Systems survey of endocytosis by multiparametric image analysis

Transcription

1 Vol Mrch 21 doi:138/nture8779 Systems survey of endocytosis y multiprmetric imge nlysis ARTICLES Cludio Collinet 1, Mrtin Stöter 2, Chrles R. Brdshw 1, Nikoly Smusik 1, Jochen C. Rink, Denise Kenski, Binc Hermnn 1, Frnk Buchholz 1, Roert Henschel 3, Mtthis S. Mueller 3, Wolfgng E. Ngel 3, Eugenio Fv 2, Ynnis Klidzidis 1,4 & Mrino Zeril 1 Endocytosis is complex process fulfilling mny cellulr nd developmentl functions. Understnding how it is regulted nd integrted with other cellulr processes requires comprehensive nlysis of its moleculr constituents nd generl design principles. Here, we developed new strtegy to phenotypiclly profile the humn genome with respect to trnsferrin (TF) nd epiderml growth fctor (EGF) endocytosis y comining RNA interference, utomted high-resolution confocl microscopy, quntittive multiprmetric imge nlysis nd high-performnce computing. We identified severl novel components of endocytic trfficking, including genes implicted in humn diseses. We found tht signlling pthwys such s Wnt, integrin/cell dhesion, trnsforming growth fctor (TGF)- nd Notch regulte the endocytic system, nd identified new genes involved in crgo sorting to suset of signlling endosomes. A systems nlysis y Byesin networks further showed tht the numer, size, concentrtion of crgo nd intrcellulr position of endosomes re not determined rndomly ut re suject to specific regultion, thus uncovering novel properties of the endocytic system. Endocytosis is fundmentl process supporting mny functions such s nutrient uptke, intrcellulr signlling 1, morphogenesis during development 2 nd defence ginst pthogens 3. Dysfunctions of the endocytic system led to severe metolic 4, infectious 5 nd neurodegenertive diseses 6. Receptors re internlized y oth Clthrin-dependent nd -independent mechnisms into erly endosomes nd, from here, follow different intrcellulr routes. For exmple, trnsferrin (TF) is recycled ck to the plsm memrne, wheres epiderml growth fctor (EGF) is routed to lte endosomes/ lysosomes for degrdtion. Despite gret del of moleculr insights 7 9 our understnding of the endocytic mchinery remins frgmentry. Furthermore, the design principles of the endocytic pthwy, its integrtion in the overll cellulr system nd high-order control re still lrgely unknown. Addressing these prolems requires methods cple of seizing the complexity t the system level (tht is, mesuring mny key prmeters). Here, we designed new nlyticl pproch imed t phenotypiclly profiling genes with respect to endocytosis y genome-wide RNAi screening through the quntittive ssessment of mny system prmeters. Multiprmetric profiling of the endocytic system We visulized endocytosis of fluorescent TF nd EGF in HeL cells efore we could pply our pltform to more disese-relevnt systems. After smll interfering RNA (sirna) trnsfection, cells internlized the two crgo mrkers simultneously for 1 min, were fixed, stined with 49,6-dimidino-2-phenylindole (DAPI)/SYTO42 lue for nuclei nd cytoplsm detection, nd imged (12 imges per well) y triplecolour high-resolution utomted confocl microscopy (Fig. 1). Insted of performing phenotypic evlution on the sis of single or few prmeters s commonly done in previous screens 1,11,we developed quntittive multiprmetric imge nlysis (QMPIA) pltform, on the sis of the custom-designed imge nlysis softwre MotionTrcking 12. First, imges were segmented to determine morphologicl nd positionl descriptors for ech endosome (Fig. 1). Second, totl set of 62 prmeters were extrcted, 4 serving s qulity control to reject su-optiml imges nd 58 descriing defined iologicl properties of the endocytic system, such s totl internl crgo, numer, size (men pprent re) nd position of endosomes (Fig. 1c, Supplementry Tle 1 nd Supplementry Informtion for detils). The ltter were processed to suppress plte-to-plte rndom vritions, normlized (see Supplementry Informtion) nd the 46 most roust prmeters (see Supplementry Informtion nd Supplementry Tle 1) were comined into multiprmetric profiles (Fig. 1d) descriing quntittively the endocytic phenotypes. In preprtion for the genomic screen, we vlidted the qulity nd reproduciility of ssy nd QMPIA. First, we verified tht the depletion of estlished components of the endocytic mchinery, such s Clthrin (CLTC), Dynmin-2 (DNM2), EGF receptor (EGFR), Trnsferrin receptor (TFRC), erly endosome ntigen 1 (EEA1) nd others y RNAi (Supplementry Figs 1 nd 2), yielded the expected ltertions in phenotypic profiles. For exmple, silencing of TFRC reduced the vlues of numer of endosomes, totl internl crgo nd men crgo lod per endosome for TF ut not EGF (Fig. 1d nd Supplementry Fig. 1e). Conversely, EGFR downregultion pertured the prmeters of EGF ut not of TF (Supplementry Fig. 1e). Second, we vlidted the use of single confocl section s representtive of the three-dimensionl endosoml popultion of the entire cell (Supplementry Fig. 3). Third, we vlidted the reproduciility of the oserved phenotypic profiles using the sme nd different sirnas trgeting the sme gene for set of 78 genes nd 468 sirna (6 sirnas per gene). We found tht profiles of the sme sirnas were highly reproducile (Fig. 2) (men Person correltion coefficient s.e.m.), compred with other monoprmetric screens 13 1 Mx Plnck Institute for Moleculr Cell Biology nd Genetics, 2 High-Throughput Technology Development Studio, MPI-CBG, Pfotenhuerstrsse 18, 137 Dresden, Germny. 3 Center for Informtion Services nd High Performnce Computing (ZIH), Dresden University of Technology, D-162 Dresden, Germny. 4 Belozersky Institute of Physico-Chemicl Biology, Moscow Stte University, , Moscow, Russi. Present ddresses: University of Uth School of Medicine, 41 MREB, 2 North 19 Est, Slt Lke City, Uth , USA (J.C.R.); Sirn Therpeutics Inc., Sn Frncisco, Cliforni 94158, USA (D.K.). 21 Mcmilln Pulishers Limited. All rights reserved 243

2 ARTICLES NATURE Vol Mrch 21 EGF TF DAPI Synthetic 2 μm 2 μm s P1 P2P3 P4 P5P6 P7 P8P9 C12orf4 runs P1 P11 P12P13P14 P15 P2 P23 P3 P31P32 P33 P34P35 P36 P37P38 P39 P4P41 P42 P43P44 P45P47 P5 P52 P55P56 P57P58 c Multi-prmetric nlysis Endosome prmeter groups/crgo. Numer of endosomes (1).. Crgo uptke (2).. Endosome re (3).. Vesicle elongtion (3).. Endosoml crgo concentrtion (3).. Endosoml crgo content (3).. Endosome distnce from nucleus (3).. Endosome clustering (9). Other endocytic prmeters. Bckground intensity (2).. Colocliztion (2). Qulity control prmeters - Numer of nuclei. - Nucler size. - Nucleus intensity. - Focus score. d 3 Control 2 Internlized crgo Crgo lod TFRC 1 Numer Crgo density P13 P14P15 P1 P11P12 P7 P8P9 P4 P5P6 P1 P2P3 P2 P16 P17P18 P56 P57P58 P5 P52P55 P44 P45P47 P41 P42P43 P38 P39P4 P35 P36P37 P32 P33P34 P29 P3P31 P23P26P28 Prmeter Figure 1 Multiprmetric imge nlysis., Exmple of three-colour highresolution imges collected with the utomted spinning disk confocl OPERA microscope (left). Nuclei nd cytoplsm re pseudo-coloured in lue, Alex-488-lelled EGF is pseudo-coloured in red nd Alex-647-lelled TF is pseudo-coloured in green. The synthetic imge (right) ws otined fter ckground intensity sutrction nd modelling of endosoml structures (see Supplementry Informtion)., Close-ups show the model structure of individul endosomes. c, List of quntittive prmeters divided into groups ( ). Ten groups of prmeters (numer of prmeters in ech group indicted in rckets) were used to mesure endosoml fetures nd four prmeters s imge qulity control. See Supplementry Tle 1 nd Supplementry Informtion for further description on the QMPIA prmeters. d, Exmple of QMPIA profile (TFRC; see pictures in Supplementry Fig. 1c). The 46 most specific prmeters (listed in Supplementry Tle 1 nd Supplementry Informtion) re ligned on the x-xis nd normlized z-vlues re plotted on the y-xis. s () nd prmeter numers (P1 P58) re indicted corresponding to the nomenclture in c nd to the enumertion in Supplementry Tle 1, respectively. Blck continuous lines seprte the prmeter groups nd grey dshed lines seprte the prmeters. The horizontl rs indicte the endocytic mrkers which the prmeters refer to. Key prmeters ffected in the TFRC profile re indicted y rrows. (see Supplementry Informtion), indicting tht the experimentl noise is not incresed y the higher complexity of the ssy. In contrst, different sirnas trgeting the sme gene frequently yielded different profiles (Fig. 2) (men Person correltion coefficient s.e.m.), consistent with high incidence of RNAi off-trget Aswe excluded vritions in protein depletion nd silencing of lterntively spliced trnscripts (dt not shown), the mjor cuse of incoherence etween profiles of different sirnas is most proly off-trget effects. s P1 P2P3 P4 P5P6 P7 P8P9 C12orf4 ll sirnas P1 P11 P12P13P14 P16P17P18 P18P2P23 P15 P16P17 P26P28P29 Prmeter P26 P28P29P3 Prmeter Figure 2 Reproduciility of multiprmetric profiles y the sme sirna nd different sirnas/gene., Reproduciility of multiprmetric profiles of single sirna silencing the C12orf4 gene. The sme sirna (ID: 2847) ws trnsfected in seven independent experiments (runs) nd multiprmetric profiles (differently coloured for ech experiment) were clculted nd plotted s descried in Fig. 1., Reproduciility of multiprmetric profiles of different sirnas trgeting the sme gene (C12orf4). C12orf4 ws silenced with six different sirnas (IDs: 13385, 13386, 13387, 28285, 28378, 2847) in the sme experiment nd corresponding multiprmetric profiles were clculted nd plotted s descried in Fig. 1. proilistic pproch to infer gene-specific phenotypes y verging the profiles of different sirnas trgeting the sme gene. We defined the gene profile s the most prole profile clculted s mode of the posteriori joint proility distriution considering ll individul sirna profiles for given gene. Notly, here every sirna trgeting given gene contriutes, to different extent, to the determintion of gene profiles (see Supplementry Fig. 4), circumventing the need of similrity thresholds to define supporting sirnas (2 of 3, 3 of 3, nd so on) 18. Gene profiles insted of individul sirna profiles were susequently used for phenotypic clustering nd gene clssifiction. As test, we screened 1, genes including 35 known endocytic regultors (Supplementry Tle 2) nd 92 kinses identified in previous screen 19 (Supplementry Tle 3) using seven sirnas per gene. On the sis of the QMPIA, 31 of the 35 endocytic regultor genes (,9%) nd 62 out of 92 kinse genes presented x 2 proility (1. 2 P) $.95 corresponding to sttisticlly significnt enrichment (P nd P , respectively) in the list of scores. Interestingly, only 7 out of the 35 endocytic regultors would hve scored considering only two prmeters (EGF nd TF totl internl crgo). Altogether, these dt indicte tht the QMPIA hs the dvntge of higher comprehensiveness nd specificity over monoprmetric red-outs in the identifiction of genes regulting endocytic trfficking. P31 P32 P33 P34P35 P36 P37P38 P39 P4P41 P42 P43P44 P45 P47P5 P52 P55P56 P57P58 Gene phenotypic profiling Assuming tht ech sirna profile hs oth on-trget nd offtrget phenotypic components, we exploited the QMPIA to suppress s much s possile the ltter contminnt. For this, we developed Mcmilln Pulishers Limited. All rights reserved Genome-wide survey of EGF nd TF endocytosis To improve ccurcy nd coverge in the survey, we screened directly in the primry ssy three genomic lirries, two sirna nd one endorionuclese-prepred sirna (esirna) 15, corresponding to

3 NATURE Vol Mrch 21 ARTICLES 7 8 si/esirnas per gene for totl of 161,492 si/esirnas (see Methods). This pproch presents two mjor dvntges. First, the high numer of si/esirnas nlysed y QMPIA reduces the impct of flse negtives owing to disgreement etween sirnas. Second, reproduciility of phenotypes is ssessed lredy in the primry screen for ll genes (not just the hits). We cquired 12 imges per smple to otin sttisticlly significnt dt, ccounting for totl of, imges. Qulity controls for trnsfection efficiency, QMPIA nd toxicity (see Methods) were included in ech plte (see Supplementry Fig. 5). Imge nlysis nd gene profile determintion required, CPU hours of clcultion on 2,584-cores computer cluster. Phenotypic clustering nd gene scoring As our im ws not to identify few hit genes ut rther to conduct systems nlysis, we exploited the multidimensionlity of gene profiles. First, we clustered ll genes of the humn genome using men-shift clustering 2 (see Supplementry Informtion). We identified 14 phenotypic cluster groups (Fig. 3) stle over wide rnge of the lgorithm prmeters, indicting tht the endocytic system is under constrints nd chnges etween discrete set of possile sttes in response to perturtions. Second, genes were scored on the sis of two criteri, phenotype mplitude (proility of x 2 ) nd presence of specific phenotypic trits (defined s the cluster group profiles; pheno-score ; Supplementry Informtion), s shown in Supplementry Fig. 6, providing oth strong nd mild, ut specific, phenotypes. A list of 4,69 genes, 3,84 with strong nd 85 with mild phenotype ws compiled (Supplementry Tle 4; imges nd profiles ville t Considering oth the design of the ssy nd the integrtion of endocytosis with mny cellulr processes, such lrge numer ws expected ecuse, esides the core endocytic mchinery, mny genes hve direct or indirect role in endocytosis. For comprison with the common prctices in the field, we estimted tht screening single lirry (2 3 sirnas per gene) with typicl two-prmeter ssy nd vlidting the hits with n independent set of si/esirnas (4 5) would yield only 336 genes (dt not shown), indicting tht the lrge numer of genes in the list results from (1) roder window of detection (mny prmeters) nd (2) high numer of si/esirnas screened rther thn permissive criteri for selection. To estimte the reproduciility of the hits, we re-screened twice the kinome-phosphtome (9,33 sirnas, 1,459 genes), including lrge frction of the gene hits (235). We oserved 84% reproduciility for strong nd 72.5% for mild phenotypes. Nevertheless, to ssign etter proilistic vlue to the scores, the genes in the list will e rescreened (the we dtse will e updted). The silencing efficcy of the lirries ws ssessed y the mnufcturers y quntittive RT PCR (qrt PCR) on set of sirnas nd we independently confirmed it y western lotting (see Methods nd Supplementry Fig. 7). The lrgest portion of scored genes (,34%, 1,593; P-vlue of enrichment P ) encoded for components of metolic c e g Cluster 1 Cluster 2 Cluster 5 Cluster 6 d Cluster 3 Cluster 4 Cluster 7 Cluster 8 f Cluster 9 Cluster 1 Cluster 11 h EGF TF Other Cluster 12 Cluster 13 Figure 3 Phenotypic cluster groups profiles. i, Men-shift genomic clustering of the gene profiles (see text nd Supplementry Informtion) yielded 14 cluster groups. Multiprmetric profiles of the cluster groups re plotted s in Fig. 1. The mplitude of the phenotype vectors ws normlized to 1., Cluster group 1 (lck) nd cluster group 2 (lue) present selective nd opposite effects, tht is, incresed vs decresed internl TF, respectively., Cluster group 3 (lck) nd cluster group 4 (lue) present specific nd opposite effects on the sucellulr locliztion of endosomes ( distnce from nucleus ); tht is, endosomes re either clustered in the perinucler region (cluster group 3) or dispersed in the periphery (cluster group 4). c, Cluster group 5 (lck) nd cluster group 6 (lue) is pir of opposite clusters presenting ech opposite effects with respect to EGF nd TF endocytosis: incresed EGF endocytosis with concomitntly decresed TF endocytosis (cluster group 5) nd vice-vers (cluster group 6). d, Cluster group 7 (lck) i k Cluster group Cluster Mcmilln Pulishers Limited. All rights reserved Prmeters list Endosome prmeter groups/crgo. Numer of endosomes (1).. Crgo uptke (2).. Endosome re (3).. Vesicle elongtion (3).. Endosoml crgo concentrtion (3).. Endosoml crgo content (3).. Endosome distnce from nucleus (3).. Endosome clustering (9). Other endocytic prmeters. Bckground intensity (2).. Colocliztion (2). Description Hits Totl size 1 Selective up-regultion of TF endocytosis 932 5,258 2 Selective down-regultion of TF endocytosis Specific effect on sucellulr locliztion: endosomes pper clustered in the cell centre 81 2,99 4 Specific effect on sucellulr locliztion: endosomes pper dispersed in the cell periphery 26 1,55 5 Opposite effects on EGF nd TF endocytosis: EGF endocytosis is incresed nd TF endocytosis is decresed 224 1,158 6 Opposite effects on EGF nd TF endocytosis: EGF endocytosis is decresed nd TF endocytosis is incresed Effects on endocytosis of oth mrkers: incresed EGF nd TF endocytosis Effects on endocytosis of oth mrkers: decresed EGF nd TF endocytosis 799 2,256 9 Selective up-regultion of EGF endocytosis Selective down-regultion of EGF endocytosis 324 1, Selective up-regultion of EGF endocytosis with ccumultion of endosomes in cell centre 271 1,12 12 Reduced TF endocytosis with endosomes ccumulted in the cell centre Selective increse in EGF endosomes numer nd elongtion Increse in elongtion of TF endosomes with mild increse of TF endocytosis nd cluster group 8 (lue) present opposite effects on endocytosis of oth EGF nd TF (oth incresed for cluster group 7, oth decresed for cluster group 8). e, Cluster group 9 (lck) nd cluster group 1 (lue) present selective nd opposite effects on EGF endocytosis: incresed nd decresed, respectively. f, Cluster group 11 presents incresed EGF endocytosis nd endosome ccumultion in the perinucler region. g, Cluster group 12 presents reduced TF endocytosis nd endosomes ccumulted in the cell centre. h, Cluster group 13 presents selective increse in EGF endosomes numer nd their elongtion. i, Cluster group 14 presents minly n increse in TF endosomes elongtion. j, List of prmeter-groups compiled s in Fig. 1c. k, Tle providing summry of the most representtive phenotypes in ech group ( Description ). The numer of hits nd the totl size of ech group re lso indicted. j 245

4 ARTICLES NATURE Vol Mrch 21 nd signlling pthwys; 1% (468, P ) could ioinformticlly e ssigned to the core endocytic mchinery, emrcing estlished (CLTC, DNM2, EEA1) nd predicted genes on the sis of trfficking-relevnt functionl domins, such s VPS9 (8/1; P ), TBC (13/38; P 5 8), FYVE-finger (9/27; P 5 7) 21, PX (14/41; P 5 7) or BAR (4/15; P 5.59) 22 domins. Some genes (17%; 798; P ) were unknown owing to sence of curted nnottions. We found severl novel potentil endocytic regultors such s four TBC domin-contining proteins (MGC16169 (lso known s TBCK), TBC1D9, TBC1D9B, TBC1D21), two FYVE-finger proteins (MTMR4 nd WDFY1), nd protein with VPS9 domin (RINL). Notly, the list ws enriched in genes implicted in different humn diseses (Kyoto Encyclopedi of Genes nd Genomes dtse; see Supplementry Informtion) such s different cncers (for exmple, gliom (P ), colorectl (P ), pncretic cncer (P ) etc.), neurodegenertive nd metolic disorders (for exmple, denttorurl-pllidoluysin trophy; P ), Huntington disese (P 5 26) nd type II dietes (P ). Regultion y metolic nd signlling pthwys The cluster nlysis provided insights into how endocytosis is integrted with other cellulr functions nd mde predictions concerning the function of unchrcterized genes. Mny cluster groups were nti-correlted (Fig. 3 e nd Supplementry Fig. 8, ), reflecting positive nd negtive regultion on the endocytic system. For exmple, ltion of genes in cluster groups 1 nd 2 incresed nd decresed the mount of internlized TF, respectively (Fig. 3 nd Supplementry Fig. 8). Interestingly, genes of metolic pthwys were highly enriched in cluster group 1, most proly reflecting compenstory response to metolic deficit y generlly incresing endocytosis of nutrients. Another exmple of nti-correlted phenotypes is provided y cluster groups 3 nd 4 (Fig. 3 nd Supplementry Fig. 8), where the endosomes were loclized either closer to or more distnt from the nuclei, respectively. We found enrichment (P 5 22 in cluster group 3 nd P in cluster group 4) of components of the ctin nd tuulin cytoskeleton (for exmple, RAC1, ROCK2, CAPZB, microtuule nd ctin motors KIF2A nd MYH9). Interestingly, in cluster group 4 we found genes involved in cytokinesis such s CEP55 (ref. 23) nd KIF2A (ref. 24), strengthening the link with endocytosis 25. Cluster groups 8 nd 1 contined severl known endocytic components. Cluster group 8 (decresed internl TF nd EGF; Fig. 3d) contined mny genes encoding components required for endocytic uptke, such s CLTC, DNM2 nd phosphtidylinositol kinses/ phosphtses (INPP4A, INPP5B, PIB5PA, INPPL1; P ). Cluster group 1 contined severl endosoml regultors such s Rnkyrin-5 (ANKFY1) 26, Renosyn-5 (ZFYVE2) 27 nd Hrs (HGS), indicting tht EGF endocytosis is prticulrly susceptile to depletion of PI(3)P effectors. On the sis of these results, we predict tht mny new components of the endocytic core mchinery re in these cluster groups. Signlling pthwys exert distinct effects on the endocytic system. Here we confirmed the ctivity of mitogen-ctivted protein kinse (MAPK), C 21, integrin/cell dhesion nd mtor signlling pthwys 19,28 nd uncovered the ctivity of severl new ones, TGF-/ ctivin, Wnt nd Notch. Some (MAPK, C 21, mtor, Wnt) hd wide rnge of effects on the endocytic system (enriched in severl cluster groups; not shown) ut others produced very specific effects (enriched in single or opposite groups). For exmple, in cluster groups 3 nd 4 we found enrichment in the integrin/cell dhesion pthwy (P 5 12 in cluster group 3 nd P in cluster group 4; ITGA5, ITGB1, ITGA9, PAK1, MAPK9, MAPK1). Interestingly, the Notch nd TGF- pthwys were enriched (P 5 26 nd P , respectively) in cluster group 8 (reduced totl internl crgo for oth EGF nd TF). As we excluded Mcmilln Pulishers Limited. All rights reserved the possiility tht the reduction in totl internl crgo ws due to downregultion of surfce receptors (Supplementry Fig. 9), we suggest new role of these pthwys in the regultion of endocytic uptke. We found for the Notch pthwy NOTCH4, PSENEN nd PSEN2, MAML1 nd RBPJ nd for the TGF- pthwy severl genes encoding for ctivin receptors (ACVR1B, ACVR2A, ACVR2B nd ACVRL1). In conclusion, the cluster nlysis ssigns new functions to metolic nd signlling pthwys nd further predicts functions for new genes in endocytosis. Differentil regultion of TF nd EGF trfficking To revel generl design principles of the endocytic system we considered ll,16, knockdown conditions s independent pertured sttes of the system nd determined correltions etween prmeters. A first interesting spect emerging from such nlysis ws high divergence in EGF nd TF endocytosis. Despite the good correltion etween the size of EGF- nd TF-positive endosomes (Fig. 4) nd their intrcellulr locliztion (Fig. 4), we found n unexpected low correltion etween the levels of totl internl crgo (Fig. 4c) of the two mrkers. These results strongly indicte tht lthough the moleculr regultors of endosome size nd distnce from nucleus re essentilly shred y EGF nd TF, the mchineries controlling internliztion nd sorting re lrgely distinct. We exclude mjor contriution of lterntive non-clthrin-dependent entry pthwys 29 under our experimentl conditions, ecuse depletion of CLTC lmost completely locked the uptke of oth crgo mrkers (Supplementry Fig. 1). One possile interprettion is tht EGF nd TF my preferentilly e internlized into endocytic vesicles sujected to different regultion. For exmple, depletion of AP-2 complex specificlly reduces TF ut not EGF internliztion (Supplementry Fig. 1) 3, wheres knockdown of other genes such s PLEKHB2, CCDC33 nd RCHY1 specificlly downregulted EGF internliztion. Interestingly, silencing of certin genes resulted in the dispersl of EGF into smll, peripherl endosomes (Fig. 4e), pttern reminiscent of APPL1-endosomes 31. These endosomes re lrgely distinct from cnonicl EEA1-positive erly endosomes 31 nd ct s signlling pltforms 32. Knockdown of ZFYVE2, ANKFY1, PIK3R1, PIK3R2 long with novel genes such s CCDC128 (lso known s KLRAQ1), C12orf4, MGC16169 (lso known s TBCK) specificlly ffected EEA1-positive endosomes (Fig. 4d e) nd incresed the colocliztion of EGF with APPL1 (Fig. 4h i). Conversely, knockdown of INPP5B nd RABEP1 specificlly ffected APPL1-positive endosomes (Fig. 4d f) nd concomitntly decresed their colocliztion with EGF (Fig. 4h i). Altogether, we identified new components tht regulte the sorting of EGF to APPL1-positive endosomes. New design principles of the endocytic system Our nlysis lso showed tht vrious endosoml prmeters re significntly correlted nd, therefore, functionlly linked. For oth EGF (Fig. 5 c) nd TF, we oserved negtive correltion etween men size nd numer of endosomes, positive correltion etween the numer of endosomes nd their men distnce from nucleus nd negtive correltion etween endosome size nd distnce from nucleus. Totl internl crgo nd crgo concentrtion per endosome (Fig. 5d) s well s numer of endosomes (Fig. 5e) were lso strongly correlted positively. In contrst, there ws no correltion etween totl internl crgo nd endosome size (Fig. 5f) nd their distnce from nucleus (not shown). The nlysis of pir correltions indicte tht the endocytic prmeters re functionlly linked, ut does not infer cuse effect reltionships. To infer directionlity in these reltionships nd lern generl rules underlying the orgniztion of the endocytic system, we performed Byesin network nlysis. The structure of Byesin networks is directed cyclic grph where nodes denote the vriles of interest nd edges correspond to logicl dependencies etween

5 Distnce from nucleus TF NATURE Vol Mrch 21 ARTICLES Men size TF 3 Correltion coefficient = Men size EGF c Totl internl crgo TF 4 Correltion coefficient = Totl internl crgo EGF d e f RABEP1/EGF 2 μm Control/EGF 2 μm CCDC128/EGF 2 μm RABEP1/EEA Correltion coefficient = Distnce from nucleus EGF g Percentge colocliztion i h.6 Control.5 EGF>EEA1 EGF>APPL Percentge colocliztion Percentge colocliztion.6 CCDC128.5 EGF>EEA1 EGF>APPL.6 RABEP1.5 EGF>EEA1 EGF>APPL 2 μm Control/EEA1 2 μm CCDC128/EEA1 2 μm Control/APPL1 CCDC128/APPL1 RABEP1/APPL Figure 4 Regultors of trnsport of crgo to EEA1 vs APPL endosomes. c, Correltion etween endosome size (), endosome distnce from nuclei () nd totl internl crgo (c) for oth EGF nd TF. The dt re represented s sctter plots of z-score vlues. The density of the dt point is indicted in lue for low- nd in red for high-density vlues. Person correltion vlues (correltion coefficient) re indicted for ech sctter plot. d f, Exmples of knockdown of genes ffecting the ccumultion of EGF in APPL1-positive endosomes. HeL cells silenced for the indicted genes were fixed nd stined for the indicted mrkers. d, Mock-trnsfected control cells. e, Silencing of CCDC128 selectively ffected EEA1 endosomes (enlrged endosomes in the perinucler region), determining the ccumultion of EGF in APPL1-positive endosomes. f, Silencing of RABEP1 them 33. We constructed two Byesin networks for EGF nd TF (Fig. 5g h), where the nodes correspond to the key phenotypic prmeters nd the edges to their pir correltion. The two networks shred similr structure. Surprisingly, for oth EGF nd TF the prmeter mesuring endosome distnce from nucleus ws found to influence other prmeters, indicting key regultory role in reduced the endosoml stining of APPL1 determining n increse of the cytoplsmic stining. Insets show higher mgnifictions to visulize etter the endosoml (co)-locliztion. Merge insets show lwys EGF overlid on the indicted endosoml mrker. Scle rs, 1 mm. White rrows indicte the presence of EGF in the APPL1 nd EEA1-positive endosomes in the merge imges. Green rrows indicte the presence of EEA1 on the indicted endosomes in the EGF nd EEA1 insets. Light lue rrows indicte the presence of APPL1 on the indicted endosomes in the EGF nd APPL1 insets. g i, Quntifiction of EGF colocliztion with the mrkers indicted. EGF-to-APPL colocliztion is incresed upon silencing of CCDC128 (h) nd decresed upon silencing of RABEP1 (i) compred to control cells (g). Error rs indicte stndrd error of the men (s.e.m.). the endocytic system. The dependencies etween the triplet endosome distnce from nucleus numer size reflect progression from mny smll peripherl to few lrge perinucler endosomes. Genes involved in protein uiquitintion were found enriched mong the regultors of this process (cluster group 4 P 5 32). We found RBX1 nd DDB1, encoding components of multiprotein complex Men size Distnce from nuclei c Distnce from nuclei d 3 Correltion coefficient = Numer of endosomes e 3 Correltion coefficient = Numer of endosomes f Correltion coefficient = Men size Totl internl crgo Totl internl crgo Totl internl crgo Correltion coefficient = Crgo concentrtion Correltion coefficient = Numer of endosomes Correltion coefficient = Endosome size g.79 Totl internl crgo Crgo.79 concentrtion Figure 5 Design principles of the endocytic system. f, Anlysis of correltions etween different endocytic prmeters. The dt re represented s in Fig. 4 nd only correltions etween EGF prmeters re shown: negtive correltion etween endosome size nd numer (), positive correltion etween numer nd distnce from nuclei (), negtive correltion etween endosome size nd endosomes distnce from nuclei (c), positive correltion etween totl internl crgo nd men endosome concentrtion (d), positive correltion etween totl internl crgo nd Crgo content 3 5 EGF.61 3 Endosome size.61 Distnce from nucleus 7 Endosome numer 21 Mcmilln Pulishers Limited. All rights reserved Crgo concentrtion Totl internl crgo.88 3 Crgo content TF 6 Endosome size 2.68 Distnce from nucleus 8 Endosome numer endosome numer (e) nd wek correltion etween totl internl crgo nd endosome size (f). g h, Byesin networks of the endocytic prmeters for the genome-wide nlysis of EGF nd TF endocytosis. The nodes represent individul endocytic prmeters mesuring EGF or TF endocytosis nd the numers indicte the Person correltion vlues. In lck the direct correltions shred y the two networks, in red (present) nd grey (sent) correltions tht present differences etween the two networks. h 247

6 ARTICLES NATURE Vol Mrch 21 including cullins ctlysing uiquitin ligtion 34 nd other genes such s FBXO22, FBXO3, FBXO46, FBXO6, USP31 nd HECTD3, indicting new role of protein uiquitintion in the endosoml sptiotemporl progression. However, the two networks lso presented some interesting differences. The negtive dependency of endosome crgo content on distnce from nucleus reflects progressive concentrtion of EGF in endosomes, in contrst to TF tht is continuously removed nd recycled to the surfce. A comprison of the triplet concentrtion size content reveled some unexpected properties of the endocytic system. The TF network hs structure typicl of pssively sorted memrne proteins, where crgo content depends on size nd concentrtion of crgo per endosome. In contrst, the men concentrtion of EGF in the endosomes is djusted so tht the men totl content of EGF per endosome remins constnt, independently of endosome size (negtive dependence of the size on the concentrtion nd no correltion etween size nd content). Such mechnism cnnot e explined only y sorting of EGF into intr-luminl vesicles 9, ecuse (1) this is inconsistent with the high correltion etween size of EGF nd TF endosomes nd (2) the knockdown of HGS (Hrs) nd TS1 (ref. 9) does not lter the size concentrtion content reltionship. These dt rther indicte the existence of qusi-deterministic mechnism tightly coupling fusion nd fission events to regulte the concentrtion of EGF in ech endosome. Finlly, the cell exerts tighter control on the numer of EGF- thn TF-positive endosomes, s indicted y the stronger dependence on totl internl crgo (corr EGF 5.61 vs corr TF 5 6). In other words, to increse crgo cpcity the system in ddition to concentrting EGF in endosomes (s for TF) lso genertes new endosomes. Altogether, these findings indicte tht the cell regultes the numer, size, loding nd, therefore, the intrcellulr density of EGF-ering endosomes. These findings hve profound implictions for the role of endosomes in signlling 31,35,36. Discussion As first step towrds systems nlysis of endocytosis, we profiled the humn genome reltive to set of 58 prmeters tht quntittively descrie endocytic fetures. From this survey we could ttriute new functions to known nd unknown (798) genes with respect to EGF nd TF endocytosis. Furthermore, we reveled unexpected properties of the endocytic system. First, we uncovered the effect of vrious signlling pthwys, notly integrin/cell dhesion, TGF-, Notch nd Wnt, on oth TF nd EGF endocytosis. The ctivity of these pthwys depends on endocytosis 37 39, ut our dt further indicte tht they exert feedck mechnism on the endocytic mchinery. Interestingly, TGF-, Notch nd Wnt pthwys hve een implicted in the regultion of epithelil mesenchyml trnsition (EMT) 4, cell-fte determintion 41 nd tissue morphogenesis 42 nd given the estlished role of endocytosis in these events 2,43,we propose tht prt of their morphogenetic ctivity depends on their effects on the endocytic pthwy. Second, surprising result ws the low correltion etween genes regulting TF nd EGF endocytosis (Figs 5g h nd 4 c). Differences in the moleculr mchinery regulting crgo uptke nd recycling hve een reported 3, ut such degree of diversity is unprecedented. Third, our nlysis provided insights into the design principles underlying the endocytic system. The fct tht the endosome distnce from the nucleus is key regultory prmeter demonstrtes directly tht the sptio-temporl progression of endosomes oserved in living cells 12 is not n epiphenomenon ut hs fundmentl role in the function of the endocytic system. The mny components of the ctin nd tuulin cytoskeleton identified in the screen (cluster groups 3 nd 4) provide insights into the moleculr regultion underlying this mechnism. An interesting spect emerging from the comprison of the EGF nd TF Byesin networks is tht the cell specificlly regultes the numer of EGF-positive endosomes, nd strictly couples size nd concentrtion of crgo in endosomes. Given tht endosomes cn Mcmilln Pulishers Limited. All rights reserved ct s intrcellulr signlling pltforms 32,35, the system might regulte the concentrtion of EGF nd the numer of EGF-positive endosomes to quntittively control signlling responses, for exmple, for different cell-fte decisions, s shown for the MAPK network 44. Regulting the density of receptor tyrosine kinses, nd numer nd distriution of endosomes my e necessry to clirte the concentrtion of signlling molecules nd their signlling output. In identifying genes tht selectively ffect crgo trnsport to APPL1-positive endosomes, we unveiled sorting mechnism to distriute crgo to this comprtment. This is the eginning of more detiled moleculr chrcteriztion of APPL1 endosomes, their regultion of trnsport nd their role in signl propgtion 32,35. However, limittion of our pproch remins the poor identifiction of highly homologous nd functionlly redundnt genes (for exmple, RAB5A, B, C nd so on). This could e overcome y comintions of severl sirnas, s for CBL, CBLB nd CBLC (Supplementry Tle 4). This explortory endevour required uncommon resources for primry screen. Nevertheless, it ws necessry to estlish the conditions for stremlining the technology nd rendering such genomic profiling esier nd more ffordle in the future. Severl genes identified in the screen re ssocited with humn diseses, confirming the fundmentl role of endocytosis in the pthogenesis of mny humn diseses nd providing novel potentil drug trgets. It will e importnt to determine which prmeters of the QMPIA might ct s physiopthologicl indictors reflecting disese-relevnt ltertions. When pplied to primry cells nd disese model systems, the screening pltform descried here will revel its true potentil for drug discovery, through the identifiction of novel therpeutic trgets, new mechnisms of ction, nd for discerning therpeutic effects of smll molecules from toxic ones. METHODS SUMMARY RNAi screen for endocytosis. HeL cells were reverse trnsfected with three genomic lirries (Amion, Qigen nd custom-mde esirna lirry 15 ) using 2 nm sirnas or 25 ng per well esirnas. Seventy-two hours fter trnsfection, cells were incuted with 1 ng ml 21 EGF-Alex 488 nd 5 mgml 21 Trnsferrin Alex 647 (Moleculr Proes) in serum-free medium for 1 min t 37 uc efore fixtion with formldehyde. Therefter, nuclei nd cytoplsm were stined with mgml 21 DAPI nd mm SYTO42 lue (Moleculr Proes). Automted imge cquisition. Triple-colour imges were cquired in fully utomted nd unised mnner using spinning disk confocl microscope (OPERA, Evotec Technologies-PerkinElmer) nd 34 wter immersion ojective (NA 5.9). Twelve imges per well were collected to otin sufficient numer of cells for relile sttisticl nlysis. Imge correction nd imge nlysis were performed using custom designed imge nlysis softwre (MotionTrcking; see Supplementry Informtion). EEA1-APPL1 endosome dissection of EGF trfficking. Genes of interest (78) were screened to detect ltertions in the EEA1- or APPL1-positive endosomes with respect to EGF trfficking. Following sirna trnsfection nd EGF Alex 488 internliztion, cells were stined with mouse monoclonl nti-eea1 (BD Biosciences; Phrmingen) nd rit polyclonl nti-appl1 (ref. 31) ntiodies nd qudruple-colour imges were cquired s descried (Methods). Detection of surfce receptors. To exclude chnges in the level of surfce receptors upon downregultion of Notch nd TGF- signlling pthwys, cells were trnsfected with si/esirnas trgeting these genes nd 1,34 rndomly selected si/esirnas, incuted with 1 ng ml 21 EGF Alex 488 nd 5 mgml 21 TF Alex 647 (Moleculr Proes) in serum-free medium for 3 min on ice, fixed nd stined with DAPI-SYTO42 s descried erlier. Signls were enhnced y stining with rit nti-alex 488 (Moleculr Proes) nd mouse monoclonl ntiodies ginst the ectodomin of humn TFRC (BD Biosciences; Phrmingen) without permeiliztion. Triple-colour imges (ten per well) were collected with 32 wter immersion ojective (NA 5.7) s descried erlier. Full Methods nd ny ssocited references re ville in the online version of the pper t Received 1 Decemer 28; ccepted 17 Decemer 29. Pulished online 28 Ferury; corrected 11 Mrch 21 (see full-text HTML version for detils). 1. Polo, S. & Di Fiore, P. P. Endocytosis conducts the cell signling orchestr. Cell 124, (26).

7 NATURE Vol Mrch 21 ARTICLES 2. Dudu, V., Pntzis, P. & González-Gitán, M. Memrne trffic during emryonic development: epithelil formtion, cell fte decisions nd differentition. Curr. Opin. Cell Biol. 16, (24). 3. Burgdorf, S. & Kurts, C. Endocytosis mechnisms nd the cell iology of ntigen presenttion. Curr. Opin. Immunol. 2, (28). 4. Pgno, R. E., Puri, V., Dominguez, M. & Mrks, D. L. Memrne trffic in sphingolipid storge diseses. Trffic 1, (2). 5. Gruenerg, J. & vn der Goot, F. G. Mechnisms of pthogen entry through the endosoml comprtments. Nture Rev. Mol. Cell Biol. 7, (26). 6. Bskys, A., Byzitov, I., Zhu, E., Fng, L. & Wng, R. R-medited endocytosis: linking neurodegenertion, neuroprotection, nd synptic plsticity? Ann. NY Acd. Sci. 1122, (27). 7. Mrsh, M. & McMhon, H. T. The structurl er of endocytosis. Science 285, (1999). 8. Stenmrk, H. R GTPses s coordintors of vesicle trffic. Nture Rev. Mol. Cell Biol. 1, (29). 9. Piper, R. C. & Ktzmnn, D. J. Biogenesis nd function of multivesiculr odies. Annu. Rev. Cell Dev. Biol. 23, (27). 1. Perrimon, N. & Mthey-Prevot, B. Applictions of high-throughput RNA interference screens to prolems in cell nd developmentl iology. Genetics 175, 7 16 (27). 11. Friedmn, A. & Perrimon, N. A functionl RNAi screen for regultors of receptor tyrosine kinse nd ERK signlling. Nture 444, (26). 12. Rink, J., Ghigo, E., Klidzidis, Y. & Zeril, M. R conversion s mechnism of progression from erly to lte endosomes. Cell 122, (25). 13. König, R. et l. A proility-sed pproch for the nlysis of lrge-scle RNAi screens. Nture Methods 4, (27). 14. Kulkrni, M. M. et l. Evidence of off-trget effects ssocited with long dsrnas in Drosophil melnogster cell-sed ssys. Nture Methods 3, (26). 15. Kittler, R. et l. Genome-wide resources of endorionuclese-prepred short interfering RNAs for specific loss-of-function studies. Nture Methods 4, (27). 16. Birminghm, A. et l. 39 UTR seed mtches, ut not overll identity, re ssocited with RNAi off-trgets. Nture Methods 3, (26). 17. Cullen, B. R. Enhncing nd confirming the specificity of RNAi experiments. Nture Methods 3, (26). 18. Echeverri, C. J. et l. Minimizing the risk of reporting flse positives in lrge-scle RNAi screens. Nture Methods 3, (26). 19. Pelkmns, L. et l. Genome-wide nlysis of humn kinses in clthrin- nd cveole/rft-medited endocytosis. Nture 436, (25). 2. Fukung, K. & Hostetler, L. The estimtion of the grdient of density function, with pplictions in pttern recognition. IEEE Trns. Inf. Theory 21, 32 4 (1975). 21. Hykw, A., Hyes, S., Leonrd, D., Lmright, D. & Corver, S. Evolutionrily conserved structurl nd functionl roles of the FYVE domin. Biochem. Soc. Symp. 74, 955 (27). 22. Hermnn, B. The BAR-domin fmily of proteins: cse of ending nd inding? EMBO Rep. 5, (24). 23. Crlton, J. G. & Mrtin-Serrno, J. Prllels etween cytokinesis nd retrovirl udding: role for the ESCRT mchinery. Science 316, (27). 24. Hill, E., Clrke, M. & Brr, F. A. The R6-inding kinesin, R6-KIFL, is required for cytokinesis. EMBO J. 19, (2). 25. Blusk, F., Menzel, D. & Brlow, P. W. Cytokinesis in plnt nd niml cells: endosomes shut the door. Dev. Biol. 294, 1 (26). 26. Schntwinkel, C. et l. The R5 effector Rnkyrin-5 regultes nd coordintes different endocytic mechnisms. PLoS Biol. 2, e261 (24). 27. Nielsen, E. et l. Renosyn-5, novel R5 effector, is complexed with hvps45 nd recruited to endosomes through FYVE finger domin. J. Cell Biol. 151, (2). 28. Glvez, T. et l. sirna screen of the humn signling proteome identifies the PtdIns(3,4,5)P3-mTOR signling pthwy s primry regultor of trnsferrin uptke. Genome Biol. 8, R142 (27). 29. Sigismund, S. et l. Clthrin-independent endocytosis of uiquitinted crgos. Proc. Ntl Acd. Sci. USA 12, (25). 3. Motley, A., Bright, N. A., Semn, M. N. & Roinson, M. S. Clthrin-medited endocytosis in AP-2-depleted cells. J. Cell Biol. 162, (23). 31. Miczynsk, M. et l. APPL proteins link R5 to nucler signl trnsduction vi n endosoml comprtment. Cell 116, (24). 32. Schenck, A. et l. The endosoml protein Appl1 medites Akt sustrte specificity nd cell survivl in verterte development. Cell 133, (28). 33. Perl, J. Proilistic Resoning in Intelligent Systems: Networks of Plusile Inference (Morgn Kufmnn, 1988). 34. Petroski, M. D. & Deshies, R. J. Function nd regultion of cullin-ring uiquitin ligses. Nture Rev. Mol. Cell Biol. 6, 9 2 (25). 35. Miczynsk, M., Pelkmns, L. & Zeril, M. Not just sink: endosomes in control of signl trnsduction. Curr. Opin. Cell Biol. 16, 4 46 (24). 36. Kermorgnt, S. & Prker, P. J. Receptor trfficking controls wek signl delivery: strtegy used y c-met for STAT3 nucler ccumultion. J. Cell Biol. 182, (28). 37. Hung, S. S. & Hung, J. S. TGF- control of cell prolifertion. J. Cell. Biochem. 96, (25). 38. Le Borgne, R. Regultion of Notch signlling y endocytosis nd endosoml sorting. Curr. Opin. Cell Biol. 18, (26). 39. Gglirdi, M., Piddini, E. & Vincent, J. P. Endocytosis: positive or negtive influence on Wnt signlling? Trffic 9, 1 9 (28). 4. Heldin, C. H., Lndström, M. & Moustks, A. Mechnism of TGF- signling to growth rrest, poptosis, nd epithelil-mesenchyml trnsition. Curr. Opin. Cell Biol. 21, (29). 41. Blnk, U., Krlsson, G. & Krlsson, S. Signling pthwys governing stem-cell fte. Blood 111, (28). 42. Weinmster, G. Notch signl trnsduction: rel rip nd more. Curr. Opin. Genet. Dev. 1, (2). 43. Entchev, E. V. & González-Gitán, M. A. Morphogen grdient formtion nd vesiculr trfficking. Trffic 3, 989 (22). 44. Sntos, S. D., Verveer, P. J. & Bstiens, P. I. Growth fctor-induced MAPK network topology shpes Erk response determining PC-12 cell fte. Nture Cell Biol. 9, (27). Supplementry Informtion is linked to the online version of the pper t Acknowledgements We cknowledge T. Glvez nd G. Mrsico, memers of the Zeril group for discussions nd scientific support. We re prticulrly indeted to K. Korn who developed trnsfection protocols nd orgnized vrious preprtory steps nd logistics efore the screen nd E. Krusz for the mngement of the HT-TDS, the MPI-CBG screening fcility. We thnk J. Schmitt, A. Lohmnn, S. Christ, N. Tomschke, A. Niederlein, J. Wgner, M. Gierth, E. Krusz for technicl, rootics nd computtionl ssistnce, nd M. Boes nd J. Oegem for IT support tht mde the lrge-scle computtionl nlysis possile. We cknowledge I. C. Bines, T. Glvez, J. Howrd, T. Hymn, M. McShne, G. O Sullivn nd P. Tomnck for comments on the mnuscript. This work ws finncilly supported y the Mx Plnck Society (MPG) nd y the systems iology network HeptoSys of the Germn Ministry for Eduction nd Reserch BMBF, the DFG nd the Gottlie Dimler und Krl Benz Stiftung. This work is lso prt of the project Endotrck, which received reserch funding from the Europen Community s Sixth Frmework Progrmme. Author Contriutions C.C., Y.K. nd M.Z. conceived the project nd M.Z. directed it. C.C. nd D.K. developed the endocytosis ssy under the guidnce of M.Z. nd E.F.; M.S. performed the screen nd imge cquisition. Y.K. developed the QMPIA with the help of C.C. nd J.C.R. nd performed ll computtionl nlysis of the screening dt. N.S. developed the clustering nd pheno-score lgorithms nd together with C.C. performed the nlysis of the clusters. C.R.B., under the supervision of B.H., performed the si/esirnas rempping nd ioinformtics nlysis. CC. nd Y.K performed the Byesin Network nlysis. F.B. provided the esirna lirry. R.H., under the supervision of M.S.M. nd W.E.N., provided IT support for the use of the computer cluster locted t the TUD. C.C.,Y.K. nd M.Z. wrote the mnuscript. Author Informtion Reprints nd permissions informtion is ville t The uthors declre no competing finncil interests. Correspondence nd requests for mterils should e ddressed to M.Z. (zeril@mpi-cg.de). 21 Mcmilln Pulishers Limited. All rights reserved 249

8 doi:138/nture8779 METHODS sirna nd esirna lirries nd rempping to RefSeq 28. Three genomic RNAi lirries, two commercil sirna lirries (Amion Silencer Genome wide sirna Lirry V.3 nd Qigen Humn Whole Genome sirna Lirry V.1) nd custom-mde esirna lirry 15 were screened. The Amion lirry included 51,88 sirnas trgeting 22,527 genes in the humn genome nd ws designed prtly on the sis of the RefSeq dtse Relese 15 nd prtly on the ENSEMBL dtse. The Qigen lirry, which included 91,956 sirnas trgeting 22,732 genes, ws entirely designed on the sis of the RefSeq dtse (kinse/ phosphtse nd gpcr susets: Refseq Relese 18, Druggle Genome Set: Refseq Relese 17, the rest of the Whole Genome Set nd the Predicted Genes Set RefSeq Relese 12). The esirna lirry ws prepred s descried previously 13 nd included 17,188 esirnas trgeting 16,722 humn genes, designed on the sis of the ENSEMBL (Relese 36) dtse. Amion nd Qigen vlidted 768 nd 3,73 sirnas y qrt PCR, respectively. Out of these, 74 (Amion) nd 3,72 (Qigen) were shown to e efficient (.7% reduction) t silencing their corresponding trget mrnas, providing n overll estimted efficiency of.95% for the Amion nd.99% for the Qigen lirries, respectively. The sequences of ll 3 lirries were rempped on the RefSeq Relese 28 dtse y BLAST. The sequence of n sirna or esirna ws mpped to given gene on the sis of the following three conditions: (1) 19/19 ses mtch on region on the mrna, (2) mtching with ll trnscripts (including differentilly spliced mrnas) of given gene, (3) sence of mtching with ny other trnscripts from other genes. All sirnas/esirnas tht did not meet these conditions were discrded from further nlysis. After re-mpping to RefSeq Relese 28, the 3 lirries covered 33,223 genes of which 28,279 encode for proteins or contin n open reding frme, corresponding to 62% coverge of the trnscriptome nd 73% of the proteome. HT RNAi trnsfection nd endocytic ssy. HeL cells were reverse trnsfected with 2 nm sirnas or 25 ng per well esirnas (finl concentrtion) with ml per well Oligofectmine (Invitrogen) in totl volume of 5 ml. For HT-trnsfection Freedom Evo pipetting root (Tecn) equipped with 384-fixed-needle-hed, n 8-needle-pipettor nd gripper ws used under customized iologicl sfety cinet. A fully utomted workflow prepred trnsfection complexes in 384 wells t throughput of 5 pltes/5 min (mnuscript sumitted). For cell seeding nd dispensing the ssy solutions the MTP dispensers WellMte nd Multidrop 384 (Thermo Scientific) were used; wsh procedures were performed using the Power Wsher 384 (Tecn). Briefly, 1 ml of 2 nm sirnas or 5 ng ml 21 esirnas were mixed with 1 mlof pre-diluted Oligofectmine (4 mlml 21 OptiMem) for 15 min to llow complex formtion. Susequently, 1 ml of this mixture were plced in 384-well CellCrrier pltes (Evotec Technologies). HeL cells (1 cells per well, counted with CASY Model TT, Schärfe System) were seeded in 4 ml per well. Pltes were incuted for 72 h t 37 uc nd 5% CO 2. Therefter, the cells were incuted with 1 ng ml 21 EGF Alex 488 nd 5 mgml 21 trnsferrin Alex 647 (Moleculr Proes) in serum-free medium for 1 min t 37 uc. Cells were susequently wshed with PBS nd fixed with 3.7% formldehyde. Nuclei nd cytoplsm were stined with mgml 21 DAPI nd mm SYTO42 lue (Moleculr Proes). For the nlysis of protein downregultion, cells were grown in 96-well pltes nd trnsfected with 2 nm sirnas (finl concentrtion) with ml per well Oligofectmine (Invitrogen) in totl volume of 1 ml. After 72 h totl protein extrcts were prepred nd nlysed y quntittive western lotting using ntiodies ginst EEA1, ZFYVE2 nd ANKFY1. Automted spinning disk confocl imging. Twelve imges per well were collected with 34 wter immersion ojective (NA 5.9) in fully utomted nd unised mnner with spinning disk confocl microscope (OPERA, Evotec Technologies-Perkin Elmer). Triple-colour imges were cquired in two sequentil exposures y three different CCD cmers: the Alex 488 signl ws collected in first exposure using 488 nm lser light nd 54/75 nd-pss filter; Alex 647 nd DAPI-SYTO 42 lue signls were collected in second exposure using 635 nm nd 45 nm lser light nd with 7/9 nd 45/5 nd-pss filters, respectively. Imges were susequently corrected (MotionTrcking; see Supplementry Informtion) for uneven field illumintion nd chromtic errtions ccording to reference imges. For reference, imges of 2.5 mm multicolour eds, nd imges of reference dyes nd drk field imges were collected using the Oper Adjustment Plte (Evotec Technologies-Perkin Elmer). Three-dimensionl imging of the endosome distriution of the cell ws performed on smll set of 27 sirnas (trgeting 1 genes of the humn genome). Using the sme microscope nd utomted imging protocols s ove 1 focl plnes with.5 mm steps were cquired. EEA1-APPL1 endosome dissection of EGF trfficking. A smller set of 78 genes ws further screened to detect ltertions in the EEA1- or APPL1-positive endosomes with respect to EGF trfficking. For this, cells were trnsfected, EGF Alex 488 ws internlized nd cells were fixed s ove. Cells were then permeilized with % sponin nd % fish geltin (s locking gent) nd susequently stined with mouse monoclonl nti-eea1 (BD Biosciences Phrmingen) nd rit polyclonl nti-appl1 (ref. 31) ntiodies. Fluorescently conjugted got nti-rit Alex 568 nd got nti-mouse Alex 647 secondry ntiodies (Moleculr Proes) reveled the ntigen signls nd DAPI-SYTO 42 lue stined the nuclei nd cytoplsm. Qudruple-colour confocl imges were collected (OPERA, Evotec Technologies-PerkinElmer) in two sequentil exposures: Alex 488 nd Alex 647 signls were collected in first exposure using 488 nm nd 635 nm lser light nd with 54/75 nd 69/5 nd pss filters; Alex 568 nd DAPI-SYTO 42 lue signls were collected in second exposure using 45 nm nd 561 nm lser light nd with 45/5 nd 65/4 nd-pss filters, respectively. Reference imges were cquired nd imge correction performed s efore. 21 Mcmilln Pulishers Limited. All rights reserved