Determining the architectures of macromolecular assemblies

Transcription

1 Vol Novemer 2007 doi: /nture06404 Determining the rchitectures of mcromoleculr ssemlies ARTICLES Frnk Aler 1 *, Svetln Dokudovsky 2 *{, Lieseth M. Veenhoff 2 *{, Wenzhu Zhng 3, Juli Kipper 2 {, Dmien Devos 1 {, Adisetyntri Suprpto 2 {, Orit KrniSchmidt 2 {, Rosemry Willims 2, Brin T. Chit 3, Michel P. Rout 2 & Andrej Sli 1 To understnd the workings of living cell, we need to know the rchitectures of its mcromoleculr ssemlies. Here we show how proteomic dt cn e used to determine such structures. The process involves the collection of sufficient nd diverse highqulity dt, trnsltion of these dt into sptil restrints, nd n optimiztion tht uses the restrints to generte n ensemle of structures consistent with the dt. Anlysis of the ensemle produces detiled rchitecturl mp of the ssemly. We developed our pproch on chllenging model system, the nucler pore complex (NPC). The NPC cts s dynmic rrier, controlling ccess to nd from the nucleus, nd in yest is 50 MD ssemly of 456 proteins. The resulting structure, presented in n ccompnying pper, revels the configurtion of the proteins in the NPC, providing insights into its evolution nd rchitecturl principles. The present pproch should e pplicle to mny other mcromoleculr ssemlies. A mechnistic understnding of the cell requires the structurl chrcteriztion of the thousnds of its constituent iologicl ssemlies 1. So fr, conventionl pproches hve provided vlule ut limited window into the structures of these ssemlies. For exmple, Xry crystllogrphy nd nucler mgnetic resonnce (NMR) spectroscopy cn resolve the tomic detils of individul proteins nd smll complexes, wheres electron microscopy produces morphologicl mps ut cn lck the ility to identify nd detil specific components in the mp of the whole ssemly. As result, we do not yet hve tomicresolution structures, or even lowresolution representtions, for the vst mjority of complexes in the cell. How, then, re we to resolve the moleculr rchitectures of these ssemlies? In n ttempt to ddress this prolem, we hve tken the yest (Scchromyces cerevisie) nucler pore complex (NPC) s cse in point. The NPC is mong the lrgest mcromoleculr ssemlies in the cell, mediting the exchnge of molecules tht pss etween the nucler nd cytoplsmic comprtments. Yest NPCs re,50 MD structures uilt of multiple copies of some 30 different proteins (nucleoporins), totlling t lest 456 protein molecules 2. Ech NPC is plstic structure emedded in the nucler envelope nd is composed of eight morphologiclly similr spokes surrounding centrl tue 3 6. Filling this tue nd projecting into oth the cytoplsmic nd nucler sides re flexile filmentous domins from proteins termed FG (phenyllnineglycine) repet nucleoporins; these domins form the docking sites for trnsport fctors tht crry mcromoleculr crgoes through the NPC. The NPC represents significnt chllenge for conventionl structure determintion pproches owing to its lrge size nd the high degree of flexiility of the complex nd its components. Thus, lthough electron microscopy hs provided vlule insights into the overll shpe of the NPC, its moleculr rchitecture (tht is, the sptil configurtion of its component proteins) hs yet to e reveled, nd tomic structures hve only een solved for domins covering,5% of its component protein sequences 7. The NPC therefore encpsultes mny of the ostcles tht will e encountered in the detiled structurl exmintion of other mcromoleculr ssemlies. We descrie here set of proteomics experiments nd computtionl pltform for converting the resulting dt into the structures of mcromoleculr ssemlies. Centrl to this pproch is the reliztion tht mny kinds of iophysicl nd proteomic dt contin vlule structurl informtion out ssemlies. Overview of integrtive structure determintion Our pproch to structure determintion cn e seen s n itertive series of four steps: dt genertion y experiment, trnsltion of the dt into sptil restrints, clcultion of n ensemle of structures y stisfction of these restrints, nd n nlysis of the ensemle to produce the finl structure (Fig. 1). The structure clcultion prt of this process is expressed s n optimiztion prolem, solution of which requires three min components: (1) representtion of the ssemly in terms of its constituent prts; (2) scoring function, consisting of individul sptil restrints tht encode ll the dt; nd (3) n optimiztion of the scoring function, which ims to yield structures tht stisfy the restrints. Formlly, our pproch is similr to the determintion of protein structures y NMR spectroscopy, in which the folding of the polypeptide chin is determined y stisfying distnce restrints etween pirs of toms 8. As with NMR spectroscopy, structure is computtionlly determined from experimentl dt. Here, toms 1 Deprtment of Bioengineering nd Therpeutic Sciences, Deprtment of Phrmceuticl Chemistry, nd Cliforni Institute for Quntittive Biosciences, Byers Hll, Suite 503B, th Street, University of Cliforni t Sn Frncisco, Sn Frncisco, Cliforni , USA. 2 Lortory of Cellulr nd Structurl Biology, nd 3 Lortory of Mss Spectrometry nd Gseous Ion Chemistry, The Rockefeller University, 1230 York Avenue, New York, New York 10065, USA. {Present ddresses: Lortory of Nucleocytoplsmic Trnsport, Institut Jcques Monod, 2 plce Jussieu, Tour 43, Pris 75251, Frnce (S.D.); Deprtment of Biochemistry, University of Groningen, Nijenorgh 4, 9747 AG Groningen, The Netherlnds (L.M.V.); Germn Aerospce Center (PTDLR), HeinrichKonenStrsse 1, D53227 Bonn, Germny (J.K.); Structurl Bioinformtics, EMBL, Meyerhofstrsse 1, D69117 Heidelerg, Germny (D.D.); Office of Technology Trnsfer, The Rockefeller University, 1230 York Avenue, New York, New York 10065, USA (A.S.); Herert Irving Comprehensive Cncer Center, Columi University, 1130 St Nichols Avenue, New York, New York 10032, USA (O.K.S.). *These uthors contriuted eqully to this work. 683

2 ARTICLES NATURE Vol Novemer 2007 re replced y proteins, nd their positions nd reltive proximities re restrined on the sis of dt from vriety of proteomics nd other experiments, including ffinity purifiction, ultrcentrifugtion, electron microscopy nd immunoelectron microscopy (immunoem). Dt genertion. The most importnt spect of our pproch is its potentil to use simultneously lmost ny conceivle type of informtion to determine ssemly structures. For exmple, sedimenttion nlysis of the isolted proteins cn e used to infer their shpes; immunoem cn give n pproximte locliztion of ech protein in the ssemly; nd ffinity purifiction of tgged proteins nd protein complexes cn yield informtion out the rrngement nd interctions of proteins within the ssemly. These dt cn e of kind not normlly used for structure determintion (for exmple, complexes identified y ffinity purifiction), cn refer to different levels in the structurl hierrchy (for exmple, protein domin, whole protein, or protein complex), nd cn e miguous in terms of their structurl interprettion (for exmple, the uncertinty s to which copy of the protein is involved in n interction, when multiple copies exist). The use of such dt for structure determintion presented us with four mjor chllenges. First, lrge mounts of suitle dt must e collected to give sufficient sptil informtion to define structures; fortuntely, the proteomic revolution hs provided methodologies tht llow us to grner enough informtion. Second, much of the dt cn e of reltively low precision; thus, to void overinterprettion, pproprite tolernces must e used in its structurl interprettion. Third, the possiility of flsepositive dt must e minimized nd tken into considertion. Fourth, miguity of the dt in terms of its structurl interprettion must e treted when multiple copies of the sme protein re present in n ssemly nd the experiment does not determine which specific instnce of protein is detected. All of these chllenges cn e ddressed y n integrtive pproch tht incorportes informtion vrying gretly in terms of its ccurcy nd precision; limittions of ny prticulr type of dt cn e overcome y the use of lrge nd diverse dt sets derived from synergistic experimentl methods 1,9. Dt trnsltion into sptil restrints. The dt cn e used to restrin mny different fetures of the ssemly, such s the positions of proteins, protein contcts, proximity etween proteins, nd the shpe nd symmetry of the whole ssemly. A restrint specifies vlues of the restrined feture tht re consistent with the experimentl informtion out it; perfectly stisfied restrint is indicted here y 0, wheres vlues lrger thn 0 correspond to violted restrint. Thus, restrint encodes our uncertinty in the restrined feture. In essence, restrints cn e thought of s generting force on ech component in the ssemly, to mould them into configurtion tht stisfies the dt used to define the restrints. Optimiztion. All the restrints re summed to otin scoring function, which determines the degree of consistency etween the restrined sptil fetures in structure nd the experimentl informtion; perfect structure is indicted y 0, reflecting the summed vlues of perfectly stisfied restrints, wheres vlues lrger thn 0 correspond to structure tht incresingly violtes restrints. The scoring function is then optimized to clculte structure tht minimizes violtions of the restrints. It is necessry to generte mny such structures to provide good smpling of structures tht re consistent with the dt (tht is, the ensemle ). Ensemle nlysis. All of the structures tht stisfy the input restrints re clustered into distinct sets, on the sis of their similrities. There re three possile outcomes of such clustering. First, if only single cluster of structures stisfies ll the input informtion, Dt genertion Ultrcentrifugtion 30 Svlues 1 Svlue Quntittive immunolotting 30 reltive undnces Affinity purifiction Overly ssy 75 composites 7 contcts Electron microscopy Electron microscopy mp Immunoelectron microscopy 10,615 gold prticles Bioinformtics nd memrne frctiontion 30 protein sequences Dt trnsltion into sptil restrints shpe Complex shpe stoichiometry connectivity in composites Z R contcts Z R NPC symmetry Z R Nucler envelope excluded volume locliztion Z R Nucler envelope surfce locliztion excluded volume Optimiztion Produce n ensemle of solutions tht stisfy the input restrints, strting from mny different rndom configurtions Ensemle nlysis positions contcts Gle Figure 1 Determining the rchitecture of the NPC y integrting sptil restrints from proteomic dt. First, structurl dt (red) re generted y vrious experiments (lck). Second, the dt re trnslted into sptil restrints. Third, n ensemle of structurl solutions tht stisfy the dt re 684 Up82 p configurtion p120 up188 Nu Nup N Derive the structure from the ensemle Assess the structure otined y minimizing the violtions of the sptil restrints, strting from mny different rndom configurtions. Fourth, the ensemle is clustered into distinct sets of structures on the sis of their similrities, nd nlysed in terms of protein positions, contcts nd configurtion.

3 NATURE Vol Novemer 2007 ARTICLES there is proly sufficient dt for determining the unique ntive stte. Second, if different clusters re consistent with the input informtion, either the dt re insufficient to define the single ntive stte or there re multiple ntive structures. If the numer of clusters is smll, the structurl differences etween them my suggest dditionl experiments so s to nrrow down the possile solutions. Third, if no structures stisfy ll input informtion, either the dt or their interprettion in terms of the restrints is incorrect. Given the first two outcomes, the ensemle cn e nlysed to determine different spects of the ntive stte, such s protein positions, contcts nd configurtion. The vriility of the ensemle provides n estimte of the precision of the structure determintion. We illustrte our pproch y determining the configurtion of the protein components in the NPC from the yest S. cerevisie (Fig. 1). Dt genertion As no single experimentl technique hs een sufficient to solve the moleculr rchitecture of the NPC, we used vriety of techniques, ech of which gve different nd synergistic informtion out the structure; the techniques were chosen to generte the needed structurl informtion with defined level of ccurcy. An NPC component list. To determine ny structure, we must first define its prts (Fig. 2). In the cse of the NPC, we hve lredy determined tht some 30 nucleoporins constitute the ssemly 2. Although the exct composition is still uncertin ecuse some proteins interct reltively trnsiently with the NPC, potentil omission of smll frction of such trnsient components is unlikely to interfere with structure determintion. The stoichiometry of ech component in the NPC. The stoichiometry of ech nucleoporin in ech hlfspoke hs een previously estlished 2. However, hving found the stoichiometry of to e miguous, we reexmined it with new strins nd found tht is present in two copies per spoke (Fig. 3 nd Supplementry Fig. 7). The shpe nd size of ech component. Next, we must represent the structures of the constituent nucleoporins. Becuse tomic structures hve not yet een solved for most nucleoporins, we estimted their shpes sed primrily on their sedimenttion coefficients determined y ultrcentrifugtion of the purified proteins (Fig. 3 nd Supplementry Informtion). The sedimenttion ehviour of most FG nucleoporins grees with their predicted filmentous, ntive disordered structure 10,11., n integrl memrne component, ppered to e highly elongted structure, consistent with its multiple domins modelled s cdherinlike folds 7. Most of the other nucleoporins pper to hve reltively compct tertiry structure tht is gin in greement with their predicted fold ssignments 7,12. The sevenprotein complex 13 could e seprted into two smller complexes on sedimenttion: n elongted tetrmer (composite 30, see elow) nd n elongted hexmer (composite 45, see elow), consistent with their elongted ppernce when visulized y electron microscopy 14. The size, shpe nd symmetry of the NPC. It is lso helpful to hve some informtion on the overll shpe nd symmetry of the NPC. The position of the nucler envelope memrne reltive to the NPC nd the NPC s symmetry re sed on our electron microscopy nd cryoelectron microscopy (cryoem) dt 5. These studies hve reveled n eightfold rottionl symmetry of the yest NPC nd n pproximte twofold rottionl symmetry etween the nucleoplsmic nd cytosolic hlves of the NPC, defining the hlfspoke s 16fold pseudosymmetry unit of the NPC (Fig. 2). We hve lso previously shown tht heprin tretment of isolted NPCs produced ringlike sustructure ( Pom rings ), which is ssocited with the pore memrne nd perinucler spce in the intct NPC 15. We isolted nd exmined these rings (Supplementry Informtion), nd found tht they hd mximum dimeter of,106 nm, consistent with the mesured mximum NPC dimeter of,97 nm 5. The locliztion of ech component in the NPC. We hve previously otined the corse locliztion of most nucleoporins within the NPC y immunoem, relying on goldlelled ntiody tht specificlly intercted with the loclized protein through its croxyterminl PrA tg (Fig. 4) 2. We hve now generted more ccurte nd complete immunolocliztion mp of the NPC, in which its constituent proteins, except, hve een loclized using lrger dt set nd improved nlysis (Fig. 4 nd Supplementry Informtion). Inherent limittions in the immunoem method llow it to provide only rod rnge of llowed xil nd rdil vlues for ech nucleoporin. Nevertheless, these rnges re smller thn the dimensions of the hlfspoke nd so re still informtive. Notly, most nucleoporins re found on oth the nucler nd cytoplsmic sides of the NPC nd re tightly pcked within region djcent to the nucler memrne (Fig. 4). Most of the FG nucleoporins re found on oth sides of the NPC, with smll numer found exclusively on the cytoplsmic or nucler side; for simplicity, we consider nd Nup100 to e cytoplsmiclly disposed nd to e nucleoplsmiclly disposed, lthough,20% of the signl of ech is found on the opposite side. Most of the nonfg nucleoporins re lso found on oth sides. The memrne proteins re found close to the nucler envelope memrne, nd PrA is loclized to the lumen of the nucler envelope. Our immunoem mp grees lmost entirely with independent locliztions performed y other groups. For exmple, nd hve previously een shown to e restricted to the peripherl cytoplsmic fce 16 ; Nup1 ws found on the peripherl nucler fce 17 ; nd,, Nup53 nd Nup59 were shown to loclize proximlly to oth sides of the A R Z 30 nm U Y 108 nm Pore side 78 nm X Equtoril plne Figure 2 Structurl representtion of the NPC., Hierrchicl representtion of the NPC tht fcilittes the expression of the experimentl dt in terms of sptil restrints. Formlly, we define the whole NPC ssemly A s set of symmetry units U of two different types with eight instnces ech, referred to s hlfspokes. Hlfspokes of the first type (green) reside t the cytoplsmic side nd hlfspokes of the second type (red) reside t the nucleoplsmic side of the nucler envelope. Two djcent hlfspokes, one of ech type, form spoke. Ech of the 16 NPC hlfspokes consists of set of proteins P tht re descried y their type nd index. Ech protein is represented y flexile string of eds B in the root representtion k 5 1. Additionl representtions k. 1 cn e derived from the root representtion (for exmple, y omitting some eds s in k 5 2or y comining eds s in k 5 3). For the NPC, ech protein is descried with up to nine different representtions., Top pnel: the dimensions of the nucler envelope, s tken from cryoem imges (ref. 5). Bottomleft pnel: the coordinte system we use hs the origin t the centre of the nucler envelope pore. The nucler envelope is indicted in grey. Bottomright pnel: the eightfold (C8) nd twofold (C2) symmetry xes of the NPC, s reveled primrily y cryoem 5. We pply the twofold symmetry only to proteins tht pper with identicl stoichiometry in oth the nucleoplsmic nd cytoplsmic hlfspokes. P Perinucler side 4.5 nm C8 B κ=1 κ=2 κ =3 C2 685

4 ARTICLES NATURE Vol Novemer 2007 NPC 18 (other independent locliztions re listed in Supplementry Tle 10). How the NPC components fit together. The corse shpe, pproximte position nd stoichiometry of ech nucleoporin re not c Frction: Alumin Ovlumin * * Nup1 140 Nup1* Nup / Gle Nup Nup Nup Nup Nup Nup Gle Pom Complex Complex Sedimenttion coefficient (S) 10 8 Amylse ADH 6 4 Alumin Ovlumin Pek frction Moleculr mss (kd) (protein +PrA) S os S mx /S os Stoichiometry (cyt.) Stoichiometry (nucl.) Bed numer Amylse ADH Bed representtion Bed rdius (nm) Figure 3 shpe nd stoichiometry informtion., shpe from hydrodynmic experiments. Purified ntive PrAtgged nucleoporins were sedimented on sucrose grdients, together with set of iotinlelled mrker proteins. Frctions were collected nd nlysed y immunolotting of the iotin nd PrA tgs. An immunolot of frctions from typicl sedimenttion nlysis is shown, indicting the position of the tgged protein ( PrA) together with the mrkers ovlumin (3.6 S), ovine serum lumin (4.3 S), lcohol dehydrogense (ADH, 7.4 S) nd mylse (8.9 S)., Pek positions for the sedimenting proteins were determined nd liner regression ws used to clirte the sedimenttion coefficients of the PrAtgged nucleoporin. c, Bed representtions k 5 1 of the NPC proteins nd their stoichiometries per hlfspoke. The stoichiometry of protein in the cytoplsmic (cyt.) nd nucleoplsmic (nucl.) hlfspoke, s mesured y quntittive immunolotting 2,isshown.S mx vlues were clculted sed on the moleculr mss (kd) of ech protein; S mx /S os, 1.4 indictes gloulr protein; , modertely elongted;.2, highly elongted 45.Ansterisk indictes tht Cterminl frgments were mesured. Also shown is visuliztion of the protein s flexile ed chin (shown here in its most extended configurtion), which is sed on sedimenttion nlysis, identifiction of domins y sequence comprison nd secondry structure prediction. 686 enough to uild n ccurte picture of the NPC: rther like the pieces in jigsw puzzle, we lso need informtion on the interctions etween nucleoporins. We otined this informtion from lrge numer of overly ssys nd ffinity purifiction experiments, s well s from the composition of the Pom rings (consisting of Pom34 nd ). An overly ssy identifies pir of proteins tht interct with ech other, wheres n ffinity purifiction identifies one or more proteins tht interct directly or indirectly with the it protein (Figs 5 nd 6 nd Supplementry Informtion). An ffinity purifiction produces distinctive set of coisolting proteins, which we term composite. A composite my represent single complex of physiclly intercting proteins or mixture of such complexes overlpping t lest t the tgged protein. We only used overly nd ffinity purifiction dt with signltonoise rtio ove demnding threshold (Supplementry Informtion). We designed severl ffinity purifiction methods to otin lrge nd diverse set of composites (Supplementry Informtion). PrA ws used s highffinity Cterminl purifiction tg on ech nucleoporin. Different cell frctions from the tgged strins served s strting mterils, lthough most frctions were produced y wholecell cryolysis, which proved to e rpid nd convenient, yielding high mounts of ech complex with miniml losses nd proteolytic dmge. We generted,20 vrints of extrction uffers with diverse properties to relese different kinds of complexes from the frctions. Complexes were isolted vi the tgged nucleoporins using ntiody conjugted to either Sephrose or mgnetic eds, lthough we preferred mgnetic eds s it permitted rpid, highyield isoltions, nd eliminted n upper size limit on the purified complexes (Supplementry Informtion). We lso performed ffinity purifictions from diploid compred with hploid strins to detect potentil second, untgged copy of given nucleoporin in the complex strong indiction of homotypic interction for tht nucleoporin; PrA nd PrA were the only two nucleoporins giving composites contining second untgged copy. Although Z (nm) Cyt. Nucl. 0 R (nm) Z Cyt. Nucl. R 60 Nucler envelope NPC Nup42 Nup49 Nup100 Nup59 Nup53 Pom34 Nup57 Nup60 Nup1 Pom NonFG Nup FG Nup FG Nup (Cyt. only) FG Nup (Nucl. only) Figure 4 Locliztion of proteins y immunoem., ImmunoEM montges for PrA nuclei nd PrA nucler envelopes. Scle rs re grduted in 10nm intervls using the coordinte system defined in Fig. 2. The mjor fetures in ech montge re shown schemticlly t the right, showing how the position of every gold prticle in ech montge ws mesured from oth the centrl Zxis of the NPC (R) nd from the equtoril plne of the nucler envelope (Z)., Estimted position of the C terminus of ech protein in the NPC reltive to the centrl Zxis of the NPC (R) nd the equtoril plne (Z) superimposed on the protein density mp of crosssection of the yest NPC otined y cryoem 5. The verge llowed rnges long the R nd Z coordintes (68 nm nd 64.5 nm, respectively) re indicted y the rown rs in the ottom right corner.

5 NATURE Vol Novemer 2007 ARTICLES we originlly designed our pproches for the purifiction of NPC complexes, they hve proved to e useful for the isoltion of mny types of complexes from different cells 7,12, Identifiction of proteins ws performed y mss spectrometry 25,26. Generlly, the most vicinl ssocites of the tgged protein should e pproching stochiometric mounts in the purified complexes; conversely, distlly ssociting proteins my e less undnt. By concentrting on only Coomssiestinle SDS polycrylmide gel electrophoresis (PAGE) nds, we ensured tht we identified only the more undnt proteins in ny given ffinity purifiction nd voided trce residuls (Fig. 5). Polypeptides elow,20 kd were excluded from this nlysis for technicl resons 27 ; however, due to their smll volume, their exclusion is not likely to significntly ffect structure determintion. Affinity purifictions of tgged versions of ll yest nucleoporins, s well s the NPCssocited messenger RNA trnsport fctors Gle1 nd Gle2 (refs 28, 29), yielded 73 distinct composites; together with overly ssys nd Pom ring dt, we hve defined totl of 82 composites (Fig. 6 nd Supplementry Informtion). The composites vried in complexity from dimers to those contining 20 proteins (composite 82) nd, importntly, shred significnt overlp in composition (Fig. 6). Therefore, we expect considerle synergy mong the composites when used to mp the rchitecture of the whole ssemly. A good exmple of the compositionl overlp is the complex (Fig. 5, ) 13,14,30. The smllest uilding locks of this complex re heterodimers (Fig. 5, composites 7, 14, 15). Under different isoltion conditions, these dimers cn e purified with n incresing numer of dditionl proteins, such s trimers (25, 20), tetrmer (33), pentmer (39), hexmers (44, 45, 51), nd the full septmeric complex (53, 54, 57). This full complex intercts with (63, 66) nd (60). Finlly, the entire complex coprecipittes together with the complex nd n contining complex (79). Our dt lso gree with composites generted y other groups. For exmple, the composites 13,14,30, Sec31 Sec31 Sec31 Sec31 / Sec24 Ur2 Iml1 / Yol138 / Ydr128 / Yl104 / / / / / / / / / / / / / / Sec23 / / / Mex67 Cdc19 IgG Cdc19 Cdc19 Tef1 Tef1 Tef1 Tef1 Tef1 Tef1 Eno2 Adh1 Tdh3 Adh1 Lsp1 Adh1 Tdh Figure 5 interctions of the complex., A smple of ffinity purifictions contining complex proteins. Affinitypurified PrAtgged proteins nd intercting proteins were resolved y SDS PAGE nd visulized with Coomssie lue. The nme of the PrAtgged protein together with corresponding identifiction numer for the composite is indicted ove ech lne (Supplementry Informtion). Moleculr mss stndrds (kd) re indicted to the left of the pnel. The nds mrked y filled circles t the left of the gel lnes were identified y mss spectrometry (either of the exmple shown here or of prllel version; Supplementry Informtion). The identity of the copurifying proteins is indicted in order elow ech lne; PrAtgged proteins re indicted in lue, copurifying nucleoporins in lck, NPCssocited proteins in grey, nd other proteins (including contminnts) in red. Ech individul gel imge ws differentilly scled long its length so tht its moleculr mss stndrds ligned to single reference set of moleculr mss stndrds, nd contrstdjusted to improve visiility., The mutul rrngement of the complexssocited proteins s visulized y their locliztion volumes. The locliztion volumes, otined from the finl NPC structure (Fig. 9), llow visul interprettion of the reltive proximities of the proteins. 687

6 ARTICLES NATURE Vol Novemer 2007 composite 31, composite 18, Nup42 Gle1 dimer 29, composite 32 nd others (Supplementry Tle 9) hve een previously descried, nd re completely consistent with the composites identified here. Dt trnsltion into sptil restrints The next step is to trnslte the experimentl dt out the NPC structure into sptil restrints (Fig. 1). These restrints were numerous, overlpping nd vried in type, nd thus were expected to e sufficient for defining the rchitecture of the NPC. Restrints nd the scoring function. Structure determintion is enled y expressing informtion s scoring function, the glol optimum of which corresponds to the structure of the ntive Pom rings Nup1 Nup100 * Gle1 Nup60 Nup59 Nup57 Nup53 Nup49 Nup42 Gle2 Pom34 Frequency Composite size Nup1 Nup100 Gle1 Nup60 Nup59 Nup57 Nup53 Nup49 Nup42 Gle2 Frequency 1, Similrity Pom Composite identifier Figure 6 proximity y ffinity purifiction., Composites determined y ffinity purifiction. The ffinitypurified nucleoporin PrA is indicted on the verticl xis, nd the corresponding nucleoporins in ech composite re shown on the horizontl xis. Composite identifiers re indicted to the right. Presence of nucleoporin in composite is indicted y lck ox, nd the tgged nucleoporin is indicted y light grey ox. In composite 64 () nd in composites 31 nd 61 (), second untgged copy of corresponding protein is present, indicted y lck ox. A direct interction determined y overly ssy is indicted y drk grey ox. The sterisk for indictes tht the dt were otined with GFPtgged., Distriutions of composite size (left) nd composite similrity (right). The similrity etween two composites is defined y 2/ (2 1 1 c), where is the numer of proteins tht occur in oth composites, is the numer of proteins present only in the first composite, nd c is the numer of proteins present only in the second composite. 688 ssemly 33. One such function is joint proility density function (PDF) of protein positions, given the ville informtion I out the system, p(c/i), where C 5 (c 1,c 2,,c n ) is the list of the crtesin coordintes (c i ) of the n component proteins in the ssemly (tht is, the configurtion of the proteins). This joint PDF gives the proility density tht component i of the ntive configurtion is positioned very close to c i, given the informtion I we wish to consider in the clcultion. In generl, I my include ny structurl informtion from experiments, physicl theories, or sttisticl preferences. The complete joint PDF is generlly unknown, ut cn e pproximted s product of PDFs p f tht descrie individul ssemly fetures (for exmple, distnces or reltive orienttions of proteins): pc=i ð Þ~ P p f (C=I f ) f The scoring function F(C) is then defined s the logrithm of the joint PDF: F(C)~{ ln P p f (C=I f )~ X r f (C) f f For convenience, we refer to the logrithm of feture PDF s restrint r f nd the scoring function is therefore the sum of the individul restrints. Setting up the representtion of the NPC. To define restrints on the components of n ssemly, we must first specify the symmetry unit of the ssemly (tht is, the hlfspoke in the cse of the NPC) (Fig. 2) nd the stoichiometry of its components (Fig. 3). In ddition, we must define the representtions of the components. Ech nucleoporin ws represented y flexile chin consisting of smll numer of connected eds (Figs 2 nd 3). The numer nd rdii of the eds were chosen to reproduce the protein msses nd the sedimenttion coefficients 34. The flexiility of the representtion nd the low grnulrity of the NPC structure re sufficient to ccommodte uncertinties in the mesured Svlues nd their interprettion. For the FG nucleoporins, no restrints other thn the chin connectivity nd excluded volume were imposed on the eds representing the FGrepet regions. Given the symmetry unit nd the protein representtions, we cn formlly represent the NPC with fourlevel hierrchy corresponding to the whole NPC, the hlfspokes, proteins nd eds representing ech protein (Fig. 2). In ddition, the nucler envelope ws represented s rigid surfce of mny smll eds, providing mould in which the NPC forms (Fig. 2). Symmetry of the NPC. The eightfold nd pproximte twofold rottionl symmetries of the NPC (Fig. 2) were imposed y requiring essentilly identicl configurtions of the proteins in common within ech hlfspoke; the corresponding restrint is formlly the rootmensqure of the differences etween equivlent intrhlfspoke distnces. Although ny individul NPC ssemly my e pertured from this perfect symmetry t ny given point in time, restrints on the symmetry re nevertheless justified y the reltively lowresolution structure reported here, our intent to chrcterize the verge structure, nd exclusion of the FGrepet regions from the symmetry restrints. positions from immunoem. To reflect the uncertinty in the immunoem dt, we do not restrin protein to specific position. Insted, the Cterminl ed of ech protein, corresponding to the tg position, ws restrined y imposing lower nd upper hrmonic ounds on its Z nd R coordintes (Fig. 2), corresponding to the rnges llowed y the immunoem dt. On verge, the llowed re spns 16 nd 9 nm long the R nd Z coordinte, respectively (Fig. 4 Supplementry Tles 2 nd 7, nd Supplementry Fig. 8). With such lrge llowed rnges, the immunoem dt provide little more informtion to the structure clcultion thn which side of the nucler envelope ech nucleoporin is on, nd whether it is close to or distl from the NPC equtoril plne nd the NPC xis.

7 NATURE Vol Novemer 2007 ARTICLES positions using the nucler envelope s mould. The trnsmemrnespnning helices of the three memrne proteins, nd Pom34 were predicted y the progrm TMHH 35. The corresponding eds were then restrined to the surfce of the nucler envelope y hrmonic positionl restrints. In ddition, the terminl regions of ech protein were restrined either to the pore or perinucler sides of the nucler envelope, on the sis of the immunoem dt nd the numer of predicted trnsmemrne helices 2. proximities from overly ssys nd ffinity purifictions. The overly ssys nd ffinity purifictions crry informtion out protein proximities, nd so re encoded y the sme type of sptil restrint. These dt provide the richest set of restrints for our NPC structure. To interpret ech composite in terms of sptil restrint, we must consider three miguities. First, there is n miguity s to wht contcts re present in composite when it contins more thn two proteins. A composite implies only tht copy of ech protein in the composite must directly interct with t lest one copy of nother protein in the composite; ny structure tht stisfies this condition is consistent with the oserved composite. In other words, composite of n proteins implies t lest n21 such interctions etween proteins of ll types in the composite. Thus, ech llowed comintion of protein interctions corresponds to spnning tree of composite grph (s explined in Fig. 7). Second, when there re multiple copies of the sme protein in the ssemly, there is n miguity s to which copy is involved in given type of interction (Fig. 7). A mesured interction implies only tht t lest one copy of the protein is involved in tht interction. Third, when multiple eds re used to represent protein, there is n miguity s to which ed is involved in the interction (Fig. 2). A mesured interction implies only tht t lest one ed of the protein is involved in tht interction. As result of these three miguities, we need to encode composite y conditionl restrint, ensuring tht ll llowed comintions of lterntive ssignments of intercting ed pirs re considered (Fig. 7). Finding the ssignment of interctions to specific eds tht stisfies the dt ecomes prt of the optimiztion process (see elow). Other minor restrints were lso derived from O RS O RS O RS O RS O RS O RS Composite grph Spnning trees Miniml spnning tree O MST Figure 7 Amiguity in dt interprettion nd conditionl restrints., The miguity for protein interction etween proteins of green nd yellow types is illustrted. The miguity results from the presence of multiple copies of the sme protein in the sme or neighouring symmetry unit. In our NPC clcultions, oth neighouring hlfspokes on the cytoplsmic nd nucleoplsmic sides re considered, for totl of four neighouring hlfspokes (not shown)., The conditionl restrint is illustrted y n exmple of composite of four protein types (yellow, lue, red, green), derived from n ssemly contining single copy of the yellow, lue, nd red protein nd two copies of the green protein; proteins re represented y single ed (lue protein), pir of eds (green nd red proteins), nd string of three eds (yellow protein) (right pnel). This composite implies tht t lest three of the following six possile types of interction must occur: lue red, lue yellow, lue green, red green, red yellow nd yellow green. In ddition, (1) the three selected interctions must form spnning tree of the composite grph (defined elow); (2) ech type of interction cn involve either copy of the green protein (in generl, ll lterntives must e considered s illustrted in ); nd (3) ech protein cn interct through ny of its eds. These considertions cn e encoded through treelike evlution of the conditionl restrint. At the top level, ll optionl ed ed interctions etween ll protein copies re clustered y protein types. Ech lterntive ed interction is restrined y hrmonic upper ound on the distnce etween the eds; these re optionl restrints, ecuse only suset is selected for contriution to the finl vlue of the conditionl restrint. Next, rnkndselect opertor (O RS ) selects only the lest violted optionl restrint from ech interction type, resulting in six restrints (thick red line) t the middle level of the tree. Finlly, the miniml spnning tree opertor (O MST ) finds the comintion of three restrints tht re most consistent with the composite dt (thick red line); here the edge weights in the miniml spnning tree (defined elow) correspond to the restrint vlues given the current ssemly structure. The column on the right shows structurl interprettion of the composite with proteins represented y their coloured eds nd lterntive interctions indicted y edges etween them. The composite grph (shown on the left) is fully connected grph tht consists of nodes for ll identified protein types nd edges for ll pirwise interctions etween protein types; in the context of the conditionl restrint, the edge weights correspond to the restrint vlues. Five of the sixteen possile spnning trees re lso shown. A spnning tree is grph with the smllest possile numer of edges tht connect ll nodes. The miniml spnning tree is the spnning tree with the miniml sum of edge weights. This restrint evlution process is executed t ech optimiztion step sed on the current configurtion, thus resulting in possily different susets of selected optionl restrints t ech step. 689

8 ARTICLES NATURE Vol Novemer 2007 the overly ssy nd ffinity purifiction dt (Supplementry Informtion). Optimiztion With the scoring function in hnd, the positions of the proteins re determined y optimiztion of the scoring function (Supplementry Informtion), resulting in structures tht re consistent with the dt (Fig. 1). The optimiztion strts with rndom configurtion of the constituent proteins eds, nd then itertively moves them so s to minimize violtions of the restrints (Fig. 8). In essence, the restrints cooperte to slowly pull together the proteins into goodscoring configurtion. We use stndrd methods of conjugte grdients nd moleculr dynmics with simulted nneling (Supplementry Informtion). These methods llow the evolving structure some rething room to explore the scoring function lndscpe, minimizing the likelihood of getting cught in locl scoring function minim (Fig. 8). To comprehensively smple structures consistent with the dt, independent optimiztions of rndomly generted initil configurtions were performed until n ensemle of 1,000 Contct similrity Score Numer of configurtions ,000 4,000 Score Figure 8 Clcultion of the NPC ed structure y stisfction of sptil restrints., Representtion of the optimiztion process s it progresses from n initil rndom configurtion to n optiml structure. The grph shows the reltionship etween the score ( mesure of the consistency etween the configurtion nd the input dt) nd the verge contct similrity. The contct similrity quntifies how similr two configurtions re in terms of the numer nd types of their protein contcts; contct etween two proteins occurs if the distnce etween their closest eds is less thn 1.4 times the sum of the ed rdii (Supplementry Informtion). The verge contct similrity t given score is determined from the contct similrities etween the lowest scoring configurtion nd smple of 100 configurtions with the given score. Error rs indicte stndrd devition. Representtive configurtions t vrious stges of the optimiztion process from left (very lrge scores) to right (with score of 0) re shown ove the grph; score of 0 indictes tht ll input restrints hve een stisfied. As the score pproches zero, the contct similrity increses, showing tht there is only single cluster of closely relted configurtions tht stisfy the input dt., Distriution of configurtion scores. The presence of configurtions with the score close to 0 demonstrtes tht our smpling procedure finds configurtions consistent with the input dt. These configurtions stisfy ll the input restrints within the experimentl error. 690 structures stisfying the input restrints ws otined (pproximtely 200,000 trils were required, running for pproximtely 30 dys on 200 CPUs) (Fig. 8). Ensemle interprettion We nlysed the ensemle of 1,000 structures tht stisfy the input dt (Fig. 8) in terms of protein positions, contcts nd configurtion (Figs 9 nd 10). positions. These 1,000 structures were first superposed (Fig. 9) (Supplementry Informtion). Next, the superposed structures were converted into the proility of ny volume element eing occupied y given protein (tht is, the locliztion proility ) (Fig. 9). The spred round the mximum locliztion proility of ech protein descries how precisely its position ws defined y the input dt. The positions tht hve single nrrow mximum in their proility distriution in the ensemle re determined most precisely. When multiple mxim re present in the distriution t the precision of interest, the input restrints re insufficient to define the single ntive stte of tht protein (or there re multiple ntive sttes). The ctul locliztion proilities yielded single pronounced mxim for lmost ll proteins, demonstrting tht the input restrints define one predominnt structure. The verge stndrd devition for the distnce etween neighouring protein centroids is 5 nm; the precision of the lrger, centrlly positioned proteins seems to e higher thn tht of the nchor domins of some FG nucleoporins. This level of precision defines region smller thn the dimeters of mny nucleoporins. Thus, our mp is sufficient to determine the reltive positions of proteins in the NPC; we do not interpret fetures smller thn this precision. On the sis of the locliztion proilities (Fig. 9), we lso define the volume most likely occupied y ech protein, termed the locliztion volume (Figs 9c nd 10). The locliztion volumes of the proteins overlp only to smll degree, such tht only 10% of the NPC volume is ssigned to two or more proteins, gin underscoring how well the position of ech nucleoporin is resolved. On the sis of our current dt, we re not le to distinguish etween the two possile mirrorsymmetric structures; here, we present one of them. contcts. The proximities of ny two proteins in the structure cn e mesured y their reltive contct frequency, which is defined y how often the two proteins contct ech other in the ensemle (Fig. 10). Contcts re highly conserved mong the ensemle structures, despite some vriility; 32 protein pirs hve contct frequency higher thn 65%. Of ll the 435 contct frequencies, 7% re high (65 100%) nd 73% re low (0 25%); this gin demonstrtes tht the structure is well defined, s n ensemle of vried structures would yield minly medium contct frequencies. Notly, few highcontct frequencies re seen etween proteins of the sme type, indicting tht the NPC is held together primrily y heterotypic interctions. We cn improve our determintion of contcts y considering not only the contct frequencies ut lso the composite dt (Fig. 10c). More specificlly, we define two proteins to e djcent if their reltive contct frequency is lrger thn 65% or if they pper in the mximl spnning tree of ny composite grph whose edge weights correspond to contct frequencies (s explined in Fig. 10c). If two proteins re djcent, they re more likely to interct with ech other in the ntive NPC structure thn when they re not djcent 36. In totl, 51 types of djcencies were found (Fig. 10d). A prticulrly lrge numer of djcencies re oserved for nd, which oth pper in two copies per symmetry unit, s well s for the core proteins nd. Wheres the ltter two proteins ridge the ulk of the NPC to the memrne proteins nd lso provide nchor sites for FG nucleoporins, ridges mjor ring structures of the NPC nd lso serves s n nchor site for FG nucleoporins 37. Most FG nucleoporins re peripherlly locted nd therefore show only few djcencies.

9 NATURE Vol Novemer 2007 ARTICLES configurtion. We cn now comine the protein positions nd djcencies into configurtion of the NPC proteins (Fig. 10e, f). This representtion llows us to deconvolute the composites into their constituent complexes (for exmple, see Figs 5 nd 10g). Synergy mong restrints. How our dt ct synergisticlly is est demonstrted y the progressive increse in the certinty out the protein positions, s result of n incrementl ddition of informtion (Fig. 11). Hence, the vriility mong the 1,000 structures is significntly smller thn the uncertinties in ny of the originl dt. For exmple, the llowed rnges for protein locliztion y immuno EM re reduced from 64.5 nd 68 nm long the Zxis nd the rdil coordinte, respectively, to 62 nd 63 nm in the ensemle, s direct result of dt integrtion. Similrly, dt integrtion lso improves the prediction of protein interctions (Fig. 11). Assessment of precision nd ccurcy The ccurcy of model is defined s the difference etween the model nd the ntive structure. Therefore, it is currently impossile to know with certinty the ccurcy of the determined NPC structure. Nevertheless, five lines of evidence indicte tht the ccurcy of our structure is similr to its precision, nd thus representtive of the true configurtion of the NPC. Selfconsistency of the experimentl dt. Inconsistencies in the experimentl dt or its interprettion cn e identified when the optimiztion genertes only frustrted structures tht do not stisfy the input restrints. This is not the cse for our NPC clcultions; we find only single cluster of NPC structures tht stisfy ll the input restrints. To show tht it is not trivil to find structures stisfying ll restrints, we repeted the clcultions with comprle, ut prtly incorrect set of restrints (Supplementry Informtion). Specificlly, ll untgged proteins were rndomly swpped etween composites, leving the numer of composites, the numer of proteins in ech composite, nd ll other restrints unchnged. An optimiztion using this modified restrint set filed to produce ny structures tht stisfied ll restrints. Vriility in the ensemle. We hve confirmed tht the ensemle of 1,000 structures is sufficiently lrge for the precision of the NPC rchitecture to e determined relily: the reproduciility of contct frequencies clculted from rndom susets of the ensemle ws plotted s function of the suset size (Supplementry Informtion). The similrity etween two sets of contct frequencies converges for rndom susets of,100 structures. The ility of restrint set to define ntive stte. We hve previously descried n pproch to test whether or not given restrint set is sufficient to reconstruct known ntive stte 36. In this pproch, ntive structure is ssumed, the restrints to e tested re simulted from this structure, the structure is then reconstructed sed only on these restrints, nd finlly the reconstruction is compred to the originl ssumed structure. Using this pproch, we hve simulted composite restrints sed on our NPC structure, reproducing the numer of composites nd the distriution of their size in the originl dt set; ll other restrints were kept the sme s in the rel ppliction. The ccurcy of the reconstructed model ws comprle to the precision of the current NPC model. Ptterns unlikely to occur y chnce. The distriution of nucleoporins in our structure is expected to reflect their functionlity nd evolution, nd so should e decidedly nonrndom. Indeed, s discussed t length in the ccompnying pper 37, there is striking cosegregtion of proteins y fold type to prticulr loctions in the Bed models Locliztion proility c Locliztion volume + + Nup Figure 9 Bed model, ensemle, locliztion proility nd locliztion volume., Top: two representtive ed models of the NPC (excluding the FGrepet regions) from the ensemle of 1,000 superposed structures stisfying ll restrints (Fig. 8). The eight positions of three smple proteins (, Nup57 nd ) on the cytoplsmic side re shown, with detiled view of the ed representtion of one copy of t the ottom., The locliztion proility for ech protein type is otined y converting the ensemle into the proility of ny volume element eing occupied y the protein. Shown re contour mps of the crosssections in the plne prllel to the equtoril plne tht contins the mximum vlue of the protein locliztion proility. c, The locliztion volume of the smple proteins, derived from the locliztion proility. The volume elements re first sorted y their locliztion proility vlues. The locliztion volume then corresponds to the toprnked elements, the volume of which sums to the protein volume, estimted from its moleculr mss. The locliztion volume of protein revels its most prole locliztion. Becuse of the limited precision of the informtion used here, the locliztion volume of protein should not e mistken for its density mp, such s tht derived y cryoem. 691

10 ARTICLES NATURE Vol Novemer 2007 NPC, lthough no fold informtion (except for the trnsmemrne domins) ws used in the genertion of the structure. Experimentl dt not used in the clcultion of the model. Finlly, our structure cn e most directly tested y compring it to experimentlly determined dt tht were not included in the structure clcultion. First, our structure is roust, in the sense tht omission of rndomly chosen suset of 10% of the protein interction dt still results in structures with contct frequencies essentilly identicl to those derived from the complete dt set. Second, the shpe of our NPC structure 37 strongly resemles the pulished electron microscopy mps of the NPC 5,38 42, even though these dt were not used here (Supplementry Fig. 22). Third, the dimeter of the trnsport chnnel in our structure is,38 nm (excluding the FGrepet regions), in good greement with the experimentlly reported mximl dimeter of trnsported prticles 43. Fourth,, which hs een experimentlly shown to interct with highly curved memrnes vi its ALPSlike motif, is djcent to the nucler envelope in our structure 44. Moreover, three of the four dditionl scffold nucleoporins tht re predicted to contin the ALPSlike motif re lso close to the nucler envelope. Finlly, perhps the est exmple is tht of the complex. Our configurtion for this complex (Fig. 5) 37 is completely consistent with previous results 13,14,30. Specificlly, nd form dimer tht together with forms the trimeric hed of the complex, consistent with the top two rms of the Y shped complex (Fig. 5) 14. Similrly,,, nd form the til in Cytoplsmic side Nucleoplsmic side Cytoplsm 38 nm Nucleoplsm 98 nm Pom34 Gle2 Nup42 Nup49 Nup53 Nup57 Nup59 Nup60 Gle1 Nup100 Nup1 Nup1 Nup100 Gle1 Nup60 Nup59 Nup57 Nup53 Nup49 Nup42 Gle2 Pom c d Gle Nup42 Gle Nup100 Nup Nup Nup Nup Nup Nup60 Nup Pom Nup59 e f g Nup Figure 10 Ensemle interprettion in terms of protein positions, contcts nd configurtion., Locliztion volumes of ll 456 proteins in the NPC (excluding the FGrepet regions) in four different views. The dimeter of the trnsport chnnel nd the NPC re lso indicted. The proteins re colourcoded ccording to their ssignment to the six NPC modules 37., Contct frequencies for ll pirs of proteins. The contct frequency of pir of protein types is the frction of structures in the ensemle tht contins t lest one protein contct etween ny protein instnces of the two types. c, Contct frequencies etween proteins in composite 40. s re nodes connected y edges with the oserved contct frequency s the edge weight (indicted y its thickness). Edges tht re prt of the mximl spnning tree re shown y thick lue lines; the mximl spnning tree is the spnning tree tht mximizes the sum of the edge weights. All edges with sttisticlly significnt reduction in contct frequency from their initil vlues implied y the composite dt lone (Pvlue, ; Supplementry Informtion) re indicted y dotted lines with contct frequencies shown in red. d, djcencies for the whole NPC, with proteins s nodes nd edges connecting proteins tht re determined to e djcent to ech other. The edge weight is the oserved contct frequency. e, Configurtion of the proteins in composite 40. The loction of protein corresponds to the verge position of the eds representing nonfg repets of the protein. f, Configurtion of nd the NPC scffold proteins. g, Locliztion volume of nd the NPC scffold proteins

11 NATURE Vol Novemer 2007 ARTICLES Nucler envelope pore volume ImmunoEM Ultrcentrifugtion Overly ssys Nucleoporin stoichiometry Affinity purifictions NPC symmetry Single composite Nucleoporin stoichiometry NPC symmetry All dditionl composites Figure 11 The structure is incresingly specified y the ddition of different types of synergistic experimentl informtion., positions. As n exmple, ech pnel illustrtes the locliztion of 16 copies of in the ensemle of NPC structures, generted using the dt sets indicted elow. The locliztion proility is contoured t 65% of its mximl vlue (red). The smller the volume, the etter loclized re the proteins. The NPC structure is therefore essentilly moulded into shpe y the lrge mount of diverse experimentl dt., contcts. Prediction of protein interctions from contct frequencies improves s more dt re used. As n exmple, ech pnel illustrtes the contct frequencies etween proteins found in composite 34. Contct frequencies re shown s edge weights nd indicted y the thickness of the lines oth our structure nd the Yshped complex (Fig. 5) 14. Here, we resolve the reltive positions of the proteins in this complex nd show how the complex is integrted into the rchitecture of the entire NPC. Together these ssessments indicte tht our dt re sufficient to determine the configurtion of the proteins comprising the NPC. Indeed, it is hrd to conceive of ny comintion of errors tht could hve ised our structure towrds single solution tht resemles known NPC fetures in so mny wys. Conclusions We hve devised n integrtive pproch to solve the structure of the NPC using diverse iophysicl nd proteomic dt. This pproch hs severl dvntges. First, it enefits from the synergy mong the input dt. Dt integrtion is in fct necessry for structure determintion, ecuse none of the individul dt sets contins sufficient sptil informtion on its own. Despite the little structurl informtion in ech individul restrint, the concurrent stisfction of ll restrints derived from independent experiments mrkedly reduces the degenercy of the finl structures. Second, the integrtive pproch cn potentilly survey ll the structures tht re consistent with the dt. Alterntively, if no structure is consistent with the dt, then some experiments or their interprettions re incorrect. Third, this pproch cn mke the process of structure determintion more efficient, y indicting which mesurements would e most informtive. Fourth, the pproch cn, in principle, incorporte essentilly ny structurl informtion out given ssemly. Thus, it is strightforwrd to dpt it for clculting higher resolution + + Nucler envelope pore volume Ultrcentrifugtion ImmunoEM connecting the proteins. Left: when only single composite is used (together with stoichiometry nd symmetry informtion), ll interctions re eqully likely (initil contct frequency, Supplementry Informtion). Middle: when the highest likelihood of interction etween prticulr protein pir from ll composites is used, the uncertinty out the interctions is reduced. Right: when ll dt re used, the contct frequencies re either very high (.0.65) or very low (,0.25), thus llowing strong prediction of protein interctions. Contct frequencies reflect the likelihood tht protein interction is formed given the dt considered nd re clculted from the ensemle of optimized structures. Numers in red indicte finl contct frequencies tht significntly decresed (t Pvlue,10 23 ) from their initil vlues (Supplementry Informtion). structures y including dditionl sptil restrints from higher resolution dt sets, such s tomic structures of proteins, chemicl crosslinking, footprinting, smll ngle Xry scttering (SAXS) nd cryo EM. It is conceivle tht these dditionl dt sets might llow us to determine pseudotomic structures of ssemlies s complex s the NPC. Furthermore, y otining detiled structurl informtion concerning different stges of dynmic process, our pproch my nimte the NPC s ssemly nd trnsport mechnisms 6. The moleculr rchitecture of mny mcromoleculr complexes could, in principle, e resolved using similr integrtive pproch. With regrds to the NPC, the resulting structure hs lredy provided undnt insights into the function nd evolution of the cell 37. METHODS SUMMARY See Supplementry Informtion for detiled description of our Methods. The experimentl dt, the Integrtive Modelling Pltform softwre nd the NPC structurl model re ville t Received 30 August; ccepted 22 Octoer Sli, A., Gleser, R., Ernest, T. & Bumeister, W. From words to literture in structurl proteomics. Nture 422, (2003). 2. Rout, M. P. et l. The yest nucler pore complex: composition, rchitecture, nd trnsport mechnism. J. Cell Biol. 148, (2000). 3. Mcr, I. G. Trnsport into nd out of the nucleus. Microiol. Mol. Biol. Rev. 65, (2001). 4. Weis, K. Nucleocytoplsmic trnsport: crgo trfficking cross the order. Curr. Opin. Cell Biol. 14, (2002). 693