First-passage time to clear the way for receptor-ligand binding in a crowded environment
FFirst-passage time to clear the way forreceptor-ligand binding in a crowdedenvironment
Jay Newby ∗ Jun Allard † March 8, 2016
Abstract
Certain biological reactions, such as receptor-ligand binding at cell-cell interfaces and macromolecules binding to biopolymers, require manysmaller molecules crowding a reaction site to be cleared. Examples includethe T cell interface, a key player in immunological information processing.Diffusion sets a limit for such cavitation to occur spontaneously, therebydefining a timescale below which active mechanisms must take over. Weconsider N independent diffusing particles in a closed domain, contain-ing a sub-region with N particles, on average. We investigate the timeuntil the sub-region is empty, allowing a subsequent reaction to proceed.The first passage time is computed using an efficient exact simulationalgorithm and an asymptotic approximation in the limit that cavitationis rare. In this limit, we find that the mean first passage time is sub-exponential, T ∝ e N /N . For the case of T cell receptors, we find thatstochastic cavitation is exceedingly slow, seconds at physiological den-sities, however can be accelerated to occur within 5 second with only afour-fold dilution. Diffusion drives many biological processes, both positively, by deliveringcargo to a target, and negatively, by removal of cargo from a region of in-terest (ROI). While the temporal dynamics of diffusional delivery have beenextensively studied [5, 4, 24, 12], diffusion-driven removal has been less charac-terized experimentally or theoretically [3]. Removal is of particular interest inthe crowded environment of cells, where large biomolecules and cellular struc-tures require the displacement of smaller molecules, a phenomenon we termstochastic cavitation.A specific example arises in the study of cell-cell interfaces including theT-cell/antigen-presenting-cell interface [22, 2, 29, 8] (see Fig. 1). A fundamen-tal question for all cell-cell interfaces is how receptors and ligands come into ∗ Department of Mathematics, University of North Carolina, Chapel Hill, 329 Phillips Hall,Chapel Hill, NC 27599 † Department of Mathematics, University of California, Irvine, 340 Rowland Hall, Irvine,CA 92697 a r X i v : . [ q - b i o . S C ] M a r ontact, despite being separated by large molecules, the extracellular fluid, andother structures in the glycocalyx. On either cell surface, large molecules suchas CD45 and LFA-1 undergo 2D diffusion in the cell membrane with a diffusioncoefficient of D ∼ . µ m /s [27, 6]. These large molecules impair interactionsbetween smaller pairs of molecules, such as the T cell receptor and its ligand—akey step in immunological information processing and decision-making. It hasbeen estimated that a region of radius ∼ , devoid of large molecules, isnecessary for spontaneous T cell receptor interaction [2], which is occupied byon average ∼ particles at equilibrium. A natural question is whether thisempty region can form spontaneously in a biologically relevant time. Under-standing contact formation will address cell-cell interactions in the crowded,heterogeneous environment inside organisms and what machinery is necessaryfor general cell-cell interactions [15, 31].Examples of diffusional cavitation in biology also arise in other dimension-alities. In 1D, microtubules (inflexible polymers of the protein tubulin) aredecorated by hundreds of microtubule-associated proteins [1, 28]. These pro-teins exhibit significant crowding [11] and lateral diffusion along the microtubulelattice [13, 17]. Large microtubule-binding molecules may therefore have to waitfor a region to be clear before binding. What is the mean time for such clearance,and is it the rate-limiting step in microtubule binding? A similar situation oc-curs for DNA and the myriad of DNA-binding molecules, some of which undergolateral diffusion across base pairs [16, 25]. A significant waiting-time for largeDNA-binding molecules has potential implications for the study of the chemicalmodification of DNA and RNA, all of which require an enzyme to attach to thepolymer.For some of the above scenarios, it has been hypothesized that clearance ofthe target region requires an active process [2, 18]. To address the feasibilityof passive diffusion-driven cavitation, a theoretical assessment of the timescalesinvolved is needed. In other words, can diffusion-driven cavitation reliably oc-cur on biologically relevant timescales? To address this question, we consider N independent particles undergoing simple diffusion in either the 1D domainFigure 1: Cell-cell interface formation between a T-cell and an antigen-presenting cell. CD45 molecules (black) block the receptor-ligand (purple) bondfrom forming while they inhabit the ROI (orange).2 − L, L ) or the 2D domain ( − L, L ) . In each case, we study the first-passagetime until a smaller region, a disk of radius L , is empty.The cavitation event can be rare (i.e., the first passage time can be verylarge compared to the diffusion timescale L /D ) under certain circumstances.To understand this, consider the 1D domain with N particles. At equilibrium,each particle has a uniformly distributed postion within the domain. In thelimit L → ∞ and N → ∞ with the average particle density ϕ = N/ (2 L ) fixed, the equilibrium probability of finding a region of radius L containing noparticles is small, P = e − N , where N = 2 Lϕ . Therefore, when N (cid:29) weexpect cavitation to be a rare event. Although we might expect the 1D meanfirst-passage time (MFPT) to scale as T ∝ P − ∼ e N , we instead we find anasymptotic scaling of T ∝ e N /N .In this Letter, we develop a simulation algorithm to efficiently generate ex-act realizations of the first passage time, based on Green’s function reactiondynamics [26]. For situations where cavitation is a rare event and computa-tion becomes unfeasible, i.e., when N is very large or the ROI occupies most ofthe explorable area, we develop an asymptotic approximation of the mean firstpassage time.Consider N independent random walkers Y n ( t ) , with n = 1 , · · · , N , that areconfined to the interval − L < y < L . The ROI is the inner domain centered atthe origin with radius L < L . The event we wish to characterize is the first timeat which the ROI is empty (i.e., min n { Y n ( t ) } = L ). We first nondimensionalizethe problem using the space scale L and the time scale L /D , where D is thediffusion coefficient. We define the nondimensional distances R n = | Y n | /L , (cid:15) = 1 − L /L , and l = L /L . Then, a given particle is inside the ROI if < R n < l .By formulating a simulation algorithm, we can generate exact samples of thefirst passage time. We take advantage of explicit formulas for the probability dis-tributions that govern single particle Brownian motion in a closed domain. Notethat even though we focus on the 1D and 2D cavitation problem in this letter,exact distributions are also known for 3D Brownian motion [7]. The algorithmproceeds as follows. Given a set of random starting positions { R n ( t ) } ≤ n ≤ N ,select a particle that is inside the ROI and closest to the origin. That is, select R m = min { R n } < l . The first step is to compute the first time τ at whichthe selected particle leaves the ROI (i.e., R m ( t + τ ) = l ). Once τ has beencomputed, set t (cid:48) = t + τ . The cavitation event cannot have occurred beforetime t (cid:48) because we are certain that R m ( t ) < l for all t < t < t (cid:48) . Therefore,the position of the other particles between time t and time t (cid:48) is irrelevant, weneed only generate the random position for each of the remaining particles attime t (cid:48) . Once all positions have been updated, select a new R m = min { R n ( t (cid:48) ) } .We know that the cavitation event has occurred if R m ≥ l . If R m < l , then This problem is equivalent to the diffusion of a particle in ND dimensions (the productof number of particles and dimensionality of space). Since ND (cid:29) , this Brownian motionis not recurrent, so we might naively expect the system to be well-mixed in ND -dimensionalphase space, and the rate of first passage would be the attempt rate times the probability ofbeing in the target state. t = t (cid:48) and repeat the above procedure.At each step, the jump times τ can be sampled from the exact distribution f ( τ | r ) , obtained from the fundamental solution to the diffusion equation witha reflecting boundary at r = 0 and an absorbing boundary at r = l . Therandom positions can be sampled from the distribution p ( r | r , τ ) , satisfyingthe diffusion equation with reflecting boundaries at r = 0 and r = 1 . Anefficient way of sampling from p is to use a rejection method, similar to theone described in [26]. For the jump time sampled from f , we found that therejection method could not easily be adapted to our situation. Instead, wesample the jump time using a root finding algorithm. Additional details areprovided in Supplementary Material. The simulation algorithm is maximallyfast in the sense that only the (average) slowest particle determines the nextevent time, allowing us to efficiently access densities around φL ∼ . We findthis is sufficiently high to validate our asymptotic approximations.To obtain a complete picture of cavitation in the rare event limits, we developan asymptotic approximation for the MFPT, ¯ T . The approximation is derivedfor 1D cavitation, and based on simulations, we observe that in the limit L → ∞ with a fixed particle density, the approximation is also surprisingly accurate for2D cavitation. We first state the main results (Eqs. 46-37) and then summarizetheir derivation.For fixed N , the first term in the asymptotic approximation for < (cid:15) (cid:28) of the MFPT, averaged over a uniformly distributed initial position for each ofthe N particles, is given by ¯ T ∼ N A N ( C N (cid:15) ) N − + O (1) , N ≥ (1)where A N = Γ( N )2 π N ( N − . (2)The constant C N is the Newtonian capacitance of a hypercube in R N ; as ex-plained below, it determines the far field behavior of certain solutions to Laplace’sequation [32]. An explicit formula for the Newtonian capacitance of a cube for N > is unknown. However, a good approximation for N = 3 is C ≈ . [19]. The (cid:15) (cid:28) approximation (solid line) is compared to simulations (symbols)in Fig. 2. For N = 3 , we find good agreement between simulation and the inde-pendently derived estimate for C from [19]. From physical arguments detailedat the end of this letter, we have determined an expansion of the Newtoniancapacitance for large N given by C N ∼ (cid:114) Nπe (cid:18) N N + α N + O ( N − ) (cid:19) . (3)The unknown constant in the above expansion is independent of all parame-ters. Using the exact simulation algorithm, we obtain the numerical estimate, α ≈ − . . Our MFPT calculation thus provides an approximation for thecapacitance C N , which otherwise remains challenging to compute [19].4 .5 0.6 0.7 0.8 0.9 1.0 L /L -1 M F P T ¯ T N =5 N =4 N =3 Figure 2: The MFPT (in nondimensional units) as a function of L /L = 1 − (cid:15) .The small (cid:15) approximation (solid lines) is compared to simulations (symbols),using C = 1 . , C = 1 . , and C = 1 . .For fixed < (cid:15) < , an asymptotic expansion for N (cid:29) is given by ¯ T ∼ κ N (cid:15) N − , N (cid:29) , (4)where κ ≈ . depends only on α (via Eq. 19). The N (cid:29) MFPT approxi-mation is compared to simulations in Fig. 3. N -2 -1 M F P T ¯ T L /L =0 . L /L =0 . L /L =0 . Figure 3: The large N MFPT approximation (solid lines) compared to simula-tions (symbols).Finally, we consider the case the radius of the ROI L is fixed and L → ∞ with a fixed number of particles per unit length ϕ = N/ (2 L ) . Let N be theaverage number of particles in the ROI. The L → ∞ MFPT approximation (indimensional units) is T ∞ ∼ κ L e N N D , N (cid:29) . (5)5he MFPT is shown in Fig. 4 as functions of N , for different values of L .The approximation (37) matches closely with the L/L = 33 . simulations for N M F P T ( L / L ) ¯ T L/L =4 . L/L =8 . L/L =25 L/L =33 . L/L = ∞ Figure 4: The MFPT vs N , the average number of particles in the ROI. Thesymbols indicate averaged simulations; the 1D simulations are shown as cir-cles and the 2D simulations are shown as diamonds. Also shown is the L → ∞ approximation (dashed curve) for both 1D and 2D. Note that time is nondimen-sional using the L /D timescale. N > . Hence, the effect of a small domain size compared to the ROI isto increase the MFPT, making the cavitation event more rare. This providesa quantitative measure of when the domain size L no longer influences thecavitation event, which is relevant when, for example, considering cavitation ona relatively long strand of DNA compared to a shorter plasmid. Microtubulefilaments also vary in length.Simulations of 2D cavitation are also shown in Fig. 4 as diamond symbols.Although the asymptotic approximation (37) is derived for 1D, we find that it isa remarkably good fit to the simulation data after changing a single parameter:the prefactor κ ≈ . . We therefore infer that cavitation is roughly threetimes faster in 2D than in 1D. Heuristically, this speed-up occurs because themean time for a random walker to escape a spherical region decreases withdimensionality.The asymptotic approximations (46)-(37) for 1D cavitation are derived asfollows. Because all of the N walkers are independent, the problem can bereformulated as the first passage time of a single random walker in a N di-mensional domain. Define the domain Ω ≡ (0 , N , and let Ω (cid:15) ≡ (1 − (cid:15), N be the small target domain. The random process R ( t ) ∈ Ω \ Ω (cid:15) representsthe original process with R ( t ) = ( R ( t ) , · · · , R N ( t )) . Define the MFPT as6 ≡ (cid:104) inf { t > ≤ n ≤ N R n ( t ) = l }(cid:105) ; it satisfies N (cid:88) n =1 ∂ T∂r n = − , r ∈ Ω , (6) ∂ η T ( r ) = 0 , r ∈ ∂ Ω , (7) T ( r ) = 0 , r ∈ ∂ Ω (cid:15) . (8)An approximate solution to (6) can be obtained using the method of matchedasymptotics [33, 10, 30, 9, 20]. We split the solution into two parts: an inner andouter solution. The inner solution satisfies the absorbing boundary conditionon ∂ Ω (cid:15) and ignores the reflecting boundary. The outer solution satisfies thereflecting boundary on ∂ Ω and is singular as r → (1 , · · · , . The two solutionsare then matched to obtain a uniformly accurate approximation using the Van–Dyke matching principle [23].Define the inner coordinates z = r − r b (cid:15) , and let z = (cid:107) z (cid:107) . The inner solutionsatisfies ∆ z w = 0 , w ( z ∈ ∂ Z N ) = 0 , (9)where Z N is the unit hypercube. The exact solution to the inner problem forarbitrary N is unknown. However, from electrostatics [21], for large z , the innersolution has the two term expansion, w ∼ B N ( (cid:15) ) (cid:104) ( z/C N ) − N − (cid:105) , N ≥ . (10)where B N is a constant determined by matching to the outer solution. Theconstant C N , called the Newtonian capacitance, is a boundary dependent termdiscussed below.Up to an unknown constant ˜ T , the outer solution is T out ∼ − G N ( r , r b ) + ˜ T , (11)where the Green’s function G N satisfies, N (cid:88) n =1 ∂ G N ∂r n = 1 − δ ( r − r (cid:48) ) , r ∈ Ω , (12) ∂ η G N ( r , r (cid:48) ) = 0 , r ∈ ∂ Ω , (13) (cid:90) Ω G N ( r , r (cid:48) ) d r = 0 . (14)By integrating (11) over Ω using (36), we find that ˜ T is the MFPT averagedover a uniformly distributed set of initial positions, i.e., ˜ T = ¯ T . Again fromelectrostatics [21], in the limit r n → with r (cid:48) n = 1 and (cid:107) r − r (cid:48) (cid:107) = (cid:15)z , theGreen’s function scales like G N ∼ N A N ( (cid:15)z ) − N + O (1) , N ≥ , (15)7here A N is given by (45).Matching the inner and outer solutions we find that the z dependent termsmatch provided that B N ( (cid:15) ) = − N A N (cid:15) − N for N ≥ . The remaining unknownterm ¯ T yields the approximation (46).In order to access the rare event limit where both N (cid:29) and (cid:15) (cid:28) , wemust find how the Newtonian capacitance C N scales with N . This problem hasno known exact solution for N > [19].If the cuboid boundary ∂ Ω (cid:15) were replaced by a spheroid with the samehypervolume, then the Newtonian capacitance is known for general N , C N ≈ √ π Γ (cid:18) N (cid:19) /N ∼ (cid:114) Nπe . (16)We therefore propose a general expansion of C N (for the present case of cuboidboundary) having the same form as the large- N expansion of (16), C N ∼ (cid:114) Nπe (cid:18) α log NN + α N + O ( N − ) (cid:19) . (17)Note that (16) and (17) have the same leading-order term.To elucidate how the unknown constants α , affect the large N MFPTapproximation, we use Stirling’s formula, leading to N A N C N − N ∼ κ N β , N (cid:29) , (18)where β = α + 1 / , κ = 2 √ πe α +1 . (19)In dimensional units, the MFPT approximation is T ∼ L κ N β D (cid:18) − L L (cid:19) − N . (20)We determine the value of α by exploiting a physical constraint as follows. As L → ∞ with the density of particles ϕ = N/ (2 L ) held constant, the MFPTmust converge to a finite value. Substituting L = N/ (2 ϕ ) and N = 2 L ϕ into(20) yields T ∼ N − β κ ϕ D (cid:18) − N N (cid:19) − N . (21)Since lim N →∞ (cid:0) − N N (cid:1) − N = e N , we must have that β = 2 (and therefore α = 3 / ) in order for (44) to converge to a finite, nonzero value in the limit ( L, N ) → ∞ . We also find that β = 2 is supported by numerical simulations (seeSupplementary Material). The limiting result is the approximation Eq. (37).While the approximation matches well with simulations in 2D, a more sys-tematic asymptotic analysis for the 2D case should be feasible. For small (cid:15) and8nite L , the leading order in (46) holds in 2D. A notable feature of our 1D caseis that there are no terms in the expansion between the leading order term andthe O (1) term, making our 1D approximation converge particularly fast. Thisfeature is lacking in 2D, where there are other terms singular in (cid:15) , therefore weexpect this approximation to converge more slowly. In 2D for large L at con-stant density, a different scaling between N and L prevents the approximationin (44) from converging, necessitating an alternative strategy that will be thesubject of future research.Returning to the specific question of cell-cell contact at T cell interfaces,large diffusing molecules such as CD45 disfavor proximity between receptorsand ligands on apposing cells. These molecules have diffusion coefficients of D ≈ . µ m /s [27] and density such that on average there are N = 30 moleculesin the 100-nanometer ROI [2]. The approximation (37), using the prefactor κ = 0 . from the numerical fit to simulations, yields an estimate of T ≈ seconds. In contrast, the MFPT for a single particle to escape a circulardomain is T = L / (4 D ) = 0 . seconds. Since T cell receptor triggeringoccurs within seconds [14], the above calculation predicts that receptor-ligandbinding must involve a mechanism faster than passive diffusion. We thereforesuggest the alternative hypothesis that an active force drives receptor-ligandproximity [2]. To obtain an empty ROI spontaneously in less than five seconds,we would require N ≤ , corresponding to a four-fold dilution, which could beexperimentally accessible. The biological system is complicated by interactionsof large molecules within and between molecular species, lipid heterogeneity,and transient immobilization, all of which could be exploited to dynamicallytune the rate of ligand binding and will be studied by expanding the presentframework. JN was supported by a NSF-funded postdoctoral fellowship (NSF DMS-1100281,DMS-1462992). JA was supported by a NSF CAREER award (DMS-1454739).
A Simulation algorithm
The exact simulation algorithm makes use of two solutions to the 1D diffusionequation. Let p a,r ( x, x , t ) be solutions to ∂p∂t = ∂ p∂x , < x, x < x a,r (22) ∂p∂x = 0 , x = 0 (23) p ( x, x ,
0) = δ ( x − x ) , (24)with two different right boundary conditions. Let p a be the solution with anabsorbing BC at x a = l = 1 − (cid:15) . This solution is used to derive f , the jump time9istribution. Let p r be the solution with a reflecting BC at x r = 1 . This solutionis used to generate the random position of each particle given a jump time. Inboth cases, the solution is represented as an infinite series. Two different seriesrepresentations are derived for each solution: one that converges quickly forshort times and one for long times.Let a n = πl ( n − / , b n = nπ. (25)For large times, we have p a ( x, x , t ) = 2 l ∞ (cid:88) n =1 cos( a n x ) cos( a n x ) e − a n t , (26)and p r ( x | x , t ) = 1 + 2 ∞ (cid:88) n =1 cos( b n x ) cos( b n x ) e − b n t . (27)For short times we have p a ( x, x , t ) = 1 √ πt ∞ (cid:88) n =0 ( − n (cid:18) e − (2 l n +( x + x t − e − (2 l n +1) − ( x + x t + e − (2 l n + | x − x | )24 t − e − (2 l n +1) −| x − x | )24 t (cid:19) (28)and p r ( x, x , t ) = 1 √ πt ∞ (cid:88) n =0 (cid:18) e − (2 n +( x + x t + e − (2( n +1) − ( x + x t + e − (2 n + | x − x | )24 t + e − (2( n +1) −| x − x | )24 t (cid:19) (29)For short times, the first passage time density is f ( t | x ) = − ∂∂x p a ( l , x , t )= 4 π (4 πt ) / ∞ (cid:88) n =0 ( − n (cid:26) ( l (2 n + 1) + x ) e − ( l n +1)+ x t +( l (2 n + 1) − x ) e − ( l n +1) − x t (cid:27) (30)with the cumulative distribution, F ( t | x ) = 1 + ∞ (cid:88) n =0 ( − n (cid:26) erf( l (2 n + 1) + x √ t ) + erf( l (2 n + 1) − x √ t ) (cid:27) (31)10or long times, the first passage time density and cumulative distribution are f ( t | x ) = − l ∞ (cid:88) n =1 ( − n a n cos( a n x ) e − a n t , (32) F ( t | x ) = 1 + 2 l ∞ (cid:88) n =1 ( − n cos( a n x ) e − a n t a n . (33)The jump time is sampled using a standard root finding algorithm. Given auniform random variable U , the jump time is the unique solution to F ( t | x ) − U = 0 . (34)The distribution p r can be sampled using a rejection method as follows. Amajoring function C ( x ) must be chosen such that C ( x ) > p ( x | x , t ) for all x ∈ (0 , . A tentative value X is sampled from the distribution P ( x ) = C ( x ) (cid:82) C ( x ) dx . (35)A second random variable is drawn according to Y = C ( X ) U , where U is a unituniform random variable. If Y > p r ( X | x , t ) , then the sample X is rejected.The procedure is repeated until a sample is accepted.For the long time expansion (27) we select X to be a uniform random variablein (0 , and set Y = (cid:32) e − π t − e − π t (cid:33) U. (36)For the short time expansion (29) we select X to be a normal random variablewith mean x and variance √ t . Note that care must be taken to ensure that < X < . In this case, Y = e − ( X − x ) / (4 t ) U √ πt . (37) A.1 2D simulations
For 2D cavitation, the outer boundary is a square of side length L . Thisgeometry allows us to reuse the jump propagator from the 1D algorithm toupdate positions. The x and y coordinate of each particle are updated fromseparate samples of the 1D propagator p r as described in the previous section.The jump times are generated from the 2D distribution of first passage timesto the boundary of a circle.The 2D first passage time problem is ∂∂t p ( r, t | r ) = 1 r ∂∂r (cid:18) r ∂p∂r (cid:19) , (38) p ( r, | r ) = δ ( r − r )2 πr , (39) p ( l , t | r ) = 0 . (40)11he solution can be written as an expansion in Bessel functions. The solutionis p ( r, t | r ) = 2 l ∞ (cid:88) j =1 J ( rβ n ) J ( r β n ) J ( l β n ) e − β n t , (41)where α n are the roots of J ( α n ) = 0 , and β n = α n /l . The jump time densityfunction is f ( t | r ) = − ∂∂r p ( l, t | r ) = 2 l ∞ (cid:88) j =1 β n J ( r β n ) J ( l β n ) e − β n t , (42)and the cumulative distribution is F ( t | r ) = 1 − l ∞ (cid:88) j =1 J ( r β n ) β n J ( l β n ) e − β n t . (43)We use a root finding method to sample the jump time. B Parameter estimation
We use maximum likelihood to estimate parameter values in the large N expan-sion of the Newtonian capacitance of a hypercube. We exploit the one to onecorrespondance between α and β and between α and κ . The parameters β and κ are estimated using realizations of the first passage time. The likelihoodfunction is computed by assuming that the first passage time is an exponentiallydistributed random variable with mean ¯ T ∼ κN β (cid:15) N − . (44)This assumption is valid asymptotically as N → ∞ when the first passage timeis a rare event. The likelihood function for β and κ from n iid samples { τ k } , k = 1 , · · · , n is given by P ( { τ k } | β, κ ) = exp (cid:20) − n (cid:18) T n ¯ T β,κ + log ¯ T β,κ (cid:19)(cid:21) , (45)where T n = 1 n n (cid:88) k =1 τ k . (46)Two data sets were generated for N = 20 , , , · · · , , with (cid:15) = 10 − /N and (cid:15) = 10 − /N . A value of T n was generated for each parameter set using samples of the first passage time from the exact simulation algorithm. Wenumerically computed the maximum of the product of the likelihood functionsfrom all parameter values. The likelihood functions were computed on a × grid for . < β < . and . < κ < . The resulting maximizers were12 .00 0.02 0.04 0.06 0.08 0.10 /N T † =10 − /N † =10 − /N /N -5 -4 -3 -2 -1 r e l a t i v e e rr o r Figure 5: The MFPT approximation compared to Monte Carlo simulation esti-mates. Each symbol shows the sample mean of simulations. β ≈ . and κ ≈ . . As shown in Fig. 5, ¯ T and T n are in good agreement withthese parameter values. The corresponding parameter values in the capacitanceexpansion are α ≈ / and α ≈ − . . The capacitance approximation isshown in Fig. 6 compared to numerical estimates. From the expansion, weexpect the error (given by the absolute difference divided by √ N ) to scale like /N as N → ∞ . We find good agreement between the error and /N .
20 40 60 80 100 120 N C N asymptoticnumerical
20 40 60 80 100 120 N -4 -3 -2 -1 error /N Figure 6: The Newtonian capacitance of a hypercube.
References [1] B Alberts, A Johnson, J Lewis, M Raff, and K Roberts.
Molecular Biologyof the Cell . Garland, 2014.[2] Jun F Allard, Omer Dushek, Daniel Coombs, and P Anton Van Der Merwe.Mechanical Modulation of Receptor-Ligand Interactions at Cell-Cell Inter-faces.
Biophys J , 102(6):1265–1273, March 2012.133] E Ben-Naim and P L Krapivsky. First-passage exponents of multiple ran-dom walks.
J Phys A , 43(49):495008–17, November 2010.[4] H C Berg.
Random walks in biology . 1993.[5] Paul Bressloff and Jay Newby. Stochastic models of intracellular transport.
Rev. Mod. Phys. , pages 1–69, August 2012.[6] C W Cairo, R Das, A Albohy, Q J Baca, D Pradhan, J S Morrow,D Coombs, and D E Golan. Dynamic Regulation of CD45 Lateral Mo-bility by the Spectrin-Ankyrin Cytoskeleton of T Cells.
J Biol Chem ,285(15):11392–11401, April 2010.[7] H. S Carslaw and J. C Jaeger.
Conduction of heat in solids . ClarendonPress, Oxford, 2nd ed edition, 1959.[8] Amit K Chattopadhyay and Nigel J Burroughs. Close contact fluctuations:The seeding of signalling domains in the immunological synapse.
Euro PhysLett , 77(4):48003, 2007.[9] A. F. Cheviakov and M. J. Ward. Optimizing the principal eigenvalue ofthe laplacian in a sphere with interior traps.
Mathematical and ComputerModelling , 53(7-8):1394–1409, 2011.[10] S. Condamin, O. Bénichou, and M. Moreau. Random walks and brown-ian motion: A method of computation for first-passage times and relatedquantities in confined geometries.
Phys. Rev. E , 75:021111, Feb 2007.[11] Leslie Conway, Derek Wood, Erkan Tuzel, and Jennifer L Ross. Motortransport of self-assembled cargos in crowded environments.
Proc. NatlAcad. Sci , 109(51):20814–20819, December 2012.[12] Daniel Coombs, Ronny Straube, and Michael Ward. Diffusion on a Spherewith Localized Traps: Mean First Passage Time, Eigenvalue Asymptotics,and Fekete Points.
SIAM J Applied Math , 70(1):302–332, January 2009.[13] R Dixit, J L Ross, Y E Goldman, and E L F Holzbaur. Differen-tial Regulation of Dynein and Kinesin Motor Proteins by Tau.
Science ,319(5866):1086–1089, February 2008.[14] Omer Dushek, Raibatak Das, and Daniel Coombs. A role for rebind-ing in rapid and reliable t cell responses to antigen.
PLoS Comput Biol ,5(11):e1000578–e1000578, 2009.[15] Helen S Goodridge, Christopher N Reyes, Courtney A Becker, Tamiko RKatsumoto, Jun Ma, Andrea J Wolf, Nandita Bose, Anissa SH Chan, An-drew S Magee, Michael E Danielson, et al. Activation of the innate im-mune receptor dectin-1 upon formation of a/phagocytic synapse/’.
Nature ,472(7344):471–475, 2011. 1416] P Hammar, P Leroy, A Mahmutovic, E G Marklund, O G Berg, and J Elf.The lac Repressor Displays Facilitated Diffusion in Living Cells.
Science ,336(6088):1595–1598, June 2012.[17] Jonne Helenius, Gary Brouhard, Yannis Kalaidzidis, Stefan Diez, andJonathon Howard. The depolymerizing kinesin MCAK uses lattice dif-fusion to rapidly target microtubule ends.
Nature , 441(7089):115–119, May2006.[18] Frédéric J Hoerndli, Dane A Maxfield, Penelope J Brockie, Jerry E Mellem,Erica Jensen, Rui Wang, David M Madsen, and Andres V Maricq. Kinesin-1 Regulates Synaptic Strength by Mediating the Delivery, Removal, andRedistribution of AMPA Receptors.
Neuron , 80(6):1421–1437, December2013.[19] Chi-Ok Hwang, Michael Mascagni, and Taeyoung Won. Monte carlo meth-ods for computing the capacitance of the unit cube.
Mathematics andComputers in Simulation , 80(6):1089 – 1095, 2010.[20] Samuel A. Isaacson and Jay Newby. Uniform asymptotic approximation ofdiffusion to a small target.
Phys. Rev. E , 88:012820, Jul 2013.[21] John David Jackson.
Classical Electrodynamics . John Wiley and Sons,1962.[22] Yoshihisa Kaizuka, Adam D Douglass, Rajat Varma, Michael L Dustin, andRonald D Vale. Mechanisms for segregating T cell receptor and adhesionmolecules during immunological synapse formation in Jurkat T cells.
Proc.Natl Acad. Sci , 104(51):20296–20301, December 2007.[23] James P Keener.
Principles of Applied Mathematics . Perseus Books, Cam-bridge, Mass., 2000.[24] Konstantin V Klenin, Holger Merlitz, Jörg Langowski, and Chen-XuWu. Facilitated Diffusion of DNA-Binding Proteins.
Phys. Rev. Lett ,96(1):018104–4, January 2006.[25] Kristine McKinney, Melissa Mattia, Vanesa Gottifredi, and Carol Prives.p53 Linear Diffusion along DNA Requires Its C Terminus.
Molecular Cell ,16(3):413–424, November 2004.[26] Tomas Oppelstrup, Vasily V. Bulatov, Aleksandar Donev, Malvin H. Kalos,George H. Gilmer, and Babak Sadigh. First-passage kinetic monte carlomethod.
Phys. Rev. E , 80(6):066701, Dec 2009.[27] Vishaal Rajani, Gustavo Carrero, David E Golan, Gerda de Vries, andChristopher W Cairo. Analysis of Molecular Diffusion by First-PassageTime Variance Identifies the Size of Confinement Zones.
Biophys J ,100(6):1463–1472, March 2011. 1528] Roman Rouzier, Radhika Rajan, Peter Wagner, Kenneth R Hess, David LGold, James Stec, Mark Ayers, Jeffrey S Ross, Peter Zhang, Thomas ABuchholz, Henry Kuerer, Marjorie Green, Banu Arun, Gabriel N Horto-bagyi, W Fraser Symmans, and Lajos Pusztai. Microtubule-associatedprotein tau: a marker of paclitaxel sensitivity in breast cancer.
Proc. NatlAcad. Sci , 102(23):8315–8320, June 2005.[29] B Rozycki, R Lipowsky, and T R Weikl. Segregation of receptor-ligandcomplexes in cell adhesion zones: phase diagrams and the role of thermalmembrane roughness.
New J Phys , 12:095003, 2010.[30] Z. Schuss, A. Singer, and D. Holcman. The narrow escape problem fordiffusion in cellular microdomains.
Proc. Natl Acad. Sci , 104(41):16098–16103, OCT 9 2007.[31] Bhupinder Shergill, Laurence Meloty-Kapella, Abdiwahab A Musse, GerryWeinmaster, and Elliot Botvinick. Optical tweezers studies on notch:single-molecule interaction strength is independent of ligand endocytosis.
Developmental cell , 22(6):1313–1320, 2012.[32] Frank Spitzer. Electrostatic capacity, heat flow, and brownian motion.
Probability theory and related fields , 3(2):110–121, 1964.[33] MJ Ward and JB Keller. Strong localized perturbations of eigenvalue prob-lems.