Supertertiary structure of the synaptic MAGuKscaffold proteins is conservedJames J. McCanna, Liqiang Zhengb, Daniel Rohrbeckc, Suren Felekyanc, Ralf Kühnemuthc, R. Bryan Suttond,Claus A. M. Seidelc, and Mark E. Bowenb,1

aDepartments of Pharmacological Sciences, and bPhysiology and Biophysics, Stony Brook University, Stony Brook, NY 11794; cInstitut für PhysikalischeChemie, Lehrstuhl für Molekulare Physikalische Chemie, Heinrich-Heine-Universität, Universitätsstraβe 1, Geb 26.32, 40225 Düsseldorf, Germany; anddCell Physiology and Molecular Biophysics, Texas Tech University Health Sciences Center, Lubbock, TX 79430

Edited by Taekjip Ha, University of Illinois at Urbana-Champaign, Urbana, IL, and accepted by the Editorial Board August 15, 2012 (received for review January6, 2012)

Scaffold proteins form a framework to organize signal transduc-tion by binding multiple partners within a signaling pathway.This shapes the output of signal responses as well as providingspecificity and localization. The Membrane Associated GuanylateKinases (MAGuKs) are scaffold proteins at cellular junctions thatlocalize cell surface receptors and link them to downstream signal-ing enzymes. Scaffold proteins often contain protein-bindingdomains that are connected in series by disordered linkers. Thetertiary structure of the folded domains is well understood, but de-scribing the dynamic inter-domain interactions (the superteritarystructure) of such multidomain proteins remains a challenge tostructural biology. We used 65 distance restraints from single-molecule fluorescence resonance energy transfer (smFRET) to de-scribe the superteritary structure of the canonical MAGuK scaffoldprotein PSD-95. By combining multiple fluorescence techniques,the conformational dynamics of PSD-95 could be characterizedacross the biologically relevant timescales for protein domainmotions. Relying only on a qualitative interpretation of FRET data,we were able to distinguish stable interdomain interactions fromfreely orienting domains. This revealed that the five domains inPSD-95 partitioned into two independent supramodules: PDZ1-PDZ2 and PDZ3-SH3-GuK. We used our smFRET data for hybridstructural refinement to model the PDZ3-SH3-GuK supramoduleand include explicit dye simulations to provide complete character-ization of potential uncertainties inherent to quantitative inter-pretation of FRET as distance. Comparative structural analysis ofsynaptic MAGuK homologues showed a conservation of this super-tertiary structure. Our approach represents a general solution todescribing the supertertiary structure of multidomain proteins.

intrinsic disorder ∣ protein structure ∣ single molecule fluorescence ∣fluorescence lifetime ∣ fluorescence correlation spectroscopy

Nature relies on scaffold proteins to provide the physical con-straints necessary for efficient signal transduction. Scaffold

proteins interact with multiple pathway components to hold thesignal transduction machinery in close proximity. Scaffolds areoften composed of modular, protein-binding domains linkedtogether in series by intrinsically disordered linkers (1). The pre-sence of disorder may be a defining feature for scaffolds andother proteins that interact with multiple binding partners (2).

The high effective concentrations brought about by domaintethering can give rise to unexpected interactions between theprotein-binding domains. In some cases, canonical protein-bind-ing domains fold together into an inseparable structural supramo-dule (3). While much has been learned by studying truncatedfragments, we need to put the pieces back together. Such ques-tions are difficult to address because disorder presents a funda-mental challenge to structural biology.

The Membrane-Associated Guanylate Kinase (MAGuK) scaf-fold proteins regulate signaling at cellular junctions (4). There arefour MAGuKs in excitatory synapses (PSD-95, PSD-93 SAP97and SAP102), which share the same arrangement of three PDZ

domains and an SH3 domain preceding an inactive GuanylateKinase domain (GuK). The sequence identity is greater than70% in the binding-domains, but falls to less than 20% within thelinkers. Physiological studies have suggested that the four MA-GuKs are not functionally redundant (5, 6) but their individualprotein-binding domains are structurally identical.

In this paper, we used single-molecule, fluorescence resonanceenergy transfer (smFRET) to describe the supertertiary structure(7) of full-length PSD-95 in solution. FRET has been long usedto define the 3D relationship between individual components ordomain fragments (8–11). We found that the five domains ofPSD-95 partition into two independent “supramodules.” Whilethe interdomain linkers permit dynamics, they do not impart iso-tropic domain positioning. Homologous measurements in SAP97and SAP102 showed that this organization is conserved in theother synaptic MAGuKs. This study represents the first unambig-uous assignment of domain positioning in a full-length scaffoldprotein and the most extensive characterization of a dynamic pro-tein structure with smFRET.

ResultsEnsemble Measurement of Compaction in PSD-95. The elution timein analytical size exclusion chromatography (SEC) is sensitive toboth protein shape and the density of domain packing. From theelution time, we calculated the apparent molecular weight forPSD-95 relative to globular protein standards (Fig. 1). The appar-ent molecular weight for full-length PSD-95 was larger thanexpected, indicating that the domains are not tightly packed. Toprobe domain interactions, we compared the difference betweenthe actual and apparent molecular weight (ΔMW) for a seriesof PSD-95 truncation constructs. The single PDZ domainshad a ΔMW of zero as expected for compact, folded domains(Fig. 1 E–G). For SH3-GuK and PDZ3-SH3-GuK, ΔMW wasslightly negative, indicating that PDZ3 localizes closely to theSH3-GuK domain (Fig. 1 H–I). In contrast, the PDZ1–2 tandemshows a positive ΔMW (Fig. 1D). Structures of the PDZ tandemhave revealed an extended rod-like conformation (12). Con-structs containing the N-terminus or the PDZ2–3 linker showedsignificantly higher ΔMW (Fig. 1 B and C). The deviation for thefull-length protein is primarily attributable to these two linkers,which are either disordered or in an extended conformation.

Author contributions: M.E.B. designed research; J.J.M., L.Z., D.R., S.F., R.K., R.B.S., andM.E.B. performed research; J.J.M., L.Z., and M.E.B. contributed new reagents/analytictools; J.J.M., D.R., S.F., R.K., R.B.S., C.A.M.S., andM.E.B. analyzed data; and J.J.M., D.R., R.K.,C.A.M.S., and M.E.B. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission. T.H. is a guest editor invited by the Editorial Board.1To whom correspondence should be addressed. E-mail: mark.bowen@sunysb.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1200254109/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1200254109 PNAS ∣ September 25, 2012 ∣ vol. 109 ∣ no. 39 ∣ 15775–15780

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

Dow

nloa

ded

by g

uest

on

Oct

ober

2, 2

021

smFRET Analysis of Domain Organization in PSD-95. We introducedand characterized 30 different labeling sites in the full-lengthprotein (Fig. S1). We made 65 FRET measurements to sampleall possible domain combinations multiple times using differentlabeling sites (Table S1). PSD-95 was labeled with Alexa555 andAlexa647 and measured with a Total Internal Reflection Fluor-escence (TIRF) microscope equipped with an EMCCD camerarunning at 100 ms∕frame. Proteins were encapsulated in immo-bilized phospholipid vesicles. A representative smFRET experi-ment, from raw data to computational analysis, is included in theSI Text (Fig. S2).

Single PSD-95 molecules showed stable FRET levels untilphotobleaching (Fig. S2A). Variations in the smFRET signal canindicate dynamics (13), but PSD-95 showed little signal variance(Fig. S2B). Most smFRET histograms showed single Gaussianpeaks (Fig. S2C and Table S1). The narrow peak widths are simi-lar to what we have measured in DNA (14) but also in randomcoil proteins with rapid dynamics (15). Thus, individual measure-ments are compatible with either a static structure or dynamics.Only 7 out of 65 histograms deviated from this by showing either awidth greater than 0.2 or a systematic deviation from a Gaussianpeak shape (asterisks, Table S1).

Distinguishing between a static and dynamic conformationis not possible with one EMCCD smFRET measurement. How-ever, “oversampling” the domain position using different labelingsite combinations did reveal dynamics. There is no time averagingin a static structure, so measurements using different labelingsites show a dispersion of mean FRET. Time-averaging fromdynamics makes FRET insensitive to labeling position. Similarly,limited dispersion is also seen in NMR resonances of disorderedproteins (16). To analyze FRET dispersion, we calculated thevariance of mean FRET (VMF) using either PDZ1 (Fig. 2A) orGuK (Fig. 2B) as the reference domain. VMF compares mea-surements using different labeling sites as opposed to variancein the smFRET signal.

The domains in PSD-95 partitioned into two categories basedon VMF. Domains with a well-defined conformation such asthe PDZ1–2 tandem and SH3-GuK, which have both been crys-tallized (12, 17, 18), showed a large VMF. The VMF for PDZ3to GuK was similar to that for SH3-GuK. This agrees with ourSEC data, which also suggested that PDZ3 associates with SH3or GuK. In contrast, only PDZ2 showed a high VMF relative toPDZ1. The low VMF of PDZ1 to all the other domains suggeststhat the PDZ1–2 tandem undergoes isotropic motion relative tothe rest of PSD-95. The narrow smFRET peak widths and lackof signal variance suggest rapid dynamics across the PDZ2–3linker (13).

Fig. 2. Variance of mean FRET distinguishes static and dynamic structures.(A) Variance of mean FRET (VMF) for measurements between PDZ1 and theother domains in PSD-95 (indicated beneath the panel). Measurementsbetween PDZ1 and PDZ2 was measured previously (44). VMF characterizesreplicate measurements using different labeling sites to oversample thedomain position. VMF between PDZ1 and PDZ2 was significantly higher thanmeasurements to PDZ3 or GuK (* p ¼ 0.046, Brown-Forsythe test for unequalvariance) (B) VMF for measurements between GuK and the other domains inPSD-95 (indicated beneath the panel). The VMF between GuK and PDZ3 orSH3 were indistinguishable from each other but significantly higher thanmeasurements to PDZ1 or PDZ2. (** p ¼ 0.048).

Fig. 3. PDZ3 has a defined position relative to the SH3 and GuK domains.(A) Residues used for labeling in smFRET measurements to PDZ3 are shown asspheres in the carton representation of the GuK domain. Coloring denoteswhether these sites showed high FRET (red, K591, Q621 and A640), mid FRET(green, R671 and H702) or low FRET (blue, E572 and S606). (B) Mean FRET foreach measurement between PDZ3 and GuK plotted against the groupedlabeling positions from panel A. Error bars indicate the standard deviationsfor each set of measurements. The mean FRETof these groups are statisticallydifferent (* p ¼ 0.033, one-way ANOVA). (C) Labeling sites in SH3 (left) andPDZ3 (right) used in interdomain smFRET measurements. Residues used aslabeling-sites are shown as spheres. Residues in SH3 are colored according totheir FRET dependence in measurements to PDZ3 (cyan; R492 and C445;magenta, T482). (D) Dependence of mean smFRET efficiency on labeling siteposition for measurements between SH3 and PDZ3. Curves are coloredaccording to the labeling sites as depicted in panel C. The PDZ3 site is denotedbeneath the plot.

Fig. 1. Probing domain organization in PSD-95 with analytical size exclusionchromatography. (A) Schematic representation of the domain order in PSD-95. The truncated PSD-95 constructs used in the SEC experiments are depictedin the table. Filled cells indicate the domain composition. For the SEC analysis,each fragment was designated by a letter in line with the associated row.(B) Logarithm of the molecular weight (MW) is plotted against the normal-ized elution volume (Ve∕Vo). Fragments are identified by the letter shownnext to each data point. Error bars indicate replicate measurement errors.The black line was calculated using globular protein standards. (C) The dif-ference between apparent molecular weight from SEC and formula weight(ΔMW) is shown for each PSD-95 fragment. Fragments are identified by theletters beneath the plot.

15776 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1200254109 McCann et al.

Dow

nloa

ded

by g

uest

on

Oct

ober

2, 2

021

PDZ3 Localizes Near the SH3-GuK Interface in Full-length PSD-95.Based on their location within the GuK domain (Fig. 3A), thesmFRET measurements between PDZ3 and GuK could begrouped into three clusters with statistically different meanFRET efficiencies (Fig. 3B). Measurements between 3 labelingsites in SH3 and 3 labeling sites in PDZ3 (Fig. 3C) showed a com-plex dependence on labeling position that is not compatible withisotropic rotation (Fig. 3D).

The fluorescent dyes have flexible linkers that minimize inter-actions with PSD-95, resulting in similar fluorescence anisotropyfor labeled protein and free dye (Table S1). The downside of thelinkers is uncertainty about the dye position and orientation re-lative to the protein backbone. To address this, we ran 1000 trialsof simulated annealing for each labeling site (Fig. S2D). This gen-erated a family of sterically allowed positions, which showed anisotropic distribution of orientations within a cone (Fig. S2E).The mean dye position was used as a fixed pseudo-atom to applyour distance restraints (10, 11, 19). We calculated an empiricalFörster radius (R0) for each FRET pair based upon measuredphotophysical parameters. Initially, we adopted the standard 2∕3approximation for the orientation factor (κ2). Anisotropy valuesin the range we measured are typically assumed to indicate suffi-cient dye mobility to use this assumption (20–23).

There are several approaches to combine low-resolution dis-tance restraints from FRET data with high-resolution structuresof individual components for hybrid structure determination (10,19, 24, 25). We used the Crystallography and NMR System(CNS), which is well established for macromolecular refinementand was recently adapted to use smFRET data (11, 26). For dock-ing, PDZ3 and GuK were held as rigid bodies along with the coreof the SH3 domain. From 500 trials with randomly assigned start-ing orientations, we identified a best-fit model with a root-mean-square error (ERMS) of 7.8 Å between dye positions and smFRETdistance restraints (Fig. 4A and Table S2). Lower scoring modelsshowed the same position for PDZ3, indicating a high precisionwith respect to discriminating from other possible binding sites(Fig. S3A) (27). PDZ3 is positioned close to the interface be-tween SH3 and GuK, which does not involve the canonical SH3or GuK ligand-binding sites (Fig. S3A). This position for PDZ3 issubstantially different than that observed in the distantly relatedMAGUK, zonula occludens 1 (ZO-1) (Fig. S3B) (28). Nonethe-less, our data confirms that a PDZ-SH3-GuK supramodule is thefundamental structural unit in the extended MAGuK family.

Once we had an initial model, we used our dye simulations toestimate the distributions of interdye distance (Fig. S2F) and dyeorientation (Fig. S2G) for each FRET pair. This allowed us totest our initial assumptions about κ2 and the use of a single pseu-doatom position. The mean κ2 from all simulations was 0.64,which is close to the standard κ2 ¼ 2∕3 approximation. The meanκ2 varied for each dye pair (Table S2), so we calculated anindividual R0 for each FRET measurement. Using simulatedκ2 only changed R0 by 1.2� 0.9 Å relative to using κ2 ¼ 2∕3,

and did not affect the model (Fig. S4B). Our model was also re-latively unchanged by the omission of any single distance restraint(Fig. S4C).

We can assess the reliability and quality of our model usingestablished benchmarks (29). As noted in that work, experimentalerror is small (typically ΔE < 0.01 FRET units) and contributeslittle. For any individual FRET measurement, the uncertaintyassociated with dye position and κ2 is high (Fig. S4D). Despitethis, the uncertainty for the model is much lower. The large num-ber of restraints oversamples the domain position, leaving fewsolutions to the docking calculation (29). If outlying dye confor-mations from the simulations were stably populated, the apparentdye positions would be significantly different than our calculatedaverages (Fig. S2H). This represents the worst-case scenarioin which fluorescent dyes maintain unfavorable positions andorientations. By incorporating these worst-case estimates into ourrefinement, we can illustrate the model dependence on uncer-tainties in the dye position and in κ2 (Fig. S4 E–G). The worst-case scenario associated with both dye position and orientationresults in a model RMSD of 10.2 Å. This level of uncertainty re-presents an unlikely scenario, since our anisotropy data indicatesthat the dyes are reorienting on the nanosecond timescale(see below).

A Model for the Native State Conformational Ensemble of Full-lengthPSD-95.Surprisingly, a comparison of 10 measurements in PSD-95truncations found no differences in smFRET compared to thefull-length protein (Fig. S5). Interactions between nonsequentialdomains appear to play no role in the organization of PSD-95.The four interdomain interfaces within PSD-95 span a continuumof stability. The stability of domain interactions scaled with thelinker length. The PDZ tandem and SH3-GuK have the shortestlinkers and are the most stable, while PDZ2–3 has the longestlinker and shows random coil dynamics.

The PDZ1–2 tandem and the PDZ3-SH3-GuK supramoduleappear independent. Using the Gaussian chain model, we pre-dicted a measured FRETof approximately 0.4 for domains con-nected by a 67 residue random-coil linker. The average FRETvalue for all measurements between PDZ3 and PDZ1 or PDZ2was 0.41� 0.08 (Table S1). Thus, the PDZ2–3 linker imparts themean domain separation of a relaxed, random coil. Our data sug-gests that full extension is not an intrinsic property of the linkersin PSD-95. Measurements between the N-terminus and PDZ3suggest that the longest average dimension of PSD-95 in solutionwould be approximately 13.5 nm. Our smFRET model is ingood agreement with SAXS and EM studies (30, 31), which alsosuggested that PSD-95 adopts a compact configuration in solu-tion (Fig. 4B).

Conformational Dynamics of PSD-95.None of our measurements be-tween PDZ3 and SH3-GuK showed FRET greater than 0.78, sothe closest point of approach for PDZ3 in our model is 5 Å from

Fig. 4. Amodel for full-length PSD-95 in solution. (A) Best-fit model from rigid body docking using smFRET restraints to position PDZ3 (orange) relative to SH3(red) and GuK (purple). The SH3-GuK supramodule is shown as surface representation, while PDZ3 is depicted as cartoon representation. Arrow indicates theSH3 HOOK domain. The left and right images are related by a 90° rotation as indicated. (B) Cartoon model for the supertertiary structure of full-length PSD-95.Domain positioning and the average dimensions (indicated in the panel) are in accordance with our smFRET restraints. This represents one snapshot image of adynamic protein that changes conformation on the submillisecond timescale.

McCann et al. PNAS ∣ September 25, 2012 ∣ vol. 109 ∣ no. 39 ∣ 15777

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

Dow

nloa

ded

by g

uest

on

Oct

ober

2, 2

021

SH3. The lack of direct contact suggests that PDZ3 undergoesmotional averaging. Some measurements to PDZ3 showed wideor irregularly shaped smFRET distributions (Table S1), which isconsistent with dynamics (13). Dynamics between PDZ3 andSH3 were also suggested by a recent NMR study on a PDZ3-SH3 fragment (32).

To analyze dynamics between PDZ3 and GuK, SH3 or PDZ2(Samples 34, 44, and 53, respectively; Table S1), we used confocalmicroscopy with multiparameter fluorescence detection (MFD),which allows FRET to be calculated based on both fluorescenceintensity and fluorescence lifetime (33). Comparing these twoparameters of FRET using 2D plots of the ratio of donor toacceptor fluorescence (FD∕FA) against donor lifetime (τDðAÞ)can distinguish static and dynamic structures (34). Owing to therandom labeling, we found two species in all samples: D-only(FD∕FA ≈ 40) and a FRET-population (highlighted in red). Two-color pulsed interleaved excitation (PIE) confirmed that only theFRET species had an active acceptor.

For all samples, the FRET species fell on the dynamic FRETline (solid line) rather than the theoretical static FRET line(dashed line) (Fig. 5 A–C, upper panels). The associated bottompanels contain corresponding 2D plots of donor anisotropiesrscatter against τDðAÞ. Comparison to the Perrin equation indicatesthat the majority of the donor populations have mean rotationalcorrelation times (ρ) ranging between 2 and 6 ns. Together withresidual anisotropies (Table S3), this indicates the dyes are notimmobile and do undergo dynamic reorientation (ρ ≈ τ). Evenif complete isotropic rotation does not occur, dye motions withinthe sterically allowed volume result in κ2 values close to 2∕3 (35).

The samples differed in FRETefficiency, peak width and theirlimiting FRET states as taken from fits to the subensemble life-time decays (Fig. 5D and Table S3). For PDZ3 relative to SH3-GuK (Fig. 5 A and B) the peaks are very broad, which indicatescomplex dynamics with a slow relaxation component on the orderof the diffusion time (1.7 ms). The equilibrium is shifted towardsthe high FRETstate as the peak is closer to τ1. We conclude that

PDZ3 undergoes dynamic binding to a defined site near theSH3-GuK interface with a mean occupancy of the position de-fined by our model.

In contrast, the FRET peak for PDZ2–3 (Fig. 5C) is narrower,indicating a faster relaxation time. The limiting donor lifetimessuggest a compact high-FRET state τ1 ¼ 0.8 ns (RDA ≈ 41 Å)and a more extended low-FRET state τ2 ¼ 3.4 ns (RDA ≈ 72 Å)with the equilibrium shifted towards the extended conformationconfirming the VMF analysis. The fluorescence cross-correlationfunctions have a prominent anticorrelation signature in the sub-millisecond time range (Fig. 5E). A formal fit (SI Text, Eq. S16)revealed relaxation times of approximately 1 μs, 100 μs, and anadditional third slow term of 1.7 ms The existence of multiplescales for domain motions is a hallmark of protein dynamics (36).

Modeling the Dynamic Loops of the SH3 Domain.The extended loopsin the PSD-95 SH3 domain were poorly resolved in the crystalstructure (18). We performed torsion angle refinements usingsmFRET restraints to describe the missing polypeptide. Refine-ments identified a lowest energy configuration of the pseudo-atom positions that is consistent with a family of loop conforma-tions. To validate our model for the SH3 loops, we performed a100 nanosecond, fully solvated, all-atom molecular dynamicssimulation based on the SH3-GuK crystal structure. The SH3domain showed rapid reorganization, while the GuK domain re-mained relatively unchanged (Fig. 6A). The SH3 domain showedlimited motion within the folded core, but the elongated loopswere highly dynamic (Fig. 6B). Both our family of smFRET mod-els and the conformational ensemble from the MD simulationshow fluctuations of a similar magnitude (Fig. 6B). In particular,the HOOK loop (491 to 522) adopted a mean configuration sig-nificantly different from the extended conformation partiallymodeled in the crystal structure.

Conservation of Supertertiary Structure in Synaptic MAGuK Homolo-gues. To identify major structural differences in the synaptic

Fig. 5. Probing PDZ3 dynamics with multiparamater fluorescence detection. PSD-95 was labeled in PDZ3 and (A) GuK (sample #34), (B) SH3 (#44) or (C) PDZ2(#53) (labeling scheme Table S1). 2D-MFD plots in the upper panel show the FRET indicator, FD∕FA (ratio of D and A fluorescence) against the donor fluor-escence lifetime in the presence of the acceptor (τDðAÞ, inset plot, top). The expected curve for static FRET is represented by the dashed line, while that fordynamic FRET is shown in black. The FRET populations identified by PIE are shown in red. The equations for the FRET lines are given in the SI Text, (Eq. S15B). (D)Subensemble lifetime fit for the FRET population confirms the existence of two limiting FRET states with τ1 ¼ 0.8 ns (34%) and τ2 ¼ 3.4 ns (66%). (E) Fluor-escence cross correlation spectroscopy (FCCS) curve for the FRET bursts confirm the multitime scale dynamics. The data was fit to a diffusion and relaxationmodel (blue line; see SI Text, Eq. S16). The pure diffusion component is illustrated by the dashed line.

15778 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1200254109 McCann et al.

Dow

nloa

ded

by g

uest

on

Oct

ober

2, 2

021

MAGuKs, we introduced labeling sites into homologous posi-tions in both SAP97 and SAP102. All four measurementsbetween PDZ1 and PDZ2 were identical in the three MAGuKshomologues (Fig. 7A). This implies a nearly identical structurefor the PDZ1–2 tandem. Measurements between PDZ1 orPDZ2 and PDZ3 indicate a similar separation between the supra-modules in PSD-95 and SAP102, with SAP97 showing a largerseparation (Fig. 7B). Measurements between PDZ3 and theGuK domain revealed differences between MAGuK proteins(Fig. 7C). This suggests structural differences in the mean posi-tion for PDZ3. However, the magnitude of the difference interms of distance (4.7 Å and 2.3 Å for SAP97 and SAP102, re-spectively) wouldn’t suggest a wholly different binding site forPDZ3 in SAP97 or SAP102. Thus, the organization of domainsinto two supramodules appears to be conserved.

DiscussionPSD-95 and the other synaptic MAGuKs play critical roles insynaptogenesis, receptor clustering, and the modulation of synap-tic plasticity (4). The available high-resolution structures fromPSD-95 represent the most tightly associated domains with theshortest interdomain linkers. Moving beyond the study of trun-cated fragments remains a challenge for proteins with intrinsicdisorder.

Signal transduction proteins are macromolecular entities, sothe supertertiary structure of scaffolds could impose geometricalconstraints on complex formation and activity. Numerous studieshave shown that linker sequence (37), linker length (38), domainorder (39), and interdomain interactions can alter scaffoldingactivity (40, 41). This suggests that although scaffolds contain dis-order, their supertertiary structure is a critical element of theirbiological activity. PSD-95 is a dynamic assembly of domains.Their connection assures the interaction in a crowded cellularenvironment.

Electron tomographic reconstructions of the synapse identifiedmembrane-associated filaments containing PSD-95 (42). Their20 nm length would imply a complete extension of a PSD-95monomer. However, EM immuno-localization in synapses alsomeasured the distances between PDZ3 or a C-terminal tag andthe synaptic cleft (42, 43). The similarity of these measurementssuggests that PDZ3 and GuK are equidistant from PDZ1. The8 nm separation measured between PDZ1 and PDZ3 also agreeswell with our smFRET measurements. Thus, immuno-EM mea-surements in synapses are compatible with our model for PSD-95.If PSD-95 retains a compact configuration in synapses, then thevertical filaments may not correspond to PSD-95 monomers butrather complexes containing PSD-95.

Our measurements in neutral phospholipid vesicles capturePSD-95 in its lowest energy “ground state.” Although PSD-95might not maintain this state in the synapse, this represents thesupertertiary structure encoded in the primary sequence. Nativeinteractions which are not present in our experiments could alterthe conformation. Such effects are unpredictable and couldexplain the biochemical differences between the closely relatedMAGuK isoforms.

MethodsAdditional details are provided in the SI Text.

Single Molecule TIRF FRET Experiments. Recombinant protein samples werepurified and fluorescently labeled with a mixture of Alexa Fluor 555 C2 mal-eimide and Alexa Fluor 647 C2 maleimide as described previously (44). PSD-95was encapsulated in 100 nm liposomes, which were immobilized to a quartzslide for Total Internal Reflection illumination with alternating laser excita-tion at 532 nm (approximately 7.5 mW before prism) and 633 nm (approxi-mately 3 mW before prism). Emission was collected using an Andor iXonEMCCD camera (Andor Technologies). To correct for distortions in the mea-sured donor and acceptor intensities, we used per molecule gamma normal-ization based upon acceptor photobleaching events (14).

Multiparameter Fluorescence Detection. PSD-95 was labeled with Alexa Fluor488 C5 maleimide and Alexa Fluor 647 C2 maleimide. Samples were excited at496 nm (40 μW) using a pulsed, linearly polarized argon-ion laser (Sabre®,Coherent) focused by an 60× 1.2 NA water immersion objective (Olympus).Fluorescence emission was detected by 4 single photon counting avalanchephotodiodes (SPCMAQR-14, Laser Components, for the acceptor, andPDM050CTC, MPD, for the donor). Laser beam diameter and a pinhole inthe detection path provided a detection volume of about 2 fL. Detector out-puts were registered by SPC 132 counting boards (Becker & Hickl).

Molecular Dynamics Simulations. Molecular dynamics simulations of the SH3-GuK structure (45) were run at the High Performance Computing Center(HPCC) facilities at Texas Tech University using GROMACS 4.5 (46). Simulationswere performed in a cubic box of water using boundary conditions at constanttemperature and pressure (NPT). The simulation was run for 100 ns in total.

Fig. 7. Supertertiary structure is conserved across the MAGUK family ofscaffold proteins. Each panel shows smFRET histograms for a set of measure-ments made between homologous positions in PSD-95 (black), SAP97(dashed) and SAP102 (grey). When shown, the number in each panel indi-cates the labeling site combination in PSD-95 according to Table S1. Measure-ments made between (A) PDZ1 and PDZ2, (B) either PDZ1 or PDZ2 and PDZ3and (C) PDZ3 and the GuK domain. Mean smFRETand peak widths from theseexperiments are listed in Table S4.

Fig. 6. Resolving SH3 domain loop configurations with molecular dynamicsand smFRET. (A) The rootmean squared deviation (RMSD) as a function of timeis plotted for the protein backbone relative to the starting structure during an100 ns all-atom molecular dynamics simulation of the SH3-GuK domain frag-ment (grey, SH3 domain; black, GuK domain). The SH3 domain showed a rapidreorganization that stabilized after approximately 30 nanoseconds. (B) Theroot mean squared fluctuation (RMSF) for each Cα atom in the SH3 domainduring the molecular dynamics simulation (black line) is plotted along withthe RMSF from our ensemble of smFRET-derived models (grey diamonds). Be-cause the core of the SH3 domain was held rigid in our docking calculations,the RMSF can only be calculated for freely moving atoms within the extendedloops. The regions containing secondary structural elements (α-helices as cylin-ders and β-sheets as arrows) are indicated above the panel.

McCann et al. PNAS ∣ September 25, 2012 ∣ vol. 109 ∣ no. 39 ∣ 15779

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

Dow

nloa

ded

by g

uest

on

Oct

ober

2, 2

021

Rigid Body Structural Refinement. As we have done previously (44), the dis-tance restraints were applied to the mean dye position, which was estimatedusing CNS version 1.3 (47). FRET restraint were modeled as a harmonic squarewell potential, with values closer to R0 having less uncertainty (11). Theenergy function included a repulsive term for nonbonded interactions andthe distance restraints, but lacked electrostatics and the attractive van derWaals terms (48). Cluster analysis of 500 simulations using different randomlygenerated starting positions was used to isolate similar structures. The best-

fit model was identified based upon the difference between FRET distancerestraints and the model results (26).

ACKNOWLEDGMENTS. We thank Suzanne Scarlata for assistance with fluoro-metry. We benefited from conversations with Axel Brunger, Steven Kaiser,and Urszula Golebiewska. We acknowledge the National Institutes of Healthfor funding to M.E.B. (MH081923) and the German Science Foundation forfunding to C.A.M.S. (SFB590).

1. Good MC, Zalatan JG, Lim WA (2011) Scaffold proteins: Hubs for controlling the flowof cellular information. Science 332:680–686.

2. Cortese MS, Uversky VN, Dunker AK (2008) Intrinsic disorder in scaffold proteins:Getting more from less. Prog Biophys Mol Biol 98:85–106.

3. FengW, ZhangM (2009) Organization and dynamics of PDZ-domain-related supramo-dules in the postsynaptic density. Nat Rev Neurosci 10:87–99.

4. Funke L, Dakoji S, Bredt DS (2005) Membrane-associated guanylate kinases regulateadhesion and plasticity at cell junctions. Annu Rev Biochem 74:219–245.

5. Bassand P, Bernard A, Rafiki A, Gayet D, Khrestchatisky M (1999) Differential interac-tion of the tSXVmotifs of the NR1 andNR2ANMDA receptor subunits with PSD-95 andSAP97. Eur J Neurosci 11:2031–2043.

6. Kim E, ShengM (1996) Differential K+ channel clustering activity of PSD-95 and SAP97,two related membrane-associated putative guanylate kinases. Neuropharmacology35:993–1000.

7. Tompa P (2012) On the supertertiary structure of proteins. Nat Chem Biol 8:597–600.8. Margittai M, et al. (2003) Single-molecule fluorescence resonance energy transfer

reveals a dynamic equilibrium between closed and open conformations of syntaxin1. Proc Natl Acad Sci USA 100:15516–15521.

9. Rasnik I, Myong S, Cheng W, Lohman TM, Ha T (2004) DNA-binding orientationand domain conformation of the E. coli rep helicase monomer bound to a partialduplex junction: Single-molecule studies of fluorescently labeled enzymes. J Mol Biol336:395–408.

10. Mekler V, et al. (2002) Structural organization of bacterial RNA polymerase holoen-zyme and the RNA polymerase-promoter open complex. Cell 108:599–614.

11. Choi UB, et al. (2010) Single-molecule FRET-derived model of the synaptotagmin1-SNARE fusion complex. Nat Struct Mol Biol 17:318–324.

12. Sainlos M, et al. (2011) Biomimetic divalent ligands for the acute disruption of synapticAMPAR stabilization. Nat Chem Biol 7:81–91.

13. Gopich IV, Szabo A (2007) Single-molecule FRET with diffusion and conformationaldynamics. J Phys Chem B 111:12925–12932.

14. McCann JJ, Choi UB, Zheng L, Weninger K, Bowen ME (2010) Optimizing methodsto recover absolute FRET efficiency from immobilized single molecules. Biophys J99:961–970.

15. Choi UB, McCann JJ, Weninger KR, BowenME (2011) Beyond the random coil: Stochas-tic conformational switching in intrinsically disordered proteins. Structure 19:566–576.

16. Dyson HJ, Wright PE (2004) Unfolded proteins and protein folding studied by NMR.Chem Rev 104:3607–3622.

17. McGee AW, et al. (2001) Structure of the SH3-guanylate kinase module from PSD-95suggests a mechanism for regulated assembly of MAGUK scaffolding proteins. MolCell 8:1291–1301.

18. Tavares GA, Panepucci EH, Brunger AT (2001) Structural characterization of theintramolecular interaction between the SH3 and guanylate kinase domains ofPSD-95. Mol Cell 8:1313–1325.

19. Schröder GF, Grubmüller H, Seidel CAM, Oesterhelt F (2008) Single-molecule FRETmeasures bends and kinks in DNA. Proc Natl Acad Sci USA 105:18337–18342.

20. Chakraborty A, et al. (2012) Opening and closing of the bacterial RNA polymeraseclamp. Science 337:591–595.

21. Ivanov V, Li M, Mizuuchi K (2009) Impact of emission anisotropy on fluorescencespectroscopy and FRET distance measurements. Biophys J 97:922–929.

22. Wu P, Brand L (1992) Orientation factor in steady-state and time-resolved resonanceenergy transfer measurements. Biochemistry 31:7939–7947.

23. Dale RE, Eisinger J, Blumberg WE (1979) The orientational freedom of molecularprobes. The orientation factor in intramolecular energy transfer. Biophys J 26:161–193.

24. Muschielok A, et al. (2008) A nano-positioning system for macromolecular structuralanalysis. Nat Methods 5:965–971.

25. Schröder GF, Grubmüller H (2004) FRETsg: Biomolecular structure model building frommultiple FRET experiments. Comput Phys Commun 158:150–157.

26. Vrljic M, et al. (2010)Molecularmechanism of the synaptotagmin-SNARE interaction inCa2+-triggered vesicle fusion. Nat Struct Mol Biol 17:325–331.

27. Korkin D, et al. (2006) Structural modeling of protein interactions by analogy: Appli-cation to PSD-95. PLoS Comput Biol 2:e153.

28. Pan L, Chen J, Yu J, Yu H, Zhang M (2011) The structure of the PDZ3-SH3-GuK tandemof ZO-1 protein suggests a supramodular organization of the membrane-associatedguanylate kinase (MAGUK) family scaffold protein core. J Biol Chem 286:40069–40074.

29. Knight JL, Mekler V, Mukhopadhyay J, Ebright RH, Levy RM (2005) distance-restraineddocking of rifampicin and rifamycin SV to RNA polymerase using systematic FRETmeasurements: Developing benchmarks of model quality and reliability. Biophys J88:925–938.

30. Fomina S, et al. (2011) Self-directed assembly and clustering of the cytoplasmicdomains of inwardly rectifying Kir2. 1 potassium channels on association withPSD-. Biochim Biophys Acta 1808:2374–2389.

31. Nakagawa T, et al. (2004) Quaternary structure, protein dynamics, and synaptic func-tion of SAP97 controlled by L27 domain interactions. Neuron 44:453–467.

32. Zhang J, Petit CM, King DS, Lee AL (2011) Phosphorylation of a PDZ domain extensionmodulates binding affinity and interdomain interactions in the PSD-95 MAGUK.J Biol Chem.

33. Sisamakis E, Valeri A, Kalinin S, Rothwell PJ, Seidel CAM (2010) Chapter 18—Accuratesingle-molecule FRET studies using multiparameter fluorescence detection. Methodsin Enzymology, ed GW Nils (Elsevier Academic Press, San Diego), Vol 475, pp 455–514.

34. Kalinin S, Valeri A, Antonik M, Felekyan S, Seidel CAM (2010) Detection of structuraldynamics by FRET: A photon distribution and fluorescence lifetime analysis of systemswith multiple states. J Phys Chem B 114:7983–7995.

35. Hoefling M, et al. (2011) Structural heterogeneity and quantitative FRET efficiencydistributions of polyprolines through a hybrid atomistic simulation and Monte Carloapproach. PLoS One 6:e19791.

36. Henzler-Wildman K, Kern D (2007) Dynamic personalities of proteins. Nature450:964–972.

37. Uversky VN, Oldfield CJ, Dunker AK (2005) Showing your ID: Intrinsic disorder as an IDfor recognition, regulation and cell signaling. J Mol Recognit 18:343–384.

38. Gerber HP, et al. (1994) Transcriptional activation modulated by homopolymericglutamine and proline stretches. Science 263:808–811.

39. Imamura F, Maeda S, Doi T, Fujiyoshi Y (2002) Ligand binding of the second PDZdomain regulates clustering of PSD-95 with the Kv1.4 potassium channel. J Biol Chem277:3640–3646.

40. Qian Y, Prehoda KE (2006) Interdomain interactions in the tumor suppressor discs largeregulate binding to the synaptic protein GukHolder. J Biol Chem 281:35757–35763.

41. WuH, et al. (2000) Intramolecular interactions regulate SAP97 binding to GKAP. EMBOJ 19:5740–5751.

42. Chen X, et al. (2008) Organization of the core structure of the postsynaptic density.Proc Natl Acad Sci USA 105:4453–4458.

43. Chen X, et al. (2011) PSD-95 is required to sustain the molecular organization of thepostsynaptic density. J Neurosci 31:6329–6338.

44. McCann JJ, Zheng L, Chiantia S, Bowen ME (2011) Domain orientation in the N-Term-inal PDZ tandem from PSD-95 is maintained in the full-length protein. Structure19:810–820.

45. McGee AW, Bredt DS (1999) Identification of an intramolecular interaction betweenthe SH3 and guanylate kinase domains of PSD-95. J Biol Chem 274:17431–17436.

46. Hess B, Kutzner C, van der Spoel D, Lindahl E (2008) GROMACS 4: Algorithms for highlyefficient, load-balanced, and scalable molecular simulation. J Chem Theory Comput4:435–447.

47. Brunger AT, et al. (1998) Crystallography and NMR system: A new software suitefor macromolecular structure determination. Acta Crystallogr D Biol Crystallogr54:905–921.

48. Brunger AT, Strop P, Vrljic M, Chu S, Weninger KR (2011) Three-dimensional molecularmodeling with single molecule FRET. J Struct Biol 173:497–505.

15780 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1200254109 McCann et al.

Dow

nloa

ded

by g

uest

on

Oct

ober

2, 2

021