DOI: 10.1126/science.1251358, 711 (2014);344 Science

et al.Alexander M. JonesArabidopsisA Membrane-Linked Interactome of −−Border Control

This copy is for your personal, non-commercial use only.

clicking here.colleagues, clients, or customers by , you can order high-quality copies for yourIf you wish to distribute this article to others

  here.following the guidelines

can be obtained byPermission to republish or repurpose articles or portions of articles

  ): May 15, 2014 www.sciencemag.org (this information is current as of

The following resources related to this article are available online at

http://www.sciencemag.org/content/344/6185/711.full.htmlversion of this article at:

including high-resolution figures, can be found in the onlineUpdated information and services,

http://www.sciencemag.org/content/suppl/2014/05/14/344.6185.711.DC1.html can be found at: Supporting Online Material

http://www.sciencemag.org/content/344/6185/711.full.html#ref-list-1, 36 of which can be accessed free:cites 80 articlesThis article

http://www.sciencemag.org/cgi/collection/botanyBotany

subject collections:This article appears in the following

registered trademark of AAAS. is aScience2014 by the American Association for the Advancement of Science; all rights reserved. The title

CopyrightAmerican Association for the Advancement of Science, 1200 New York Avenue NW, Washington, DC 20005. (print ISSN 0036-8075; online ISSN 1095-9203) is published weekly, except the last week in December, by theScience

on

May

16,

201

4w

ww

.sci

ence

mag

.org

Dow

nloa

ded

from

o

n M

ay 1

6, 2

014

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fr

om

on

May

16,

201

4w

ww

.sci

ence

mag

.org

Dow

nloa

ded

from

o

n M

ay 1

6, 2

014

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fr

om

on

May

16,

201

4w

ww

.sci

ence

mag

.org

Dow

nloa

ded

from

o

n M

ay 1

6, 2

014

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fr

om

on

May

16,

201

4w

ww

.sci

ence

mag

.org

Dow

nloa

ded

from

Border Control—A Membrane-LinkedInteractome of ArabidopsisAlexander M. Jones,1 Yuanhu Xuan,1* Meng Xu,1 Rui-Sheng Wang,2 Cheng-Hsun Ho,1

Sylvie Lalonde,1 Chang Hun You,1 Maria I. Sardi,1† Saman A. Parsa,1 Erika Smith-Valle,1‡Tianying Su,1§ Keith A. Frazer,1 Guillaume Pilot,1,3 Réjane Pratelli,1,3 Guido Grossmann,1||Biswa R. Acharya,4 Heng-Cheng Hu,5¶ Cawas Engineer,6 Florent Villiers,5 Chuanli Ju,5

Kouji Takeda,5 Zhao Su,4 Qunfeng Dong,7 Sarah M. Assmann,4 Jin Chen,1,8 June M. Kwak,5,9

Julian I. Schroeder,6 Reka Albert,2 Seung Y. Rhee,1# Wolf B. Frommer1#

Cellular membranes act as signaling platforms and control solute transport. Membrane receptors,transporters, and enzymes communicate with intracellular processes through protein-proteininteractions. Using a split-ubiquitin yeast two-hybrid screen that covers a test-space of 6.4 × 106

pairs, we identified 12,102 membrane/signaling protein interactions from Arabidopsis. Besidesconfirmation of expected interactions such as heterotrimeric G protein subunit interactions andaquaporin oligomerization, >99% of the interactions were previously unknown. Interactions wereconfirmed at a rate of 32% in orthogonal in planta split–green flourescent protein interaction assays,which was statistically indistinguishable from the confirmation rate for known interactions collectedfrom literature (38%). Regulatory associations in membrane protein trafficking, turnover, andphosphorylation include regulation of potassium channel activity through abscisic acid signaling,transporter activity by a WNK kinase, and a brassinolide receptor kinase by trafficking-related proteins.These examples underscore the utility of the membrane/signaling protein interaction network for genediscovery and hypothesis generation in plants and other organisms.

Genome projects have provided the inven-tories of genes and predicted proteins,yet we can only begin to understand how

organisms function or interact with their envi-ronment once we understand the functions of theproteins in the proteome and their wiring di-

agram. Plants, as sessile organisms, are particu-larly efficient in acclimating to the dynamics oftheir environments. Plant growth, development,homeostasis, and acclimation require mecha-nisms that monitor changes in the environment,coordinate cells and compartments, and adjusttransport of ions and metabolites across cellularmembranes. Sensing, signaling, and transporterregulation are mediated through interactionswith membrane proteins. Largely because of tech-nical challenges caused by the hydrophobicityof membrane proteins, only a small number ofmembrane protein interactions are known. Mem-branes contain thousands of proteins whose bio-chemical or physiological functions have not beenidentified experimentally and are thus classifiedas “unknowns.” Identification of genetic and mo-lecular interactions is a promising way to assignfunctions to the unknowns in the genome (1–3).

To uncover membrane protein interactions,we performed a systematic binary interactionscreen using a yeast two-hybrid system specif-ically developed for membrane protein interac-tions: the mating-based split-ubiquitin system(mbSUS) (4, 5). The split-ubiquitin system iden-tified interactions between integral membraneproteins that form homo- and hetero-oligomericcomplexes such as sucrose and ammonium trans-porters, potassium channels, and aquaporins(4–6). The mbSUS was also instrumental foridentifying regulators of yeast ammonium andamino acid transporters (7), interactions amongyeast membrane proteins (8), and interactionsbetween plant proteins and pathogen effectorproteins (9). Here, we report a binary screen,

covering more than 6.4 million potential interac-tion pairs from a library of 3286 membrane andsignaling proteins from Arabidopsis thaliana(Arabidopsis) (Fig. 1 and fig. S1) (5). Proteinsfrom the Gene Ontology (GO) biological pro-cess categories of transport, signal transduction,and response to abiotic or biotic stimulus arewell represented. Our analysis verified expectedinteractions, such as those between heterotri-meric G protein subunits (10) and aquaporinisoforms (11), and also identified over 12,000new protein-protein interactions. Of these, 3849interactions connected proteins lacking exper-imentally derived functional annotations to in-teraction partners with experimentally derivedfunctional annotations; these connections mayassist in elucidating the functions of the un-known proteins. The results are freely accessibleonline (www.associomics.org).

mbSUS Screen for Membrane-LinkedProtein-Protein InteractionsA primary mbSUS screen probed pair-wise in-teractions among 3286 distinct proteins (Fig. 1,table S1, and supplementary materials). For in-teraction tests in mbSUS, N- and C-terminalhalves of ubiquitin (Nub and Cub) are separatelyfused to proteins of interest (Fig. 1A) (4, 5). Thekey advantage of mbSUS is its ability to detectprotein interactions that occur outside of theyeast nucleus, specifically at membranes thatinterface the cytosol (5). Besides interactionsamong membrane proteins, we tested interac-tions of membrane proteins with soluble pro-teins predicted to be involved in signaling (asNub-fusions). Cub-fusions with soluble proteinsare prone to self-activation and were thereforeexcluded. An initial screen identified “auto-activating” or “nonactivatable” Cub-fusions forexclusion, further reducing the number of inter-action tests to be performed. The resulting matrixof 3233 Nub-fusions against 1070 Cub-fusionscomprised >3 × 106 interaction tests performedin duplicate and at two stringency conditions (atotal of four assays per pair), with a quantifi-able readout of yeast colony growth assayedfor complementation of histidine auxotrophy(Fig. 1B). We created imaging and statistical anal-ysis pipelines based on distribution model-basedclassification so as to determine interaction con-fidence scores and thresholds for positive in-teractions (figs. S2 and S3). A total of 30,426interactions, less than 1% of the pairs interro-gated, were defined as positive in the primaryscreen based on stringent criteria (supplementarymaterials 2). We evaluated all positive interac-tions from the primary screen in a secondary screenwith 12 additional interaction tests in yeast: sixreplicated tests for complementation of histidineauxotrophy at two stringency conditions (fig. S4).A network of 12,102 interactions between 1523 pro-teins tested positive consistently at both stringencyconditions and was termed MIND1 (Membrane-linked Interactome Database version 1) (fig. S5).

RESEARCHARTICLES

1Department of Plant Biology, Carnegie Institution for Science,CA 94305, USA. 2Department of Physics, Pennsylvania StateUniversity, University Park, PA 16802, USA. 3Department ofPlant Pathology, Physiology, and Weed Science, VirginiaPolytechnic University and State University, Blacksburg, VA24061, USA. 4Department of Biology, Pennsylvania StateUniversity, University Park, PA 16802, USA. 5Department ofCell Biology and Molecular Genetics, University of Maryland,College Park, MD 20742, USA. 6Cell and Developmental BiologySection, Division of Biological Sciences, University of California,San Diego, La Jolla, CA 92093, USA. 7Department of BiologicalSciences, University of North Texas, Denton, TX 76203, USA.8Michigan State University–U.S. Department of Energy (MSU-DOE) Plant Research Laboratory and Department of ComputerScience and Engineering, Michigan State University, EastLansing, MI 48824, USA. 9Center for Plant Aging Research,Institute for Basic Science, Department of New Biology, DaeguGyeongbuk Institute of Science and Technology, Daegu 711-873, Republic of Korea.*Present address: College of Pharmaceutical Sciences, WenzhouMedical University, Wenzhou, Zhejiang, 325035 China.†Present address: Department of Microbiology Doctoral TrainingProgram, University of Wisconsin-Madison, Madison, WI 53715, USA.‡Present address: Five Prime Therapeutics, South San Francisco,CA 94080, USA.§Present address: Department of Biology, Stanford University,Stanford, CA 94305, USA.||Present address: Centre for Organismal Studies, UniversitätHeidelberg, 69120 Heidelberg, Germany, Cluster of ExcellenceCellNetworks, 69120 Heidelberg, Germany.¶Present address: Department of Agronomy, Iowa StateUniversity, Ames, IA 50011, USA.#Corresponding author. E-mail: wfrommer@stanford.edu(W.B.F.); srhee@carnegiescience.edu (S.Y.R.)

www.sciencemag.org SCIENCE VOL 344 16 MAY 2014 711

To evaluate the reliability of MIND1, a ran-dom subset of 7770 positive interactions fromthe primary screen was reanalyzed in a tertiaryscreen (fig. S4). Results from the secondaryscreen were largely confirmed by the tertiaryscreen, which included two histidine auxotro-phy assays and a LacZ activity assay; only 12%

of secondary screen positives failed in all threetertiary screen assays, whereas 80% were pos-itive in at least two assays (fig. S5). Becausewe tested the interactions of Arabidopsis pro-teins in a single-cell heterologous system, theobserved associations are not necessarily rele-vant in planta. Therefore, we also tested 195

interactions in planta using the orthogonal, low-throughput split–green flourescent protein (GFP)protein interaction assay (Fig. 1C and supple-mentary materials 2.6). These pairs tested pos-itive in 31.8% of the split-GFP interaction tests[95% confidence interval (CI) (25.6%, 38.5%),bootstrap analysis]. We estimated the in planta

1,070 proteinstested as

Cub-fusions

3,233 proteins tested as Nub-fusions

Nub Cub

TF

TF Release

Marker Activation

uas

TF

TF

CubNub

A

D

BMIND1 membrane protein-membrane protein interactionMIND1 membrane protein-soluble protein interaction

Transporter

Trafficking

Receptor-like Kinase

Kinase/PhosphataseEnzym

e

Unknown - m

embrane

Other - m

embrane

Ubiquitination

Small G

TPase Signaling

Receptor

Other - membrane

Unknown - membrane

Trafficking

Enzyme

Ubiquitination

Kinase/Phosphatase

Receptor-like Kinase

Receptor

Transporter

E

Unknown - soluble

Other - soluble

Other - soluble

Unknown - soluble

Split-GFP assay

NRSPRS

195

Per

cent

age

of p

ositi

ve te

sts

MIND1

49120

# of interactionstested

C

peroxisome

mitochondrion

golginucleus

vacuole

extracellular

cytosol

endoplasmic reticulum

plastid

plasma membrane

Unknown

ThioredoxinATPase

Trafficking

TetraspanninCystein rich repeatsecretory protein

Chloroplast

Phosphatase

Protease

Transporter

AGP

Ankyrin

Calcium binding

Kinase

Small GTPasesignaling

Receptor

Peptidase

Enzyme

E3 ligase

Transcription factor

0%

10%

20%

30%

40%

50%

60%

Fig. 1. The Arabidopsis Membrane-linked Interactome (MIND1). (A) Apositive result in the mbSUS screen is the result of four molecular steps: Theinteraction of the proteins of interest brings into proximity the Nub and Cubmoieties, reconstituting a ubiquitin protein (UBQ) that is then recognized by anendogenous ubiquitin specific protease that cleaves the Cub fusion protein torelease a transcription factor (TF). The TF is then free to enter the nucleus and

activate marker genes whose expression indicates physical interaction between the proteins of interest. (B) 12,102 interactions in MIND1 generated from two rounds ofmbSUS screen. Chart displays protein-protein interactions between a membrane protein with another membrane protein or a soluble protein. (C) The results oforthogonal validation of the MIND1 protein-protein interactions by retesting in the split-GFP assay. Error bars show 95% CI estimated with bootstrap analysis(supplementary materials 2.6). (D) Protein family interaction network. Nodes represent family types and edges indicate interactions between two families. Only familiesthat have more than two proteins are included, and only those family pairs that have 10 or more interactions are displayed. (E) Cellular localization interactionnetwork of MIND1NH. Nodes represent cellular components; edges indicate interactions between or among proteins localized in the respective compartments.

16 MAY 2014 VOL 344 SCIENCE www.sciencemag.org712

RESEARCH ARTICLES

false-positive rate of MIND1 by comparing itssplit-GFP validation rate to the rate of split-GFP validation for two reference sets of proteinpairs (supplementary materials 2.6). A posi-

tive reference set of 49 independently reportedinteractions—randomly sampled from the subsetof primary screen protein pairs that were re-ported to physically interact in the literature—was

used to estimate the true-positive rate of inter-actions tested in the split-GFP assay, which was38.8% [95%CI (24.5%, 53.1%), bootstrap anal-ysis]. A negative reference set of interactions—randomly sampled from the primary screen proteinpairs—was used to estimate the false-positiverate of interactions tested in the split-GFP assay,which was 2.5% [95% CI (0, 5.8%), bootstrapanalysis]. MIND1 interactions tested positiveat a rate that is not significantly different fromthe independently reported positive interactions(positive reference set, P = 0.364, permutationtest) and over 12-fold higher than noninteractingprotein pairs (negative reference set, P value = 0,permutation test). The MIND1 in planta false-positive rate was estimated to be 19.2% [95%CI(0, 46.6%), bootstrap analysis], although the smallfraction of interactions tested in the split-GFPassay precludes a statistically reliable point esti-mate of the false-positive rate.

The Potential of Low Stringency Interactionsfor DiscoveryMIND1 does not include interaction pairs thattested positive in only one or two out of the fourtests performed in the primary screen. However,we provide these interactions online (www.associomics.org) because they could be biolog-ically relevant. For example, four different SWEETtransporter hetero-oligomers were detected inonly one out of the four primary screen repli-cate tests. On the basis of this information, we

MIND1 BP-based-network

lipid biosynthesisfatty acid biosynthesis

fatty acid metabolism

organic acid metabolism

cation transport metal ion transport

fruit development

multicellular organismaldevelopment

reproductive structure development

anatomical structure development

reproduction protein modification

RLK signaling pathway

phosphorylation

protein catabolism

catabolismcell communication

drug transport

macromolecule biosynthesis

primary metabolism

gene expression

transmembrane transport

establishment of localization

N−terminal protein lipidation

protein processing

signal peptide processing

BP-BP association enriched in MIND1 BP-BP association enriched in MIND1NH

Metabolism

Protein

Transport

Growth anddevelopment

Regulation ofprocess

regulation

Fig. 2. Enriched connections between GO biological process groups in MIND1. Nodes representdifferent GO groups and are color-coded by group type as shown. Edges represent statistically enrichedconnections between GO groups in the complete MIND1 (solid lines) or in MIND1NH (dashed lines)determined by comparing with 10,000 randomized networks and using a false discovery rate Q < 0.05 asthreshold.

Evidence for ABA association: Genetic Evidence Gene Ontology Gene ExpressionMIND1NH ABA Signaling subnetwork

DUF26−Ib

AT3G56370

AT5G47530

AT5G10290

HAE

AT1G71940

MYB33

DGL1

PYR1

TTH1

DUF26−IbMHX1

PK3ERD4

CPK6

AAP3AT1G12500

AT2G01680

CAS

BCB

HOS3−1ROP10

AT1G72300

DUF26−IfAT3G53190

TBL36AT2G37050

erd6−like1

RABB1BAT4G12000

ELIP1

FEI1

erd6−like5 WAT1

ADS2

NRT1.8

AT5G13400

ANNAT1CYP707A2

ProT1

CERK1

CESA8

RBOHD

VSR1AT4G13530

AT5G19130 AT5G42420

AT1G62330ABCG10ABCG16

AT2G01970ABCG20

AMT1;1

ELIP2PERK15

AT5G10020

AT1G49230

AT1G54540

HHP1

AT3G19100

OCP3

LECRKA4.1

CPK3

ANAC089

AT3G28040

AT2G17972

NRT1.9

AT3G14860

ETR1

AT1G68400

MRP5

CPD

BAT5NEK3

HA2

AT2G32380

SWEET1

AT5G38990

NCRK

CORI1CYP705A4

PAD3

AT4G31830

KAT1

AP2C1

AAP6

AT1G51800

AT5G66440

AT1G04360

ALMT1

CPK15

AT1G67470

PIP2;5

DUF26−IbPIP1;4

AT4G03820

GRH1

AGP16

AT2G42390

DTX28

CPK30

AT3G06470

ERD2

RAB11A

NRAMP3

AT5G35460

AT5G49760

AT5G17210

AT1G66760

NHL1

LRR1

AT5G07340

Node degree: 701Node membrane association: SolubleMembrane

Fig. 3. Interactions between abscisic acid signaling-related pro-teins. Three types of evidence (genetic evidence, GO annotation,and abscisic acid responsive gene expression) were used to classify

MIND1NH nodes as related to abscisic acid signaling. Interactionsinvolving hubs and unconnected abscisic acid–related proteins arenot included.

www.sciencemag.org SCIENCE VOL 344 16 MAY 2014 713

RESEARCH ARTICLES

performed a systematic analysis to detect homo-and hetero-oligomerization among all 17 ArabidopsisSWEETs and found 8 homo- and 47 hetero-oligomers using the split-ubiquitin system (12).Several of these pairs were confirmed in plantaby using the split-GFP assay, suggesting that oligo-merization is a general feature of SWEETs andhighlighting that pairs that did not yield highlyreproducible results may be valuable for inclu-sion in specific hypothesis-driven studies.

MIND1 Network CharacteristicsMIND1, like other biological networks (13, 14),exhibits small-world properties (short distanceamong proteins coupled with local order indicatedby high clustering coefficient) (supplementarymaterials 3) and follows a heavy-tailed degreedistribution with a reliance on high-degree hubs(proteins with many interactions) (fig. S6). Be-cause hubs have many interaction partners, in-dividual interactions of a hub have low specificity.Thus, hub proteins with degree >70were removedto generate a non-Hub network (MIND1NH) of3354 interactions (fig. S7). Interacting proteinsoften share similar expression patterns (15);therefore, we compared expression correlation forMIND1- and MIND1NH-interacting protein pairs.Relative to noninteracting pairs, only MIND1NHpairs were enriched for expression correlation(fig. S8). However, bothMIND1- andMIND1NH-interacting protein pairs were enriched for GOfunctional similarity (fig. S9), indicating that ahub protein was more likely to share a function-al annotation with its interaction partners thanan expression pattern. MIND1 interactions arecomplementary to existing interactome net-works from Arabidopsis in that over 99% ofthe interactions had not been previouslyreported (13). This is not unexpected becausedifferent interactome assays typically yieldcomplementary data sets (14, 16–18) andbecause the clone sets analyzed were different(only 212 MIND1 proteins were also screened

by the Arabidopsis Interactome Mapping Con-sortium) (fig. S10) (13).

MIND1-Derived Functional PredictionsBecause 327 proteins (21%) in MIND1 lackGO biological process annotation, the networkcan potentially contribute to functional pre-dictions (fig. S11). For example, the small, sin-gle transmembrane-domain–containing proteinAT5G61630 [a potential Ras guanosine triphos-phatase (GTPase); www.greenphyl.org] hadthree interactors that were annotated “waterchannel activity” (GO:0015250), intimating arole in regulating water transport (table S2). Wesystematically generated functional predictionsfor the proteins in MIND1NH by identifyingfunctional annotations that were overrepre-sented among each protein’s interacting partnersrelative to the proteins present in the network(table S2). For the above example, the anno-tation “water channel activity” is overrepre-sented in the interaction partners of AT5G61630(P = 5.68 × 10–6, Fisher’s exact test). We alsogenerated functional predictions for the proteinsin MIND1 by first isolating highly connectednetwork clusters (putative functional modules)and then identifying annotations enriched withinthese clusters. Hierarchical clustering of MIND1resulted in a large super-cluster dominated by in-teractions involving hub proteins, whereas hierar-chical clustering ofMIND1NH resulted in numeroussmaller clusters (fig. S13). Two additional moduledetection methods (supplementary materials 3.7)identifiedMIND1NH clusters with enriched GOannotations, potentially indicating functionalmodules (table S4 and fig. S14). For example,the unknown gene described above, AT5G61630,belongs to a cluster that contains five aquaporinisoforms and is enriched for “water transport”(GO:0006833, Bonferroni-corrected P value =1.36 × 10–4, Fisher’s exact test), a function thatmaypertain to other members of the cluster, includ-ing AT5G61630.

Individual interactions, irrespective of thenetwork context, also guide functional predic-tions. For example, we found interactions betweenthe nitrate transceptor CHL1 (AT1G12110) andthe potassium transporter KT2 (AT2G40540) aswell as the lysine-deficient protein kinaseWNK8(AT5G41990). CHL1’s transport activity or con-formation was affected by these interactions, asdemonstrated with the fluorescent transport ac-tivity sensor NiTrac1 (19). WNK8 is a key playerinvolved in dose- and duration-dependent sugarsignaling, which controls endocytosis of the Gprotein–coupled receptor RGS1 (20). Interestingparallels exist in the animal kingdom, in whichWNK kinases play important roles in controllingion transport processes (21).

Characterization of Hubs in MIND1Disruption of network hubs typically has pleio-tropic effects (22). Purifying selection acts morestrongly on genes with functions essential forbiological fitness and results in reduced evolu-tionary rates (23). Consistently, MIND1 hubsshowed evidence of purifying selection in thatKa/Ks [the ratio of the number of nonsynon-ymous substitutions per potential nonsynony-mous site (Ka) to the number of synonymoussubstitutions per potential synonymous site(Ks)] based on Arabidopsis lyrata orthologswas low (mean Ka/Ks = 0.15) and significantlylower than Ka/Ks for all Arabidopsis proteinswith A. lyrata orthologs (mean Ka/Ks = 0.21, Pvalue = 0.028, Student’s t test) (supplementarymaterials 3.8 and table S3). Although we cannotexclude the possibility that the high number ofprotein interactions for an individual MIND1hub could be an artifact, the functions of thehubs are consistent with general roles thatrequire interaction with a large number of targetproteins; out of the 33 MIND1 hubs with knownor inferred molecular function, 21 have putativefunctions in protein modification (chaperones/thioredoxins, signal peptidases, or proteases) orprotein sorting [soluble N-ethylmaleimide–sensitive factor (NSF) attachment protein (SNAP)receptors (SNAREs) or cornichon] (table S3).Moreover, subcellular localization of 16 MIND1hub GFP-fusions, including nine unknowns,showed that all but one localized to endomem-branes, which is consistent with roles in mem-brane protein modification or sorting (fig. S15and table S3).

Analysis of MIND1 at the Level of ProteinFamilies and Biological ProcessesAnalysis of interactions at the protein familylevel can be used to predict genetic redundancies,interaction motifs, and potential interactions withparalogs. A gene family–based network (Fig. 1D,figs. S16 and S17, and table S5) revealed over-representation of interactions involving proteinsof unknown function. The unknowns showednumerous interactionswith transporters, indicatingundiscovered transport or transport-regulatoryfunctions among these proteins (Fig. 1D).

0

KAT1KAT1+AP2C1AP2C1water-injected

5

10

15

20

-40-80-120-160-200 (mV)

(µA)

PYR1

KAT1

AP2C1

AT5G66440

ABA signaling ABA regulated

Node degree: 701

SolubleMembrane

A B

Fig. 4. AP2C1 regulates the potassium channel KAT1. (A) TheBlondel clustering algorithm revealed an abscisic acid–relatedcluster (Blondel cluster 93) in MIND1NH that includes KAT1 and AP2C1. (B) Current-voltage rela-tionship of the current in Xenopus laevis oocytes injected with KAT1 or AP2C1 or with both cRNAs uponperfusion with 100mMKCl. In all recordings, the holding potential was set at –40 mV, and voltage stepswere applied to potentials ranging between –40 and –200 mV with –20-mV decrements. Data aremeans T SE from at least three different individual oocytes.

16 MAY 2014 VOL 344 SCIENCE www.sciencemag.org714

RESEARCH ARTICLES

Aggregating interactions by predicted subcellularlocalization yielded a location-based network ofconnections between different cellular compart-ments predicting higher-level association patterns(Fig. 1E). MIND1NH proteins are predominantlyplasma membrane–localized, and these plasmamembrane proteins show abundant interactionswith other plasma membrane proteins and withcytosolic partners (Fig. 1E).

To investigate potential interaction motifsamong biological processes, we examined inter-actions at the level of GO biological processes.Intrabiological process interactions are lessabundant in MIND1 as compared with otherpublished networks (fig. S18) (13). We generateda meta-network comprising enriched interac-tions between biological processes. This networkrevealed potential interaction motifs such as“Receptor-like Kinase/Pelle (RLK) signalingpathway” linking to “protein catabolism” (Fig. 2).

Regulation of Transport Activityby HormonesTo strengthen functional predictions fromMIND1,the protein interaction network was overlaid

with transcriptome and other annotation data.For example, 311 proteins in MIND1 are relatedto the hormone abscisic acid according to tran-scriptomic, genetic, or other experimental evi-dence (24). MIND1 interactions connected manyof these proteins into an abscisic acid–relatedsubnetwork (Fig. 3 and fig. S19). Similar sub-networks obtained for other hormones alsorepresent interaction sets that are strengthenedby multiple data sources (figs. S19 to S26).

Transport is highly regulated at the transcrip-tional and posttranscriptional levels, and hormonesignaling often targets membrane-related pro-cesses. The activity of potassium channels—which contribute to potassium acquisition fromsoil, regulation of enzyme activities, adjustmentof membrane potential and turgor, and regulationof cellular homeostasis and electrical signaling—is regulated by protein kinases and phosphatases(25). A key mechanism by which abscisic acidpromotes stomatal closure is through inhibitionof transporters such as the potassium channelKAT1 (26). Abscisic acid responses are mediatedby a co-receptor complex that consists of abscisicacid receptors (PYR/PYL/RCAR proteins) and

clade-A protein phosphatase 2Cs (PP2C) (27).Although much of abscisic acid signaling pro-ceeds through clade-A PP2C regulation of SnRK2kinases and SnRK2 regulation of transcriptionfactors, cladeA PP2Cs and SnRK2 kinases canalso directly regulate the activity of K+ channels(28, 29).

Abscisic acid signaling proteins in MIND1share extensive connections with transportersand othermembrane proteins (Fig. 3 and fig. S19).For example, the clade-B PP2C, AP2C1, inter-acts with the K+ channel KAT1 in a cluster en-riched for abscisic acid–related GO annotations(Fig. 4A); Blondel cluster 93 (table S4) is en-riched for GO:0009738—“abscisic acid signal-ing pathway” (Bonferroni-corrected P = 1.25 ×10–2, Fisher’s exact test)—and GO:0071215—“cellular response to abscisic acid stimulus”(Bonferroni-corrected P = 1.41 × 10–2, Fisher’sexact test). Although clade-B PP2Cs are notcanonical abscisic acid coreceptor componentsand are not known to regulate potassium chan-nels, when coexpressed with KAT1 in oocytes,AP2C1 completely inhibited channel activity(Fig. 4B). KAT1 functions in stomatal opening;

Fig. 5. Functional inter-action of BRI1 with traf-ficking proteins. Split-GFPassay for VAMP727, SYP22,andBRI1, andbrassinolide (BL)hyposensitivity of vamp727 –/+;syp22 and VAMP727 over-expressed (OX) plants. (A)Yellow fluorescent protein(YFP) fluorescence and brightfield images (left, fluores-cence channel; right, brightfield). (a and b) Reconstitu-tion of YFP fluorescence fromnYFP-VAMP727+ BRI1-cCFP;(c and d) nYFP-SYP22+BRI1-cCFP; (e and f) BRI1-cCFP+nYFP (negative control). Scalebars, 20 mm. (B) VAMP727overexpression plants weregrown under indicated con-centrations of BL, and hypo-cotyl lengths were measured.(C) Seedlings were grownunder various concentrationsof BL, and hypocotyl lengthswere measured. (B and C)Error bars indicate SE. Experi-ments were repeated at leastthree times. (D) Wild typeand vamp727−/+;syp22 weregrown for5dayswith100nMor without BL. BL-mediatedhypocotyl elongation and pri-mary root growth inhibi-tion (coiling) were affectedin vamp727−/+;syp22 com-pared with wild-type plants.

BRI1-cCFP/ nYFP-VAMP727

GFP DIC

BRI1-cCFP/ nYFP-SYP22

BRI1-cCFP/ nYFP WT vamp727-/+/syp22

BL - + - +

0

50

100

150

200

0 10 100 200 nM BL

Col-0 OX8 OX9

Hyp

ocot

yl le

ngth

(%

) H

ypoc

otyl

leng

th (

%)

0

50

100

150

200

0 10 100 200 nM BL

WT vamp727-/+/syp22

a b

c d

e f

A B

C

D

www.sciencemag.org SCIENCE VOL 344 16 MAY 2014 715

RESEARCH ARTICLES

thus, AP2C1 could promote stomatal closure anddrought responses by inhibiting KAT1. Indeed,an ap2c1,pp2c5 double mutant has increasedstomatal aperture (30), and we speculate that clade-B PP2Cs such as AP2C1 and PP2C5may reducestomatal opening by directly modifying KAT1.

Receptor Kinase Signaling and RegulationThe RLK family of transmembrane receptors islarger in plants as compared with animals andlikely serves as a predominant mechanism ofcommunication across membranes. However,characterization of RLKs is complicated by ap-parent genetic redundancy (31). MIND1 contains554 interactions for 175 full-length RLK proteins,extending RLK associations within membranesand with soluble signaling proteins (fig. S27).MIND1NH contains >20 RLK interactions withsmall GTPases (fig. S27). At present, no canon-ical pathway for intracellular RLK signalinganalogous to the metazoan G protein–coupledreceptor (GPCR) transmembrane signaling throughintracellular heterotrimeric G proteins has beendefined in plants. However, small GTPases andrelated proteins have been identified as intra-cellular RLK signaling components (32). RLK-GTPase signaling could act in several ways—forexample, via direct signal transduction from RLKactivation to ROP activation. We observed nu-merous RLK interactions with Rab-GTPases,proteins involved in trafficking (33), potentiallyindicatingRab involvement in trafficking of RLKsas shown for RabF2b regulation of FLS2 endo-cytosis (34). ExtendingRLK-GTPase associationsby usingMIND1 and additional interactome datasets (13) reveals additional potential GTPaseinteraction motifs that could help advance theunderstanding of the role of small GTPases(fig. S27).

Regulation of Brassinosteroid Signaling byTrafficking ProteinsMembrane proteins are maintained at appropriatesteady-state levels by a balance of delivery tothe plasma membrane and recycling throughendocytosis. Receptor endocytosis plays a keyrole in controlling activation and signal termi-nation (35). For example, upon activation byligands, mammalian RLKs are subject to ac-celerated lysosomal degradation (36). In Arabi-dopsis, brassinosteroids such as brassinolide areperceived by the RLK BRI1 and its coreceptorBAK1. Ligand-independent trafficking of BRI1between plasma membrane and early endo-somes and degradation in the vacuole has beenobserved (37, 38), and the Membrane SteroidBinding Protein (MSBP1) was shown to triggerBAK1 endocytosis in a brassinolide-independentmanner (39). Although considerable evidenceimplicates BRI1-trafficking in brassinosteroidsignaling, the underlying mechanisms are stillambiguous (40), and the interactions found inMIND1 may help solve some of the open ques-tions regarding RLK trafficking. Here, we foundthat BRI1 interactswith theR-SNAREVAMP727,

a candidate for regulating BRI1 trafficking (tableS1). We confirmed this interaction using the split-GFP assay and also identified SYP22, aVAMP727interacting Q-SNARE (41), as a BRI1-interactor(Fig. 5A). VAMP727 and SYP22 accumulate atlate endosomes/multivesicular bodies (41). Over-expression of VAMP727 in Arabidopsis resultedin partial insensitivity to brassinolide (Fig. 5B);overexpression of SYP22 in bri1-5, a weak bri1mutant, enhanced its dwarf phenotype (fig. S28).The phenotypes of VAMP727 and SYP22 over-expressors are consistent with inhibitory roles inbrassinosteroid signaling, perhaps by promotingBRI1-trafficking to the vacuole for degradation.However, although vamp727 and syp22 singlemutants did not show altered brassinolide re-sponses (fig. S29), vamp727−/+;syp22 mutantswere partially insensitive to brassinolide (Fig. 5,C and D), indicating a role for VAMP727 andSYP22 in promoting brassinosteroid signaling.We hypothesize that overexpression and muta-tion of VAMP727 and SYP22 might result inaberrant BRI1 localization because both resultin an apparent reduction in brassinosteroid sig-naling. VAMP727 had also been observed at theplasma membrane (42), implicating additionalfunctions of VAMP727 in trafficking from plasmamembrane to vacuole. Although further exper-iments are required to fully understand the roleof trafficking in BRI1 endocytosis and signaling,we provide genetic and molecular evidence thatVAMP727 and SYP22 interact with BRI1 and af-fect brassinosteroid signaling, potentially throughmodulation of steady-state levels of BRI1 at theplasma membrane or in endosomes (fig. S29).

Membrane Protein and Signaling ProteinInteractome for Functional GenomicsMembrane proteins are central components ofmany cellular processes, often through coordi-nated action with either membrane or solubleinteraction partners. The 12,102 protein-proteininteractions in MIND1 (www.associomics.org)expand the functional genomics knowledge basefor the reference plant Arabidopsis and serve asa resource for gene discovery and hypothesisgeneration. Analysis of MIND1 in conjunctionwith other data sources has not only confirmedknown interactions but also has uncovered con-nections that shed light on how membrane pro-teins are regulated at the levels of trafficking,accumulation, and activity.

References and Notes1. S. Lalonde et al., Plant J. 53, 610–635 (2008).2. C. Boone, H. Bussey, B. J. Andrews, Nat. Rev. Genet. 8,

437–449 (2007).3. M. C. Jonikas et al., Science 323, 1693–1697

(2009).4. P. Obrdlik et al., Proc. Natl. Acad. Sci. U.S.A. 101,

12242–12247 (2004).5. S. Lalonde et al., Front. Physiol 1, 24 (2010).6. W. X. Schulze, A. Reinders, J. Ward, S. Lalonde,

W. B. Frommer, BMC Biochem. 4, 3 (2003).7. G. Van Zeebroeck, M. Kimpe, P. Vandormael,

J. M. Thevelein, PLOS ONE 6, e24275 (2011).8. J. P. Miller et al., Proc. Natl. Acad. Sci. U.S.A. 102,

12123–12128 (2005).

9. J. D. Lewis et al., BMC Genomics 13, 8 (2012).10. Y. Trusov et al., Plant Cell 19, 1235–1250 (2007).11. G. P. Bienert et al., Biochem. J. 445, 101–111 (2012).12. Y. H. Xuan et al., Proc. Natl. Acad. Sci. U.S.A. 110,

E3685–E3694 (2013).13. M. Dreze et al., Science 333, 601–607 (2011).14. H. Yu et al., Science 322, 104–110 (2008).15. J. L. Sevilla et al., IEEE/ACM Trans. Comput. Biol.

Bioinformatics 2, 330–338 (2005).16. K. Tarassov et al., Science 320, 1465–1470 (2008).17. J. R. Parrish, K. D. Gulyas, R. L. Finley Jr., Curr. Opin.

Biotechnol. 17, 387–393 (2006).18. N. J. Krogan et al., Nature 440, 637–643 (2006).19. C. H. Ho, W. B. Frommer, eLife 3, e01917 (2014).20. Y. Fu et al., Cell 156, 1084–1095 (2014).21. D. Takahashi et al., Biosci. Rep. (2014).22. H. Jeong, S. P. Mason, A. L. Barabási, Z. N. Oltvai,

Nature 411, 41–42 (2001).23. R. Sterken et al., Plant Cell 21, 2987–2998 (2009).24. Z. Jiang et al., Nucleic Acids Res. 39, D1123–D1129 (2011).25. U. Anschütz, D. Becker, S. Shabala, J. Plant Physiol. (2014).26. T. H. Kim, M. Böhmer, H. Hu, N. Nishimura, J. I. Schroeder,

Annu. Rev. Plant Biol. 61, 561–591 (2010).27. S. R. Cutler, P. L. Rodriguez, R. R. Finkelstein,

S. R. Abrams, Annu. Rev. Plant Biol. 61, 651–679 (2010).28. J. Xu et al., Cell 125, 1347–1360 (2006).29. A. Sato et al., Biochem. J. 424, 439–448 (2009).30. A. K. Brock et al., Plant Physiol. 153, 1098–1111 (2010).31. L. A. Gish, S. E. Clark, Plant J. 66, 117–127 (2011).32. P. Bommert, B. I. Je, A. Goldshmidt, D. Jackson, Nature

502, 555–558 (2013).33. A. H. Hutagalung, P. J. Novick, Physiol. Rev. 91,

119–149 (2011).34. M. Beck, J. Zhou, C. Faulkner, D. MacLean, S. Robatzek,

Plant Cell 24, 4205–4219 (2012).35. E. R. Andersson, Cell. Mol. Life Sci. 69, 1755–1771 (2012).36. L. K. Goh, A. Sorkin, Cold Spring Harb. Perspect. Biol. 5,

a017459 (2013).37. N. Geldner, D. L. Hyman, X. Wang, K. Schumacher,

J. Chory, Genes Dev. 21, 1598–1602 (2007).38. C. Viotti et al., Plant Cell 22, 1344–1357 (2010).39. L. Song, Q. M. Shi, X. H. Yang, Z. H. Xu, H. W. Xue,

Cell Res. 19, 864–876 (2009).40. A. L. Contento, D. C. Bassham, J. Cell Sci. 125,

3511–3518 (2012).41. K. Ebine et al., Plant Cell 20, 3006–3021 (2008).42. K. Ebine et al., Nat. Cell Biol. 13, 853–859 (2011).

Acknowledgments: We thank W. Monsell (Pennsylvania StateUniversity) and Y. S. Han (University of Maryland) for technicalassistance and K. Revanna and V. Desu (University of NorthTexas) for assistance with the MIND web portal development.This work was made possible by a National Science Foundation(NSF) Arabidopsis 2010 grant (MCB-0618402) to W.B.F.,S.M.A., R.A., J.M.K., S.Y.R., and J.I.S. and NSF Arabidopsis 2010grant (MCB-1052348) to W.B.F. and S.Y.R., with partialsupport from NSF-MCB-1021677 (W.B.F.), NSF-MCB-0918220( J.I.S.) and NSF-MCB-1121612 (S.M.A. and R.A.). Because ofthe large number of constructs, we are not able to distributeindividual clones. We make split-ubiquitin system vectorsand an Arabidopsis ORF collection available through theArabidopsis Biological Resource Center (https://abrc.osu.edu)and split-GFP assay vectors available through AddGene(www.addgene.org). Instructions for acquiring other materialsas well as a materials transfer agreement governing plasmidsare provided online at www.associomics.org. Additionalmaterials are included in the suplementary materials. Theauthors declare that the research was conducted in theabsence of any commercial or financial relationships thatcould be construed as a potential conflict of interest.

Supplementary Materialswww.sciencemag.org/content/344/6185/711/suppl/DC1Materials and MethodsSupplementary TextFigs. S1 to S30References (43–83)Tables S1 to S6

27 January 2014; accepted 16 April 201410.1126/science.1251358

16 MAY 2014 VOL 344 SCIENCE www.sciencemag.org716

RESEARCH ARTICLES