PHASE SEPARATION

A molecular assembly phasetransition and kinetic proofreadingmodulate Ras activation by SOSWilliam Y. C. Huang1*, Steven Alvarez2†, Yasushi Kondo1,3,4†, Young Kwang Lee1,Jean K. Chung1, Hiu Yue Monatrice Lam1, Kabir H. Biswas5‡,John Kuriyan1,3,4,6,7, Jay T. Groves1,4,7§

The guanine nucleotide exchange factor (GEF) Son of Sevenless (SOS) is a key Ras activatorthat is autoinhibited in the cytosol and activates upon membrane recruitment. Autoinhibitionrelease involves structural rearrangements of the protein at the membrane and thusintroduces a delay between initial recruitment and activation. In this study, we designed asingle-molecule assay to resolve the time between initial receptor-mediated membranerecruitment and the initiation of GEF activity of individual SOS molecules on microarrays ofRas-functionalized supported membranes. The rise-and-fall shape of the measured SOSactivation time distribution and the long mean time scale to activation (~50 seconds)establish a basis for kinetic proofreading in the receptor-mediated activation of Ras. Wefurther demonstrate that this kinetic proofreading is modulated by the LAT (linker foractivation of T cells)–Grb2–SOS phosphotyrosine-driven phase transition at the membrane.

Akey step in the mitogen-activated proteinkinase (MAPK) pathway is the activationof the membrane-anchored Ras guanosinetriphosphatase (GTPase) by guanine nucleo-tide exchange factors (GEFs), including

Son of Sevenless (SOS) (1–5). SOS, which is animportant Ras GEF in T cell receptor (TCR)and epidermal growth factor receptor (EGFR)signaling, is recruited via Grb2 to activated re-ceptors or scaffold proteins on the membrane,where it subsequently activates Ras (5). The in-fluence of SOS dynamics in MAPK signaling isunderscored by the relatively low SOS proteincopy number in the EGFR-MAPK pathway (6).In addition to its GEF activity, SOS serves as across-linking molecule that, by interacting withtwo Grb2 molecules, forms linkages between re-ceptor or scaffold molecules (7, 8). In the caseof the scaffold protein linker for activation ofT cells (LAT), multivalency of the phosphorylatedtyrosine (pY) sites on LAT, which serve as Grb2binding sites, enables the networked assembly ofLAT-Grb2-SOS into a condensed two-dimensionalgel or liquid phase on the membrane surface

(9–13). Although similarproteinassembly–mediatedphase transitions are emerging as a general themein several different signaling systems (14, 15), littleis known about how they regulate signaling.We performed a detailed analysis of the mo-

lecular mechanism by which SOS autoinhibitionis released after receptor-mediated recruitmentto the membrane. We reconstituted purifiedfull-length SOS (SOSFL) along with Grb2 on sup-ported membranes containing phosphorylatedLAT and Ras to mimic the signaling geometry ofthe inner leaflet of the T cell plasma membrane.By developing a single-molecule activation assayin a membrane microarray format (16, 17), weresolved the timing process from the initial mem-brane association to the initiation of GEF activityfor each individual SOS molecule. The results re-veal a long mean time delay (~50 s) between ini-tial recruitment and the activation of SOS GEFactivity. Additionally, the delay time distribution—exhibiting the distinctive rise and fall of a gammadistribution—reveals the existence of rate-limitingkinetic intermediates in the activation of SOS.These features of SOS autoinhibition release,along with the system’s being intrinsically out ofequilibrium, provide the basic requirements for akinetic proofreading mechanism (18) in the ac-tivation of Ras (10). In such a mechanism, short-dwelling SOS molecules at the membrane mayfail to achieve release of autoinhibition and thusfail to activate any Ras, whereas long-dwellingSOS molecules would exhibit disproportionatelyhigher probabilities of activation. Such a screen-ing process may be especially important for SOSbecause, once activated, SOS remains trapped atthe membrane, where it can processively activatehundreds of Ras molecules (16, 19, 20).Lastly, we experimentally demonstrate that the

LAT-Grb2-SOS pY-mediated phase transition,which has been shown to extend SOS dwell times

(10), strongly modulates the amount of Ras ac-tivated by a group of SOSmolecules. These obser-vations suggest a mechanism by which proteinassembly phase transitions, such as LAT-Grb2-SOS, modulate downstream signaling activity viakinetic discrimination or proofreading processes.Receptor-mediated activation of SOS is tightly

regulated (3, 5, 10, 20–24). The catalytic coreof SOS, consisting of the Ras exchanger motifand a CDC25 domain, is flanked by a C-terminalproline-rich (PR) domain and N-terminal auto-inhibitory domains consisting of the histone fold,Dbl homology, and Pleckstrin homology domains(Fig. 1A). Mutations in the N-terminal domainsof SOS can lead to the dysregulation of auto-inhibition and the development of Noonan syn-drome disorder (25). The PR domain binds theSrc homology 3 (SH3) domains of Grb2, whereasthe SH2 domain of Grb2 interacts with pY resi-dues on membrane receptors such as LAT orEGFR, thereby localizing SOS onto membranesurfaces in response to receptor activation (4, 26).Upon membrane recruitment, interactions ofthe SOS N-terminal domains with anionic lipids,such as phosphatidylinositol 4,5-bisphosphate(PIP2), facilitate the release of autoinhibition(21, 27). The catalytic domains of SOS containtwo binding sites for Ras: catalytic and allosteric(28). With Ras engaged in the allosteric site, SOSremains on membranes for extended periodsof time (>1 min) and enters a highly processivemode in which it catalyzes hundreds of Rasguanosine triphosphate (GTP)–loading events(16, 19, 20). Although there is growing under-standing that a multitude of molecular interac-tions are critical for efficient SOS activation onmembranes (5), real-time monitoring of receptor-mediated SOS activation has been inaccessible,in part because of challenges in SOS purificationand the development of a membrane assay capa-ble of resolving such features.We developed a single-molecule activation as-

say on supported membranes to resolve the tim-ing of SOS activation frommembrane recruitmentto the initiation of nucleotide turnover in Ras (Fig.1B). Supported membranes consisting primarilyof DOPC (1,2-dioleoyl-sn-glycero-3-phosphocholine)lipids doped with 2% PIP2, 2% MCC-DOPE {1,2-dioleoyl-sn-glycero-3-phosphoethanolamine-N-[4-(p-maleimidomethyl)cyclohexane-carboxamide]},and 4% Ni2+–NTA-DOGS {1,2-dioleoyl-sn-glycero-3-[(N-(5-amino-1-carboxypentyl)iminodiaceticacid)succinyl]} were deposited on a glass sub-strate by vesicle fusion. Cytoplasmic domainsof the Src kinase Hck and the adaptor proteinLAT, both with six-histidine (His6) tags, weretethered to the supported membrane throughHis6 interactions with nickel-chelating lipids(29). Although Hck is not an efficient kinase forthe phosphorylation of LAT (30), it phosphor-ylates LAT to completion, given sufficient time.H-Ras molecules were covalently attached tomembranes via maleimide chemistry with theMCC-DOPE lipids (31). All of these proteins re-constituted on supported membranes were lat-erally mobile, with typical densities of <100,~800, and ~500 mm−2 for Hck, LAT, and Ras,

RESEARCH

Huang et al., Science 363, 1098–1103 (2019) 8 March 2019 1 of 6

1Department of Chemistry, University of California, Berkeley,CA 94720, USA. 2Department of Materials Science andEngineering, University of California, Berkeley, CA 94720,USA. 3Department of Molecular and Cell Biology, Universityof California, Berkeley, CA 94720, USA. 4California Institutefor Quantitative Biosciences, University of California,Berkeley, CA 94720, USA. 5Mechanobiology Institute,National University of Singapore, Singapore 117411,Singapore. 6Howard Hughes Medical Institute, University ofCalifornia, Berkeley, CA 94720, USA. 7Divisions of MolecularBiophysics and Integrated Bioimaging, Lawrence BerkeleyNational Laboratory, Berkeley, CA 94720, USA.*Present address: Department of Chemical and Systems Biology,Stanford University, Stanford, CA 94305, USA. †These authorscontributed equally to this work. ‡Present address: School ofMaterials Science and Engineering, Nanyang Technological Uni-versity, Singapore 639798, Singapore.§Corresponding author. Email: jtgroves@lbl.gov

respectively. The LAT densities in the reconsti-tution are comparable to those measured in cells(32). Although Ras local densities in cells canvary widely, we used densities similar to thoseof Ras clusters (>1000 mm−2) in cells (33, 34).LATwas fully phosphorylated byHck before theaddition of SOS, asmonitored by the recruitmentof fluorescent Grb2 (10). SOSFL was prepared byintein ligation (35) of the N-terminal flankedcatalytic domain and the C-terminal PR domainof human SOS1. By adding ~1 nM SOSFL and20 nM Grb2 in solution [the concentrations arecomparable to recent quantifications of proteinabundance in the EGFR-MAPK pathway (6);dissociation constant Kd ~ 10 mM (36)], single-molecule membrane recruitment of SOS viaGrb2 can be clearly resolved by total internal

reflection fluorescence (TIRF) microscopy. IfGrb2 was not included, we observed ~90% lowerSOS recruitment. On membranes, SOS catalyzesthe nucleotide exchange of Ras [preloaded withguanosine diphosphate (GDP)] with GTP insolution. Ras-GTP levels are read out via therapidly reversible binding of a fluorescentlylabeled Ras-binding domain (RBD) (40 nM insolution). The RBD used in this study is amutant[Lys65→Glu (K65E)] derived from Raf-1; it hasfaster binding kinetics than the wild-type RBD(37), providing real-time (~1-s) Ras-GTP densitymeasurements to precisely identify the momentof SOS activation.We resolve the activity of single SOSFL mo-

lecule recruitment events by localizing the re-actions into small corrals separating patches of

fluid membranes (16) (Fig. 1C). This strategyallows precise assignment of the amount andtiming of Ras activation to individual SOS re-cruitment events. In practice, this was achieved bycorralling the supported membranes with 1-mmby 1-mm or 2-mm by 2-mm grid arrays of chem-ically inert chromium barriers, which are pre-fabricated onto the underlying substrate beforemembrane deposition (17). The chromium barriersare typically ~10 nm in height, and although theyare effective barriers to the mixing of membrane-associated components, they pose essentially nobarrier to the flow of molecules in the adjoiningsolution (fig. S1). Both LAT and Ras showedsmall variations (<10%) across different arrays(fig. S1). This membrane microarray strategyalso facilitates simultaneous sampling of large

Huang et al., Science 363, 1098–1103 (2019) 8 March 2019 2 of 6

600

500

400

300

600

500

400

300

).u.a(ytis net nI

SO

S

). u.a (ytisnetnI

SO

S

1 501005 00time (s) time (s)

6000

5000

4000

3000

2000

4000

3000

2000

1000).u.

a(yti

snet

nID

BR

) .u .

a(y ti

s net

nID

BR

6 040200-20

B

microarray supported membranes

D E

activation timerejection time

SOS-Alexa Fluor 555RBD-Alexa Fluor 647LAT-Atto 488C

SOS

Grb2

phos.kinase

GTP GDP RBDErk activation

receptor triggeringpLAT

RasGDPRasGTP

activationrelease ofautoinhibition

rejection time

activation time

p

pp

p

pp

PIP2

p

pp

p

pp

membrane

PIP2 binding site

Ras allosteric site

Ras catalytic site

HF DH PH REM CDC25 PR

1 198 440 605 776 1068 1333

Grb2Ras allosteric sitePIP2

RasGDP RasGTP

SOSFL

SOSFL

SOScatPR

N-terminus autoinhibitionA

Fig. 1. Single-molecule activation assay of SOSFL on supportedmembranes. (A) Domain architecture (left) and a cartoon model (right)of SOSFL. The crystal structure was rendered with Protein Data Bank entry3KSY. HF, histone fold; DH, Dbl homology domain; PH, Pleckstrinhomology domain; REM, Ras exchanger motif. Numbers indicate aminoacid positions. (B) Schematic of the reconstitution of Grb2-mediatedSOS activation. LAT phosphorylated by membrane-bound kinase Hck(pLAT) and Ras preloaded with GDP were decorated on the supportedmembranes corralled by 1-mm by 1-mm or 2-mm by 2-mm chromium grids.

The injection of Grb2, SOSFL-Alexa Flour 555, RBD-K65E–Alexa Fluor647, and GTP into the solution triggered SOS recruitment, subsequentrelease of autoinhibition, and activation. Receptor triggering anddownstream activation (in gray) are shown to illustrate the signalingpathway, and not included in the experiments. phos., phosphorylation.(C) Snapshot images of SOS (green) and RBD (red) recruitmentin microarray supported membranes. Scale bar, 10 mm. t, time.(D) Definition of activation time. (E) Definition of rejection time.[(D) and (E)] Scale bars, 2 mm. a.u., arbitrary units.

RESEARCH | REPORT

numbers (~1000) of nearly identical membranecorrals, all in contact with the same solution.The simultaneous imaging of SOSFL–Alexa

Fluor 555 and RBD-K65E–Alexa Fluor 647 at aframe rate of 0.5 Hz recorded when and whereSOS recruitment and activation occurred (Fig. 1,C to E). Once activated, SOS processively cata-lyzes nucleotide exchange in Ras at a relativelyrapid rate, providing a distinctly discernabletransition between inactive and active states ofSOS (see fluorescence intensity traces in Fig. 1D).Additionally, SOS tends to stay active for anextended period of time (16, 19). Specifically, weparsed out corrals with a single SOS recruit-ment event and then measured the time to theonset of RBD recruitment, indicating the acti-vation of SOS. The SOS recruitment-activationtrajectories can be classified into two types:recruitment of SOS followed by activation (Fig.1D) and recruitment and dissociation of SOSwithout activation (Fig. 1E). The first type oftrajectory reveals the time interval between ini-tial membrane engagement and the release ofSOS autoinhibition, enabling GEF activity, whichwe define as the activation time. The latter tra-jectory indicates the membrane dwell times forSOS molecules that unbind from the membranebefore activating, which we define as the rejec-tion time. The shape of the activation time dis-tribution is especially informative. The existenceof rate-limiting intermediates, through which

SOS must pass en route to activation, will createa distinctive rise-and-fall shape (gammadistribu-tion), whereas an exponential activation timedistribution will indicate that autoinhibitionrelease follows a first-order process. The impli-cation of this kinetic detail is that kinetic proof-readingmechanisms, under which long-dwellingmolecules are disproportionately more likely toactivate, can operate only in the first case, withrate-limiting kinetic intermediates (10, 18).A single SOS recruitment event results in

either activation or rejection. By collecting en-sembles of these single-molecule recruitment-activation trajectories, we compiled histogramsof the activation and rejection time distribu-tions, which also reveal the relative probabilitiesof these two outcomes (Fig. 2, A and B). Theactivation time distribution exhibits the rise-and-fall shape of a gamma distribution, with arather long mean delay of 55 s and a standarddeviation of 44 s (Fig. 2A). The time resolutionof detectable activation in the current experi-ment is <2 s, evident because RBD-K56E bind-ing to Ras-GTP has fast binding kinetics (dwelltime, ~100 ms) and some trajectories show rapidRBD recruitment (<2 s) after SOS binding tomembranes (fig. S3). The rejection time distribu-tion is roughly exponential in shape, with a meanof 30 s and a standard deviation of 29 s (Fig. 2B).The shapes of these distributions reveal under-lying mechanistic features of SOS regulation.

To quantitatively analyze thesemeasured prob-ability distributions, we construct a simple ki-netic model in which the activation of SOS onmembranes is described by the competing kineticsbetween activation and dissociation from mem-brane surfaces (Fig. 2C) (for derivations, see ma-terials and methods) (10, 38, 39). In this model,SOS binds tomembranes initially in an autoinhibi-tion state (SOS0), and the release of autoinhibitionis achieved after overcoming a series of kineticintermediates. Alternatively, SOS may unbindfrom the membrane, going back into solution,where it eventually relaxes back to its native auto-inhibited state (SOS0). Given N number of kineticintermediate(s) in the activation pathway, theactivation time distribution has the form

pactðtÞ ¼ pN ðtÞ � e�k�1t

where pN(t) is the activation time distributionfor a linear multistep process, arising from theconvolution of (N + 1) single Poisson steps withrate constants ki, and k−1 is the dissociationrate constant. The overall probability of SOSactivation, Pact, depends on both the transitionrate (ki) and the unbinding rate (k−1) and is

given by Pact ¼ ∫∞

0pactðtÞdt. A similar expression

defines the overall probability of rejection, Prej, interms of the rejection time distribution, prej(t), andthe integrated sum of these two distributionsnormalizes to one.In a case with a single rate-limiting step or

multiple steps with similar rates, pN(t) approachesa gamma distribution, pN ðtÞ ¼ kNþ1

N tNe�kN t=N !.This approximation provides a useful intuitionabout the shape of the activation time distri-bution: The existence of intermediate(s) predictsa rise-and-fall shape to the distribution (Fig. 2D).In the alternative case, without any kinetic inter-mediates (N = 0), the activation time distributionis strictly exponential (Fig. 2D). The observeddata (Fig. 2A) are consistent with the first case,indicating that SOS activation on membranesinvolves progressing through at least one rate-limiting kinetic intermediate, with a low transi-tion rate (kN) of 0.02 s−1. In practice, the kineticobservation will be dominated by the slowestkinetic step for a unidirectional activationmech-anism (20). Thus, we used N = 1 to analyze thedistributions, i.e.,pactðtÞ ¼ k2N te

�ðkNþk�1Þt. If thismodel is sufficient to describe SOS activation,the fitted values from the activation time dis-tribution should, with no additional fit param-eters, predict the rejection time distribution (Fig.2B), which is given by

prejðtÞ ¼ GðN ; tÞN !

� k�1e�k�1t

where GðN ; tÞ ¼ ∫∞

kN tt0Ne�t ′dt ′, the upper incom-

plete gamma function. In the case of N = 1,prejðtÞ ¼ k�1ðkN t þ 1Þe�ðkNþk�1Þt . The experi-mentally measured rejection time distributionis well described by the prediction based onparameters measured from the activation timedistribution (Fig. 2B). The activation time dis-tribution of SOS is indicative of an activation

Huang et al., Science 363, 1098–1103 (2019) 8 March 2019 3 of 6

A

B15x10

-3

10

5

0

pjer(t

)

2001000rejection time (s)

fitting

prediction

non-receptortriggering

C

D

ptca(t

)

activation time activation time

no intermediate: with intermediate:

4x10-3

3

2

1

0

ptca(t

)

2001000activation time (s)

Pact = 0.35

Activation

-1

activation lag time

-1

rejection time

membranesolution

[Grb2 mediated] [release of autoinhibition]

Fig. 2. Activation time distribution of SOS. (A and B) Histograms of the activation time andrejection time of Grb2-mediated SOSFL recruitment from the single-molecule activation assay. Thesolid line is fitting to the model in (C); the fitted values are kN = 0.02 s−1 and k−1 = 0.016 s−1. Thedashed line represents the prediction by the model using fitted values from the activation timedistribution. (C) A simple model of SOSFL activation. kN denotes the transition rate constantsfor the kinetic intermediates, and k−1 represents the dissociation rate constants from membranes.SOSsoln, SOS in solution. (D) Without a kinetic intermediate (N = 0), the activation time distribution is anexponential distribution peaked at t = 0. In contrast, the existence of at least one intermediate producesthe characteristic rise-and-decay shape for the activation time distribution.

RESEARCH | REPORT

mechanism dominated by a single rate-limitingkinetic intermediate. This model does not ex-haust all possible routes of activation but de-lineates the main pathway. In addition, it ispossible that SOS dissociation from membranesis a multistep process, but it is likely much fasterthan the activation time scale such that the dataare well described by the simple model.The activation time of SOS is determined, in

part, by the release of autoinhibition of thecatalytic domains by the N-terminal domains.We compared Grb2-mediated SOS recruitmentand activation with three other orthogonal acti-vation profiles (Fig. 3 and fig. S2): (i) receptor-independent activation without Grb2 (Fig. 3B),

(ii) activation without N-terminal autoinhibi-tion using the construct SOScatPR (Figs. 1A and3C), and (iii) hindered activation without PIP2(Fig. 3D). Although the recruitment rate of SOSis markedly enhanced by Grb2 (Fig. 3E), thesethree cases exhibit fundamental differences in theactivation time compared with Grb2-dependentactivation (Fig. 3, F to I). Cases (i) and (ii) resultin single-step activation (i.e., the activation timedistribution is exponentially shaped) over theexperimental time scale of 1 to 200 s. WithoutGrb2, spontaneous activation of SOSFL is (rarely)observed as a result of SOS apparently fluctuat-ing into an active configuration before interact-ing with the membrane. In this state, SOSFL

binds to Ras directly and begins to processivelyactivate Ras without the need to pass throughthe kinetic intermediate. This detail is revealedby the exponential shape of the activationtime distribution for these spontaneous acti-vation events. Before successful purification ofSOSFL, these are the only activation events thathave been observed in prior studies (16, 20).The SOScatPR construct lacks N-terminal auto-inhibition and also apparently lacks the kineticintermediate, as revealed by its exponentiallyshaped distribution (Fig. 3, G and H). Case (iii),lacking PIP2 in the membrane, exemplifies thescenario where the activation pathway on themembrane is hindered, as revealed by a decreased

Huang et al., Science 363, 1098–1103 (2019) 8 March 2019 4 of 6

Receptor-independent activation Slower release of autoinhibitionLoss of autoinhibititionvatiot independent activctivepttor-independent acrr independent ac

p

pp

hibitier release of autoinhinhoweer release of autoielease of autoip

pp

bititiLoss of autoinhibhibLoss of autoinhLoss of autoinh

p

pp

SOScatPR: no N-terminus domains

no Grb2

no PIP2

12

8

4

02001000

pjer(t

)/p

tca(t

)

1

10

100

1000

0.1 1 10

30

20

10

02001000

prediction12

8

4

02001000

12

8

4

02001000

J K L Mdata

rate of SO

S act.

time (s) time (s) time (s) time (s)

p

pp

Receptor-mediated activation

F G Hfitting

20010002001000 2001000

15x10-3

10

5

02001000

ptca(t

)

I

activation time (s) activation time (s) activation time (s) activation time (s)

Act.

-1 -1

(slow)Act.

-1

Act.

-1

Act.

-1 -1

A B C D

6

4

2

0

rela

tive

valu

e recruitment rateactivation probabilityactivation rate

0.12 0.340.09

E

Fig. 3. The regulation of autoinhibition defines the activation timingof SOS and its kinetic proofreading capability. The top schematicsshow the experimental design (details are in fig. S2). (A to D) The kineticmodels for Grb2-mediated SOSFL activation, SOSFL activation withoutGrb2, SOScatPR activation, and SOSFL activation without PIP2. Act.,activation. (E) The relative recruitment rate, activation probability, andactivation rate for each condition. Activation rate = recruitment rate �

activation probability. The Grb2-mediated recruitment rate of SOSFL was 3 ×10−3 s−1 mm−2. The dashed line is at 1. (F to I) The activation timedistribution for each condition. Black lines are fitting to the models in (A)to (D). (F) is the same data from Fig. 2A. (J to M) The ratio betweenrejection and activation counts as a function of SOS dwell time. The insetin (J) shows the model’s extrapolation for very short dwelling SOS. Dashedlines are predictions by the models in (A) to (D).

RESEARCH | REPORT

activation rate, yet the kinetic bottleneck is re-tained, as revealed by the rise-and-fall shapeof the activation time distribution (Fig. 3I).Together, these data provide a molecular de-

scription of the autoinhibition release process ofSOSFL: The first recruited state (SOS0) corre-sponds to Grb2-mediated recruitment while underthe protection of autoinhibition. The slowestintermediate (SOS1) involves PIP2-mediated re-lease of autoinhibition (Figs. 2C and 3, A to D).Finally, fully activated SOS presumably involveslocking Ras into the allosteric pocket (16, 23),enabling the processive catalysis.A multistep activation process, such as we have

observed for SOSFL, can, in principle, lead tolonger-dwellingmolecules having disproportion-ally higher activation rates (10)—this is classicallyknown as kinetic proofreading (18). To test forthis behavior experimentally, we examined theratio prej(t)/pact(t) as a function of SOS dwell timefrom the corral experiments. This analysis com-pares the rate (probability per unit of time) of SOSactivation for different dwell times (Fig. 3, J toM).The kinetic intermediate defined by autoinhibitionrelease greatly enhances the proofreading capabil-

ity, as evident from the decaying functional formof the prej(t)/pact(t) curve for SOS

FL (Fig. 3, J andM). By contrast, a weakly discriminating processwill have a near-constant curve, as observed forSOScatPR molecules lacking kinetic intermediates(Fig. 3, K and L). These data provide evidencethat SOSFL activity is capable of being modu-lated by a kinetic proofreading mechanism.With a proofreading mechanism at play, the

rate of Ras activation depends disproportiona-tely on the membrane dwell times of SOS. Giventhat the individual pY-Grb2 interactions exhibitfast kinetics [mean dwell time, ~0.1 s (10)] com-pared with the mean activation time of SOSmeasured in this study (~55 s), we speculate thatmonovalent recruitment of SOS to the mem-brane by Grb2 is unlikely to lead to Ras activa-tion. However, SOS dwell time on themembranecan be greatly elongated when SOS interactswith more than one Grb2 molecule. Single-molecule tracking of SOS shows that the initialstable recruitment of SOS (dwell time > 1 s) ismediated primarily by Grb2 instead of PIP2 orRas (figs. S4 and S5). Furthermore, mobilityanalysis indicates that stably recruited SOS is

interacting with at least two LAT molecules viamultiple Grb2 molecules (fig. S6).Recent studies have vividly demonstrated that

LAT-Grb2-SOS can undergo a phase transition toform a condensed two-dimensional liquid or gelstate consisting of a networked molecular as-sembly on the membrane surface (9–13). A dis-tinctive feature of these condensed phases is thatmultivalent interaction of SOS within the net-work can markedly elongate SOS dwell time onthe membrane (10). To explore the effect of thisprotein assembly network on Ras activation, weadded the PR domain of SOS, which has no in-trinsic catalytic activity but allows formation ofthe assembly, into the corral experiments (Fig. 4, Aand B). If the PR domain acted solely as a com-petitor, titrating it would decrease Ras activationby outcompeting the fixed concentration of SOSFL.Instead, we observed that the rate of Ras activationincreased substantially when the PR domain wasadded (50 nM) (Fig. 4, C to E), suggesting thatLAT assemblies promote Ras activation. Nota-bly, this enhanced activation by the PR domainwas absent at low LAT densities, where assem-blies cannot form; titrating the PR domains to ahigher concentration (≥100 nM in our experiments)also abolished this enhancement by outcompet-ing SOSFL (fig. S7). Furthermore, the twofold in-crease in SOSFL recruitment with the PR domainis insufficient to explain the eightfold increase inRas activation (Fig. 4, C to E).To quantitatively account for the effect of

Grb2-mediated SOS dwell time at the mem-brane on the probability of SOS activation, wefirst evaluated the activation probability of SOS

per recruitment event, Pact ¼ ∫∞

0pactðtÞdt . Pact isintrinsically a function of the Grb2-mediatedSOS dwell time, as longer-dwelling SOS mole-cules have a higher likelihood of activationbefore unbinding. The functional dependenceof Pact on the mean dwell time is plotted in Fig.4F, on the basis of the measured activation timedistribution of SOS (Fig. 2) (N = 1 intermediate).This calculation reveals that Pact is highly sen-sitive to the SOS dwell time when the meanactivation time is substantially longer than themean dwell time, which is the case in ourreconstitution and likely in live cells as well (14).Next, we evaluated the fold increase in SOS ac-tivity per molecule, xSOS (the definition is pro-vided in materials and methods), as a functionof change in the mean dwell time. Figure 4Fenables a direct comparison of SOS activationprobabilities in the unassembled and assem-bled LAT configurations in Fig. 4, D and E.Single-molecule analysis reveals that the frac-tion of long-dwelling species of SOS increaseswith the addition of the PR domain (and con-sequent LAT assembly) in this experiment (fig.S8), although the degree ofmolecular assembly isnot as large as that observed in a fully condensedphase (10). The roughly fourfold net increase inRas activation per SOS molecule (after accountingfor recruitment) is comparable to the estimationon the basis of mean dwell time changes (from~4 to 6 s). Thus, relatively small changes in the

Huang et al., Science 363, 1098–1103 (2019) 8 March 2019 5 of 6

p

pp

p

pp

p

pp

A C

D

SOS

Grb2p

pp

p

pp

pLAT

PR

p

pp

p

pp

Ras

GTP

SOS+Grb2

Grb2+PR

SOS+Grb2+PR

membrane

solution

E

10-4

10-3

10-2

pdf

t = 0

turnover by SOS

RBD-Alexa Fluor 647

60000RasGTP (a.u.)

40002000

t = 15 min

B

F

for a fixed LAT density:[PR]

intrinsic turnover

LAT

1%

7%

64%

10-4

10-3

10-2

pdf

10-4

10-3

10-2

pdf

0.1

1

10

100

1000

0.1 1 10 100Δ mean dwell time (s)

basal = 5 sbasal = 1 s

basal = 0.2 s

0.01

0.1

1

Pac

t

200150100500mean dwell time (s)

mea

n ac

t. tim

eac

t

Fig. 4. Molecular assembly enhances Rasactivation by SOS. (A and B) LAT assembliesare tuned by the addition of Grb2 and thePR domain of SOS, which enables multivalentcross-linking but has no intrinsic catalyticactivity. Images represent LAT–Alexa Flour 555after the addition of Grb2 and the PR domain.(C to E) Snapshot images of RBD after15 min of reaction time. The histograms showthe RBD intensities per corral of the images.The dashed lines are guidelines for themaximum extent of intrinsic nucleotideturnover. The percentages show the numberof corrals with turnover above the level ofthe dashed line. Scale bar, 10 mm. pdf, proba-bility density function. (F) Calculation of theprobability of SOS activation as a functionof Grb2-mediated mean dwell time (left) andthe increase in SOS activity as a function ofthe change in mean dwell time (right). Thechange in mean dwell time is defined asDhti ¼ htiactivated � htibasal.

RESEARCH | REPORT

SOS dwell time lead to large changes in the prob-ability of activation because the system is oper-ating under conditions in which the overallprobability of activation is low (e.g., themean dwelltime is substantially shorter than the mean ac-tivation time). Monovalent recruitment interac-tions between SOS and phosphorylated LAT viaGrb2 are brief (<1 s) (10) and will thus have anexceptionally low probability of successful SOSactivation. In the fully condensed state, the LAT-Grb2-SOS assembly can elongate SOS dwell timesbymore than an order ofmagnitude (10), which iseven more than that achieved in the PR domainexperiments presented here. Collectively, thesedata suggest that LAT-Grb2-SOS assembly canexert a powerful influence over SOS activation bymodulating the SOS dwell time at the membrane.In resting T cells, LAT typically exhibits some

basal level of phosphorylation (40). A proofread-ingmechanism in SOS activation can prevent thespontaneousmembrane localization of SOS fromactivating Ras and thus prevent upstream basalpY activity from activating certain downstreampathways (fig. S9). This kinetic proofreading atthe level of SOS is different from kinetic proof-reading at the TCR, which has been proposed tocontribute to antigen discrimination on the basisof peptide major histocompatibility complex–TCR binding kinetics (41).The type of kinetic proofreading identified here

in the activation of Ras by SOS reveals a poten-tially broad mechanism by which membranerecruitment kinetics can be used to modulatedownstream signaling activity. Many membrane-associated signaling molecules activate in a multi-step fashion—for example,Vav (GEF forRho familyGTPase) (42), Raf (43), class I protein kinase C-a(44), phosphoinositide 3-kinases (45), and indu-cible T cell kinase (46). The initiation of actinnucleation by the Arp2/3 complex is also a slowmultistep process. Notably, the actin regulatorneuron Wiskott-Aldrich syndrome protein (N-WASP) and the Arp2/3 complex have recentlybeen shown to exhibit dwell time–dependent ac-tivitymodulation by nephrin–Nck–N-WASP con-

densation phase transition, in a mechanism highlyparallel to that described here for Ras and SOS(15, 47). Hence, we speculate that the timing andproofreading mechanism observed in this studyis not exclusive to SOS but is a common theme insignal regulation at the membrane.

REFERENCES AND NOTES

1. L. E. Samelson, Annu. Rev. Immunol. 20, 371–394 (2002).2. J. Schlessinger, Cell 103, 211–225 (2000).3. H. Sondermann et al., Cell 119, 393–405 (2004).4. N. W. Gale, S. Kaplan, E. J. Lowenstein, J. Schlessinger,

D. Bar-Sagi, Nature 363, 88–92 (1993).5. P. Bandaru, Y. Kondo, J. Kuriyan, Cold Spring Harb. Perspect.

Med. 9, a031534 (2019).6. T. Shi et al., Sci. Signal. 9, rs6 (2016).7. J. C. D. Houtman et al., Nat. Struct. Mol. Biol. 13, 798–805 (2006).8. R. L. Kortum et al., Sci. Signal. 6, ra99 (2013).9. A. Nag, M. I. Monine, J. R. Faeder, B. Goldstein, Biophys. J. 96,

2604–2623 (2009).10. W. Y. C. Huang et al., Proc. Natl. Acad. Sci. U.S.A. 113,

8218–8223 (2016).11. X. Su et al., Science 352, 595–599 (2016).12. W. Y. C. Huang, H. K. Chiang, J. T. Groves, Biophys. J. 113,

1807–1813 (2017).13. W. Y. C. Huang, J. A. Ditlev, H. K. Chiang, M. K. Rosen,

J. T. Groves, J. Am. Chem. Soc. 139, 18009–18015 (2017).14. B. J. Mayer, J. Yu, J. Mol. Biol. 430, 4547–4556 (2018).15. S. Banjade, M. K. Rosen, eLife 3, e04123 (2014).16. L. Iversen et al., Science 345, 50–54 (2014).17. J. T. Groves, N. Ulman, S. G. Boxer, Science 275, 651–653 (1997).18. J. J. Hopfield, Proc. Natl. Acad. Sci. U.S.A. 71, 4135–4139 (1974).19. Y. K. Lee et al., Nat. Commun. 8, 15061 (2017).20. S. M. Christensen et al., Nat. Struct. Mol. Biol. 23, 838–846

(2016).21. J. Gureasko et al., Proc. Natl. Acad. Sci. U.S.A. 107, 3430–3435

(2010).22. J. T. Groves, J. Kuriyan, Nat. Struct. Mol. Biol. 17, 659–665

(2010).23. J. Gureasko et al., Nat. Struct. Mol. Biol. 15, 452–461 (2008).24. K. K. Yadav, D. Bar-Sagi, Proc. Natl. Acad. Sci. U.S.A. 107,

3436–3440 (2010).25. M. Tartaglia et al., Nat. Genet. 39, 75–79 (2007).26. J. C. D. Houtman et al., Biochemistry 43, 4170–4178 (2004).27. C. Zhao, G. Du, K. Skowronek, M. A. Frohman, D. Bar-Sagi, Nat.

Cell Biol. 9, 707–712 (2007).28. S. M. Margarit et al., Cell 112, 685–695 (2003).29. J. A. Nye, J. T. Groves, Langmuir 24, 4145–4149 (2008).30. N. H. Shah et al., eLife 5, e20105 (2016).31. W. C. Lin et al., Proc. Natl. Acad. Sci. U.S.A. 111, 2996–3001

(2014).32. D. J. Williamson et al., Nat. Immunol. 12, 655–662 (2011).33. T. Gurry, O. Kahramanoğullari, R. G. Endres, PLOS ONE 4,

e6148 (2009).

34. K. J. Cho, J. F. Hancock, Small GTPases 4, 57–60 (2013).35. A. J. Stevens et al., J. Am. Chem. Soc. 138, 2162–2165

(2016).36. C. B. McDonald et al., Biochemistry 51, 2122–2135 (2012).37. B. K. Jaitner et al., J. Biol. Chem. 272, 29927–29933 (1997).38. J. R. Moffitt, Y. R. Chemla, C. Bustamante, Methods Enzymol.

475, 221–257 (2010).39. S. C. Kou, B. J. Cherayil, W. Min, B. P. English, X. S. Xie,

J. Phys. Chem. B 109, 19068–19081 (2005).40. J. Lin, A. Weiss, J. Biol. Chem. 276, 29588–29595 (2001).41. T. W. McKeithan, Proc. Natl. Acad. Sci. U.S.A. 92, 5042–5046

(1995).42. M. Turner, D. D. Billadeau, Nat. Rev. Immunol. 2, 476–486 (2002).43. R. E. Cutler Jr., R. M. Stephens, M. R. Saracino, D. K. Morrison,

Proc. Natl. Acad. Sci. U.S.A. 95, 9214–9219 (1998).44. B. P. Ziemba et al., Biochemistry 53, 1697–1713 (2014).45. O. Vadas, J. E. Burke, X. Zhang, A. Berndt, R. L. Williams,

Sci. Signal. 4, re2 (2011).46. A. H. Andreotti, P. L. Schwartzberg, R. E. Joseph, L. J. Berg,

Cold Spring Harb. Perspect. Biol. 2, a002287 (2010).47. L. B. Case, X. Zhang, J. A. Ditlev, M. K. Rosen, Science 363,

1093–1097 (2019).

ACKNOWLEDGMENTS

We thank M. Rosen and L. Case for their insightful commentsand for communicating results before publication. We thankH.-K. Chiang for assistance in the pilot experiments. We thankS. Hansen, M. Triplet, and other members in the Groves laband Kuriyan lab for helpful discussion. Funding: This workwas supported by National Institutes of Health (NIH) NationalCancer Institute (NCI) Physical Sciences in Oncology Network(PS-ON) project 1-U01CA202241 and by the Novo NordiskFoundation Challenge Program under the Center forGeometrically Engineered Cellular Systems. Additional supportwas provided by NIH grant PO1 A1091580. Authorcontributions: W.Y.C.H. and J.T.G. conceived the study.W.Y.C.H. designed the experiments. S.A., Y.K., H.Y.M.L., andK.H.B. prepared reagents. W.Y.C.H., Y.K.L., and J.K.C. performedthe experiments. W.Y.C.H. and S.A. analyzed data. W.Y.C.H.and J.T.G. wrote the manuscript. J.K. and J.T.G. supervisedthe project. All authors commented on the manuscript. Competinginterests: None declared. Data and materials availability: Alldata necessary to evaluate the conclusions in the paper are availablein the paper or the supplementary materials.

SUPPLEMENTARY MATERIALS

www.sciencemag.org/content/363/6431/1098/suppl/DC1Materials and MethodsSupplementary TextFigs. S1 to S9References (48–54)Movie S1

29 June 2018; accepted 10 January 201910.1126/science.aau5721

Huang et al., Science 363, 1098–1103 (2019) 8 March 2019 6 of 6

RESEARCH | REPORT