing protein structure by limited proteolysis. Acta Biochim. Pol. 51, 299–321 (2004). 19. Aceto, A. et al. Analysis by limited proteolysis of domain organization and GSH-site arrangement of bacterial glutathione transferase B1–1. Int. J. Biochem. Cell Biol. 27, 1033–1041 (1995). 20. Dieckmann, R., Pavela-Vrancic, M., von Dohren, H. & Kleinkauf, H. Probing the domain structure and ligand-induced conformational changes by limited proteolysis of tyrocidine synthetase 1. J. Mol. Biol. 288, 129–140 (1999). 21. Peng, Z.Y. & Kim, P.S. A protein dissection study of a molten globule. Biochemistry 33, 2136–2141 (1994). 22. Polverino de Laureto, P. et al. Protein aggregation and amyloid fibril formation by an SH3 domain probed by limited proteolysis. J. Mol. Biol. 334, 129–141 (2003).

nature biotechnology  advance online publication

23. Picotti, P., Bodenmiller, B., Mueller, L.N., Domon, B. & Aebersold, R. Full dynamic range proteome analysis of S. cerevisiae by targeted proteomics. Cell 138, 795–806 (2009). 24. Uversky, V.N. Neuropathology, biochemistry, and biophysics of alpha-synuclein aggregation. J. Neurochem. 103, 17–37 (2007). 25. De Franceschi, G. et al. Molecular insights into the interaction between alphasynuclein and docosahexaenoic acid. J. Mol. Biol. 394, 94–107 (2009). 26. Vilar, M. et al. The fold of alpha-synuclein fibrils. Proc. Natl. Acad. Sci. USA 105, 8637–8642 (2008). 27. Kendrew, J.C. et al. A three-dimensional model of the myoglobin molecule obtained by x-ray analysis. Nature 181, 662–666 (1958). 28. Eliezer, D. & Wright, P.E. Is apomyoglobin a molten globule? Structural characterization by NMR. J. Mol. Biol. 263, 531–538 (1996). 29. Chechik, G. et al. Activity motifs reveal principles of timing in transcriptional control of the yeast metabolic network. Nat. Biotechnol. 26, 1251–1259 (2008). 30. DeRisi, J.L., Iyer, V.R. & Brown, P.O. Exploring the metabolic and genetic control of gene expression on a genomic scale. Science 278, 680–686 (1997). 31. Ghaemmaghami, S. et al. Global analysis of protein expression in yeast. Nature 425, 737–741 (2003). 32. Zhang, Q.C. et al. Structure-based prediction of protein-protein interactions on a genome-wide scale. Nature 490, 556–560 (2012). 33. Apweiler, R. et al. UniProt: the Universal Protein knowledgebase. Nucleic Acids Res. 32, D115–D119 (2004). 34. Costenoble, R. et al. Comprehensive quantitative analysis of central carbon and amino-acid metabolism in Saccharomyces cerevisiae under multiple conditions by targeted proteomics. Mol. Syst. Biol. 7, 464 (2011). 35. Flikweert, M.T. et al. Pyruvate decarboxylase: an indispensable enzyme for growth of Saccharomyces cerevisiae on glucose. Yeast 12, 247–257 (1996). 36. Xu, Y.F. et al. Regulation of yeast pyruvate kinase by ultrasensitive allostery independent of phosphorylation. Mol. Cell 48, 52–62 (2012). 37. Kirtley, M.E. & McKay, M. Fructose-1,6-bisphosphate, a regulator of metabolism. Mol. Cell. Biochem. 18, 141–149 (1977). 38. Halfmann, R. & Lindquist, S. Epigenetics in the extreme: prions and the inheritance of environmentally acquired traits. Science 330, 629–632 (2010). 39. Veisova, D. et al. The C-terminal segment of yeast BMH proteins exhibits different structure compared to other 14–3-3 protein isoforms. Biochemistry 49, 3853–3861 (2010). 40. Thakur, S.S. et al. Deep and highly sensitive proteome coverage by LC-MS/MS without prefractionation. Mol. Cell. Proteomics 10, M110 003699 (2011). 41. Kleifeld, O. et al. Isotopic labeling of terminal amines in complex samples identifies protein N-termini and protease cleavage products. Nat. Biotechnol. 28, 281–288 (2010). 42. Amemiya, T., Koike, R., Fuchigami, S., Ikeguchi, M. & Kidera, A. Classification and annotation of the relationship between protein structural change and ligand binding. J. Mol. Biol. 408, 568–584 (2011). 43. Hwang, H., Vreven, T., Janin, J. & Weng, Z. Protein-protein docking benchmark version 4.0. Proteins 78, 3111–3114 (2010). 44. Zacharias, M. Accounting for conformational changes during protein-protein docking. Curr. Opin. Struct. Biol. 20, 180–186 (2010). 45. Narayanaswamy, R. et al. Widespread reorganization of metabolic enzymes into reversible assemblies upon nutrient starvation. Proc. Natl. Acad. Sci. USA 106, 10147–10152 (2009). 46. Oliveira, A.P. et al. Regulation of yeast central metabolism by enzyme phosphorylation. Mol. Syst. Biol. 8, 623 (2012). 47. Lomenick, B. et al. Target identification using drug affinity responsive target stability (DARTS). Proc. Natl. Acad. Sci. USA 106, 21984–21989 (2009).



© 2014 Nature America, Inc. All rights reserved.

ONLINE METHODS

Expression, purification and preparation of human α-Syn. Human α-Syn was expressed in Escherichia coli cells (strain BL21, DE3) transfected with the pET28b/α-Syn plasmid. Cells were grown in Luria-Bertani medium at 37 °C to an OD600 of 0.6, followed by induction with 0.5 mM isopropyl β-thiogalactopyranoside for 4 h. Purification of the recombinant protein was conducted as described previously25. The purity of the sample was assessed by reversed phase–high-performance liquid chromatography and the identity of the eluted material was confirmed by mass spectrometry. The purified α-Syn was lyophilized and stored at −20 °C. Monomeric α-Syn was obtained by dissolving the lyophilized protein in PBS buffer (10 mM Na2HPO4, 150 mM NaCl, 2 mM NaH2PO4, pH 7.4). Pre­existing protein aggregates were removed by centrifugation at 12,000g for 3 min and subsequent filtration with a Millex-GV, low protein binding Durapore (PVDF) membrane (Millipore, Billerica, MA, USA). Amyloid-like fibrils of α-Syn were obtained by incubating the protein in PBS at 37 °C for 15 d at a protein concentration of 1 mg/ml, under agitation at 500 r.p.m. with a Thermo-mixer (Eppendorf, Hamburg, Germany). Aliquots of the sample were subjected to the thioflavin T (ThT) binding assay, to circular dichroism and Fourier transform infrared spectroscopy measurements and were imaged by transmission electron microscopy (TEM) to confirm the formation of amyloid-like fibrils. For LiP experiments, fibrils were isolated by ultracentrifugation (380,000g, 90 min, 4 °C) and resuspended in PBS buffer. ThT assay. The ThT binding assay was performed as described previously48 using a 25 µm ThT solution in 25 mM sodium phosphate, pH 6.0. Aliquots (30 µl) of the protein samples containing amyloid aggregates were diluted into the ThT buffer and fluorescence emission was measured at 25 °C with an excitation wavelength of 440 nm. Fluorescence emission was recorded at 484 nm on a Cary Eclipse Fluorescence Spectrophotometer (Agilent Technologies Inc., Loveland, CO, USA). Transmission electron microscopy. Aliquots of the protein aggregation mixture were examined by TEM with negative staining. A droplet of the sample was placed on a Butvar-coated copper grid (400-square mesh, TAAB-Laboratories Equipment Ltd., Berks, UK) or a carbon-coated copper grid (400-square mesh, Quantifoil Micro Tools GmbH, Jena, Germany), dried, and negatively stained with a droplet of uranyl acetate solution (1%, w/v). TEM pictures were taken on a Tecnai G2 12 Twin (FEI Company, Hillsboro, OR, USA) or on a Philips CM12 microscope (Philips, Amsterdam, Netherlands) operating at an excitation voltage of 100 kV. Circular dichroism and Fourier transform infrared spectroscopy. Circular dichroism spectra were recorded on a J-710 spectropolarimeter (Jasco, Tokyo, Japan). Far-UV CD spectra were recorded using a 1 mm path-length quartz cell and a protein concentration of 0.1 mg/ml. The mean residue ellipticity [θ] (deg·cm2·dmol−1) was calculated from the formula [θ] = (θobs/10)·(MRW/lc), where θobs is the observed ellipticity in degrees, MRW is the mean residue molecular weight of the protein, l the optical path length in cm and c the protein concentration in g/ml. The spectra were recorded in PBS buffer, pH 7.4. Fourier transform infrared spectroscopy measurements were conducted on a 1,720× spectrometer (PerkinElmer Life Sciences, Waltham, MA, USA). α-Syn was suspended in D2O for 1 h to allow the hydrogen-to-deuterium exchange and then was lyophilized under vacuum. The spectrum of α-Syn in solution was recorded after dissolving the deuterated protein in 20 mM Tris·DCl, 150 mM NaCl, pH* 7.2 (uncorrected for isotopic effects) at a concentration of ~5 mg/ml. For the sample corresponding to α-Syn incubated for 15 d, the aggregated species was isolated by ultracentrifugation (380,000g) and redissolved in Tris-DCl buffer at 5 mg/ml. Samples were placed between CaF2 windows, separated by a 50-µm thick Mylar spacer. The samples and the detector compartment were thoroughly purged with N2. Spectra represent averages of 50 scans recorded between 4,000 and 1,000 cm−1 at a resolution of 2 cm−1. All spectra were base-line-corrected, blank-subtracted, and smoothed using a Savitzky-Golay filter by Grams 32 program version 4.14 (Galactic Industries Corp., Salem, NH). The second derivative of the amide I band was used to identify the different spectral components.

nature biotechnology

Myoglobin samples. Horse heart myoglobin (Sigma) was solubilized in 20 mM Hepes buffer, 150 mM KCl, pH 7.5. The solution was centrifuged at 16,000g to remove potential protein aggregates. Protein concentration was determined on the basis of heme by measuring the absorbance of the solution in the Soret region49. Apomyoglobin was obtained from holomyoglobin by removal of the heme group using the 2-butanone extraction procedure50. The possible contamination of apomyoglobin by myoglobin was assessed spectrophotometrically. No significant absorption was observed in the Soret region. Myoglobin was solubilized in 20 mM Hepes buffer, 150 mM KCl, pH 7.5 and the concentration of the apoprotein was determined by measuring the UV absorbance of the solution at 280 nm. The holo or apo form of myoglobin was spiked into a yeast background proteome (up to six replicate spike-in experiments) at a concentration of 1, 5 or 10 pmol per µg of total yeast proteins for the subsequent LiP analysis. Purification of yeast pyruvate kinase and fatty acid synthase. Pyruvate kinase from S. cerevisiae (strain BY4741) with a C-terminal streptavidin tag was expressed in E. coli (Rosetta expression strain) transfected with the pET17b plasmid. Cells were grown in LB medium containing 2% glucose, lysed by a French press and after removal of cell debris, the supernatant was used to purify the tagged protein on a pre-packed strep-tactin superflow plus Qiagen-30060 (Hilden, Germany). The purity of the sample was assessed by SDS-PAGE as well as mass spectrometry. Fatty acid synthase from S. cerevisiae was purified according to procedures described in Leibundgut et al.51 and ETH Dissertation no. 19086 (ref. 52). Yeast culture. S. cerevisiae cells (strain BY4741) for the use as background proteome were grown at 30 °C in synthetic complete (SC) liquid medium 53 containing 2% glucose (w/v) to an OD600 of ~0.8. Cells were harvested by centrifugation at 3,000g for 5 min at 4 °C and washed twice with an ice-cold lysis buffer, including 20 mM Hepes, 150 mM KCl, pH 7.5. Cells were resuspended in the same buffer and disrupted by vortexing in the presence of acid-washed glass beads in three consecutive rounds of 10 min beating and 2 min incub­ ating at 4 °C. Yeast lysates were centrifuged at 16,000g for 5 min at 4 °C to remove cellular debris, the supernatants were transferred to a fresh tube, and the protein concentration in the extracts was determined by the bicinchoninic acid assay (BCA Protein Assay Kit, Thermo Scientific, Rockford, IL, USA). Proteins were precipitated by adding six volumes of cold acetone and were stored at −20 °C until use. In the metabolic shift experiments, S. cerevisiae cells (strain BY4741) were grown in either SC-glucose (2% w/v) or ethanol (2% v/v) liquid medium at 30 °C (three independent biological replicates) and sampled when the cultures reached an OD600 of 0.8. For LiP experiments, yeast cells were harvested from the liquid culture by centrifugation at 3,000g for 3 min at 4 °C. The cell pellets were washed once with PBS and once with 50 mM Hepes, 150 mM KCl, 1 mM MgCl2, 100 mM β-glycerophosphate, and 50 mM NaF, pH 7.3. Cells were resuspended in a minimal volume of the same buffer and lysed using a Freezer Mill (6870, SPEX CertiPrep Group, Metuchen, NJ, USA). Yeast lysates were centrifuged at 3,000g, for 2 min at 4 °C to remove cellular debris, the supernatants were transferred to a fresh tube, and the protein concentration in the extracts was determined by the BCA assay. Growth of wild-type yeast (BY4711) and of the ∆bmh1 strain (EUROSCARF collection, Frankfurt, Germany) were monitored in SC complete medium with 2% glucose (w/v) or ethanol (v/v) respectively. Cells were incubated in a 96-well microtiter plate at 30 °C under constant agitation in a Biolector (ABBIS, Wiesbaum, Germany) with 150 µl of medium per well. Scattered light was measured at 620 nm every 4 min for 72 h. Final growth curves were obtained by plotting the scattered light signal along the time axis. Limited proteolysis. The protein concentration of different cell extracts was adjusted to 2 mg/ml with 20 mM Hepes, 150 mM KCl and 10 mM MgCl2. The pH was adjusted with 5 M NaOH to ~7.5. For the spike-in experiments, α-Syn or myoglobin was added to the background proteome. No purified protein was added in the metabolic shift experiment. Proteinase K from Tritirachium album (Sigma) was added to the protein extract at an enzyme/substrate (E:S) ratio of 1:100 (w/w) and incubated for 5 min at room temperature. The digestion was

doi:10.1038/nbt.2999

© 2014 Nature America, Inc. All rights reserved.

stopped by transferring the reaction mixture to a tube containing guanidine hydrochloride crystals to a final concentration of 7.4 M and by boiling for 3 min. The digestion mixtures were then subjected to complete tryptic digestion as described below. Time course LiP experiments were done using myoglobin samples as described above, with incubation times of 1, 5, 30 and 60 min, respectively, and a fixed E:S ratio of 1:100. Similarly, three different E/S ratios (1:10, 1:100 and 1:1,000) were assessed with a fixed reaction time of 5 min. For the stoichiometry experiment, different mixes of apomyoglobin and holomyoglobin were prepared, containing 90, 50 and 10% of apomyoglobin, respectively, spiked into replicates of a yeast background proteome at a fixed total myoglobin amount per µg of yeast proteins. Samples containing only apomyoglobin or only holomyoglobin in the same amount were prepared as a reference. For LiP experiments in the presence of externally administered fructose1,6-bisphosphate (FBP), FBP was added at a final concentration of 0.33 nmol per µg of total protein in the lysate. LiP experiments were carried out on both the lysate after FBP addition and on control samples where no FBP was added, after confirming that the pH of the solution was invariant. In all LiP experiments the temperature during the LiP step was strictly controlled and kept constant. LiP experiments on purified Cdc19 and fatty acid synthase were performed in the same lysis buffer as above. Concentrations of the purified proteins were measured by a BCA assay. Tryptic digestion. Proteins were reduced with 12.5 mM dithiotreitol for 30 min at 37 °C and alkylated with 40 mM iodoacetamide for 45 min at 25 °C, in the dark. Samples were diluted with 0.1 M NH4HCO3 to a final concentration of 0.5 M guanidinium, and sequencing-grade porcine trypsin (Promega, Madison, WI, USA) was added to a final enzyme/substrate ratio of 1:100 (w/w). Tryptic digestion was conducted at 32 °C. After 2 h, the same amount of trypsin was added again, and the digestion mixture was incubated overnight at 32 °C. The digestion was stopped by acidification with formic acid to a final pH < 3. The peptide mixtures were loaded onto Sep-Pak tC18 cartridges (Waters, Milford, MA, USA) desalted and eluted with 80% acetonitrile. All peptide samples were evaporated on a vacuum centrifuge to dryness, resolubilized in 0.1% formic acid and immediately analyzed by mass spectrometry. Shotgun proteomic analyses and large-scale relative quantitation. The peptide samples were measured on a LTQ-Orbitrap Velos or on a Q-Exactive Plus (QE+) mass spectrometer (Thermo Scientific) equipped with a nanoelectrospray ion source. Peptides were loaded and separated chromatographically on a 15-cm reversed-phase analytical column (ProntoSIL AQ, Bischoff GmBH, 3 µm) using a linear gradient from 2–35% acetonitrile in 140 min and a flow rate of 300 nl/min (Velos) or on a 30-cm long 75 µm i.d. column (Reprosil Pur C18 Aq, Dr. Maisch, 1.9 µm) with a linear gradient from 5–25% acetonitrile in 100 min and a flow rate of 300 nl/min (QE+). Precursor ion scans were performed at resolution 30,000 at 400 m/z (Velos) or 70,000 at 200 m/z (QE+). Twenty MS/MS spectra were acquired after collision induced dissociation in the linear ion trap (Velos) or after higher-energy collisional dissociation in the Orbitrap at resolution 17.500 at 200 m/z (QE+) per each fully tryptic-MS scan. One microscan was acquired per each MS/MS scan. Only peptide ions exceeding a threshold of 1,000 ion counts were allowed to trigger MS/MSscans, followed by dynamic exclusion for 25 s. Repeat count was set to 1. The collected spectra were searched against the S. cerevisiae SGD protein database with Sorcerer-SEQUEST (Thermo Electron, San Jose, CA, USA). Trypsin was set as the digesting protease with the tolerance of two missed cleavages, one nontryptic terminus and not allowing for cleavages of KP (lysine-proline) and RP (arginine-proline) peptide bonds. The monoisotopic peptide and fragment mass tolerances were set to 50 p.p.m. and 0.8 Da, respectively. Carboxyamidomethylation of cysteines (+57.0214 Da) was defined as fixed modification and oxidation of methionines (+15.99492) as a variable modification. Protein identifications were statistically analyzed with ProteinProphet (v3.0) and filtered to a cut-off of 0.9 ProteinProphet probability, which in this particular case corresponds to a FDR < 1%, calculated based on a target-decoy approach54. Initial label-free quantification data in the discovery phase were obtained by a spectral counting approach at the peptide level55,56 followed by SRM-based

doi:10.1038/nbt.2999

validation (Supplementary Note). The spectral count for each peptide was calculated as the sum of MS/MS spectra assigned to that peptide across three biological replicates. Spectral counts obtained for each peptide from the doubly digested samples were normalized to the spectral counts for the same peptide measured in the corresponding trypsin-only control sample, to correct for trypsin unspecificity and endogenous protease activity. Data were expressed as spectral counts from the proteome digests of yeast grown in ethanol with respect to yeast grown in glucose. Peptides were classified as fully tryptic or half-tryptic, based on the number of tryptic termini (two and one, respectively). To identify LiP cleavage sites in the yeast proteome, only half-tryptic peptides, unique to one protein and devoid of missed cleavages, with a spectral count fold change of at least twofold in the ethanol- relative to the glucosegrowth condition were considered. When a peptide was not detectable in one of the two conditions, it was considered only if it was detected in the other condition with at least two spectral counts. For the analysis of the structural properties of LiP cleavage sites in the proteome of yeast grown in glucose, we focused on a subset of half-tryptic peptides produced upon LiP, devoid of missed cleavages, that map to a unique protein and appeared upon proteinase K digestion with an intensity at least fivefold higher in the doubly digested sample, relative to the control. The target peptides were selected based on spectral count data and their precise abundance change assessed by multiplexed SRM measurements. To compare proteolytic patterns upon addition of FBP to the cell lysate in the discovery experiment, label-free quantitation of LiP-MS data acquired on the QE+ was performed using the Progenesis software (Nonlinear Dynamics, Version 4) essentially as described below. Acquired raw LC-MS/MS files were imported directly into Progenesis for analysis. The same raw files were converted into mzXML format with MSconvert57 and used for database search with Sorcerer-SEQUEST with parameters described above. The resulted pepXML files containing peptides identified in the search were filtered with a FDR of 1% based on decoy counts and subsequently imported into Progenesis. Peptide identifications from MS/MS spectra were mapped to the corresponding peptide ions detected in MS1 spectra, according to their accurate m/z and retention time and areas under the extracted ion chromatograms of each peptide were calculated. Based on this annotation process, a list containing quantified peptides was produced, which was used as input for the R-based SafeQuant script58 (version 1.1) to calculate the abundance changes between FBP-treated and control samples and their corresponding q values. Further, we retained only proteotypic peptides changing abundance at least fivefold and with a q value < 0.02, mapping to proteins that could be localized in the cytosol (where FBP is produced) based on localization information from the Uniprot and SGD databases and that showed a change in their LiP pattern upon the glucose to ethanol metabolic transition. Protein hits were ranked based on the number of significantly changing fully tryptic and halftryptic peptides. SRM measurements. For each target peptide we selected a set of SRM transitions consisting of doubly or triply charged precursor ions and singly charged fragment ions of the y-series. For peptides for which no fragment ion spectrum was available, we selected all possible fragment ions of descending m/z, starting from 1,250 Da, using Skyline59 (v1.3 and v1.4, MacCoss Lab Software, Seattle, WA, USA). For peptides observed in our shotgun proteomic analysis, we selected up to eight intense fragment ions from the fragment ion spectra recorded for each peptide. For assay refinement, we experimentally tested the set of transitions in SRM mode and selected up to five most intense transitions devoid of obvious interferences for quantitation experiments. We also annotated the peptide retention times for scheduled SRM acquisition. Samples were measured on a triple quadrupole/ion trap mass spectrometer (5500 QTrap, ABSciex) equipped with a nano-electrospray ion source and operated in SRM mode. On-line chromatographic separation of the peptides was achieved with a Eksigent 1D-plus Nano liquid chromatography system (Eksigent/ABSciex, Foster City, CA, USA) equipped with a 18-cm fused silica column with 75-µm inner diameter (New Objective, Woburn, MA, USA), packed in-house with Magic C18 AQ 5-µm beads (Michrom Bioresources, Leonberg, Germany). The peptide mixtures (~3 µg) were loaded from a autosampler cooled to 4 °C (Eksigent/ABSciex) and separated with a linear gradient from 1 to 35% or from 5 to 35% acetonitrile in 30 min. SRM analysis was conducted with Q1

nature biotechnology

© 2014 Nature America, Inc. All rights reserved.

and Q3 operated at unit resolution (0.7 m/z half maximum peak width) with up to 80 transitions per run (dwell time: 20–50 ms; cycle time