Clusters of bioactive compounds target dynamicendomembrane networks in vivoGeorgia Drakakakia,1,2, Stéphanie Robertb,c,1,3, Anna-Maria Szatmarib,c,1, Michelle Q. Browna, Shingo Nagawaa,Daniel Van Dammeb,c, Marilyn Leonardd, Zhenbiao Yanga, Thomas Girkea, Sandra L. Schmidd, Eugenia Russinovab,c,Jirí Frimlb,c, Natasha V. Raikhela, and Glenn R. Hicksa,4

aCenter for Plant Cell Biology and Department of Botany and Plant Sciences, University of California, Riverside, CA 92521; bDepartment of Plant SystemsBiology, University of Ghent, Flanders Institute for Biotechnology (VIB), 9052 Ghent, Belgium; cDepartment of Plant Biotechnology and Genetics,University of Ghent, 9052 Ghent, Belgium; and dDepartment of Cell Biology, The Scripps Research Institute, La Jolla, CA 92037

Edited by Joseph R. Ecker, Salk Institute, La Jolla, CA, and approved September 16, 2011 (received for review May 27, 2011)

Endomembrane trafficking relies on the coordination of a highlycomplex, dynamic network of intracellular vesicles. Understandingthe network will require a dissection of cargo and vesicle dynamicsat the cellular level in vivo. This is also a key to establishing a linkbetween vesicular networks and their functional roles in de-velopment. We used a high-content intracellular screen to discoversmall molecules targeting endomembrane trafficking in vivo in acomplex eukaryote, Arabidopsis thaliana. Tens of thousands ofmolecules were prescreened and a selected subset was interro-gated against a panel of plasma membrane (PM) and other endo-membrane compartment markers to identify molecules that alteredvesicle trafficking. The extensive image dataset was transformedby a flexible algorithm into a marker-by-phenotype-by-treatmenttime matrix and revealed groups of molecules that induced similarsubcellular fingerprints (clusters). This matrix provides a platformfor a systems view of trafficking. Molecules from distinct clusterspresented avenues and enabled an entry point to dissect recyclingat the PM, vacuolar sorting, and cell-plate maturation. Bioactivityin human cells indicated the value of the approach to identifyingsmall molecules that are active in diverse organisms for biologyand drug discovery.

chemical genomics | high content screen | endosidin | endosome

The coordination of multicellular growth to establish andmaintain the morphology of eukaryotic organisms during

development is orchestrated by complex regulatory processes inwhich vesicular trafficking within the endomembrane system isessential (1, 2). The endomembrane system is a network ofinterconnected pathways required to establish signaling, cell-to-cell communication, cell polarity, and cell shape in response todevelopmental or environmental stimuli (3). Eukaryotic cellspossess the ability to internalize their plasma membrane (PM)and thus rapidly remodel their protein content (4), which is es-sential for polar growth, cytokinesis, hormone perception andtransport, response to pathogens, and metal detoxification (5–8).To dissect the endomembrane network from a systems per-

spective and to understand protein functions within the network, itis necessary to perturb trafficking in a controlled fashion and toexamine the consequences on growth and development. Theability to induce and evaluate subcellular phenotypes at a signifi-cant scale based on combinations of multiple endomembrane-specific markers is critical for characterization of the network. Inthis regard, small molecules are promising for the efficient dis-covery and evaluation of complex intracellular phenotypes be-cause, for anymolecule, lines expressing different endomembranefluorescent protein markers can be efficiently interrogated inparallel (3).In a previous pilot screen we identified endosidin 1 (ES1),

a compound that arrests PIN2 and BRI1 in SYP61 aggregates,named ES1 bodies (9). We have used a modified high-throughputconfocal-based screen focused on the rapid recycling of PMmarkers to discover molecules that target endomembrane traf-ficking in vivo in a complex eukaryote, Arabidopsis thaliana. Animage-based high-content screening dataset was generated

through the interrogation of compounds for their effects on thetrafficking of multiple PM markers. The data were transformedinto a matrix of shared phenotypic profiles of molecules (i.e.,clusters of bioactive molecules), which could be used to un-derstand pathways associated with complex multicomponentphenotypes. To demonstrate the power of our bioactive clusterapproach to guide rational and comprehensive dissection ofendomembrane trafficking in complex plants and animals, we in-vestigated representative molecules (endosidins) within distinctclusters.Unique features ofmajor pathways of the endomembranenetwork were revealed, including cellular polarity, endosomalrecycling and vacuole targeting, and cell-plate formation.

ResultsHigh-Content Cellular Screen for Molecules Affects the Recycling PMProteins. To dissect the endomembrane system in a systems-wideapproach, we performed a screen for molecules resulting in ab-errant intracellular phenotypes. We used an automated primaryscreen based on the germination and growth of free-living to-bacco pollen (9) to rapidly interrogate chemical libraries con-taining a total of 46,418 compounds (SI Appendix, Fig. S1). Wehypothesized that the system could identify molecules bioactivein different species, owing to evolutionary conservation. Here,our screen resulted in the identification of 360 (0.77%) inhibitorsof pollen germination.We established a secondary screen based on laser scanning

confocal microscopy (LSCM) for disruptors of trafficking fromthe PM to dissect endocytic processes. As model cargoes, weused the polar PM markers PIN2::PIN1:GFP (10) and PIN2::PIN2:GFP (11) and the nonpolar PM receptor BRI1::BRI1:GFP(12) expressed in Arabidopsis root tips (SI Appendix, Fig. S1).These markers are targeted to the PM through the secretorypathway and actively recycle through endosomes, which arepoorly defined in plants (13, 14). To focus on acute effects ontrafficking, we examined cells after 2 h of compound treatment.In addition, a longer-term 24-h treatment was implemented forpolar-localized PIN markers to examine effects on slower traf-ficking processes. The secondary screen resulted in about 7,500LSCM images representing the behavior of 18,570 seedlings.Overall, the LSCM secondary screen identified 123 small mole-

Author contributions: G.D., S.R., A.-M.S., M.Q.B., S.L.S., N.V.R., and G.R.H. designed re-search; G.D., S.R., A.-M.S., M.Q.B., S.N., D.V.D., and M.L. performed research; Z.Y., S.L.S.,E.R., J.F., and N.V.R. contributed new reagents/analytic tools; G.D., S.R., A.-M.S., M.Q.B.,S.N., D.V.D., M.L., T.G., S.L.S., N.V.R., and G.R.H. analyzed data; and G.D., S.R., and G.R.H.wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.1G.D., S.R., and A.-M.S. contributed equally to this work.2Present address: Department of Plant Sciences, University of California, Davis, CA 95616.3Present address: SLU/Umeå Plant Science Center, Departments of Forest Genetics andPlant Physiology, 901 83 Umeå, Sweden.

4To whom correspondence should be addressed. E-mail: ghicks@ucr.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1108581108/-/DCSupplemental.

17850–17855 | PNAS | October 25, 2011 | vol. 108 | no. 43 www.pnas.org/cgi/doi/10.1073/pnas.1108581108

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

29, 2

020

cules out of 360 (34%) that affected the localization of one ormore of the three PM markers analyzed in Arabidopsis root tip,underlying the versatility of the screen. We named this collectionof 123 molecules plasma membrane recycling compound set A(PMRA) (SI Appendix, Fig. S1).

High-Content Analysis of Induced Subcellular Phenotypes. To eval-uate and characterize the intracellular effects induced by the 360molecules in the secondary LSCM screen, a scoring system ofchemically induced subcellular phenotypes was developed. Thescoring system assigned a number and a corresponding colorcode to each subcellular phenotype. On the basis of distinct in-duced subcellular localization of the markers, we identified 18classes of compounds. Of these, 15 classes represented distinctsubcellular phenotypes induced by chemical treatment (Fig. 1Aand SI Appendix, Table S1). Class 0 represents “no chemicaleffect,” “autofluorescent compounds” are represented by class16, and “no data collected” represents class 17.To further characterize the 123 PMRA molecules, we exam-

ined the nonpolar recycling PM marker UBQ::NPSN12:YFP

(15) and the trans-Golgi network (TGN)/early endosome (EE)marker VHA-a1::VHA-a1:GFP (16) (SI Appendix, Figs. S1 andS2). The results were incorporated into a frequency plot of the18 chemical classes for all markers and treatments (SI Appendix,Fig. S2). Most of the phenotypes corresponded to an accumu-lation of the fluorescent markers into intracellular bodies, sug-gesting a decrease in trafficking efficiency. Several chemicalsaffected multiple markers, i.e., PINs and BRI1, indicatinga crosstalk between auxin efflux carriers and brassinosteroid re-ceptor trafficking pathways. Overall, the range of subcellularphenotypes demonstrated the dynamic nature of endomembranecompartments and the ability of small molecules to perturb orarrest them at different intersecting pathways.

Analysis of Subcellular Phenotypes and Their Frequency. A largenumber of compounds induced cargo accumulation in bodiessimilar to ES1-like compartments (Fig. 1A phenotype 2 and SIAppendix, Fig. S2). Only brefeldin A (BFA) resulted in cargoaccumulation in characteristic BFA bodies (17) (SI Appendix,Fig. S2, P3), indicating the unique nature of this macrocyclic

1 2 5 6 7

9 10 11 12 1513

A

BRI1::BRI1_2 PIN2::PIN1_2 PIN2::PIN2_2

B

3

PFG: 1a- A

HV:: 1a- AHV

1NIPα

PIN2::PIN1:GFP

151 2 3 4 5 6 7 8 9 10 11 12 13 14 16 170

Color Code

4

8 14

0

ES3

ES1

Fig. 1. Bioactive molecules induce specific profiles of subcellular phenotypes. (A) Codes P1–P15 correspond to subcellular phenotypes with P0 as a control (SIAppendix, Table S1). Representative images of PIN2::PIN1:GFP are shown as an example. Note that P13 describes increased levels at the PM or aberrantlocalization of endosomal markers such as VHAa1::VHAa1:GFP to the PM (Inset, 13). Class 15 describes changes in polarity of PIN2::PIN1 as shown by LSCM orimmunostaining using anti-PIN antibody (Inset, 15). Arrow depicts PIN2::PIN1:GFP protein apical localization. (Scale bars, 10 μm.) (B) Analysis of the PMRAcompound set reveals clusters of bioactive molecules. Clustering of the 123 PMRA bioactive compounds is based on the subcellular phenotypes. Major clustersof bioactive molecules are shown. Each cluster is labeled on the basis of a subcellular phenotype (P1–P15) and also indicates a characterized molecule. The ES1,P2 cluster represents small molecules that cause ES1-like bodies (9) on the subcellular markers. ES3 was grouped within the same cluster. P2P2_2H (PIN2::PIN2:GFP, 2-h treatment), P2P1_2H (PIN2::PIN1:GFP, 2-h treatment), B1_2H (BRI1::BRI1:GFP, 2-h treatment). (Note: In SI Appendix, Table S3, a numerical pre-sentation mirroring the map is presented. It is color coded as in this figure to indicate the relationships.)

Drakakaki et al. PNAS | October 25, 2011 | vol. 108 | no. 43 | 17851

PLANTBIOLO

GY

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

29, 2

020

lactone (18). Phenotypes such as endomembrane bodies (P1 andP2), apparent cytoplasmic localization (P8), or vacuolar body for-mation (P12), were displayed for the examined markers in differentcombinations (SI Appendix, Fig. S2, P1–P12, P16 and P17), sug-gesting the targeting of shared or interconnected pathways.Surprisingly, changes in polar localization were observed only

for basally localized PIN2::PIN1:GFP, not for PIN2::PIN2:GFPthat localizes at the apical PM (SI Appendix, Fig. S2, P15). Thisdifferential effect may reflect distinct pathways for targeting toapical and basolateral surfaces of the cell (8). Moreover, thesechanges indicate that the maintenance of PINs at the apical sideis more robust than at the basal side. The accumulation of theendosomal marker VHA-a1::VHA-a1:GFP at the PM suggestedinhibition of the endocytic pathway (Fig. 1A and SI Appendix,Fig. S2, P13) and/or enhancement of secretory trafficking (17).From our analyses, the marker most commonly affected was

PIN2::PIN1:GFP, suggesting a highly dynamic cycling capacity ofPIN1 expressed in the epidermis, followed by BRI1::BRI1:GFPand PIN2::PIN2:GFP (Fig. 1B and SI Appendix, Fig. S3). Sig-nificantly, the maps of PIN2::PIN1:GFP and PIN2::PIN2:GFPwere highly distinct when comparing 2- and 24-h treatments (SIAppendix, Figs. S3 and S5), indicating that processes are affecteddifferentially depending upon the length of treatment. Giventhat the average cell cycle in Arabidopsis roots is about 18 h (19),this highlights the importance of examining events within theirbiological time frames.

Clustering of Bioactive Molecules. To associate molecules by phe-notypic classes, we used hierarchical clustering that treats datapoints with single and multiple phenotypes in an unbiased manner.This clustering grouped structurally diverse compounds into clus-ters that induced similar phenotypes as shown for the 123 PMRAcompounds (Fig. 1B) or the total of 360 inhibitors of pollen ger-mination (SI Appendix, Fig. S3). This indicated that diverse struc-tures could result in similar phenotypes and that a simple structure-based analysis was insufficient to describe this complex system.Clustering of compounds by their structural or property descriptorsresulted in very weak coclustering of related phenotypic profiles (SIAppendix, Table S2), probably due to the high structural diversity ofthe compounds present in the libraries.

Map to Visualize Clusters of Bioactive Chemicals. A color-codedrepresentation incorporating multiple subcellular markers chem-ically treated at different time points was developed. Our computa-tional method considered all phenotypes equally in a nonbiasedapproach for each molecule across all markers and treatmenttimes (SI Appendix, Tables S3 and S4). For ease of presentation,the color-coded bars in the visual display (Fig. 1B and SI Appendix,Fig. S3) showed the first numeric phenotype (i.e., 0 > 1> 2 >. . .>15) (SI Appendix, Tables S3 and S4). As the visual maps indicate,many molecules resulted in similar phenotypes involving eitherone or multiple PM markers visualized as colored clusters (Fig.1B). In further support of the finding that the time of treatment isimportant, the bioactive clustering of PIN2::PIN1:GFP andPIN2::PIN2:GFP resulted in maps that were highly distinct when com-paring 2- and 24-h treatments (SI Appendix, Figs. S3 and S5).Overall, the resulting bioactive clusters could serve as a uniquesignature to classify known and unknown compounds.

Endosidin 3 Alters ROP-Mediated Signaling Involved in Cell Polarity.All PMRAmolecules that were characterized in detail subsequentto secondary screening and presented here were ESs (9), definedas bioactive molecules affecting trafficking or the morphology ofendomembrane compartments. A major cluster identified mole-cules that induced cargo accumulation in endomembrane bodies(i.e., ES1-like bodies) (Fig.1B, ES3 cluster). We investigated ES3as a potential inhibitor of protein trafficking from the PM. Thetrafficking of PMmarkers was perturbed by 2 h of treatment (Fig. 2A and B; SI Appendix, Fig. S4 A–H and Table S3) without modi-fying cytoskeleton morphology (Fig. 2 C–F). ES3 affected the

TGN/EE compartment without effecting Golgi or late endosome/prevacuolar compartments (SI Appendix, Fig. S4 I–P).Interestingly, developmental phenotypes such as alteration of

leaf epidermal pavement cell (PC) shape lobes was observed inES3-treated seedlings (Fig. 2K–L and S). This was consistent withthe dependency of PC shape formation on the polarity and en-

PIN2::PIN1:GFP

DMSO

A

PIN2::PIN1:GFP

ES3 [65]

35S::ROP2:GFP

DMSO

35S::ROP6:GFP

ES3 [65]

35S::ROP6:GFP

DMSO

B

G

H

35S::MAP4:GFP

DMSO

35S::TALIN:GFP

ES3 [65]

35S::MAP4:GFP

ES3 [65]

35S::TALIN:GFP

DMSO

C E

D F

I

WS2

ES3

WS2

DMSO ric1-1

DMSO

ric1-1

ES3

ric4-1

DMSO

ric4-1

ES3

rop6-1

DMSO

rop6-1

DMSO

a

b

K M

L N

O

P

Q

R

35S::ROP2:GFP

ES3 [65]

J

0

1

2

3

4

5

Ws DMSO

Ws ES3

ric1-1 DMSO

ric1-1

ES3ric4-1 DMSO

ric4-1

ES3rop6-1

DMSO

rop6-1 ES3

S

T

Effect of ES3 on polar cell expansion

ROP2/4

RIC4

Diffuse F- actin

ROP6

RIC1

Well-ordered MTs

Outgrowth promotion(Lobing)

Outgrowth supression(Indenting)

ES3

Fig. 2. ES3 enables separation of ROP-mediated signaling pathways byphenocopying the ric1-1 and rop6-1 mutations. Induction of intracellular ac-cumulation of PIN2::PIN1:PIN1 after 2 h of ES3 treatment (B) is compared withDMSO-treated samples (A). (C–F) ES3 application did not affect the localiza-tion of cytosketetal proteins MAP4-GFP (C and D) or TALIN-GFP (E and F). (Gand H) Rho GTPase 35S::ROP6:GFP is normally localized at the PM in DMSO-treated cells; however, ES3-treatment for 2 h induced a cytoplasmic localiza-tion. Note that the PM localization is abolished in H comparedwithG. (I and J)ES3 treatment does not affect the localization of 35S::ROP2:GFP (J) comparedwith the DMSO control (I). (K–R) Effect of ES3 on directional cell expansion.Leaf epidermal PCs in ric1-1 and rop6-1 (M and Q) and ES3-treated wild-typeseedlings (L) have increased neck width compared with DMSO-treated wildtype (K) and ric4-1 (O). The ric1-1and rop6-1 seedlings remainedunchanged (Nand R), whereas ric4-1 seedlings were sensitive to ES3 treatment (P). (S) Thedirectional cell expansion (lobing)wasmeasuredas the ratio (a:b) betweenthelongitudinal cell length (L, a) and the transverse length (L, b) at theneck region.This ratio was significantly reduced in wild-type WS2 and ric4-1 mutants,suggesting increased lateral expansion, whereas it remained unchanged inric1-1 and rop6-1 (n = 122 for all treatments). Quantitative analysis of the PCphenotype shows that both ES3-treatedwild-type seedlings (t = 1.45E-14) andric4-1 (t = 6.038E-09) have reduced neck regions and that treated ric1-1 (t =1.131) and rop6-1 (t=0.568) seedlingswerenot sensitive to ES3 comparedwithwild type or ric4-1–treated seedlings (statistically significant P value<0.05). (T)Schematic representation of the potential mode of action of ES3. ES3 blocksproper PM localization of ROP6, thereby acting through the ROP6/RIC1–de-pendent signaling pathway. This consequently causes increased lateral expan-sion in wild-type plants. As a result, the ric1-1 and rop6-1mutants are resistanttoES3 incontrast to the ric4-1mutantsandcoulddistinguishbetweentheROP2/ROP4 and ROP6/RIC1 pathways. (Scale bars, 10 μm.) Error bars represent SE.

17852 | www.pnas.org/cgi/doi/10.1073/pnas.1108581108 Drakakaki et al.

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

29, 2

020

docytosis/recycling of PIN1 (20). In leaf PCs, the intercalation oflobes and indentations is coordinated by the counteraction of theROP2/RIC4 and the ROP6/RIC1 pathways (21). Both pathwaysare activated by auxin, which is sensed by the secreted receptorauxin binding protein 1 (ABP1) (20), which is also involved in PINinternalization (22). We then used mutants defective in ROP2/RIC4 and ROP6/RIC1 pathways to identify possible moleculartargets of ES3. The ratio between cell length vs. indentation widthof PCs of the ric1-1 and rop6-1 mutants was already reducedcompared with wild type; however, they were unaffected by ES3(Fig. 2M,N, R, and S). To the contrary, PCs of the ric4-1 and wildtype were sensitive to ES3 (Fig. 2 L, P, and S). This resistance ofric1-1 and rop6-1 indicated that ES3was likely targeting theROP6/RIC1 pathway, whichwas supported by the sensitivity of ric4-1 thatacts within the antagonistic ROP2/4/RIC4 pathway.One potential mode of action of ES3 was the disruption of

ROP6 association with the PM, which is required for polargrowth. To test this hypothesis, we examined the localization ofPM-associated 35S::ROP6:GFP in roots where PM markers areeasily visualized compared with PCs. Indeed, in the presence ofES3, 35S::ROP6:GFP was no longer PM localized (Fig. 2 G andH). To the contrary, the localization of 35S::ROP2:GFP (Fig. 2 Iand J) was not affected by ES3, further supporting our hypothesisthat ES3 targets the ROP6/RIC1 pathway. Thus, by dis-tinguishing between the antagonistic ROP6 and ROP2/4 path-ways, ES3 can uncover interactions between ROP signaling andPIN localization in Arabidopsis roots (Fig. 2T). In terms of ourkey objective to provide molecules to understand the endo-membrane network, the data demonstrated that clusters of bio-active molecules such as the ES3 cluster were an entry point todissect endomembrane-related developmental processes such assignaling and its effect on polar growth.

ES5 Interferes with Endomembrane-Trafficking Pathways to theVacuole. We explored our clustering analysis to reveal specificpathways to the vacuole by merging related phenotypes charac-teristic of potential defects in vacuolar localization. Thus, wecombined phenotypes (9, 11, 12) related to impaired vacuolelocalization. The resulting map displayed a significant vacuole-related cluster, represented in green (SI Appendix, Fig. S5 andTable S5). The cluster included the molecule ES5 that induced astrong accumulation of PM cargo in the vacuole (Fig. 3 A–J).ES5 also affected the localization of markers for the TGN/EE,pre-vacuolar compartment (PVC) and Golgi (SI Appendix, Fig.S6 A–J) without affecting ER, cytoskeleton morphology (SIAppendix, Fig. S6 K–P), or trafficking of soluble cargo (23) to thevacuole (SI Appendix, Fig. S6 Q–T). These data indicated thatES5 induced increased trafficking specifically of PM proteins tothe vacuole or enhanced their stability once in the vacuole.Consistent with its effects on PIN recycling and targeting to the

vacuole (24), ES5 impaired the responsiveness of roots to gravityin a dose-dependent manner (SI Appendix, Fig. S6U).

ES5 Targets Recycling in Plants and Animals. To test if increasedvacuolar targeting was due to inhibition of recycling, we per-formed BFA washout experiments (Fig. 4 A–D). BFA inhibitsexocytosis to the PM (13). In the presence of ES5, the re-versibility of the BFA effect was impaired (Fig. 4D), indicatingthat ES5 inhibited exocytosis of the PM proteins in plant. Tofurther investigate the mode of action of ES5, we examined itsactivity on recycling in HeLa cells using the well-describedtransferrin (Tfn) as a marker. Accumulation of Tfn upon ES5treatment suggested inhibition of recycling (Fig. 4E). This wasconfirmed by fluorescence microscopy that revealed an accu-mulation of Tfn in an ES5-induced endosomal compartmentwith aberrant tubular morphology (Fig. 4 F–M), similar to BFA(25). The tubule-like endosome compartments appeared within10 min of ES5 treatment (SI Appendix, Fig. S7 A–C), indicatinga sorting or trafficking defect from early endosomes to recyclingendosomes. We concluded that ES5 inhibited the recycling ofPM proteins in at least two kingdoms, indicating that its target(s)and pathways were evolutionarily conserved and demonstratedthe versatility of plant-based screens in human drug discovery.

ES7 Interferes with a RABA2a-Mediated Pathway for Cell-PlateMaturation. Apart from the bioactive clusters, some isolatedmolecules caused highly specific phenotypes. Vesicular traffick-ing is central to cell division and requires de novo formation ofa transient membrane compartment, the cell plate, whichmatures to a cell wall (26). In Arabidopsis roots (SI Appendix, Fig.S3), ES7 activity induced visible gaps in the cell-plate localizationof PM markers and KNOLLE, a syntaxin involved in plate for-mation (27) (Fig. 5 A–H). Markers for ER, Golgi, TGN/EE,PVC, and cytoskeleton were unaffected (SI Appendix, Fig. S8 A–

J35S::PIP2A:GFP

35S::PIP2A:GFP

DMSO

BRI1::BRI1:GFP

35S::BRL3:GFP

35S::BRL3:GFP

ES5 [80]

DMSO

A B C D E

F G H I

PIN2::PIN1:GFP

DMSO

PIN2::PIN1:GFP

ES5 [80]

PIN2::PIN2:GFP

ES5 [80] ES5 [80]

PIN2::PIN2:GFP

DMSO

BRI1::BRI1:GFP

DMSO

ES5 [80]

Fig. 3. ES5 enhances endosomal trafficking to the vacuole. (A–J) ES5 in-duced vacuolar localization of PIN2::PIN1:GFP (A and F), PIN2::PIN2:GFP (Band G), BRI::BRI1:GFP (C and H), 35S::PIP2A:GFP (D and I), and 35S::BRL3:GFP(E and J). (Scale bars, 10 μm.)

PIN2::PIN2:GFP

ES5[40]

PIN2::PIN2:GFP

BFA[50]

PIN2::PIN2:GFP

BFA[50] /AM

PIN2::PIN2:GFP

BFA[50] ES5[40]

A C D

D

B E

F G

K

J

IH

L

0 5 10 15 20 25 35Time (min)

0

20

40

60

80

100

120

In

tra

ce

llu

la

r T

fn

(%of

initi

al b

ound

)

K L M

ES

5C

on

tro

l

ControlES5

Fig. 4. ES5 interferes with recycling in plants and animals. (A–D) BFAwashout of PIN2::PIN2:GFP with ES5. (A) PIN2::PIN2:GFP localization inDMSO-treated Arabidopsis control root cells shows PM labeling. (B) BFAtreatment for 1 h induces accumulation of PIN2::PIN2:GFP in BFA bodies. (C)Treatment with BFA for 1 h followed by 2 h of washout (in Arabidopsismedia with DMSO) eliminated BFA body formation. (D) Similarly, treatmentwith BFA for 1 h followed by 2 h of washout with ES5-supplemented mediadid not eliminate BFA body formation. Arrow depicts internalized PIN2 inBFA-induced bodies. (E) ES5 inhibits transferrin receptor recycling in HeLacells in single-round uptake of biotinylated Tfn. In control cells, Tfnwas internalized rapidly and then recycled back to the cell surface. In the ES5-treated cells, the kinetics of internalization were unaffected but recyclingwas inhibited. (F–M) In the control cells, Tfn was rapidly recycled back to themedia (F–I). Note the disappearance of Alexa488-Tfn after 20 min (I). In ES5-treated cells, Tfn accumulated in recycling endosomes whose morphologywas altered forming interconnected tubules (J–M). (Scale bars, 10 μm.)

Drakakaki et al. PNAS | October 25, 2011 | vol. 108 | no. 43 | 17853

PLANTBIOLO

GY

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

29, 2

020

N), indicating the specificity of ES7 activity for cell-plate for-mation, cell-wall deposition or cell-plate maturation. Observa-tion of live cell-plate formation by microtubule labeling inNicotiana tabacum (tobacco) Bright Yellow-2 (BY-2) suspen-sion-cultured cells demonstrated that cell cycle progression wasunaffected and that plate formation was initiated normally in the

presence of ES7; however, cell-plate growth stopped during itsexpansion phase (Fig. 5I and Movie S1). Callose accumulation inthe lumen of the cell plate has been proposed to act asa spreading force for cell-plate expansion (17).To investigate ifES7 impacted deposition of this polysaccharide, we visualizedcallose (Fig. 5 J and K) in Arabidopsis roots. Little or no cell-plate staining was observed in ES7-treated cells (Fig. 5 J and K)compared with the control roots (Fig. 5J), indicating a deficiencyin callose synthesis or deposition. Next, we examined the TGNGTPase RABA2a, which is primarily cell plate localized duringcytokinesis (28). Unlike interface cells (SI Appendix, Fig. S8),ES7 induced large endomembrane bodies containing RABA2a::RABA2a:YFP in close proximity to the cell plate (Fig. 5 L andM) during cytokinesis. However, ES7 treatment had no detect-able effect on the distribution of other TGN/EE markersor markers for ER, Golgi, and late endosomes/PVC (SI Appen-dix, Fig. S8). These data indicated impairment of cell-platematuration through a RABA2a-mediated pathway. Alterna-tively, ES7 could prevent callose deposition, resulting in the mis-localization of RAB2A:YFP, which is also consistent with re-duced callose deposition.Previous studies have shown a role for ARF1 in maintaining

KNOLLE localization at the cell plate during late cytokinesis inArabidopsis (29). The dominant negative mutant 35S::ARF1-TN:GFP (30) was altered in cell-plate formation, a phenotype that wasenhanced by ES7, indicating an additive effect of the mutationand ES7 bioactivity (Fig. 5 N–Q). This indicates that ES7 actsindependently of ARF1 or is affecting the remaining ARF1 homo-logs in Arabidopsis. Our results indicated that ES7 perturbs a secre-tory pathway that is required for callose-dependent cell-platematuration. Overall, ES7 demonstrated that small molecules withhighly distinctive phenotypes, but residing outside of major bioactiveclusters, can be powerful for dissecting specialized pathways.

DiscussionIn this study, we demonstrated that small molecules can slow orarrest dynamic endomembrane processes, facilitating the dissectionof their molecular components within a biological time frame. Themolecules were active in different systems due to evolutionaryconservation of their targets. We successfully mapped a complexmatrix of chemically induced subcellular phenotypes across a suiteof endomembrane-specific markers in a multicellular eukaryote.We identified 123 compounds and groups of subcellular phenotypicprofiles that allowed us to separate endomembrane pathways.Clusters of chemically induced phenotypes provided entry

points and avenues of research for the study of pathways within theplant and animal networks. First, ES3 distinguished between twoantagonist ROP-signaling pathways. Posttranslational lipid modi-fication is important for membrane association of ROPs (31). ES3could affect a lipid modification that is specific to ROP6. Alter-natively, ES3 could target proteins regulating ROP activity andlocalization such as the PAN1 receptor kinase (32). It is also pos-sible that ES3 affects a downstream effector protein that in turnregulates activity and localization of ROP6 in a feedback mecha-nism. Second, ES5 demonstrated the ability of clustering to targetrecycling mechanisms that alter trafficking to the vacuole. Theidentification of ES5 targets could provide insights into recyclingmechanisms across kingdoms. Targets could include proteins in-volved in vesicle formation and budding in a similar manner withGNOMbeing a BFA target (33) The inhibition of recycling of PMproteins in a conserved mechanism between two kingdoms indi-cates the broad potential of our screening data for the discovery ofmolecules and targets that could be of value for medicine andagriculture. Third, a single molecule, ES7, displayed a capacity toarrest a secretory pathway necessary for the maturation and ex-pansion of cell walls. ES7 could target directly callose synthaseitself or proteins involved in its trafficking to the cell plate such asGTPases like RABA2a. Altogether, this demonstrated the speci-ficity of the endomembrane system in different processes such aslate cytokinesis and the ability of small molecules to dissect them.

PIN2::PIN1:GFP DMSO

PIN2::PIN1:GFP ES7 [53]

PIN2::PIN2:GFP ES7 [53]

BRI1::BRI1:GFP ES7[53]

BRI1::BRI1:GFP DMSO

PIN2::PIN2:GFP DMSO

A

B

C

D

E

F

G

H

RABA2a::RABA2a:YFPES7 [53]

35S::ARF1-TN:GFP DMSO

35S ::ARF1-TN:GFP ES7 [53]

35S:: ARF1 WT :GFPDMSO

35S::ARF1 WT:GFP ES7 [53]

RABA2a::RABA2a:YFP DMSO

J

K

L

OM

α-KNOLLE ES7 [53]

α-KNOLLE DMSO

α-callose ES7 [53]

α-callose DMSO

N P

Q

35S::GFP:MAP4 FM4-64 [4] ES7 [16 ]

0min 21min 1h06min 1h33min

2h09min 2h51min 2h57min 3h00min

3h03min 3h27min 3h54min 4h15min

I

Fig. 5. ES7 interferes with cell-plate maturation. (A–H) ES7 treatment causedcell-plate defects visualized as cell-plate gaps highlighted by the PM markersPIN2::PIN1:GFP (B), PIN2::PIN2:GFP (D), and BRI::BRI1:GFP (F) in comparison withthe DMSO-treated control roots (A, C, and E). (G and H) The syntaxin KNOLLEnormally localizes at the cell plate and endosome (G). However, 2 h of ES7treatment resulted in mis-localization of KNOLLE and cell-plate gaps (H). (I) Akinetic display of cell division in BY-2 cells expressingMAP4:GFP and stainedwithFM4-64 under ES7 treatment. Representative time intervals are indicated overa4-h, 15-minperiod. Cell-plate formationwasvisualizedbyFM4-64andoccurrednormally until about 3 h after treatment at which time gaps formed leading to“cell-plate stubs” (n = 4). (J–M) Callose deposition highlights the cell plate inDMSO-treatedcells (J).ES7 treatment resulted inanabsenceof cell-plate labeling(K). RABA2a::RABA2a:YFP localized at the cell plate in DMSO-treated dividingcells (L); however, it was mis-localized upon ES7 treatment and accumulated inES7-induced structures (M). (N–Q) ES7 effect on ARF1mutant. Shown is FM4-64staining inDMSO(N) andES7-treated35S::ARF1-wild-type:GFPseedlings (O).Theheat-shock–induced dominant negative ARF1 (35S::ARF1-T31N:CFP) displayedcell-platedefects.ES7 treatmentof35S::ARF1-T31N:CFP showedsensitivity toES7(P and Q), suggesting a synergistic effect of the mutation and ES7 bioactivity.(Scale bars, 10 μm for all except in I, where the scale bar is 20 μm.)

17854 | www.pnas.org/cgi/doi/10.1073/pnas.1108581108 Drakakaki et al.

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

29, 2

020

A major conceptual feature of our work is an approach thatpermits a more comprehensive systems view of the endomem-brane network. This is achieved by identifying an extensive col-lection of chemically induced subcellular phenotypes that arepathway specific. The approach using high-content intracellularscreening facilitates the parallel discovery and evaluation ofcomplex intracellular phenotypes. This permits options such asthe purification of intermediate compartments (34) that are in-duced by chemical treatment and labeled by one or multipleendomembrane-specific markers.Thus, our database of phenotypic clusters presents a resource of

useful phenotypes for further investigation in addition to a com-pound collection of value to the community. Furthermore, thebiological roles of compartments and cargoes can be understoodby using molecules with well-described activities in vivo to linkrelevant compartments and cargoes to associated developmentaland physiological phenotypes, which is critical for understandingprotein function. This was demonstrated for all three selectedendosidins. ES3 was linked to ROP trafficking and cell-shapedevelopment, ES5was linked to PIN cycling and gravitropism, andES7 revealed a pathway involving RABA2a and cell division. Thedirect discovery of endomembrane-defective phenotypes thatcould then be linked to developmental phenotypes would bea challenge for exclusively forward genetic screens.The efficiency of high-content screens and clustering will benefit

from fully automated microscopy and image analysis, which isemerging in the study of plants (35). The assignment of endosidinsto a range of phenotypic clusters will evolve as new markers,conditions, and phenotypes, including developmental ones, areadded to our matrix database. Thus, our database will serve as anongoing resource to be expanded to include markers and whole-organism phenotypes, permitting the correlation of complex in-tracellular and developmental phenotypes. The availability ofa range of organized complex phenotypes is powerful because onecan easily shop for chemically induced phenotypes that are bi-ologically interesting and can be scored genetically. The large

dataset that we have acquired will also provide an excellent com-putational training set of subcellular responses to enable a sys-tems-based approach of elucidating the endomembrane network.Our approach also opens the possibility of discovering native

molecules involved in plant development. Indeed, it is likely thatsubpopulations of syntheticmolecules could be structurally similarto intrinsic molecules that are not yet characterized. For example,most natural growth regulators in plants have synthetic analogs.Synthetic molecules whose cognate protein targets appear to bereceptors or transporters may mimic such natural molecules.Natural molecules could then be identified from the increasingnumber of natural molecules that are being characterized usingmetabolomics. Our approach and bioactive library will provideresearch directions that connect endomembrane trafficking withplant development and response to enviromental stimuli.

Materials and MethodsFour-day-old light-grown seedlingswere treatedwith chemicals of 5-mg/mL stocksolutions for all libraries except the library of active compound on Arabidopsis(LATCA) library (2.5 mg/mL), corresponding to a final concentration of 50–100μM based on the molecular mass of each compound. All chemicals described inthis study can be identified by using their ID presented in SI Appendix, TablesS2–S5 through ChemMine (http://bioweb.ucr.edu/ChemMineV2/). BFA-treatedseedlings were transferred to 0.5× MS medium containing 0.8% phytoagar towhich chemicals were added. For further details, see SI Appendix.

ACKNOWLEDGMENTS. The authors thank Dr. Sean Cutler for sharing theLATCA and CLIKables libraries, Dr. Gerd Jürgens for the anti KNOLLE antise-rum, and Evelien Mylle for technical support. We acknowledge NationalScience Foundation Grant MCB-0515963 (to N.V.R. and G.R.H.). T.G. wassupported by National Science Foundation Grant ABI #0957099 and M.Q.B.by National Science Foundation Grant DGE-0504249. S.L.S. was supported byNational Institutes of Health Grant MH761345. S.R. and J.F. were supportedby the Odysseus program and A.-M.S. and E.R. by Grant G.0065.08. S.R. wassupported by Vetenskapsråde and VINNOVA and D.V.D. by Flanders Re-search Foundation. G.D. was supported by University of California at Davisstart-up funds.

1. Bonifacino JS, Glick BS (2004) The mechanisms of vesicle budding and fusion. Cell 116:153–166.

2. Winckler B, Mellman I (2010) Trafficking guidance receptors. Cold Spring Harb Per-spect Biol 2:a001826.

3. Hicks GR, Raikhel NV (2010) Advances in dissecting endomembrane trafficking withsmall molecules. Curr Opin Plant Biol 13:706–713.

4. Tuvim MJ, Adachi R, Hoffenberg S, Dickey BF (2001) Traffic control: Rab GTPases andthe regulation of interorganellar transport. News Physiol Sci 16:56–61.

5. Geldner N, Robatzek S (2008) Plant receptors go endosomal: A moving view on signaltransduction. Plant Physiol 147:1565–1574.

6. Takano J, Miwa K, Yuan L, von Wirén N, Fujiwara T (2005) Endocytosis and degra-dation of BOR1, a boron transporter of Arabidopsis thaliana, regulated by boronavailability. Proc Natl Acad Sci USA 102:12276–12281.

7. Van Damme D, Inzé D, Russinova E (2008) Vesicle trafficking during somatic cytoki-nesis. Plant Physiol 147:1544–1552.

8. Kleine-Vehn J, Friml J (2008) Polar targeting and endocytic recycling in auxin-de-pendent plant development. Annu Rev Cell Dev Biol 24:447–473.

9. Robert S, et al. (2008) Endosidin1 defines a compartment involved in endocytosis ofthe brassinosteroid receptor BRI1 and the auxin transporters PIN2 and AUX1. ProcNatl Acad Sci USA 105:8464–8469.

10. Wisniewska J, et al. (2006) Polar PIN localization directs auxin flow in plants. Science312:883.

11. Müller A, et al. (1998) AtPIN2 defines a locus of Arabidopsis for root gravitropismcontrol. EMBO J 17:6903–6911.

12. Friedrichsen DM, Joazeiro CA, Li J, Hunter T, Chory J (2000) Brassinosteroid-in-sensitive-1 is a ubiquitously expressed leucine-rich repeat receptor serine/threoninekinase. Plant Physiol 123:1247–1256.

13. Geldner N, Friml J, Stierhof YD, Jürgens G, Palme K (2001) Auxin transport inhibitorsblock PIN1 cycling and vesicle trafficking. Nature 413:425–428.

14. Russinova E, et al. (2004) Heterodimerization and endocytosis of Arabidopsis brassi-nosteroid receptors BRI1 and AtSERK3 (BAK1). Plant Cell 16:3216–3229.

15. Geldner N, et al. (2009) Rapid, combinatorial analysis of membrane compartments inintact plants with a multicolor marker set. Plant J 59:169–178.

16. Dettmer J, Hong-Hermesdorf A, Stierhof YD, Schumacher K (2006) Vacuolar H+-AT-Pase activity is required for endocytic and secretory trafficking in Arabidopsis. PlantCell 18:715–730.

17. Gendre D, et al. (2011) Conserved Arabidopsis ECHIDNA protein mediates trans-Golgi-network trafficking and cell elongation. Proc Natl Acad Sci USA 108:8048–8053.

18. Harri E, Loeffler W, Sigg HP, Stahelin H, Tamm H (1963) Brefeldin A. Helv Chim Acta46:1235–1243.

19. West G, Inzé D, Beemster GT (2004) Cell cycle modulation in the response of the

primary root of Arabidopsis to salt stress. Plant Physiol 135:1050–1058.20. Xu T, et al. (2010) Cell surface- and rho GTPase-based auxin signaling controls cellular

interdigitation in Arabidopsis. Cell 143:99–110.21. Fu Y, Gu Y, Zheng Z, Wasteneys G, Yang Z (2005) Arabidopsis interdigitating cell

growth requires two antagonistic pathways with opposing action on cell morpho-

genesis. Cell 120:687–700.22. Robert S, et al. (2010) ABP1 mediates auxin inhibition of clathrin-dependent endo-

cytosis in Arabidopsis. Cell 143:111–121.23. Hunter PR, Craddock CP, Di Benedetto S, Roberts LM, Frigerio L (2007) Fluorescent

reporter proteins for the tonoplast and the vacuolar lumen identify a single vacuolar

compartment in Arabidopsis cells. Plant Physiol 145:1371–1382.24. Kleine-Vehn J, et al. (2008) Differential degradation of PIN2 auxin efflux carrier by

retromer-dependent vacuolar targeting. Proc Natl Acad Sci USA 105:17812–17817.25. de Figueiredo P, et al. (2001) Inhibition of transferrin recycling and endosome tu-

bulation by phospholipase A2 antagonists. J Biol Chem 276:47361–47370.26. Samuels AL, Giddings TH, Jr., Staehelin LA (1995) Cytokinesis in tobacco BY-2 and root tip

cells: A new model of cell plate formation in higher plants. J Cell Biol 130:1345–1357.27. Waizenegger I, et al. (2000) The Arabidopsis KNOLLE and KEULE genes interact to

promote vesicle fusion during cytokinesis. Curr Biol 10(21):1371–1374.28. Chow CM, Neto H, Foucart C, Moore I (2008) Rab-A2 and Rab-A3 GTPases define

a trans-golgi endosomal membrane domain in Arabidopsis that contributes sub-

stantially to the cell plate. Plant Cell 20:101–123.29. Boutté Y, et al. (2010) Endocytosis restricts Arabidopsis KNOLLE syntaxin to the cell

division plane during late cytokinesis. EMBO J 29:546–558.30. Xu J, Scheres B (2005) Dissection of Arabidopsis ADP-RIBOSYLATION FACTOR 1

function in epidermal cell polarity. Plant Cell 17:525–536.31. Yalovsky S, Bloch D, Sorek N, Kost B (2008) Regulation of membrane trafficking, cyto-

skeleton dynamics, and cell polarity by ROP/RAC GTPases. Plant Physiol 147:1527–1543.32. Humphries JA, et al. (2011) ROP GTPases act with the receptor-like protein PAN1 to

polarize asymmetric cell division in maize. Plant Cell 23:2273–2284.33. Geldner N, et al. (2003) The Arabidopsis GNOM ARF-GEF mediates endosomal re-

cycling, auxin transport, and auxin-dependent plant growth. Cell 112:219–230.34. Drakakaki G, et al. (2011) Isolation and proteomic analysis of the SYP61 compartment

reveal its role in exocytic trafficking in Arabidopsis. Cell Res, 10.1038/cr.2011.129.35. Salomon S, et al. (2010) High-throughput confocal imaging of intact live tissue en-

ables quantification of membrane trafficking in Arabidopsis. Plant Physiol 154:

1096–1104.

Drakakaki et al. PNAS | October 25, 2011 | vol. 108 | no. 43 | 17855

PLANTBIOLO

GY

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

29, 2

020