Resource

Features of the Chaperone

Cellular NetworkRevealed through Systematic Interaction Mapping

Graphical Abstract

Highlights

d Chaperone interaction map elucidated through integrated

proteomic approaches

d A large functional chaperone supercomplex, named NAJ

complex, is revealed

d Many chaperone interactors are found to form condensates

d The AAA+ ATPases Rvb1 and Rvb2 form condensates under

nutrient starvation

Rizzolo et al., 2017, Cell Reports 20, 2735–2748September 12, 2017 ª 2017 The Author(s).http://dx.doi.org/10.1016/j.celrep.2017.08.074

Authors

Kamran Rizzolo, Jennifer Huen,

Ashwani Kumar, ..., Charles Boone,

Mohan Babu, Walid A. Houry

Correspondencemohan.babu@uregina.ca (M.B.),walid.houry@utoronto.ca (W.A.H.)

In Brief

Rizzolo et al. use a systematic integrative

approach combining physical and

genetic interaction data to construct a

comprehensive chaperone network. This

analysis revealed the presence of a large

functional chaperone supercomplex, the

NAJ complex. Furthermore, many

chaperone interactors were found to form

condensates.

Cell Reports

Resource

Features of the Chaperone Cellular NetworkRevealed through Systematic Interaction MappingKamran Rizzolo,1,11 Jennifer Huen,1,11 Ashwani Kumar,2,11 Sadhna Phanse,3,4 James Vlasblom,4 Yoshito Kakihara,1,12

Hussein A. Zeineddine,1,13 Zoran Minic,4 Jamie Snider,3 Wen Wang,5,6 Carles Pons,7 Thiago V. Seraphim,1,4

Edgar Erik Boczek,8 Simon Alberti,8 Michael Costanzo,3 Chad L. Myers,5,6 Igor Stagljar,1,3,9 Charles Boone,3,9

Mohan Babu,4,* and Walid A. Houry1,10,14,*1Department of Biochemistry, University of Toronto, Toronto, ON M5G 1M1, Canada2Department of Computer Science, University of Regina, Regina, SK S4S 0A2, Canada3The Donnelly Centre, University of Toronto, Toronto, ON M5S 3E1, Canada4Department of Biochemistry, University of Regina, Regina, SK S4S 0A2, Canada5Department of Computer Science & Engineering, University of Minnesota-Twin Cities, Minneapolis, MN 55455, USA6Program in Bioinformatics and Computational Biology, University of Minnesota-Twin Cities, Minneapolis, MN 55455, USA7Institute for Research in Biomedicine (IRB Barcelona), The Barcelona Institute for Science and Technology, Barcelona, Catalonia, Spain8Max Planck Institute of Molecular Cell Biology and Genetics, Dresden, Germany9Department of Molecular Genetics, University of Toronto, Toronto, ON M5S 3E1, Canada10Department of Chemistry, University of Toronto, Toronto, ON M5S 3H6, Canada11These authors contributed equally12Present address: Division of Dental Pharmacology, Niigata University Graduate School of Medical and Dental Sciences,

2-5274 Gakkocho-dori, Chuo-ku, Japan13Present address: Department of Neurosurgery, The University of Texas Medical School at Houston, Houston, TX, USA14Lead Contact

*Correspondence: mohan.babu@uregina.ca (M.B.), walid.houry@utoronto.ca (W.A.H.)http://dx.doi.org/10.1016/j.celrep.2017.08.074

SUMMARY

A comprehensive view of molecular chaperone func-tion in the cell was obtained through a systematicglobal integrative network approach based onphysical (protein-protein) and genetic (gene-geneor epistatic) interaction mapping. This allowed us todecipher interactions involving all core chaperones(67) and cochaperones (15) of Saccharomyces cere-visiae. Our analysis revealed the presence of alarge chaperone functional supercomplex, whichwe named the naturally joined (NAJ) chaperone com-plex, encompassing Hsp40, Hsp70, Hsp90, AAA+,CCT, and small Hsps. We further found that manychaperones interact with proteins that form focior condensates under stress conditions. Using anin vitro reconstitution approach, we demonstratecondensate formation for the highly conservedAAA+ ATPases Rvb1 and Rvb2, which are part ofthe R2TP complex that interacts with Hsp90. Thisexpanded view of the chaperone network in the cellclearly demonstrates the distinction between chap-erones having broad versus narrow substrate speci-ficities in protein homeostasis.

INTRODUCTION

Molecular chaperones are a highly interactive group of cellular

proteins that fulfill central roles in all aspects of protein homeo-

Cell ReportThis is an open access article und

stasis, including protein folding, assembly, and unfolding of

substrates. Chaperones are typically categorized into families

based on their sequence similarity and function. Proteins of

the major chaperone families in the budding yeast Saccharo-

myces cerevisiae are 2 Hsp90s, 14 Hsp70s, 22 Hsp40s, 8

CCTs, 1 Hsp60, 1 Hsp10, 6 prefoldins, 5 ATPases associated

with diverse cellular activities (AAA+), 7 small heat shock pro-

teins (sHsps), and 1 calnexin (total of 67; Table S1; Figure 1A).

In addition, the Hsp70s and Hsp90s function with 4 and 11

partner proteins, respectively, that are termed cochaperones

(total of 15; Table S1; Figure 1A). Despite many mechanistic

and functional studies, the spectrum of cellular substrates

and many of the functions mediated by these chaperones

remain largely undetermined.

Given their involvement in many cellular protein homeostasis

processes, it is necessary to study chaperones at a global

systems level. In 2005, our group published a comprehensive

physical (protein-protein) and genetic (gene-gene or epistatic)

analysis of the yeast Hsp90 chaperone interaction network,

showing the broad role this central chaperone has in many

cellular pathways (Zhao et al., 2005). Subsequently, in 2009,

we published a physical interaction atlas for 63 yeast chaper-

ones (Gong et al., 2009). Here, genetic interaction data from syn-

thetic genetic array (SGA) technology in yeast (Costanzo et al.,

2016) was combined with protein-protein interactions (PPIs) to

systematically build a comprehensive high-fidelity chaperone

network from a total of 67 chaperones and 15 cochaperones.

The network revealed several features of the chaperone func-

tional distribution in the cell, and it indicated the presence of a

functional chaperone supercomplex required for cellular protein

homeostasis.

s 20, 2735–2748, September 12, 2017 ª 2017 The Author(s). 2735er the CC BY license (http://creativecommons.org/licenses/by/4.0/).

PCC

0.1

0.8

C

B

0.200-0.20

SGA score

ER

tra

nslo

cati

on S

EC

63

CN

E1

KA

R2

SC

J1

ECM33 ERD1 HAC1 IRE1 YFL032W GPI16 ERO1 MSL5 ADE16 PRP42 SPC3 ALG6 DIE2 YAL058C-A SSS1 YPL245W WBP1 SPC2 SEC62 SEC66 STT3

GIM

4 G

IM3

GIM

2 G

IM1

GIM

6

DAM1 ELP6 CCT1 ARP4 CCT4 CCT5 CCT6

CC

T c

hape

rone

s

CC

T4

CC

T1

CC

T5

BRX1 NOP2 QNS1 LCP5 TRF5 RRP15 NOP56 SSF1 DRS1 KRE29 PUP2 KAE1 RPL12A RRP14 RIA1 BRR2 PDR11 RPL11A DIA4 YOR012W RGI1 YER189W YHR086W-A YOR309C

Rib

osom

e bi

ogen

esis

CC

T6

62 chaperone and 15 cochaperone genes

4,58

3 ye

ast g

enes

A

2

14

22

1

86 5

7

1 14

11

Hsp90Hsp

70Hsp

40Hsp

60CCT

Prefold

inAAA+

sHsp

Calnex

inHsp

10

Hsp70

coch

apero

nes

Hsp90

coch

apero

nes

67 chaperones15 cochaperones

Num

ber o

f pro

tein

s Localization of CCo

D

threshold cutoff(|SGA score|>0.08 and P<0.05)

393,926 CCo GI profile correlation similarities

425,751 double mutants involving CCo(62 chaperones & 15 cochaperones)

22,443 final GIs (13,704 negative & 8,739 positive)

i

ii iii

CC

T3

CC

T4

CCT5

CCT6CCT7

CCT8

CCT1

GIM1

GIM2

GIM3

GIM4

GIM6APJ1

CAJ1CWC23

DJP1

HLJ1

JEM1JJJ1

JJJ3M

DJ2

PAM

18

SCJ1

SEC63SI

S1

SWA

2

XDJ1

YDJ1

SSC3KA

R2LHS1SS

A2

SSA3SSB2SSC1

SSE1

SSE2

SSZ1FES1

MGE1SIL1

SNL1HSP90-TS

HSC82

HSP82AHA1

CDC37

CNS1

CPR6

CPR7

HCH1

PIH1

PPT1

SBA1

STI1HSP104RVB1

RVB2HSP12

HSP31HSP32

HSP33H

SP42C

NE1

HSP10

HSP60M

CX1

Nucleus &Cytoplasm

ER

Mito10

13

59C

CT2

Figure 1. CCo Network Based on GI Data

(A) Bar plot showing the yeast chaperones and cochaperones divided into different families. The colors used for the bars for each CCo family are the same in all

the figures. On the right is a diagram showing the cellular localization of the CCos.

(B) Flow-gram showing the acquisition and analysis of the CCoGI data. CCoGIs were compiled and then GIs and GI profile correlation similarities were obtained.

GIs were filtered at an intermediate threshold (jSGA scorej > 0.08 and p < 0.05), and GI profile correlation similarities were computed. See also Figures S1A–S1D.

(C) Heatmap clustering generated from the GI dataset is shown where the 77 CCos are organized on the x axis as both query and array and 4,583 yeast genes on

the y axis. Below are three examples of GI clusters involving CCos: (i) GI cluster involving Sec63 (see also Figure S1E), (ii) GI cluster of the CCT and the prefoldin

families, and (iii) GI cluster showing predominantly positive GIs between CCT and proteins involved in ribosome biogenesis (see also Figures S2 and S3).

(D) Circos plot (Krzywinski et al., 2009) showing correlations in the GI profiles among CCos. CCo families are grouped and colored according to (A). Profiles were

calculated for all CCo gene pairs andmeasured based on Pearson correlation coefficients (PCCs) from the complete GImatrix. Pairs having a PCC> 0.1 threshold

are plotted. Ribbon width corresponds to the magnitude of the PCC value and the color of the ribbon corresponds to the color of the originating segment.

RESULTS

A Comprehensive High-Fidelity Genetic InteractionChaperone and Cochaperone NetworkThe genome-wide genetic interaction (GI) data to build the chap-

erone and cochaperone (CCo) GI network are described in the

Experimental Procedures, and they were acquired as part of

the effort to map the full GI network in yeast (Costanzo et al.,

2016). For essential genes (Giaever et al., 2002), both Decreased

Abundance by mRNA Perturbation (DAmP) strains (Schuldiner

et al., 2005) as well as temperature-sensitive (TS) alleles were

2736 Cell Reports 20, 2735–2748, September 12, 2017

used (Costanzo et al., 2010, 2016). The initial CCo GI raw data

with 425,751 double mutants were subsequently filtered with a

threshold score cutoff (jSGA scorej > 0.08 and p < 0.05) as

described (Costanzo et al., 2016), and they resulted in

22,443 high-confidence GIs where 13,704 were negative (SGA

score < 0; i.e., double mutants with a more severe fitness defect

than the expected multiplicative effect of combining the individ-

ual mutants, with the extreme case being synthetic lethality) and

8,739 were positive (SGA score > 0; i.e., double mutants with a

less severe defect in fitness than expected) (Table S2; Figure 1B).

The GI data included 62 chaperones and 15 cochaperones with

only 4 missing, namely, Ssa4, Ssc2, Mdj1, and Gim5 (note that

Ssa4, Ssc2, and Gim5 are present in the PPI dataset; see below).

CCo genes had an average of 293 GIs compared to 246 for non-

CCo genes (Figure S1A). Together, the current GI dataset con-

tains three times the number of GIs involving a CCo gene

compared to previous low- and high-throughput studies

compiled from BioGrid (Breitkreutz et al., 2008). Notably, this

comparison shows that our dataset includes almost double the

number of essential genes in yeast compared to previous studies.

To gain insights into the functional ranking of the chaperone

families, we looked at the total number and overlap of GIs among

CCos (Figures S1B–S1D). Hsp70, Hsp40, and Hsp90 chaperone

families tended to have the highest total number and the highest

unique number of GIs (Figures S1B and S1C). Also, the Hsp70

and Hsp40 family members had the highest number of GI inter-

actor overlap with other chaperones (Figure S1D). The Hsp70,

Hsp40, Hsp90, and to a lesser extent CCT members shared

many interactors (Figure S1D).

Figure 1C shows a global view of the positive and negative

CCo GIs encompassing 4,583 yeast genes. CCo genes within

a cluster enriched for a particular molecular pathway were

mostly connected by negative GIs (Figure 1C). For example,

genes encoding for proteins involved in endoplasmic reticulum

(ER) translocation that form a cluster with SEC63, CNE1,

KAR2, and SCJ1 were highly enriched for negative GIs (Fig-

ure 1C, lower panel i). To further verify some of these hits, we per-

formed membrane yeast two-hybrid assays (MYTH) (Snider and

Stagljar, 2016) with Sec63, the essential subunit of the ER trans-

locon containing a J domain, as a bait. Our results showed that

all 6 prey proteins interacting with Sec63 in MYTH (Figure S1E)

had a significant negative GI with SEC63, indicating that a nega-

tive GI with an essential gene is a likely indicator of physical inter-

action. These results further validated the GI data.

Genes that encode for proteins that form part of the same

complex tend to be biased toward a single type of GI (Baryshni-

kova et al., 2010; Costanzo et al., 2016). The type of GI depends

on the essentiality of the genes involved; typically, as mentioned

above, negative GIs were found with essential genes, which was

the case for CCos. For example, the CCT complex, which is

essential for cell viability, displayed mainly negative GIs with

the prefoldin (or GIM) complex (Figure 1C, lower panel ii). This

is consistent with the experimental evidence that CCT and pre-

foldin physically interact and function together to mainly fold

and assemble actin and tubulin in addition to many other sub-

strates (Dekker et al., 2008).

For the essential CCT complex, we observed striking positive

GIs with genes whose proteins are involved in ribosome biogen-

esis (Figure 1C, lower panel iii). This was also observed in stan-

dard liquid growth as shown in Figure S2A. The CCT4-TS strain

grew about 4 times slower at 30�C compared to the wild-type

(WT) strain. In comparison, a TRF5 gene knockout strain (Fig-

ure 1C, lower panel iii) exhibited a slightly slower growth profile

than WT, whereas the double-mutant strain grew better than

CCT4-TS, but not as well as trf5D or WT. TRF5 is part of the

Trf4/Air2/Mtr4 polyadenylation (TRAMP) complex that has a

poly(A) RNA polymerase activity and is involved in post-tran-

scriptional quality control mechanisms (Houseley and Tollervey,

2006). The absence of TRF5 leads to production of misas-

sembled ribosomes (Woolford and Baserga, 2013), which likely

causes a reduction in protein synthesis. This is indeed what we

observed when comparing total protein content of cells at the

same cell density for WT, single-mutant, and double-mutant

strains grown at 30�C to stationary phase (Figure S2B). The total

protein content was lowest for trf5D than for trf5DCCT4-TS dou-

ble-mutant strain, whileWT andCCT4-TS had similar higher total

protein (Figure S2C). Notably, the presence of insoluble proteins

was reduced in the double-mutant strain compared to CCT4-TS

or trf5D single-mutant strains (Figure S2D). Hence, we interpret

the presence of positive GI between the CCT complex and ribo-

some biogenesis genes as resulting from the fact that a CCT-TS

strain exhibits accumulation of toxic misfolded proteins, which is

mitigated by reducing the total number of proteins in the cell

through the reduction of ribosome biogenesis.

CCos Are Highly Biased in Positive GIs with EssentialGenesGiven the observations made with CCT, GIs of CCos against the

essential (i.e., TS or DAmP allele strains) and nonessential genes

were further evaluated. CCos were enriched in positive GIs

against the essential array when compared to the rest of the

tested genes (Figure S2E; p < 0.05). The positive:negative ratio

against the essential array was also found to be significantly

higher for CCos (Figure S2F; p < 0.05). This suggests that,

despite CCos having a larger number of negative GIs in the

network, they are more biased toward positive GIs with essential

genes compared to other genes in yeast.

By performing gene ontology (GO) bioprocess enrichment

analysis of the essential genes interacting with CCos with a

log2 positive:negative GI ratio above zero, significant cellular

processes involved in these interactions were identified (Fig-

ure S3; p < 0.05). Of the 22 CCos, 17 were found to be interacting

with genes that were significantly enriched in distinct bio-

processes (Figure S3). Some CCos displayed enrichments for

positive GIs with diverse bioprocesses, as observed for CCTs

and prefoldins. Other CCos had positive GIs with essential genes

belonging to one or more bioprocesses, such as SSC1with tran-

scription and chromatin organization and SWA2 with proteaso-

mal genes (discussed further below).

The CCo GI Profile Correlation Similarity NetworkThe GI profile of a given gene is composed of a set of positive

and negative GIs with other genes in the genome. Genes whose

GI profiles correlate tend to be part of the same complex or

function in similar pathways (Costanzo et al., 2016). This prop-

erty was used to look at the connectivity between CCos by build-

ing a GI profile correlation similarity network (Figure 1D; Table

S3). The network highlights the inherent functional organization

within and between the CCo families. Highly connected CCo

families form part of discernible biological processes, such as

the CCTs with the prefoldins or the Hsp40s with the Hsp70s (Fig-

ure 1D). We also found that the gene for the constitutive Hsp90

chaperone, HSC82, connects primarily with STI1 (known as

Hop in mammalian cells), consistent with the fact that Sti1 is

one of the main Hsp90 cofactors (Li et al., 2012). Interestingly,

HSC82 also connects with YDJ1, the main cytoplasmic Hsp40

(Figure 1D). Indeed, there are 24 genes that overlap between

Cell Reports 20, 2735–2748, September 12, 2017 2737

Enrichment P-value

1 10-8 10-16

A

BCCT family

CCT2CCT1

CCT4CCT5

CCT6CCT3

CCT7

CCT8

Bioprocesses

Hsp40:SEC63

Cell polarity & morphogenesis

Glycosylation, Protein folding/targeting,

Cell wall biogenesis

Vesicle transport

Ribosome biogenesis,rRNA & ncRNA processing

Respiration, Oxidative phosphorylation,

Mitochondrial targeting

Mitosis & chromosome segregation,

Transcription, DNA repair

Peroxisome

CPrefoldin family

GIM6GIM2

GIM1

GIM3GIM4

D

Hsp70: SSB2F AAA+: HSP104GHsp40: YDJ1E

CCo GI profile correlation similarity network

Figure 2. The GI Profile Correlation Similar-

ity Network of CCos

(A) The GI profile correlation similarity network, left

panel, was constructed using PCCs of GI profile

similarities (edges) for all CCo genes pairs (nodes).

CCos are color coded as in Figure 1A. CCo gene

pairs with profile similarity PCC > 0.1 were con-

nected and plotted using a spring-embedded

layout algorithm in Cytoscape. Genes that have a

similar GI profile are close to each other, and

genes with less similar GI profiles are positioned

farther apart. Subsequently, genes were anno-

tated using the SAFE program (Baryshnikova,

2016), right panel, identifying network regions en-

riched for similar GO bioprocess terms.

(B–G) Regions of the CCo similarity network

significantly enriched for genes with similar GI

profiles are shown forCCT (B),GIM (C), SEC63 (D),

YDJ1 (E), SSB2 (F), and HSP104 (G).

See also Figure S4.

the Hsp90/Hsp90 cochaperones and Ydj1 in the GI profile corre-

lation similarity network. This suggests that Ydj1 might be

involved in the proper regulation of diverse Hsp90 clients rather

than interacting promiscuously.

The GI Profile Correlation Similarity Network ProvidesNew Insights into CCo FunctionsWe built a GI profile correlation similarity network composed of

CCo-CCo and CCo-non-CCo correlation pairs evaluated using

a Pearson correlation coefficient (PCC) threshold >0.1, yielding

a total of 10,443 pairs (Table S3). The spatial analysis of func-

tional enrichment (SAFE) program (Baryshnikova, 2016) was

used to functionally annotate regions enriched for particular

GO cellular bioprocesses (Figure 2A). SAFE highlights regions

that are densely connected with a particular attribute, and, in

this case, the network was evaluated with 4,373 biological pro-

cess terms from GO (Ashburner et al., 2000). The CCo network

contained 402 significantly enrichedGO terms grouped into 7 re-

gions involving 726 genes (Figure 2A).

2738 Cell Reports 20, 2735–2748, September 12, 2017

The network provides insights into the

functional specialization of the CCos. In

general, CCos were found to have broad

specificity and to interact with a large

variety of substrates. Hence, many corre-

lated interactors tended to belong to mul-

tiple cellular processes, and they showed

no significant enrichment in any specific

bioprocesses in the network. However,

both the CCT and prefoldin chaperones

exhibited significant specialization in their

functions (Figures 2B and 2C). Many of

their interactors in the CCo GI profile cor-

relation similarity network were involved

in cell polarity, morphogenesis, mitosis,

chromosome segregation, glycosylation,

protein folding, and cell wall biogenesis

(Figures 2B and 2C). This specialization

might be due to the fact that CCT and prefoldins mainly func-

tion to fold and assemble actin and tubulin (Leroux and Hartl,

2000). Most other chaperones displayed either weak or no

specialization (Figures 2D–2G; Figure S4).

The Hsp90 NetworkThe enrichment landscape of genes correlated with Hsp90 and

its cofactors, including the R2TP (consisting of Rvb1, Rvb2,

Tah1, and Pih1) complex (Kakihara and Houry, 2012), is shown

in Figure 3A. Most of the Hsp90 CCos were dispersed

throughout the GI profile correlation similarity network, and

they typically fell outside the functionally enriched regions,

except for CPR7, CNS1, PIH1, PPT1, and RVB2 (Figure 3A).

To gain further insight into the Hsp90 family interactors, in

Figure 3B we show the GI profile correlation similarity network

consisting of a total of 1,303 genes interacting with Hsp90

and its cofactors. GO annotations for each interacting gene ob-

tained from the enriched regions in Figure 3A were manually

curated to a more specific biological term. The order of Hsp90

A

B

RVB1PPT1

RVB2

PIH1

CNS1

AHA1

TAH1

HCH1HSP82

SBA1

STI1

CDC37HSC82

CPR7

CPR6

Hsp90 family & R2TP

Cell polarity & Morphogenesis

Glycosylation, Protein folding/targeting,

Cell wall biogenesis

Vesicle transport

Ribosome biogenesis,rRNA & ncRNA processing

Respiration, Oxidative phosphorylation, Mitochondrial targeting

Peroxisome

Enrichment P-value

1 10-8 10-16 Mitosis & chromosome segregation, DNA replication & repair

Hsp90-TS

OtherProtein folding, PTM & Metabolism

Actin &Morphogenesis Chromatin & Nucleic acid Vesicle

traffickingRibosomal

& RNAMito Perox

CDS1

ATG29

RIB7

AAC3

PBI1

CST26

YPL276W

SMD1

SUB2

MRPL19

MRPL50

MRPL37

TIF34

GCD1

GCD2

PRP39

GCD7

ANT1

CTR1

SWA2

TAH18

BDH1

HAA1

COQ1 RPS18B

RPS4A

RPS16A

ASC1

RPS30A

RPS22B

RPS17A

RPS21A

RPS9A

RPS8A

RPS0B

PEX1

PEX14

PEX13

PEX3

PEX10

PEX5

PEX2

NMD2

DBP6

RSA3

MAK21

DCP2

NOC3

NOP8

RPL37B

RPL35A

RPL24A

RPP2B

RPL2A

RPL12A

RPL27B

BMT6

JJJ1

SSB2

MIP6

ARX1

MAK5

SBP1

RNA1

TRM1

NOG2 APS3

KXD1

APL6

VMA11

VMA8

PBI2

RAV2

TED1

LST7

PRY3

GVP36 YFR024C

YLR269C

YGL217C

YLR366W

YLR374C

YMR294W-A

YIL032C

AHT1

YLR217W

YIL025C

YLR123C

YKL200CAPM2MRM2

FCF1

APL4

APS1

NHP2

RRS1

APS2

YIP1

VMA3 MEP1 YMC2

STV1 BET1 YNL234W

GRH1 YSY6

SLY41

RCR1

PFY1

RSB1

ROX1

ISU2

UBC7

CRT10

FPR4

SFL1

FUN30

OTU1

KNS1

YGR122W

IXR1

CFD1

OPI1

RRN3

RFX1

TOM70PIK1

IMP2

TOM7ARO2

SCS3

ABZ1

TIM22

TIM23

MAS1

PTC5

CDC21

TIR4

PAU3

SRL1

NFI1

THI2

TBF1

SPN1

HSP42

SUA7

EDS1

CKS1

PHR1

YPL216W

SRL4

SNO3

WRS1

ERG26

COQ2

IRA2

BIO4

COQ8

APN1

RBA50

CAB1

SPF1

UTP30

FRD1

POL5

ADE6

MGL2

PMR1

POX1

ERG8

ERG25

DML1

CHO2YDC1

TSC10

LYS20

ALD6

THR4

BTS1

CHA1

UPS1

FIS1

CUP1-2

BNA5

SMF2

RRG1

MRS6

GDH2

ERG10

TRP1

KRS1

ATF1

BEM4

ATG21

COX11

ARO4

SHM1

IDI1

ILV5

ISF1

YHR020W

RRM3

LCB5

MIH1

WAR1

SEC39

VPS51

VPS53

SEC15

SEC6

SEC31

SEC24

SLM4

GTR2

SEC3

NPL3

RLI1

CCA1

TIS11

DXO1

EBS1

IPK1

PTH1

DBP2

VTS1

MAK11

DBP7

SQT1

BRX1

NOP12

DBP9

RBG2

FES1

REX4

COG6

COG3

GET2

GET3

COG4

COG7

GET1

COG2

COG8

COG5

TRS65

TRS23

BET3

VPS35

PEP8

VPS29

BET5

TRS31

SBE2

COF1

VTA1

BOS1

SED5

YDR338C

VBA5

SRP40

GOT1SEC22

YMR279CSLY1

LSB5YPT6

TVP15

PRM9

SEO1

SYN8

ARR3

SEC1

ENT5

ART10

VBA3

RSN1

RGT2

IWR1

SNF3

VTI1

ERP3

YML018C

YBR220C

FIT1

PIC2

GMH1

YKT6

STL1

COY1

LST4

VPS60

DBF20

CTF4

SPT10

GZF3

JIP5

YAL037W

YPR084W

URA6

GPX1

NOP16

TIR1

PHM8

MXR1

CDC8

CMR3

SPO24

YBL028C

AIM2

YPR078C

TDA8

TMC1

MGT1

FMP41

IFM1

HNT1

NRK1

MHF1

PAB1

YPT1

CAF16

TAD2

HSP150

PAU5

BLM10

SRS2

RTR1

FAS2

SUR1

SAM35

ORM1

CSH1

PIP2

PKP1

FTH1

COX5A

IDH2

LSC2

ATG16

GYL1

COX13

WHI2

CUE4

YDR249C

YBL029C-A

CSE4

ACM1

HBN1

CUE5

CTL1

RNT1

CDC1

DDR48

STN1

NOP6

ATM1

GUD1

PAU19

NST1

DIM1

MRM1

HAS1

TRM3

NOP9

MAK16

SDA1

GSP1

CEX1

RPL16A

RPL42B

RPL43A

RPL40B

RPL21B

RPL11A

RPL20B

RPL19A

RPL43B

ITT1

STP1

TIF6

YRA2

GFD1

FAF1

CDC123

RRP5

NOP2

UTP23

YJL211C

BIO2

PEX31

PIB1

PAU2

INP1

PEX29

TVP18

YGL152C

RRP4

RIX1

PRP21

IPI1

AIR1

NTC20

SHQ1

MTR4

PRP31

LEA1

PRP6

SPP381

PRP4

PRP38

PRP8

RPL22A

RPL3

RPL37A

RPL36A

RPL16B

NOP56

UTP13

UTP18

UTP15

LSM7

UTP4

RPS24A

RPS4B

RPS0A

RPS13

RPL6A

RPP0

RPS23A

YNL089C

VPS63

YHR049C-AYMR306C-A

MBB1

YER097W

YER091C-A PSY1

YKL147C

YJL175W

YIL080W

YIL141W

YKL177W

YKL097C

YDL062W

YPL205C

YPL080C

YCR102W-A

YOR366W

YDL026W

YOR318C

YDL038C

YOR309C

YCR001W

YAL016C-B

YPR197C YBR134W

YPR012WYPR130C

YCL012W

IRC13

YOR277C

YPR076W

YDR095C

YOL099C

FYV1

YDL187C

YOR015W

YOL079W

YDR344C

YDR290W

RRT16

YDR199W

YNL109W

YDR193W

YDR149C

PEX25

PEX22ALT2

ARO1

CCC2 PEX18

PEX19DFR1

SER1

MNE1

PEX21

ARG80

LCB2PHS1

YLR118C

GTT1

SDH2ASP3-3

DCG1

TRP2

TRP3PDR11

RRT5

IRC5

NSA1

SLD3

RAD54 PAU4

AMD1

YDR222W

APT1

ALO1

YFR006W

YOR011W-A

SHE1

YML007C-A

YJR149W

YOR008C-A

YPL260W GRX2

FMN1

ESF1

PDR8

NEJ1

CDC45

PDR15

HST3

ARC1

GUS1

MES1

YJR120W

MGM101

MAE1

RGI1

ENB1

FAA2

IZH1

OXP1

GYP1YJL144W

LCB3

ATG10 YDR444W

BCD1

ENP2

DCD1

HUR1

SUA5

WSS1

RTC3

LCB1

TIM18

LAT1

KGD1

SNC1

TRS20 DAN1

RCY1

EMP47

MST28

TRS85 GEA1RPL21A

RPL26BSPP382

RPL35B

SKI6

RPL24B

RPL9A

RRP42

DIS3

NUP57

PML39

RRP8

DBP3

GRC3

POA1

PET494

KRR1

ABD1PPM2

SEC27

COP1

SEC26

RET3

RET2 ARL3

SEC17

SNC2

SFT2

TUL1

PRE3

PRE2

CUE1

RAM2

CGI121

GID7

YDJ1

BET2

GPI16

GPI8

VID28

GON7

BUD32

GPI17

MAK10

SSA3

HSE1

MAK3

OCA5

OCA1

RSP5

UBP15

TIM44

SEC62

SEC11

GTB1

SBH1

SPC2

PAM16

MGE1

PSY4

YME1

MGR1

WBP1

STT3

OST3

OST1

VPS38

CCT6

PKH1

GPI2

PIL1

ERI1

GPI1

CCT5

GPI19

SSH1

ALG13

SBH2

BRE5

TPS2

BCY1

TPK3

TPK1

CCT8

AOS1

PHB1

VPS30

GIM6

ATG14

GIM2

GIM4

YLR352W

MVP1

SSK2

RCE1

STE6

PBS2

YMR085W

ADI1

DAK1

DCR2

IMH1

STR2

BAT2

YUH1

UFD4

KDX1

KKQ8

YPT52

GLG2

YPS6

SYS1

GWT1

MET3

DAL1

STS1

MNL1

EPS1

MNT3

PIG2

DPH1

HPM1

GSY2

ARV1

MIC60

HRI1

GPI13

YPS1

KIN2

RAD5

COA4

QRI1

USO1

RTK1

ERS1

CIT2

UBP1

MAL31

EFM2

PGI1

ARL1

AGP2

UBC4

AKL1

UBP14

RKM3

ALG1

RKM4

HEM1

RUB1

KIN1

TMN2

PDC2

ASK1

BCH1

CHS5

LRS4

RTT101

MMS22

DAD1

CDC7

CLB4

CLB5

DBF4

CLB6

CLN2

CDC28

CLN3

UBC5

PTP1

THI13

HEM13

TPI1

NUS1

RBS1

AST1

GDB1

YEL1

YPK3

GPI18

MAP2

PKC1

RFA2

PDS1

RFA1

DOP1

MDM12

ARC40

ARC35

ARP2

CLB1

ARC15

CTF13

CDC39

CLB2

CDC36

CBF2

CHS7

CNE1

SSE2

HSP32

SSE1

DJP1

MCX1

SPL2

YHR112C

SCY1

FMP48

MCY1

NIF3

YFR018C

IGD1

CMK1

DDI1

TMN3

ARO10

ARH1

YPQ2

ERD1

ASI2

YNR071C

PPG1

CPR8

SNZ2

DAL82

MPA43

PFA4

TLG2

ATG19

MPD2

ALG6

AKR2

DDP1

LAS17

RUD3

ALG8

YNL108C

EAR1

GPI12

PRC1

YNL010W

PKR1

ASI1

MGR3

CPR2

AAP1

INM1

DOG2

YHL018W

MAL11

YCH1

PMT6

HKR1

FPR2

PRB1

UBC8

MNN1

GDA1

RMT2

UBC6

PPM1ALG5

ISR1

CIN2

ENV7

YIG1

FLC1

RKM1

GPH1

MET16

DPM1

PPQ1

RIM20

CIN1

PEP4

GPB1

VEL1

CDC6

RFU1

TMA17

YBL111C

MIT1

DUN1

YDL057W

GIP3

TFS1

CUE3

MIX23

YNL195C

YLR161W

YJR015W

KIP2

ESS1

PMT7

ECO1

SDD3

YPL257W

PDP3

YJR141W

AFG2

VIK1

SGM1

PHO92

YBL081W

EGO2

YNR065C

MPO1

YPL225W

YCR102C

YNR040W

PIB2

YLR361C-A

YDR514C

YPL229W

TEL1

AIM25

BXI1

BUD17

YDR541C

NPA3

SPC105

MRX3

YPL199C

TMA10

IRC22

MPS1

YBR296C-A

EPO1

HSL1

YHR054C

YDR186C

YER186C

ARF3

YHR045W

CHK1

YKL183C-A

YER084W

YMR130W

CIA2

GLE1

BRN1

RPP1

POP1

POP7

NUP170

RAD23

RAD14

CET1

SMC3

EGD1

YCS4

CEG1

MCD1

SMC1

ABF1

NSE3

SMC5

NSE4

YHR086W-A

FYV4

YER189W

YML108W

LAM4

YKL075C

IES5

YCL001W-A

ASE1

YDR169C-A

BIT2

AIM36

YER076C

YHL048C-A

YOR097C

NSE1

TAF4

TAF9

TAF1

TAF5

SMC6

NSE5

TAF7

NHP10

IES1

GTR1

ISW1

ACT1

NPR2

HAT1

PSF1

ELP2

SLD5

POB3

GAL3

GAL80

SPT16

IKI1

YER053C-A

ANS1

TOF2

YMR209C

RCN2

AIM17

YMR247W-A

SUP45

SUP35

SSL2

TFC6

TFC8

TFC1

SSL1

CCL1

KIN28

TFG2

TFB3

HDA1

RNA14

PCF11

CFT2

HDA3

HDA2

DEP1

RSC3

RXT2

RSC8

RSC6

HTL1

RRN9

RSC58

CDC14RIC1

SLI15

CAB2

SCC2

TEF2

RGP1

CAB5

IPL1

YIL055C

YMR265C

BIM1

COS6

YKR015C

SPG1

YBR200W-A

GPN2

VTC5

OM14

PXA1

HTB2

ORC6

ORC3

RRN7

ORC2

ELG1

RPB2

RPC82

RPC11

RPC34

SNF8

ELA1

SRN2

ECM30

ESA1

EAF5

EAF3

YAF9

SWC4

EAF6

MCM5

MCM3

SEN1APC5

CDC23

CDH1

DPB4

MND2

SWM1

CDC27

APC4

PRI2

POL1

TEL2

DNL4

BUB2

SWD3

CDC24

HAP4

CDC48

POL3

POL32

POL31

TAP42

SLX1

SRP54

CYR1

BDF1

VPS71

CYC8

SAP155

PRI1

SIN4

MED1

NUT1

MED6

MED8

NUT2

SSN8

SRB6

MED7

MBP1

NRM1

CSE2

SSN3

NNF1

MCM7

MCM4

SNF5

RTT102

CHD1

SGV1

WHI5

POL2

RPA34

CTF8

DCC1

DPB3

CTF18

IES3

RTT109

SAP30

SDS3

SIN3

SPO20

NDI1

TUB3GIN4

YPL068CTMA23

AIM29

FMP52

EMC10

YOR376W-A

AIM19

YBR197C

RHO3

RPT2PDS5SCH9

SGF11

SGF29

ECM11 ASA1 LTE1

SSN2

AFR1

CHL4

SSO2

PAM1

IVY1

ROT1

DOM34

INP2

SEC14TAO3

DNA2

AXL2

YHR182W

FKH1

PFS1

KSP1

SPO12

SET2

FLO11

REC107

SSF1

MDH1

GIC1

YHR022C

MCK1

ENT1

HBT1

PPH22

RTN2

VAM10

PRM7

RGA1

VPS21

SHE4

RCN1

RHO4

CWP1

SHE10

MIF2

ERV14

ENT4

HSF1

VPS13

CCW12

KIP3

VPS45

FPS1

SPO74

STU2

DNM1

FPR3

YPT7

MDM30

VRP1

SPR6

MID2

BUD27

CAK1

RIM15

ACS2

IML3

SEC18

MYO2

HCM1

PDE2

IRC15

AIM44

SLK19

IST2

REI1

TOR2

TSC11

SEC4

CDC4

CPR1

CMD1

KOG1

LST8

SNF1

SGT1

CDC53

SPS2

AQY1

RDH54

JEM1

OPY2

TIP41

CDC15

KAR9

PGA3

STU1

CDC12

SPR3

PPH21

CDC10

CDC11

RTS1

MRC1

HSP104

YBR137W

GET4

MDY2

SGT2

TOF1

CKA1

CDC20

PAC11

DYN3

RAD57

HRR25

SPC98

PEA2

BNI1DDC1

CCZ1

SGS1

NUF2

SPC25

NBP1

MOB2

FAR3

VPS64

FAR7

FAR11

FAR10

GLC7

CYK3

MUS81

HYM1

DBF2

LGE1

BRE1

MLH3

MEI5

MRE11

CNB1

CMP2

UME6

TUB2

PAN1

CSM3

LAM6

YBR013C

YGR015C

MTC1

YNL181W

YDL121C

GTT3

YPL060C-A

YNL144C

YGR026W

YEL025C

YJL027C

DGR1

YBR063C

YEL023C

YPL038W-A

YLL056C

YGR035W-A

ECM8

YNL058C

UTR5

YIL175W

ULP1

FKH2

YDL157C

YBR090C

YIL174W

NNF2

MLP1

YMR317W

CHL1

YEL076C

YMR310C

YRB1

OM45

FEX1

CIN8

YIL102C

HGH1

YKR075C

ZDS1

YOR385W

TVP38

RAD61

OPY1

YGR168C

SDC25

YDL211C

ECM33

YEL043W

MTE1

PPT1 RVB1 TAH1SBA1STI1 PIH1RVB2HSC82 HSP82 Hsp90-TS AHA1CDC37 HCH1 CPR6CPR7CNS1

Manually curated bioprocesses

Figure 3. The Hsp90 Family and R2TP Networks

(A) SAFE analysis of the Hsp90 and R2TP on the CCo GI profile correlation similarity network is shown with the annotated bioprocess regions obtained from

Figure 2A. The Hsp90 family and R2TP members are overlaid.

(B) Shown are the GI profile correlation similarities (PCC > 0.1) of interactors for Hsp90 CCos and R2TP genes grouped according to manually curated bio-

processes from (A). Ontology terms were manually curated to reflect a more specific term. Edges are color coded based on the interactor’s respective bio-

process. The sizes of the nodes of Hsp90, Hsp90 cofactors, and R2TP genes reflect the number of interactors. All CCo nodes are color coded as in Figure 1A.

cochaperones based on the number of interactors from most to

least is as follows: CDC37 (291 interactors), CNS1 (250), CPR7

(223), RVB2 (172), PIH1 (131), HCH1 (106), CPR6 (83), AHA1

(67), RVB1 (66), TAH1 (61), STI1 (36), PPT1 (20), and SBA1 (19)

(Figure 3B).

The R2TP complex was initially discovered by our group as an

Hsp90-interacting protein complex in yeast (Zhao et al., 2005),

and it was found to be required for the assembly of other critical

complexes (Nano and Houry, 2013). Rvb1 and Rvb2 are highly

conserved AAA+ ATPases in eukaryotes and are essential for

cell viability. Despite advances in characterizing the R2TP com-

plex, there is no clear understanding of its cellular functions. We

found that members of R2TP have GI profile correlation similar-

ities with genes involved in several bioprocesses (Figure 3B), but

especially between RVB2 and genes encoding for cell polarity/

morphogenesis and mitosis/chromosome segregation/DNA

replication and repair (50% of its interacting genes) and between

PIH1 and ribosomal-related processes (44% of its interacting

genes).

A High-Fidelity Physical Interaction CCo NetworkTo generate a network based on physical and GI data, proteins

physically interacting with CCos were identified by collecting

PPI pairs from several large-scale PPI screens, including ours

(Gong et al., 2009), for all the CCos listed in Table S1. These

studies comprised PPI data obtained by tandem-affinity purifica-

tion (TAP) followed by mass spectrometry using large-scale

approaches to characterize soluble multiprotein complexes

Cell Reports 20, 2735–2748, September 12, 2017 2739

8,518 PPIs: 328 pairs include at least one CCo

Markov clustering (MCL)

Krogan et al.(7,874)

Gong et al.(21,687)

Gavin et al.(2,987)

Babu et al.(21,996)

889

1,813

93

118

69

6 734

36232

8921,289

316 5,134

12,54418,729

600

400

200

05 6 7 9 108

0

0.2

0.4

0.6

0.8

1

Hart-score

CC

o pr

ecis

ion

Num

ber of PP

Is with C

Co

56 CCos

Hart threshold = 6

Hart scoring

+ 189 isolated complexes

370 complexes: 15 multimeric

CCo-containing complexes

non-CCo

CCo (colors correspond to families)

300

500

AAA+ family

Hsp70 chaperonesHsp70 cochaperonesHsp40 family

CCT familysHsp family

Hsp90 chaperonesHsp90 cochaperones

The CCo NAJ complex:

PPT1

TDA10

KAR2

YBR139W

HSP42STI1

TOS1

CCT7TAP42

CCT4CCT3

PLP2

VID27

CCT2SIS1

CCT6

CAJ1

HSP78HSC82

SSA4

ZUO1

SSZ1

YEF3

SSA3

SCP160

HSP26

TIM44HSP82

SCJ1

DJP1JEM1HLJ1

YDJ1

SSC1HSP104 SSE1SSB2

SSA1

SSB1

GRE3

SSA2

ECM10SSQ1

Figure 4. Building the CCo Physical Interaction NetworkSchematic representation of the construction of the PPI CCo network. The first step is the compilation of CCo-containing interaction pairs (CCo-to-non-CCo or

CCo-to-CCo interactions) from major studies shown as a Venn diagram. The total number of interaction pairs from each study is given in parentheses. The

compiled PPIs were then scored and filtered using the indicated threshold as described in the Experimental Procedures (see also Figures S5D and S5E), resulting

in 8,518 PPIs. Subsequently, Markov clustering was applied to define multiprotein complexes. The final PPI network is shown using the spring-embedded al-

gorithm from Cytoscape. A zoom-in of the CCo supercomplex, the NAJ CCo complex, is shown with the size of each node corresponding to the degree of

connectivity. All CCo nodes in the network are colored according to Figure 1A and non-CCo proteins are in gray.

(Gavin et al., 2002; Krogan et al., 2006), membrane protein inter-

actors (Babu et al., 2012), and chaperone interactors in yeast

(Gong et al., 2009), totaling 43,020 interactions (Figure 4).

The overlap between the complete GI (positive plus negative)

dataset and the PPI dataset was significant (p = 2.2 3 10�16;

Fisher’s exact test), and this corresponded to about 5% of the

PPIs and 9%of theGIs (positive plus negative). The highest num-

ber of interaction overlap between the 43,020 CCo PPIs and the

CCo GIs was with negative GIs, followed by the GI profile corre-

lation similarity network, and lastly positive GIs with 1,236, 795,

and 774 common interactions, respectively (Figures S5A–S5C).

We found that the CCos with the highest number of common in-

teractions included theHsp70s (Sse1/2, Ssb1/2, Ssz1, andKar2),

Hsp40s (Ydj1 and Swa2), and the Hsp90 (Hsc82) (Figures S5A–

2740 Cell Reports 20, 2735–2748, September 12, 2017

S5C). Furthermore, in characterizing the physicochemical prop-

erties of CCo interactors, we found that the molecular weight

and protein abundance of interactors in the PPI dataset were

significantly greater compared to those of the GI datasets (p =

1.2 3 10�10 and 1.9 3 10�11, respectively; Mann-Whitney-Wil-

coxon test). This was likely due to a higher number of larger

(>70 kDa) abundant proteins in the cell interacting with CCos.

All other properties were similar across different datasets.

To select high-fidelity PPI interactions, several scoring

methods were evaluated. The goal was to obtain the largest

number of CCo-containing interaction pairs while maintaining

approximately equivalent precision (selected as R70%) against

a reference set of PPIs derived from the CYC2014 catalog of

manually curated protein complexes (Pu et al., 2009). We found

CCos in complexes:1. Cdc37 (Ych020) 2. Sec63 (Ych120)3. Swa2 (Ych011)4. NAJ complex: Hsp90, Hsp70, Hsp40, AAA+, CCT and sHsp- containing complex (Ych005) 5. Gim1-6 (Ych060) 6. R2TP-containing complex (Ych009)7. Sse2 (Ych076) 8. Cwc23 (Ych002) 9. Sil1 (Ych101)

10. Mge1 (Ych156)

Negative PCC(inter-complex)

Positive PCC

CCo-containing complex

Non-CCo-containing complex

|PCC|0.03

0.5

R2TP-containing complexSec63-containingcomplex

1

2

3

45

6

7 8

9

10

RNA pol III

Histone deacetylase RNA

processing

Histone methylation

Transcriptionfactor

RNA pol II

Kinetochore

Anaphasepromoting

N-termacetyltransferase

Actinnucleation

Serine/threoninekinase

DNA pol

ERAD

Myosin

DNA replication

ER transport

Histone deacetylase

Transcriptionregulation

Ribosomebiogenesis

Transport

Cytochrome-c

Mito. ribosome

Mito.enzymes

Metabolicenzymes

snRNP

U1-snRNP

Transcription

Vesicletransport

Proteasome

DNAhelicase

Tubulin

Retromercomplex

Vesicle

Translationinitiation

Elongatorcomplex

Vacuole ATPase

ER translocase

Nuclear pore

Replication factor

Gim1-6-containing complex

Cdc37, kinase & ATP synthase-containing

complex

Mito. protease

OST

Actin motility & nucleation

SSA3TDA10

STI1

SIS1

CCT2 HLJ1

CAJ1

CCT6ZUO1

GRE3 HSP82

TIM44

PPT1TOS1 SSB1

HSP42 SSA2

HSP26

SSB2 JEM1KAR2

SSZ1

CCT7

HSP78SCJ1

SSC1

CCT3

CCT4 DJP1

TAP42

PLP2

HSP104

YBR139W

YDJ1

ECM10HSC82SSE1

IES5

SWR1

KTR6

VPS71SWC3

NHP10

IES1

IES3TAH1

PIH1

ESA1

EAF1

ACT1

ARP4

EAF6

YTA6

RVB2

EAF7

YAF9

SWC4

SWC5

VPS72

SWC7

EAF5

BDF1

ARP6

RVB1

NAJ complex:Hsp90, Hsp70, Hsp40, AAA+, CCT and sHsp-containing complex

RPN12

UBP6RPN10

RPN7

RPT2

RPN5

SEM1POL4 RPN1

RPT6 RPN11RPT1

YSP2

ECM29

SLX5

RPN6

NAS6RPT3

RPT4

SWA2

RPN13

SEC18 SPG5

ESC1

ATP14

YMR118CATG8

DDR48

FIT1

YCK1CDC37

DCW1ATP7YCK2

ATP5

RAD27

SEC63

SEC72SEC66

SEC62

GIM1

BUD27

GIM6

GIM2 GIM3

GIM4

Swa2 & proteasomecap-containing complex

Figure 5. Overlay of the GI Profile Correlation Similarities onto MCL Clusters from PPIs

Shown is a nested network of the GI profile correlation similarities between CCo- (light blue) and non-CCo- (light gray) containing MCL clusters (complexes)

described in Table S5. Positive and negative correlations are shown between complexes and only positive correlations are shown within complexes. CCos

present in each complex are indicated in the bottom-right inset. Six complexes are shown in detail. The size of each node within the complexes corresponds to

the number of interactors. For simplicity, complexes containing only two proteins are not shown.

Hart scoring (Hart et al., 2007) a better predictor of CCo interac-

tions (Figures S5D and S5E). Using a Hart score threshold of 6,

we obtained a finalized network containing 8,518 non-redundant

pairwise associations (i.e., CCo-to-non-CCo, CCo-to-CCo, or

non-CCo-to-non-CCo) (Table S4). Among these, we identified

328 associations involving a CCo with 56 of the 82 CCos

captured (Figure 4; Figures S5D and S5E; Table S4). Markov

clustering (MCL) (Enright et al., 2002) was then employed to

organize the physical network. This resulted in a total of 370

complexes with 15 of them containing a CCo (Figure 4; Figures

S5F–S5K; Table S5). Strikingly, most CCos were found to cluster

together in a specific region of the proteome, specifically

the families of Hsp90, Hsp70, Hsp40, AAA+, CCT, and sHsp

(Figure 4; discussed further below). Such connectivity among

six different CCo families has not been reported before. It should

be emphasized that the MCL complexes obtained represent

functional clusters and not necessarily tightly associating phys-

ical complexes.

A Combined Physical and Genetic CCo NetworkGenes encoding physically interacting or co-complexed proteins

tend to share many GIs in common. By overlaying interactions

based on CCo GI profile correlation similarities onto the pre-

dicted protein MCL complexes obtained from the PPI network,

we built a combined network highlighting CCo complexes (Fig-

ure 5). The complexes are connected based on the average in-

ter-complex GI profile correlation similarities between the genes

in each complex pair. Together, this allows the quantitative and

nested representation of the GI profile correlation similarities be-

tween and within the predicted MCL complexes.

The network shown in Figure 5 displays 10 CCo-containing

complexes with a direct connection to 86 non-CCo-containing

Cell Reports 20, 2735–2748, September 12, 2017 2741

complexes. The Sec63-, Cdc37-kinase-ATP synthase-, R2TP-,

Swa2-proteasome cap-, Gim1-6-, and Hsp90-Hsp70-Hsp40-

AAA+-CCT-sHsp-containing complexes are highlighted in the

figure. We named the Hsp90-Hsp70-Hsp40-AAA+-CCT-sHsp

complex as the naturally joined (NAJ) CCo complex. The pres-

ence of such a chaperone supercomplex clearly suggests a

high degree of interactor overlap and functional coordination or

redundancy among these critical chaperones. The NAJ complex

also highlights the tight cross-talk among the protein homeosta-

sis machinery.

Additionally, we found that complexes that are functionally

specialized tend to have more connectivity in this network than

those that have a broader function (Figure 5). Such is the case

when comparing the Sec63-containing complex with 47 connec-

tions or the Gim1-6-containing complex with 29 connections

against the NAJ and the Cdc37-containing complexes with just

one and two connections, respectively (Figure 5). As expected,

the Sec63-containing complex is well connected with com-

plexes involving various ER processes aswell as with the protea-

some core subunit-containing complex, which likely highlights

the ER-associated degradation pathway. The Gim complex

shows functional specialization by connecting with proteins

involved in the cytoskeleton and transcriptional regulatory pro-

cesses. On the other hand, the Cdc37-containing complex inter-

actions are less specialized, involving diverse functions ranging

from several kinases both at the inter- and intra-complex level,

DNA repair proteins, and interactions with several mitochondrial

complexes such as the ATP synthase (Figure 5). The NAJ com-

plex has a single edge connecting it with actin assembly and

motility. Given that this is a multi-CCo complex with a very broad

involvement in numerous cellular pathways, this connection re-

flects the most common pathway between the NAJ CCo com-

plex and the rest of the proteins.

As shown in Figure S3, the Hsp40 protein Swa2 was found to

have a very strong enrichment for positive GIs with proteasomal

genes. We also see such a strong association between Swa2

and the proteasome cap in Figure 5.SWA2 interacts with 13 sub-

units of the proteasome (RPN1,5-7,10-13 and RPT1-4,6) and the

two proteasome assembly chaperones NAS6 and ECM29. As

well, this complex interacts with the proteasome core particle

complex. Swa2 is a multifunctional protein involved in uncoating

of clathrin-coated vesicles and in interacting with ubiquitin

chains. Together, this suggests that Swa2 may assist in the as-

sembly or regulation of the proteasome. The interaction between

Swa2 and the proteasome has not been reported before in the

literature.

The R2TP complex clusters with members of the histone

exchange complex Swr1, histone acetyltransferase complex

NuA4, and members of the chromatin remodeling complex

Ino80. We also found that the only gene interacting with all three

RVB1, RVB2, and PIH1 is the histone exchange ATPase SWR1

(Figure 5). Importantly, actin is a major interactor of this complex

through the Rvbs by both GI profile correlation similarity and

physical interactions.

The combined network in Figure 5 clearly illustrates the

different degrees of specialization of CCos. On one end, we

find that the Sec63- and Gim-containing complexes have the

highest connections followed by the Swa2-proteasome cap-

2742 Cell Reports 20, 2735–2748, September 12, 2017

containing complex, and, on the other end, we find the NAJ com-

plex with minimal connectivity.

Rvb1/2 Form Condensates in Stressed CellsTo test the predictive value of our interaction data and given the

role of chaperones in preventing protein aggregation and pro-

moting protein folding, we examined CCos that colocalize with

foci-forming proteins on the CCo GI profile correlation similarity

network (Figure S6). We compiled a list of 547 proteins that have

previously been shown either by low- or high-throughput studies

to form foci (or condensates) under heat shock or nutrient limita-

tion (Table S6) (Bolognesi et al., 2016; Narayanaswamy et al.,

2009; O’Connell et al., 2014; Shah et al., 2014; Wallace et al.,

2015).

As shown in Figure S6A, these foci-forming genes are en-

riched (p < 0.05) in regions containing many CCo families, such

as the AAA+, CCTs, prefoldins, small heat shock, Hsp40s, and

notably the Hsp90-R2TP system among others. By extracting

the subset of foci-forming proteins that interact with the

Hsp90-R2TP system, we found that these are highly enriched

in genes involved in ribosome biogenesis and translational pro-

cesses (Figure S6B). Furthermore, these interactions are largely

driven by the CDC37, CNS1, and CPR7 cochaperones and

RVB1, RVB2, and PIH1 components of the R2TP complex (Fig-

ure S6C). Hsp82, Hsc82, and some of the Hsp90 cochaperones

have been previously reported to form foci (Figure S6C; Table

S6). Hence, we investigated whether subunits of the R2TP com-

plex also form foci. No foci formation was observed for Tah1 or

Pih1, but we found Rvb1/2 to form condensates under nutrient

limitation conditions.

Initially, we investigated growth at 30�C in nutrient-rich condi-

tions (YPD). Rvb levels were constant as a function of cell growth

(Figure S7A) and cell cycle (Figure S7B), with Rvb1 levels being

higher than Rvb2.We then generated a strain, WH12, expressing

endogenously tagged RVB1-mRFP and RVB2-GFP (see the

Experimental Procedures) tomonitor Rvb1 andRvb2 localization

throughout the cell cycle. Rvb1 andRvb2 predominantly colocal-

ized to the nucleus (Figure S7C) with a nuclear-to-cytoplasmic

protein (N/C) ratio around 3 to 4 (Figure S7D). Hence, neither

protein expression nor protein localization was significantly

changed under the conditions tested.

Next, the effect of different stresses on Rvb localization was

examined (Figure 6A). Under these stress conditions, Rvb1 and

Rvb2 protein levels did not significantly change, and the two pro-

teins remained predominantly colocalized with an N/C ratio of 3

to 4 (Figure S7E, left panel), similar to that of the N/C ratio in log

phase cells (Figure S7D), with a colocalization Pearson correla-

tion coefficient of 0.81 (Figure S7E, right panel). Cells treated

with water for 1 hr were very heterogeneous, exhibiting various

morphologies with some undergoing apoptosis, and, thus, these

cells were not quantified for N/C ratios.

In carrying out the localization studies, we noticed that a pro-

portion of Rvb1 and Rvb2 forms condensates near the nucleus

under some of the stress conditions tested (indicated by white

arrows in Figure 6A), namely, stationary phase (STAT), growth

in media lacking glucose (�Glc), treatment with 2-deoxy-D-

glucose (DG), and growth in water. It should be noted that

Rvb1 has been reported to form aggregates in cells exposed

EtOH

Mock

NaCl

Rvb1-mRFP Rvb2-GFP Merge

STAT

- N

Rap

MMS

- Glc

DG

g

5 μm

A Rvb2-GFPRvb1-mRFP DAPI

RFP/DAPI GFP/DAPI RFP/GFP

1 μm

DG

Mock

Alexa Fluor DAPI

Rvb2-GFPNup49-mCherry Merge

DG

Mock

B C

D

0

10

20

30

40

50

0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5

Distance from nuclear membrane (μm)

Per

cent

of f

oci

rela

tive

to to

tal f

oci (

%)

0.05

1 μm

1 μm

0

10

20

30

40

50

60

0 2 4 6 8 10 12 14Time (mins)

E

DG (2hr)

0+0- 10’ 40’ 60’

Glc

0102030405060708090

100

DG (4hr)

F

befo

retre

atm

ent

Per

cent

cel

ls w

ith fo

ci (%

)P

erce

nt c

ells

with

foci

(%)

H2O

37˚C

Figure 6. Condensate Formation by Rvb1/2

(A) Colocalization of Rvb1 and Rvb2 under various stress conditions. Cells of the WH12 strain were grown to log phase and then treated with various stress

conditions as follows: mock treatment, 10% ethanol (EtOH), 900 mM NaCl (NaCl), heat shock at 37�C (37�C), stationary phase (STAT), minus glucose media

(�Glc), 1 mM 2-deoxy-D-glucose (DG), minus nitrogen media (�N), 0.01% methyl methanosulfonate (MMS), 10 ng/mL rapamycin (Rap), and growth in water

(H2O). White arrowheads in STAT, �Glc, DG, and water treatment point to the Rvb condensates.

(B) Condensate formation by Rvb1 and Rvb2 upon glucose deprivation. Rvb1 and Rvb2 in WH12 strain after treatment with DG for 1 hr exhibit condensates near

the nucleus. White arrow points to a condensate.

(C) Rvb1-FLAG in YK26 (RVB1-FLAG) strain treated with DG (upper panels) or mock treated (lower panels) was labeled with anti-FLAG primary antibodies and

AlexaFluor488 secondary antibodies.

(legend continued on next page)

Cell Reports 20, 2735–2748, September 12, 2017 2743

to a high temperature of 46�C (Wallace et al., 2015), which was

not a condition tested in our study.

Because the Rvb1 and Rvb2 condensates observed here ap-

peared to be specific to glucose deprivation, we examined them

in more detail (Figures 6B–6D; Movie S1). Figure S7F provides a

statistical summary of the distribution of condensates in cells in

stationary phase, �Glc, and DG conditions. Nearly 50% of all

cells in each stress condition harbored condensates, of which

most of the cells had one condensate per cell. Only 31% of cells

contained condensates in the �Glc treatment, possibly due to

the presence of other metabolizable nutrients in the rich media.

Treatment with DG for 2 hr resulted in an increase in the number

of cells harboring three condensates per cell.

We observed that condensate formation occurred for both

Rvb1 and Rvb2 and that these condensates appeared in the

cytoplasm, close to the nucleus (Figures 6A–6D). Figure 6B

shows a closer view of Rvb1 and Rvb2 colocalized at the nucleus

and in the perinuclear condensates (RFP/GFP panel). To ensure

that condensate formation is not an artifact of GFP tagging, we

performed indirect immunofluorescence labeling on the YK26

strain, expressing endogenous Rvb1-FLAG, in the presence of

DG or mock treated (Figure 6C). Immunofluorescence labeling

showed that Rvb1-FLAG also formed condensates close to the

nucleus, indicating that condensate formation is not an artifact

of endogenous GFP tagging.

We then explicitly determined that Rvb condensates resided in

the cytoplasm using the WH4 strain. The strain expresses

endogenous Rvb2-GFP and Nup49-mCherry, which is a compo-

nent of the nuclear pore complex and can be used as a marker

for the nuclear membrane. WH4 strain was treated with DG for

1 hr and subsequently imaged. Rvb condensates appeared on

the cytoplasmic side of the nuclear membrane and localized

very close to the nucleus (Figure 6D, left panel). However, treat-

ment with DG did not affect Nup49-mCherry localization, and the

distance of the condensate relative to the nuclear membrane did

not change at 30 min, 1 hr, or 2 hr of DG treatment. Most of the

condensates were 0.5 mmor less from the nuclear envelope (Fig-

ure 6D, right panel), and their formation depended on nuclear

export by Crm1 (Figure S7G). Condensate formation was rapid

and occurred within 3 min after DG addition (Figure 6E; Movie

S1). These condensates were also rapidly reversible (within

10 min) upon the addition of glucose (Figure 6F; Figure S7H).

The rapid nature of assembly and dissolution suggests that

these condensates are not aggregates but biomolecular con-

densates (Rabouille and Alberti, 2017). We named these con-

densates Rvb bodies instantly triggered by starvation (Rbits).

Rbits did not colocalize with stress granules using Pbp1-GFP

as a marker (Figure S7I) nor with P-bodies using Dhh1-mRFP as

a marker (Figure S7J). Rbits also did not colocalize with protein

(D) WH4 strain (RVB2-GFP and NUP49-mCHERRY) was grown to log phase and

membrane (upper panels) in comparison to mock treatment (lower panels) whe

distance of the condensates from the nuclear membrane of 1,410 condensates in

relative to total condensates measured.

(E) Percentage of cells of WH12 strain forming Rbits after the addition of DG (see

(F) Average percentages of cells with Rbits are represented in a bar graph obtaine

bars represent one SD.

See also Figures S6 and S7.

2744 Cell Reports 20, 2735–2748, September 12, 2017

aggregates marked by Von Hippel-Lindau factor (VHL) (Fig-

ure S7K). We also tested whether protein components in com-

plexes containing Rvb1 and Rvb2 formed condensates that

might be similar to Rbits under glucose deprivation conditions.

The following proteins were assessed for condensate formation:

Pih1, Nop58, and Bcd1, all of which are involved in box C/D small

nucleolar ribonucleoprotein (snoRNP) assembly; Ino80, Taf14,

and Nhp10, components of the Ino80 chromatin-remodeling

complex; Vps71, component of the Swr1 chromatin-remodeling

complex; Tti1, part of the ASTRA chromatin-remodeling com-

plex and of the TTT complex; and Rpb1, Rpb2, Rpb3, Rpb7,

Rpb8, and Rpb9 subunits of the RNA polymerase II machinery.

None of these components appeared to form condensates

resembling the Rbits under our tested conditions. Together

these data suggest that Rvb1/2 form stress-inducible conden-

sates independently of other proteins and stress-inducible

compartments.

Condensate Formation by Rvb1/2 In Vitro Depends onpH, ATP Levels, and CrowdingTo further investigate the properties of Rbits, we sought to

reconstitute these condensates in vitro. Rvb1/2 complex was

purified from E. coli and the proteins were labeled with a fluoro-

phore. Yeast cells respond in many ways to stress: trehalose

production is increased (Wiemken, 1990), ATP concentration

drops (Weitzel et al., 1987), there is increased crowding (Mour~ao

et al., 2014), and there is a drop in pH (Weitzel et al., 1987). Based

on these observations, we tested how these parameters influ-

ence Rvb1/2 in vitro. Purified Rvb1/2 complex was diffuse in so-

lution under neutral pH conditions (Figures 7A and 7B). However,

at pH 6.0, Rvb1/2 formed round condensates of a size of about

0.1–0.3 mm. Rvb1/2 also formed condensates upon the addition

of polyethylene glycol (PEG) of two different lengths to mimic a

crowded cellular environment (Figures 7A and 7B).We quantified

the amount of condensed material by comparing the fluores-

cence intensity of the condensates with the fluorescence inten-

sity outside of the condensates. We found that condensate

formation increased when using a longer PEG chain in compar-

ison to a shorter one (Figures 7A and 7B).

It has been reported that some proteins are able to form liquid,

viscoelastic, or solid compartments upon crowding (Patel et al.,

2015). To probe the material properties of the Rbits, we applied

fluorescence recovery after photobleaching (FRAP). When Rbits

were photobleached, there was no significant fluorescence re-

covery in vitro (Figures 7C and 7D) or in vivo (using the WH12

strain described above; Figures 7E and 7F), suggesting that

these condensates are solid.

We then analyzed whether the condensates dissolved when

the triggering condition was reversed. For the pH-induced

treated with DG for 1 hr. Rvb2-GFP forms condensates outside the nuclear

re no condensates are observed. Right panel shows the quantification of the

800 cells treated with DG for 30min, presented as a percentage of condensates

also Movie S1).

d from images shown in Figure S7H. Experiments were repeated 3 times. Error

PEG20,000

pH7.5

Rvb1/2

PEG3,350

pH6.0

pre-bleach

post-bleach30 sec 5 min 0 sec

Fluo

eres

cenc

e S

igna

lTime (min)

10 μm

no bleach

bleach

0.1 μm

Int in

/ In

t out

pH7.5

pH6.0

PEG3,350

PEG20 k

Intin

/ In

t out

concentration (mM)

10 μmpH 6.0Rvb1/2

pre-bleach

post-bleach30 sec 5 min 0 sec

Fluo

eres

cenc

e S

igna

l

Time (min)

no bleach

bleach

1 μm

A

B

C

D

pH7.5

Rvb1/2

pH6.0

pH7.5

Intin

/ In

tout

pH 7.5 6.0 6.0 7.5

10 μm

G

H

0 %PEG

Rvb1/2

10 %PEG

0 %PEG

Intin

/ In

tout

PEG 0 % 10 % 10 0 %

10 μm

I

J

Intin

/ In

tout

w/o ATP ADP TRH

10 mMATP

pH 6.0

10 mMADP

230 mMTrehalosew/o

Rvb1/210 μm

K

L

M

E

F

DIC GFP

pH =

7.0

pH =

6.0

en

ergy

dep

letio

n

N

Figure 7. Characterization of Rbits

(A) Condensation of purified Rvb1/2 complex was monitored at different pHs or in 10% PEG of different lengths.

(B) Condensate formation was quantified by plotting the fluorescence intensity inside against the intensity outside the condensates.

(C) The material turnover of a Rvb1/2 condensate was analyzed by FRAP.

(D) Quantification of the FRAP results for a bleached (blue) and a non-bleached condensate (green).

(E) The material turnover of Rbits formed in energy-depleted yeast cells was analyzed by FRAP.

(legend continued on next page)

Cell Reports 20, 2735–2748, September 12, 2017 2745

condensates of purified Rvb1/2, pHwas increased back from 6.0

to 7.5, and this caused the condensates to dissolve (Figures 7G

and 7H). Similarly, reducing crowding of PEG-induced particles

by dialysis resulted in the dissolution of the Rbits (Figures 7I

and 7J). In agreement with a recent report that ATP dissolves

condensates (Patel et al., 2017), the presence of ATP or ADP

at 10 mM prevented condensate formation at low pH (Figures

7K and 7L). A titration experiment revealed that ADP was less

potent than ATP in preventing condensate formation of Rvb1/2

(Figure 7M). Trehalose, which is strongly upregulated in

response to stress, did not prevent condensate formation even

at high concentrations (Figures 7K and 7L).

These results suggest that, in response to energy depletion

(i.e., a drop in ATP concentration), a drop in pH and an increase

in crowding could trigger Rvb1/2 condensate formation in cells.

This in turn would indicate that a drop in the cytosolic pH alone,

independent of a metabolic response, should not induce

condensate formation. To test this, we used 2,4-dinitrophenol

(DNP), a protonophore, which can be used to equilibrate the

pH inside the cells with that outside (Munder et al., 2016). Using

this approach on the WH12 yeast cells, we indeed found that

Rbits did not form at pH 6 (Figure 7N) without energy depletion.

This indicates that cytosolic acidification alone is not sufficient to

trigger condensation but rather multiple triggers must coincide

to promote Rbits formation.

DISCUSSION

In this work, we have provided a comprehensive global survey

of the protein homeostasis network of the yeast cell using

genetic and physical interaction data. This analysis offered an

overview of chaperone connectivity that is consistent with the

extensive biochemical data available from the literature (Fig-

ure 1D); it also allowed us to identify novel features of chaperone

function.

The bias of many chaperones toward positive GIs with essen-

tial genes in comparison with the rest of the genes in yeast

(Figures S2E and S2F) suggests an unexpected finding of chap-

erones preventing or reducing cell viability if an essential gene of

the cell is affected. It is possible that this feature of chaperone

function may underlie a novel mechanism by which chaperones

maintain protein homeostasis in the cell. While counterintuitive,

we propose that chaperones maintain protein homeostasis if

nonessential genes are damaged but reduce cell viability if

essential genes are affected. This observation might have direct

implications for protein evolution.

(F) Quantification of the FRAP results for a bleached (red) and a non-bleached R

(G) pH-induced Rvb1/2 condensates could be dissolved by re-adjusting to neutr

(H) Quantification of the data in (G).

(I) PEG-induced Rvb1/2 condensates could be dissolved by dialysis against buff

(J) Quantification of the data in (I).

(K) Effect of the pre-addition of ATP, ADP, or trehalose on Rvb1/2 condensation

(L) Quantification of the data in (K).

(M) Titration of ATP and ADP reveals the critical concentrations needed for the inh

same as the representative panel labeled as w/o in (K).

(N) Rbits formation in DNP-treated yeast was monitored by adding energy deple

Distributions and errors shown in the plots of (B), (H), (J), (L), and (M) are derived

2746 Cell Reports 20, 2735–2748, September 12, 2017

Our interaction network provided a clear distinction between

chaperones that act on multiple substrates and impact many

pathways versus those that have a more defined role in the cell

(Figure 2; Figure S4). Most CCos are found to be proteins with

multiple functions (pleiotropic) with only a few members being

more specialized. In this regard, it was very striking to find that

CCT and the prefoldin complex are rather specific chaperones

and act on particular cellular pathways related to cell polarity

and mitosis (Figures 2B and 2C). This specificity likely arises

because such chaperones act predominantly on a select set of

substrates, such as actin and tubulin, rather than on a multitude

of proteins. Hence, these chaperones have been optimized

through evolution to become critical players in specific cellular

pathways.

The procedure we used to obtain a high-fidelity physical inter-

action dataset (Figure 4) revealed the presence of a huge chap-

erone supercomplex that we named the NAJ complex (Figures 4

and 5). This complex reveals that chaperones are physically

and/or functionally linked in the cell. Chaperones are always in

communication with each other even when present in different

cellular compartments. The presence of such a supercomplex

is likely essential for maintaining cellular protein homeostasis

across different organelles.

Another intriguing finding, highlighting the predictive value of

our network, is the observation that many proteins that have

been reported to form condensates under different growth con-

ditions are also interactors of many CCos. Subsequently, we

confirmed that Rvb1 and Rvb2 form perinuclear condensates,

which we named Rbits, in response to nutrient deprivation (Fig-

ures 6 and 7; Figure S7). Although the physiological role of Rbits

is currently not clear, they could possibly act as protective stor-

age depots for the Rvb1/2 proteins in response to nutrient stress.

In conclusion, our comprehensive analysis suggests that a

global integrative network approach with physical and GI data

is necessary to obtain a more thorough functional overview of

such versatile and modular proteins as the chaperones and their

cochaperones. This study reveals new findings on chaperone

functions, and it serves as a resource for gaining a better under-

standing of the mechanisms that govern protein homeostasis in

the cell.

EXPERIMENTAL PROCEDURES

Collection and Analysis of CCo GIs

Experimental approaches used to obtain high-throughput GI data for all chap-

erones and cochaperones have recently been described (Costanzo et al.,

2016). GIs were extracted for the following: the nonessential 3 nonessential

bits (green).

al pH.

er containing no PEG.

.

ibition of condensate formation. Note that the left panel was chosen to be the

tion medium or phosphate buffer at varying pHs to the cells.

from three images of three biological replicates.

(NxN); essential3 nonessential (ExN); essential3 essential (ExE);DAmP data-

sets for Cct8, Cns1, Cwc23, Jac1, Sis1, and Ssz1; and the Hsp90-TS dataset

(Zhao et al., 2005). An intermediate threshold (p < 0.05 and jSGA scorej > 0.08)

was applied to the dataset. GI profile correlation similarities were obtained by

computing PCC between all chaperone query-query and query-array pairs of

strains in the nonessential, essential, DAmP, and TS GI datasets.

Analysis of Positive and Negative GI Ratios from CCos

To calculate positive GI enrichments, positive or negative GI densities of

CCos against the essential (E) and nonessential (N) arrays were computed

as follows:

Positive GI densityðE or NÞ=positive density of gene XðE or NÞ

Average positive density for all query genesðE or NÞ

The enrichment ratio of positive-to-negativeGI in each dataset (i.e., E or N) was

calculated as follows:

Positive : NegativeðE or NÞ= Positive GI densityðE or NÞNegative GI density ðE or NÞ

Positive GI bias was calculated as follows:

Positive GI biasðE or NÞ= Positive : Negative for E

Positive : Negative for N:

GI density was calculated by selecting a single allele per array gene. The

result was averaged across 100 different randomizations. GI densities of

different query strains with mutations on the same gene were averaged to

obtain a single value per query gene. GI data for pairs of genes belonging

to the same protein complex, obtained by merging Table S1 and a recent

list of protein complexes (Costanzo et al., 2016), were removed for this

analysis.

Construction of the Yeast Chaperone PPI Network

We constructed the yeast chaperone PPI network by compiling four protein

interaction datasets (Gavin et al., 2002; Krogan et al., 2006; Babu et al.,

2012; ChaperoneDB fromGong et al., 2009). To obtain the high-fidelity network

of PPI interactions given in Table S4, themass spectrometry (MS) datasets that

included MALDI Z scores or liquid chromatography-tandem mass spectrom-

etry (LC-MS/MS) confidence values were pre-filtered at a MALDI Z score

threshold R1 and/or an LC-MS/MS confidence score R70%. Next, an inte-

grated interaction score was computed by summing the hypergeometric (HG)

interaction scores (Hart et al., 2007) as Gavin_HG + Babu_HG3 0.5 + Krogan_

HG 3 0.5 + Gong_HG 3 0.08. The final network was derived by applying an

integrated score of 6, selected to maximize chaperone interaction coverage

while maintaining a precision R70% against the CYC2014-derived reference

set (Figures S5D and S5E). The resulting integrated PPI network comprises

8,518 interactions between 2,062 proteins (Figure 4; Table S4). From these

interactions, 328 involved at least one CCo capturing a total of 56 of the 82

CCos used in this study.

We compiled a gold standard reference set of chaperone complexes by

using the CYC2014 catalog (Pu et al., 2009) containing 443 manually

curated non-redundant yeast protein complexes. Using chaperone preci-

sion and number of chaperones as a proxy, a Hart score of 6 was chosen

as threshold (Figure 4; Figures S5D and S5E). Next, given that the densely

connected regions of the PPI dataset suggest associated protein units

likely having a similar function (Wodak et al., 2009), we used the MCL

method (Enright et al., 2002) to identify these regions from within the CCo

interaction network. High-confidence 8,518 PPIs were clustered over a

range of Inflation (I) parameters (1.0 to 5.2). Cluster results were bench-

marked based on overall cluster properties, for example, on number of

clusters and average cluster size (Figures S5F and S5G), CYC2014 (Pu

et al., 2009), complex coverage through precision and homogeneity metric

(Pu et al., 2007) (Figures S5H and S5I), and intra- and inter-cluster func-

tional diversity as measured by the Shannon index of GO biological process

and molecular function terms (Figure S5J). Based on the minimization

of intra-cluster average functional diversity and coverage of known

CYC2014 complex members, an inflation parameter of 1.8 (Figure S5J)

was chosen to generate the finalized clustering of the filtered interaction

network, corresponding to 327 clusters possessing 2 or more proteins (Fig-

ure S5K; Table S5).

Construction of the GI Profile Correlation Similarity Network

Overlaid onto Predicted Complexes from PPIs

Genetic profile similarity scores for all CCo GIs were overlaid onto predicted

MCL clusters. Positive and negative GI profile correlation scores were

computed using PCC. In Figure 5, inter-complex edges represent the average

GI profile correlation similarities between the interacting members from each

complex, and the intra-complex edges represent the GI profile correlation sim-

ilarities between genes within each complex. Proteins in the different MCL

complexes are given in Table S5.

Other experimental procedures are provided in the Supplemental Experi-

mental Procedures.

SUPPLEMENTAL INFORMATION

Supplemental Information includes Supplemental Experimental Procedures,

seven figures, six tables, and one movie and can be found with this article

online at http://dx.doi.org/10.1016/j.celrep.2017.08.074.

AUTHOR CONTRIBUTIONS

W.A.H. designed the study. W.A.H. and M.B. supervised the experiments and

data analyses. K.R. and A.K. collected and analyzed the data. J.H., H.A.Z., and

Y.K. carried out the experiments related to foci formation in vivo. E.E.B. carried

out the in vitro and in vivo experiments related to foci characterization with the

supervision of S.A. T.V.S. helped with strain generation. Z.M. carried out the

mass spectrometry. J.S. designed and implemented the MYTH experiments

with the supervision of I.S. S.P. and J.V. contributed to the bioinformatic ana-

lyses. W.W., C.P., C.L.M., M.C., and C.B. contributed the GI experiments and

helped with the analyses.W.A.H. and K.R. wrote the paper with help fromM.B.

and all other authors.

ACKNOWLEDGMENTS

We thank Dr. Craig Smibert (University of Toronto) for the pRS116 DHH1

plasmid, Dr. Judith Frydman (Stanford University) for pESCGFP-VHL plasmid,

Dr. Guri Giaever (University of British Columbia) for theRVB1/rvb1D andRVB2/

rvb2D diploid strains, and Dr. Grant Brown (University of Toronto) for the

NUP49-mCHERRY strain. K.R. was supported by a Canadian Institutes of

Health Research (CIHR) Training Program in Protein Folding and Interaction

Dynamics: Principles and Diseases fellowship and by a University of Toronto

Fellowship from the Department of Biochemistry. J.H. was supported by a

Natural Sciences and Engineering Research Council of Canada PGSD2 fellow-

ship. T.V.S. was supported by a CNPq-Brazil fellowship (202192/2015-6) and a

Saskatchewan Health Research Foundation postdoctoral fellowship. M.B.

holds a CIHR New Investigator award. C.B. and C.L.M. are fellows of the Ca-

nadian Institute for Advanced Research (CIFAR). This work was partially sup-

ported by the National Institutes of Health (R01HG005853 to CB and CLM and

R01HG005084 to CLM). This work was also supported by CIHR (RSN-124512,

MOP-125952, RSN-132191, and FDN-154318 toM.B., MOP-93778 toW.A.H.,

and MOP-81256 to I.S. and W.A.H.).

Received: March 7, 2017

Revised: July 21, 2017

Accepted: August 23, 2017

Published: September 12, 2017

REFERENCES

Ashburner, M., Ball, C.A., Blake, J.A., Botstein, D., Butler, H., Cherry, J.M.,

Davis, A.P., Dolinski, K., Dwight, S.S., Eppig, J.T., et al.; The Gene Ontology

Consortium (2000). Gene ontology: tool for the unification of biology. Nat.

Genet. 25, 25–29.

Cell Reports 20, 2735–2748, September 12, 2017 2747

Babu, M., Vlasblom, J., Pu, S., Guo, X., Graham, C., Bean, B.D., Burston, H.E.,

Vizeacoumar, F.J., Snider, J., Phanse, S., et al. (2012). Interaction landscape of

membrane-protein complexes in Saccharomyces cerevisiae. Nature 489,

585–589.

Baryshnikova, A. (2016). Systematic Functional Annotation and Visualization of

Biological Networks. Cell Syst. 2, 412–421.

Baryshnikova, A., Costanzo, M., Kim, Y., Ding, H., Koh, J., Toufighi, K., Youn,

J.Y., Ou, J., San Luis, B.J., Bandyopadhyay, S., et al. (2010). Quantitative anal-

ysis of fitness and genetic interactions in yeast on a genome scale. Nat.

Methods 7, 1017–1024.

Bolognesi, B., Lorenzo Gotor, N., Dhar, R., Cirillo, D., Baldrighi, M., Tartaglia,

G.G., and Lehner, B. (2016). A Concentration-Dependent Liquid Phase Sepa-

ration Can Cause Toxicity upon Increased Protein Expression. Cell Rep. 16,

222–231.

Breitkreutz,B.J., Stark, C., Reguly, T., Boucher, L., Breitkreutz, A., Livstone,M.,

Oughtred, R., Lackner, D.H., Bahler, J., Wood, V., et al. (2008). The BioGRID

Interaction Database: 2008 update. Nucleic Acids Res. 36, D637–D640.

Costanzo, M., Baryshnikova, A., Bellay, J., Kim, Y., Spear, E.D., Sevier, C.S.,

Ding, H., Koh, J.L., Toufighi, K., Mostafavi, S., et al. (2010). The genetic land-

scape of a cell. Science 327, 425–431.

Costanzo, M., VanderSluis, B., Koch, E.N., Baryshnikova, A., Pons, C., Tan, G.,

Wang,W., Usaj, M., Hanchard, J., Lee, S.D., et al. (2016). A global genetic inter-

actionnetworkmapsawiringdiagramofcellular function.Science353, aaf1420.

Dekker, C., Stirling, P.C., McCormack, E.A., Filmore, H., Paul, A., Brost, R.L.,

Costanzo, M., Boone, C., Leroux, M.R., and Willison, K.R. (2008). The interac-

tion network of the chaperonin CCT. EMBO J. 27, 1827–1839.

Enright, A.J., Van Dongen, S., and Ouzounis, C.A. (2002). An efficient algorithm

for large-scale detection of protein families. Nucleic Acids Res. 30, 1575–1584.

Gavin, A.C., Bosche, M., Krause, R., Grandi, P., Marzioch, M., Bauer, A.,

Schultz, J., Rick, J.M., Michon, A.M., Cruciat, C.M., et al. (2002). Functional or-

ganization of the yeast proteome by systematic analysis of protein complexes.

Nature 415, 141–147.

Giaever, G., Chu, A.M., Ni, L., Connelly, C., Riles, L., Veronneau, S., Dow, S.,

Lucau-Danila, A., Anderson, K., Andre, B., et al. (2002). Functional profiling of

the Saccharomyces cerevisiae genome. Nature 418, 387–391.

Gong, Y., Kakihara, Y., Krogan, N., Greenblatt, J., Emili, A., Zhang, Z., and

Houry, W.A. (2009). An atlas of chaperone-protein interactions in Saccharo-

myces cerevisiae: implications to protein folding pathways in the cell. Mol.

Syst. Biol. 5, 275.

Hart, G.T., Lee, I., and Marcotte, E.R. (2007). A high-accuracy consensus map

of yeast protein complexes reveals modular nature of gene essentiality. BMC

Bioinformatics 8, 236.

Houseley, J., and Tollervey, D. (2006). Yeast Trf5p is a nuclear poly(A) polymer-

ase. EMBO Rep. 7, 205–211.

Kakihara, Y., and Houry, W.A. (2012). The R2TP complex: discovery and func-

tions. Biochim. Biophys. Acta 1823, 101–107.

Krogan, N.J., Cagney, G., Yu, H., Zhong, G., Guo, X., Ignatchenko, A., Li, J.,

Pu, S., Datta, N., Tikuisis, A.P., et al. (2006). Global landscape of protein com-

plexes in the yeast Saccharomyces cerevisiae. Nature 440, 637–643.

Krzywinski, M., Schein, J., Birol, I., Connors, J., Gascoyne, R., Horsman, D.,

Jones, S.J., and Marra, M.A. (2009). Circos: an information aesthetic for

comparative genomics. Genome Res. 19, 1639–1645.

Leroux, M.R., and Hartl, F.U. (2000). Protein folding: versatility of the cytosolic

chaperonin TRiC/CCT. Curr. Biol. 10, R260–R264.

Li, J., Soroka, J., and Buchner, J. (2012). The Hsp90 chaperone machinery:

conformational dynamics and regulation by co-chaperones. Biochim. Bio-

phys. Acta 1823, 624–635.

2748 Cell Reports 20, 2735–2748, September 12, 2017

Mour~ao, M.A., Hakim, J.B., and Schnell, S. (2014). Connecting the dots: the

effects of macromolecular crowding on cell physiology. Biophys. J. 107,

2761–2766.

Munder, M.C., Midtvedt, D., Franzmann, T., N€uske, E., Otto, O., Herbig, M.,

Ulbricht, E., M€uller, P., Taubenberger, A., Maharana, S., et al. (2016). A pH-

driven transition of the cytoplasm from a fluid- to a solid-like state promotes

entry into dormancy. eLife 5, e09347.

Nano, N., and Houry, W.A. (2013). Chaperone-like activity of the AAA+ proteins

Rvb1 and Rvb2 in the assembly of various complexes. Philos. Trans. R. Soc.

Lond. B Biol. Sci. 368, 20110399.

Narayanaswamy, R., Levy, M., Tsechansky, M., Stovall, G.M., O’Connell, J.D.,

Mirrielees, J., Ellington, A.D., and Marcotte, E.M. (2009). Widespread reorga-

nization of metabolic enzymes into reversible assemblies upon nutrient starva-

tion. Proc. Natl. Acad. Sci. USA 106, 10147–10152.

O’Connell, J.D., Tsechansky, M., Royal, A., Boutz, D.R., Ellington, A.D., and

Marcotte, E.M. (2014). A proteomic survey of widespread protein aggregation

in yeast. Mol. Biosyst. 10, 851–861.

Patel, A., Lee, H.O., Jawerth, L., Maharana, S., Jahnel, M., Hein, M.Y., Stoy-

nov, S., Mahamid, J., Saha, S., Franzmann, T.M., et al. (2015). A Liquid-to-

Solid Phase Transition of the ALS Protein FUS Accelerated by Disease

Mutation. Cell 162, 1066–1077.

Patel, A., Malinovska, L., Saha, S., Wang, J., Alberti, S., Krishnan, Y., and Hy-

man, A.A. (2017). ATP as a biological hydrotrope. Science 356, 753–756.

Pu, S., Vlasblom, J., Emili, A., Greenblatt, J., and Wodak, S.J. (2007). Identi-

fying functional modules in the physical interactome of Saccharomyces cere-

visiae. Proteomics 7, 944–960.

Pu, S., Wong, J., Turner, B., Cho, E., and Wodak, S.J. (2009). Up-to-date cat-

alogues of yeast protein complexes. Nucleic Acids Res. 37, 825–831.

Rabouille, C., and Alberti, S. (2017). Cell adaptation upon stress: the emerging

role of membrane-less compartments. Curr. Opin. Cell Biol. 47, 34–42.

Schuldiner, M., Collins, S.R., Thompson, N.J., Denic, V., Bhamidipati, A.,

Punna, T., Ihmels, J., Andrews, B., Boone, C., Greenblatt, J.F., et al. (2005).

Exploration of the function and organization of the yeast early secretory

pathway through an epistatic miniarray profile. Cell 123, 507–519.

Shah, K.H., Nostramo, R., Zhang, B., Varia, S.N., Klett, B.M., and Herman, P.K.

(2014). Protein kinases are associated withmultiple, distinct cytoplasmic gran-

ules in quiescent yeast cells. Genetics 198, 1495–1512.

Snider, J., and Stagljar, I. (2016). Membrane Yeast Two-Hybrid (MYTH) Map-

ping of Full-Length Membrane Protein Interactions. Cold Spring Harb. Protoc.

2016, pdb.top077560.

Wallace, E.W., Kear-Scott, J.L., Pilipenko, E.V., Schwartz, M.H., Laskowski,

P.R., Rojek, A.E., Katanski, C.D., Riback, J.A., Dion, M.F., Franks, A.M.,

et al. (2015). Reversible, Specific, Active Aggregates of Endogenous Proteins

Assemble upon Heat Stress. Cell 162, 1286–1298.

Weitzel, G., Pilatus, U., and Rensing, L. (1987). The cytoplasmic pH, ATP con-

tent and total protein synthesis rate during heat-shock protein inducing treat-

ments in yeast. Exp. Cell Res. 170, 64–79.

Wiemken, A. (1990). Trehalose in yeast, stress protectant rather than reserve

carbohydrate. Antonie van Leeuwenhoek 58, 209–217.

Wodak, S.J., Pu, S., Vlasblom, J., and Seraphin, B. (2009). Challenges and re-

wards of interaction proteomics. Mol. Cell. Proteomics 8, 3–18.

Woolford, J.L., Jr., and Baserga, S.J. (2013). Ribosome biogenesis in the yeast

Saccharomyces cerevisiae. Genetics 195, 643–681.

Zhao, R., Davey, M., Hsu, Y.C., Kaplanek, P., Tong, A., Parsons, A.B., Krogan,

N., Cagney, G., Mai, D., Greenblatt, J., et al. (2005). Navigating the chaperone

network: an integrative map of physical and genetic interactions mediated by

the hsp90 chaperone. Cell 120, 715–727.