1: \documentclass[showpacs,amsmath,amssymb,floatfix,pre,twocolumn,superscriptaddress]{revtex4} 

2: \usepackage[dvips]{graphicx} 

3: \usepackage[latin1]{inputenc}

4: \usepackage{color,rotating}

5: \usepackage{amssymb} 

6: \usepackage{amsmath} 

7: \usepackage{martin_specials}

8: \usepackage{pilT_project} 

9: \usepackage[english]{babel}

10: \hyphenation{Neis-seria gonorr-hoeae Myxo-coc-cus xant-hus}

11: \newcommand{\Ref}{Ref.\ }

12: \newcommand{\figpart}[1]{{(#1)}}

13: \bibliographystyle{apsrev}

14: \date{\today}

15: \begin{document}

16: \title{Force generation in small ensembles of Brownian motors}

17: %Brownian motors in a ring and type IV pilus retraction} 

18: 

19: \author{Martin Lind\'en}

20: \email{linden@kth.se} 

21: \affiliation{Theoretical Physics, Royal Institute of Technology,

22: AlbaNova, 10691 Stockholm, Sweden}

23: \author{Tomi Tuohimaa}

24: \email{tomi.tuohimaa@biox.kth.se} 

25: \affiliation{Applied Physics, Royal Institute of Technology,

26: AlbaNova, 10691 Stockholm, Sweden}

27: 

28: \author{Ann-Beth Jonsson} 

29: \email{Ann-Beth.Jonsson@imbim.uu.se}

30: \affiliation{Department of Medical Biochemistry and Microbiology,

31: Uppsala Biomedical Center, Uppsala University, Box 582, 75123 Uppsala,

32: Sweden} 

33: 

34: \author{Mats Wallin} \email{wallin@kth.se} \affiliation{Theoretical

35: Physics, Royal Institute of Technology, AlbaNova, 10691 Stockholm,

36: Sweden}

37: 

38: \begin{abstract}

39: The motility of certain gram-negative bacteria is mediated by

40: retraction of type IV pili surface filaments, which are essential for

41: infectivity.

42: %

43: The retraction is powered by a strong molecular motor protein, PilT,

44: producing very high forces that can exceed 150 pN.

45: %

46: The molecular details of the motor mechanism are still largely

47: unknown, while other features have been identified, such as the

48: ring-shaped protein structure of the PilT motor.

49: %

50: The surprisingly high forces generated by the PilT system motivate a

51: model investigation of the generation of large forces in molecular

52: motors. We propose a simple model, involving a small ensemble of motor

53: subunits interacting through the deformations on a circular backbone

54: with finite stiffness.

55: %

56: The model describes the motor subunits in terms of diffusing particles

57: in an asymmetric, time-dependent binding potential (flashing ratchet

58: potential), roughly corresponding to the ATP hydrolysis cycle.

59: %

60: We compute force-velocity relations in a subset of the parameter space

61: and explore how the maximum force (stall force) is determined by

62: stiffness, binding strength, ensemble size, and degree of asymmetry.

63: We identify two qualitatively different regimes of operation depending

64: on the relation between ensemble size and asymmetry. In the transition

65: between these two regimes, the stall force depends nonlinearly on the

66: number of motor subunits.

67: %

68: Compared to its constituents without interactions, we find higher

69: efficiency and qualitatively different force-velocity relations.

70: %

71: The model captures several of the qualitative features obtained in

72: experiments on pilus retraction forces, such as roughly constant

73: velocity at low applied forces and insensitivity in the stall force to

74: changes in the ATP concentration.

75: \end{abstract}

76: % 87. Biological and medical physics

77: % 87.10.+e 	General theory and mathematical aspects

78: % 87.15.Aa 	Theory and modeling; computer simulation

79: 

80: % 87.16.-b      Subcellular structure and processes

81: % 87.16.Ac 	Theory and modeling; computer simulation

82: % 87.16.Nn      Motor proteins (myosin, kinesin dynein)

83: 

84: % 82. Physical chemistry and chemical physics

85: % 82.20.-w Chemical kinetics and dynamics

86: % 82.37.-j Single molecule kinetics

87: 

88: % 05. Statistical physics, thermodynamics, and nonlinear dynamical systems

89: % 05.40.-a Fluctuation phenomena, random processes, noise, and Brownian motion 

90: % 05.60.Cd Classical transport

91: % 05.70.Ln Nonequilibrium and irreversible thermodynamics

92: 

93: % 02.50.-r 	Probability theory, stochastic processes, and statistics (see also section 05 Statistical physics, thermodynamics, and nonlinear dynamical systems)

94: % 02.50.Ey 	Stochastic processes

95: % 87.16.Ac 	Theory and modeling; computer simulation

96: \pacs{87.16.-b,05.40.-a}

97: 

98: \maketitle

99: 

100: \section{Introduction}

101: Recent experimental progress has enabled remarkable quantitative

102: measurement of biological processes on the single-molecule level

103: \citep{sheetz98}.  One example is the biomechanics of force generation

104: by molecular machines such as kinesin, myosin, and dynein

105: \citep{boal,howard}.  This has stimulated considerable modeling

106: activity in order to analyze the experiments

107: \citep{howard,boal,reimann,bustamante01,schliwa03}.  In this paper we

108: are inspired by another motor protein, called PilT \citep{mattick},

109: which has interesting properties; e.g., it is the strongest known

110: molecular motor \citep{maier02}.

111: 

112: The PilT motor is responsible for the retraction of certain bacterial

113: surface filaments, and the velocities and forces generated during

114: retraction have been measured in a series of laser tweezers

115: experiments \citep{merz00,maier02,maier04}.  Theoretical analysis of

116: the retraction data has revealed interesting information about the

117: underlying retraction mechanism \citep{maier04}. It is also of

118: interest to study the question of how large the generated forces can

119: be, given the energy and length scales relevant to the PilT motor, and

120: what features are important for generation of large forces. We will

121: address this question through a simple ratchet model, which is

122: inspired by known experimental facts of the PilT system.

123: 

124: Ratchet models of particles in fluctuating potentials are commonly

125: used in theoretical studies of molecular motors

126: \citep{reimann,prost97b}. Single-particle models have been used to

127: study kinetics of ATP consumption in molecular motors

128: \citep{lattanzi01} and to describe the kinetics of kinesin

129: \citep{astumian94}. Models of particles in ratchet potentials have

130: also been employed to describe collective effects in large ensembles

131: of interacting motors

132: \citep{shu04,badoual02,lipowsky01,prost95,prost97,plischke99,juelicher97}.

133: Finite ensembles of Brownian particles have been studied to some

134: extent in the context of two-headed motor proteins

135: \citep{derenyi96,vilfan99,csahok,dan03,klumpp01} and to describe the

136: bacterial flagella motor \cite{xing06}.

137: 

138: Another approach, in which the motion of a molecular motor is

139: described in terms of transitions between discrete chemical and

140: conformational states, has also been generalized to the case of two

141: interacting motor subunits \citep{kolomeisky05}.  As is evident from

142: Refs.\

143: \citep{shu04,badoual02,lipowsky01,prost95,prost97,plischke99,juelicher97,derenyi96,vilfan99,csahok,dan03,klumpp01,xing06},

144: interactions among several motors can lead to new and nontrivial

145: behavior of the average velocity, which is not present if the

146: interaction is turned off. It is natural to ask if this is the case

147: for force generation as well.

148: 

149: In this paper, we investigate the behavior of a small ensemble of

150: interacting processive Brownian motors and focus on the effect of the

151: interaction on the generation of large forces.  We aim at a

152: prototypical, minimal model which captures certain features of the

153: PilT system. The main input is the overall structures of the filament

154: and PilT complex. When possible, we also use the experimental

155: situation to estimate model parameters, which should ideally be as few

156: as possible.  The model is a generalization of the model of two

157: elastically coupled motors studied by \citet{dan03} to larger

158: ensembles, but focuses on different properties. By varying the density

159: of motor subunits and other parameters of the system, we explore the

160: force production in different regimes.  We compare with a single

161: building block of our model to identify the effect of the interactions

162: and also compare with experimental results.  Although the detailed

163: connection between the model and the actual molecular retraction

164: mechanism is speculative, the spirit of the model is best understood

165: in light of the known facts about the PilT system. Therefore, we will

166: briefly review some facts about pilus retraction before introducing

167: the model.

168: 

169: Type IV pili are surface filaments crucial for the initial adherence

170: of certain gram-negative bacteria to target host cells, DNA uptake,

171: cell signaling, and bacterial motility \citep{mattick}. Each filament

172: consists of thousands of pilin monomers that polymerize to a helical

173: structure with outer diameter of about 6 nm, 4 nm pitch, and five

174: monomers per turn \citep{mattick,forest97}. The bacterial motility

175: associated with type IV pili, called twitching motility, is driven by

176: repeated extension, tip attachment, and retraction of the pilus

177: filament, by which the bacterium can pull itself forward on surfaces

178: like glass plates or target host cells \citep{merz00}. Type IV pili

179: are expressed by a wide range of gram-negative bacteria

180: \citep{mattick} including \bakterie{Myxococcus xanthus} \citep{sun}

181: and human pathogens \bakterie{Neisseria gonorrhoeae} \citep{merz00}

182: and \bakterie{Pseudomonas aeruginosa} \citep{skerker}.

183: 

184: The mechanism of retraction is believed to be filament disassembly

185: mediated by PilT, a member of the AAA family of motor proteins

186: \citep{mattick}, but the microscopic details of this process are not

187: known.  One might compare pilus retraction with force generation by

188: microtubules, which are multistranded filaments with effective monomer

189: lengths similar to type IV pili. The helical pitch divided by the

190: number of filament strands is 0.8 nm for the pili and 0.6 nm for

191: microtubules \citep{howard}. However, pilus retraction generates

192: forces of up to \mbox{160 pN} \citep{maier02,maier04}, which is an

193: order of magnitude larger than those observed in \emph{in vitro}

194: experiments on microtubules \cite{howard,dogterom02,janson04}. Another

195: difference is that the pilus retraction velocity is independent of

196: filament length \citep{maier02}. Since dissociated pilin monomers are

197: stored in the cell membrane waiting to be recycled in other filaments

198: \cite{skerker}, the implication is that the velocity is independent of

199: pilin concentration in the membrane. This presumably rules out simple

200: polymerization ratchet-type models, which have been proposed to

201: describe polymerization forces generated by microtubules

202: \citep{howard,dogterom02}. The experimental evidence instead favors a

203: retraction process driven by an active molecular motor

204: \citep{maier02}.

205: 

206: Pilus retraction is highly processive, and retraction velocities are

207: of the order of $0.5-1$ $\mathrm{\mu}$m/s

208: \citep{mattick,skerker,maier02,maier04}.  Generation of high forces

209: persists when the PilT concentration is reduced, suggesting that one

210: single PilT complex retracts the pilus filament \citep{maier02}. The

211: stall force (the force at which the average velocity drops to zero)

212: and the velocity at high forces are insensitive to changes in ATP

213: concentration, and the retraction velocity is roughly

214: force-independent for small applied forces ($\lesssim 50$ pN) within

215: the experimental accuracy \citep{maier02,maier04}. PilT has been shown

216: to form a ring structure with sixfold symmetry \citep{forest04}, and

217: since each subunit has an ATP binding motif, it is possible that it

218: can hydrolyze up to six ATP molecules in parallel during retraction

219: \citep{mattick}. The outer diameter of the ring is about 10 nm, and

220: the inner diameter varies in the range $2-4$ nm \citep{forest04}.

221: 

222: Pilus retraction is interesting from a technological point of view, as

223: a potential prototype for a nanomachine that can generate large

224: forces, and from a biomedical point of view since pilus retraction is

225: important for the infectivity of various severe bacterial pathogens

226: \citep{steroids}.

227: 

228: There are several proposals for how the molecular constituents of the

229: retraction machinery fit together. One of them is that PilT forms a

230: ring around the base of the pilus \citep{mattick,kaiser00}. The hole

231: in the middle of the PilT complex seems too small to let the assembled

232: filament through, but large enough for pilin monomers.  This could

233: allow interactions between the pilus and PilT via several active sites

234: (motor subunits) that work together and is the principle that we will

235: explore here. For simplicity, we assume one motor subunit per filament

236: strand and neglect possible two-dimensional effects such as angular

237: motion of the filament.

238: 

239: We stress that the purpose of this paper is not to attempt to describe

240: the detailed molecular mechanisms involved in pilus retraction, which

241: are largely unknown.  Rather we examine a new regime of a simple

242: model, whose main features are inspired by experiments. Below we

243: obtain several results from the model, such as large force generation

244: and other properties that agree well with interesting experimental

245: results on pilus retraction.  Moreover, these results are a

246: consequence of correlations and interactions between the motor

247: subunits and are strikingly different from the characteristics of the

248: single building block of the model.

249: 

250: \section{Retraction model}\label{sec_model}

251: In this section the geometry and equations of motion of the model are

252: described. We then discuss the parameters, which come in several

253: kinds: parameters that are known for the PilT system, parameters that

254: can be estimated to varying degrees of accuracy, and parameters that

255: we will explore in a systematic way.  A few parameters cannot be

256: estimated due to the lack of knowledge of the molecular details. In

257: this case, we make an arbitrary choice in order to investigate the

258: qualitative behavior of the model.

259: \begin{figure}[t!]\begin{center}

260: \includegraphics{fig1.eps}

261: \caption{\label{potential_stroke} (Color online) \figpart{a} The

262: elements of the retraction motor model consist of a flexible ring of

263: motor subunits that interact with a moving helical

264: filament. \figpart{b} Equivalent geometry after the change of

265: variables $y_i=x_i-id/M$, which places the binding potentials of the

266: filament monomers on top of each other.  The binding potential is

267: assumed to be an asymmetric ratchet potential with amplitude $U$ and

268: asymmetry $a$. The undeformed state of the motor protein complex is

269: described by the equilibrium positions $y_i^0$ of the motor subunits,

270: and the subunits are elastically confined to their equilibrium

271: positions. Due to the helical structure of the filament, the

272: equilibrium positions become evenly spread over one period. The motor

273: subunits at positions $y_i$ (black circles) interact with the filament

274: potential. The open circle represents an unbound subunit.  During a

275: successful retraction process, a motor subunit detaches from the

276: filament, relaxes in its confinement potential, and rebinds near the

277: next binding site along the filament.  \figpart{c} Distribution of the

278: unbound subunit in B. The shaded area represents the probability for

279: the subunit to bind to the left of the potential maximum and produce a

280: failed step. As the applied force increases, the filament is pulled

281: forward relative to the distribution and the probability of a failed

282: step increases. \figpart{d} Qualitatively new behavior emerges in the

283: limit of high stiffness (strong confinement) and many motor

284: subunits. In this limit, all bound motor subunits will not relax to

285: the minima in the binding potential. Instead, some subunits will

286: interact with the shaded part of the potential and oppose force

287: production. In this regime, one expects that the stall force depends

288: nonlinearly on the number of motor subunits, and nonmonotonically on

289: the stiffness.}\end{center}\end{figure}

290: The basic setup is sketched in Fig.\ \ref{potential_stroke}(a). A ring

291: of $M$ motor subunits interacts with an $M$-stranded helical filament,

292: with repeat distance $d$ of the single strands. The filament

293: coordinate $q$ decreases during retraction. We also allow for

294: deformations of the PilT ring. This is described by displacements

295: $x_i$ of the motor subunits from the undeformed state. We assume that

296: the motor subunits interact with the filament strands via identical

297: one-dimensional binding potentials with period $d$. To mimic the

298: helical structure of the filament, the potential of subunit $i$ is

299: displaced a distance $id/M$ relative to the potential of subunit $0$.

300: The interactions between the filament and subunit $i$ therefore have

301: the form $V(x_i-q-id/M)$, where $V$ is some one-dimensional binding

302: potential. We formulate the equations of motion as a system of

303: overdamped Langevin equations for $q$ and $x_i$,

304: %

305: \ekv{eom}{

306: \begin{split}

307:   \alpha \dot{x}_i &= -\km x_i - h_i(t)\frac{\partial

308:   V(x_i-q-id/M)}{\partial x_i}+\sqrt{2\alpha \kBT}\,\xi_i(t),\\ \gamma

309:   \dot{q}& = \sum_{i=0}^{M-1}h_i(t) \frac{\partial

310:   V(x_i-q-id/M)}{\partial x_i}+ \Ftrap+\sqrt{2\gamma

311:   \kBT}\,\xi_\text{q}(t),

312: \end{split}}

313: %

314: where $i=0,\ldots,M-1$, $\alpha$ and $\gamma$ are friction

315: coefficients of the motor subunits and the filament respectively,

316: $\km$ is a spring constant describing the stiffness of the PilT ring,

317: $h_i(t)=0,1$ are chemical state variables, $k_{\mathrm B}$ is

318: Boltzmann's constant, $T$ is the temperature, $\Ftrap$ is an external

319: force acting on the filament, and $M$ is the number of motor subunits

320: and filament strands. Thermal fluctuations are included through

321: independent Gaussian white noise terms $\xi_i(t)$ and

322: $\xi_\text{q}(t)$, which obey

323: \mbox{$\mean{\xi_i(t)}=\mean{\xi_\text{q}(t)}=0$},

324: \mbox{$\mean{\xi_j(t)\xi_i(t')}=\delta_{ij}\delta(t-t')$}, and

325: \mbox{$\mean{\xi_\text{q}(t)\xi_\text{q}(t')}=\delta(t-t')$} and have

326: prefactors according to the fluctuation-dissipation theorem. The

327: temperature is set to $T$= \mbox{310 K}.

328: 

329: Before discussing the terms in the equations of motion in detail, we

330: transform the subunit coordinates to place the binding potentials on

331: top of each other. The new coordinates are convenient in order to

332: analyze the model and are close to previous works with similar models

333: \citep{dan03,klumpp01,csahok}. The transformed subunit coordinates are

334: $y_i=x_i-id/M+d$. An undeformed PilT ring is now described as

335: $y_i=y_i^0=d-id/M$, which we call the equilibrium positions of the

336: subunits. The transformed equations of motions are

337: %

338: \ekv{eomYQ}{

339: \begin{split}

340:   \alpha \dot{y}_i &=-\km (y_i-y_i^0) - h_i(t)\frac{\partial

341:   V(y_i-q)}{\partial y_i}+\sqrt{2\alpha \kBT}\,\xi_i(t)\\ \gamma

342:   \dot{q}& = \sum_{i=0}^{M-1}h_i(t)\frac{\partial V(y_i-q)}{\partial

343:   y_i}+\Ftrap+\sqrt{2\gamma \kBT}\,\xi_\text{q}(t)

344: \end{split}}

345: %

346: Note that the binding potentials are all on top of each other, as if

347: all motor subunits were interacting with just one single strand.  The

348: price for this convenience is that the equilibrium positions are

349: evenly distributed over one period of the potential.  The setup in the

350: transformed variables is sketched in Fig.\ \ref{potential_stroke}(b).

351: 

352: The motor subunits can be in one of two states: unbound ($h_i=0$), in

353: which they diffuse around their respective equilibrium positions, or

354: bound ($h_i=1$), in which they also interact with the filament through

355: the binding potential $V(y_i-q)$.  We treat the number of motor

356: subunits $M$ as a parameter, and we will present results for $M=2$,

357: $3$, $5$, $6$, and $12$.

358: 

359: For simplicity, we model the binding of the motor to the filament by

360: an asymmetric sawtooth potential, shown in Fig.\

361: \ref{potential_stroke}(c).  For the asymmetry factor we take $a=0.1$,

362: if not stated otherwise. Some asymmetry is needed to give a preferred

363: direction of motion, and the sawtooth potential was selected to give a

364: simple parametrization of the (unknown) real interaction potential.

365: One possible origin of an asymmetric binding potential is surface

366: charges on the head of the pilin monomer \citep{parge95} -- for

367: example as described in Ref.\ \citep{boal}. Another alternative is

368: that the asymmetry can be viewed as an effective description of some

369: asymmetry somewhere else in the system, e.g., in the direction of the

370: motor steps (power strokes). Based on the helical structure of the

371: pilus filament, we take the periodicity of the potential to be $d=4$

372: nm \citep{forest97,mattick}, which we use for all values of $M$.

373: 

374: The amplitude $\dE$ of the potential is the maximal energy required to

375: break the bond between the filament and the active site. Pilus

376: retraction is powered by hydrolysis of one or a few ATP per retracted

377: pilin monomer \citep{maier02}, which sets the energy scale for the

378: potential. Depending on conditions, the free energy yield from

379: hydrolysis of one ATP in a cell is about $80-100$ pN nm

380: \citep{howard}. The motor subunits are bound together to form the

381: motor complex, and we model their confinement with a harmonic

382: restoring force $-\km x_i=-\km(y_i-y_i^0)$. This linear approximation

383: is reasonable if the deformations $x_i=y_i-y_i^0$ are small, which

384: will be verified below.

385: 

386: For the binding processes $h_i(t)$, we restrict ourselves to a

387: sequential reaction scheme with $M$ chemical states.  We define state

388: $j$ as $h_k(t)=1-\delta_{kj}$; i.e., subunit $j$ is unbound and the

389: other subunits are bound.  The states are visited in ascending order,

390: and the (constant) transition rate from state $j$ to $j+1(\text{mod}

391: M)$ is $\lambda$. The bound subunits spend most of their time near a

392: minimum in the binding potential, and several geometrical

393: configurations are compatible with each chemical state. The main

394: pathway for efficient retraction in the model is that the subunits

395: take turns to hop forward to the next minimum as they release and

396: rebind to the filament (successful steps). When the retraction is less

397: efficient -- for example at large applied force -- the subunits

398: sometimes do not hop forward (failed step), which leads to geometrical

399: configurations outside the main pathway. During a successful step, a

400: motor subunit goes through the following sequence of events:

401: \begin{enumerate}

402:   \item[(i)] The subunit is released from a minimum in the binding

403:     potential.

404:   \item[(ii)] The released subunit relaxes to its equilibrium position. The

405:     filament relaxes in the opposite direction due to the forces from

406:     the other motor subunits.

407:   \item[(iii)] The subunit rebinds, close to the next minimum in the binding

408:     potential (otherwise the step fails). At the same time, the next

409:     motor subunit enters step (i).

410:   \item[(iv)] The subunit stays bound and pulls on the filament as the

411:     other $M-1$ subunits go through steps (i)--(iii). After each

412:     successful rebinding event, the filament retracts a distance

413:     $d/M$.

414: \end{enumerate}

415: This mechanism relies on the asymmetry of the potential and is similar

416: to the mechanisms studied earlier for two elastically coupled

417: particles \citep{dan03,klumpp01}.

418: %The correlation that we assume between consecutive binding events is

419: %not supported by direct evidence in the PilT system.

420: We argue below that generation of strong forces

421: in this model relies on a binding process that always keeps several

422: motor subunits bound to the filament, but the binding order is less

423: important.

424: 

425: We now discuss the parameters of the model.  In the laser tweezers

426: experiments \citep{maier04,maier02,merz00}, the outer filament tip

427: binds to an external latex bead with diameter $1-2$ $\mu$m.  Using

428: Stokes law, $\gamma=6\pi\eta R$, the approximate viscosity

429: $\eta=10^{-8}$ pN s/nm$^2$ of the bulk solution surrounding the cell

430: (somewhere between $10^{-9}$ for water and $8\times10^{-7}$ for

431: glycerin seems reasonable) and $R=1$ $\mu$m, we get $\gamma \approx

432: 2\times 10^{-4}$ pN s/nm for the bead. As a first approximation, we

433: neglect the elasticity and friction of the pilus filament itself. For

434: the internal friction coefficient, we use $\alpha = 0.5 \times

435: 10^{-3}$ pN s/nm $\ll\km/\lambda$. This sets the time scale for

436: internal relaxation $\alpha/\km$ much smaller than the typical time

437: $(M-1)\lambda^{-1}$ between binding and release of individual motor

438: subunits and lets the motor subunits reach thermal equilibrium between

439: transitions. This is consistent with estimates of thermal relaxation

440: times over length scales on the order of 10 nm \citep{juelicher97},

441: which is the size of the PilT ring. Another time scale for internal

442: relaxation is given by the time to slide down to a potential minimum,

443: $\alpha d^2/\dE$, which we also keep smaller than $(M-1)\lambda^{-1}$.

444: We then expect the velocity to be proportional to $\lambda$, and we

445: will restrict ourselves to this quasistatic regime for two

446: reasons. First, this is the biologically relevant regime where we

447: expect the stall force to be independent of $\lambda$, which

448: corresponds to the experimental observation that the stall force is

449: independent of ATP concentration \citep{maier02}. Second, the exact

450: value of $\alpha$ is not critical for the results in this regime, and

451: since it is difficult to estimate $\alpha$ accurately, we can avoid

452: making our results depend strongly on an unknown parameter.

453: 

454: Having found useful values for the potential period $d$ and the drag

455: coefficients $\alpha_i$ and $\gamma$, we go on to investigate the

456: model behavior as a function of the remaining parameters

457: $a,M,\km,\dE$, and $\lambda$ and properties of the binding process.

458: 

459: \section{Results}

460: \subsection{Methods}

461: Retraction of the filament means that $q$ decreases, so it is natural

462: to study the retraction velocity $v=-dq/dt$. In the laser tweezers

463: experiment, the tip of the bead is held by a static laser trap which

464: is to good approximation a harmonic potential -- i.e.,

465: \mbox{$\Ftrap=-\ktrap q$}, with $\ktrap$ on the order of $0.1$ pN/nm.

466: Numerical solution of Eq.\ \eqref{eom} using a standard method, known

467: as the Milstein scheme \citep{kloedenplaten}, produces a deflection

468: trajectory similar to the experimental ones. We calculate the

469: retraction velocity by fitting a second-order polynomial to a small

470: time interval around a point $q=-F/\ktrap$ and take the velocity

471: $v(F)$ as the derivative of the polynomial. This is similar in spirit

472: to how the experimental data was analyzed

473: \citep{merz00,maier02,maier04}. The retracted distance $-q(t)$

474: increases from the initial value towards a steady-state, corresponding

475: to the maximal applied force (stall force), which we define as the

476: mean applied force in the steady state.  To check our simulation code,

477: we reproduced analytical results for the steady state current in a

478: flashing ratchet model \citep{parrondo98}.

479: \subsection{Stall force and force-velocity 

480: relation}\label{qualitativeArguments} The stall force and

481: force-velocity relation of the motor is determined by several

482: competing mechanisms, which we now describe qualitatively.

483: 

484: In the case of a few motor subunits, stalling of the retraction is

485: controlled by two different mechanisms. One comes from the finite

486: binding energy between the filament and the subunits. Due to the

487: simple shape of the potential, we can estimate an upper limit for the

488: stall force, using force$=\Delta V/\Delta q$. The maximum force that

489: each subunit can exert on the filament during retraction against an

490: opposing (positive) force is $\dE/(1-a)d$, so $M-1$ bound motor

491: subunits give an upper limit of $\dE(M-1)/d(1-a)$ for the stall force.

492: At finite temperatures the upper limit is not reached, since the motor

493: subunits diffuse and can pass between potential minima by thermal

494: excitation. We think of these thermally assisted transitions as

495: slipping events, and they occur more often when the subunits are far

496: from their equilibrium positions and experience a large confining

497: force. This mechanism tends to increase the stall force with increased

498: binding strength $\dE$.

499: 

500: The other mechanism has to do with the stiffness $k$ and the

501: probability for a step to fail. This probability depends on the

502: distribution of the filament position relative to the unbound subunit,

503: which is illustrated in Fig.\ \ref{potential_stroke}(c). The shaded

504: area represents the probability of a failed step.  Higher probability

505: of failure gives lower velocity. Increasing the applied force $F$ at

506: constant stiffness pulls the filament to the right, increases the

507: fraction of failed steps, and decreases the velocity. If all steps

508: fail, no retraction takes place. The width of the position

509: distribution of a subunit in the harmonic confinement is

510: $\sqrt{\kBT/\km}$, but the distribution in Fig.\

511: \ref{potential_stroke}(c) is broader, since the filament also

512: fluctuates. The average relative position $\Delta$ is roughly

513: proportional to $F/\km$. At very low stiffness, the distribution is

514: broad enough for some steps to fail without applied load, and the

515: average relative position varies strongly with applied force. This

516: gives a monotonically decreasing force-velocity relation and a low

517: stall force limited by the stiffness. At very high stiffness, the

518: distribution is narrow and the average relative position is less

519: sensitive to the applied force. Almost no steps fail without applied

520: load, and it takes some threshold force before steps start to fail

521: significantly. We get a force-velocity relation that is almost

522: independent of force at low forces, and the stall force is mainly

523: limited by slipping events.

524: \begin{figure}[t]\begin{center}

525:     \includegraphics{fig2.eps}

526:     \caption{\label{vFrelations} (Color online) Force-velocity

527:     relations for high (triangles), intermediate (squares) and low

528:     stiffness (circles) calculated for $M=5$. The y axis is velocity

529:     normalized by $\lambda d/M$, which corresponds to the velocity for

530:     $M$ motor subunits if all steps were successful.}

531: \end{center}\end{figure}

532: Figure \ref{vFrelations} shows examples of normalized force-velocity

533: relations of the model for strong, intermediate, and weak stiffness,

534: compared to the binding strength. The curves illustrate the

535: qualitative arguments of the preceding paragraphs. We also find that

536: the stall force is insensitive to changes in the reaction rate

537: $\lambda$ (not shown). Since the reaction rate corresponds to the ATP

538: concentration, this is in qualitative agreement with the experimental

539: result that the stall force is insensitive to the ATP concentration

540: \citep{maier02}. This is expected in the quasistatic regime that

541: results from the choice of time scales discussed at the end of Sec.\

542: \ref{sec_model}. Stiff systems ($\km d^2/\dE\gg 1$) have a plateau in

543: the retraction velocity at low forces, which is also the general

544: experimental trend \citep{maier02,maier04}. For the parameter regime

545: we have investigated, Fig.\ \ref{vFrelations} gives the qualitative

546: shape of the force-velocity relation as a function of $\km/\dE$. The

547: size of the plateau is roughly proportional to the stall force, with a

548: proportionality constant that is to first approximation a function

549: only of $\km/\dE$.

550: 

551: As the number of motor subunits and filament strands increases, high

552: stiffness can also have a destructive effect on the stall force, as

553: illustrated in Fig.\ \ref{potential_stroke}D. This occurs when the

554: distance between equilibrium positions is shorter than the region of

555: the binding potential with backward slope (shaded) -- i.e.,

556: $M>1/a$. In that case, some of the bound motor subunits will tend to

557: interact with the shaded region of the potential if the stiffness is

558: high. There they act with a negative force on the filament and

559: contribute negatively to the force production. This effect is enhanced

560: by increased stiffness, and we therefore expect the stall force to

561: have a maximum as a function of stiffness. This is a qualitatively

562: different behavior than with only a few motor subunits ($M<1/a$) and

563: makes the stall force depend in a highly nonlinear way on $M$.

564: 

565: \subsection{Parametrization of the stall force}

566: Equation \eqref{eomYQ} suggests that the stall force might depend on

567: the ratio $\km /\dE$, instead of $\km$ and $\dE$ independently. We

568: will use this observation to further analyze force production in the

569: model.  A parametrization of the stall force is obtained from a

570: combination of the estimated upper limit $\dE(M-1)/d(1-a)$ for the

571: stall force with an function of $\km/\dE$.  Using this (unknown)

572: function $f_M$, which also depends on $M$, we describe the effect of

573: stiffness in the following way:

574: \ekv{scaleEq}{\begin{array}{cc}

575: \Fst(\km,\dE)=\Finf\,\fscale{M}{\frac{\km d^2}{\dE}},&

576: \Finf=\frac{M-1}{d(1-a)}\left(\dE -\dE_M\right),

577:   \end{array}}

578: where $\dE_M$ is a free $M$-dependent parameter, independent of $\dE$

579: and $\km$, and $d^2$ was inserted to make the argument of $f_M$

580: dimensionless.  One can obtain $\fscale{M}{x}$ by plotting

581: $\Fst/\Finf$ against $\km d^2/\dE$ and adjusting $\dE_M$. Such a plot

582: is shown in Fig.\ \ref{scaleFig}(a), with the best fit values of

583: $\dE_M=25.4$, $25.9$, $28.9$, $28.3$, and $33.5$ pN nm for $M$= 2,

584: 3, 5, 6, and 12, respectively. For $M$=3, 5, and 6, the data points

585: fall on a single curve for each $M$ to good approximation, while $M$=2

586: and 12 show some scattering.

587: \begin{figure}[t]\begin{center}

588: \includegraphics{fig3.eps}

589: \caption{\label{scaleFig} (Color online) \figpart{a} Stall force as a

590: function of $\km/\dE$ in the range \mbox{$5<\km<250$ pN/nm} and

591: \mbox{$80<\dE<200$ pN nm}. The normalization $\Finf$ is chosen

592: according to Eq.\ \eqref{scaleEq}. Lines are guides to the eye.  The

593: case of many subunits ($M=12$) is qualitatively different from the

594: other cases. Here, the stall force depends nonmonotonically on

595: $\km/\dE$. \figpart{b} Stall force per bound motor subunit in the

596: stiff limit $\km\to\infty$ as a function of $\dE-\dE_M$.  For $M\le 6$

597: the simulations are consistent with $f_M(\infty)=1$ in Eq.\

598: \eqref{scaleEq} (solid line). $M=12$ gives a lower stall force that

599: does not fit Eq.\ \eqref{scaleEq}, which again illustrates the

600: nonlinear behavior of the stall force on $M$.  Error bars in both

601: graphs are smaller than the symbols.}

602: \end{center}\end{figure}

603: We can understand the ansatz \eqref{scaleEq} and the results in Fig.\

604: \ref{scaleFig} in light of the qualitative arguments in Sec.\

605: \ref{qualitativeArguments}.  We interpret $\Finf$ as the stall force

606: in the stiff limit $\km\to\infty$ and $\dE_M$ as the effect of thermal

607: fluctuations, which induce slipping events and thereby lower the

608: maximal force that the binding potential can support.  The stiff limit

609: $\km\to\infty$ can be simulated by locking the motor subunits to their

610: equilibrium positions. The results are shown in Fig.\

611: \ref{scaleFig}(b), using the above values for $\dE_M$.  The stall

612: force per bound motor subunit is consistent with Eq.\ \eqref{scaleEq}

613: and $\lim_{x\to\infty}\fscale{M}{x}=1$ for $M=2$, $3$, $5$, and

614: $6$.  The case $M=12$ also gives a straight line for the stall force

615: as a function of $\dE$ in the stiff limit, but neither consistent with

616: $\lim_{x\to\infty}\fscale{12}{x}=1$ nor with the value of $\dE_{12}$

617: from finite stiffness. This illustrates the effect of dense motor

618: subunits ($M>1/a$) in a stiff system.

619: 

620: The function $\fscale{M}{\km d^2/\dE}$ is a normalized stall force per

621: motor subunit. As is evident from Fig.\ \ref{scaleFig}(a), the force

622: production per motor subunit varies strongly with stiffness and also

623: with the number of subunits, $M$. Clearly, the interaction between the

624: motor subunits is important for the force generation. The stall force

625: is more complicated than a sum of contributions from the individual

626: parts. According to the arguments in Sec.\ \ref{qualitativeArguments},

627: one expects $\fscale{M}{x}$ to be an increasing function of $x$ for

628: $M<1/a$, since larger stiffness decreases the probability that the

629: unbound motor subunit binds to the wrong potential well. For $M>1/a$,

630: the stall force is expected to have a maximum as a function of

631: stiffness, reflecting the inset of the qualitatively different

632: behavior at high stiffness.  These expectations are consistent with

633: Fig.\ \ref{scaleFig}, where $\fscale{2}{x}$, $\fscale{3}{x}$,

634: $\fscale{5}{x}$, and $\fscale{6}{x}$ are monotonically increasing in

635: the simulated region, whereas $\fscale{12}{x}$ has a maximum around

636: $1.5$. To confirm that dense motor subunits give lower stall force

637: also for lower values of $M$, we performed simulations with $a=0.3$

638: and found that with $\dE=200$ pN nm and $M=5$, a maximum stall force

639: occurs for stiffness between 40 and 150 pN/nm (not shown), as

640: expected.

641: 

642: \subsection{Comparison with an individual motor subunit}

643: To further highlight the effect of interactions and correlations, it

644: is interesting to compare the interacting ring model to an isolated

645: motor subunit. This is a single particle in a flashing ratchet

646: potential and a special case of the interacting model, with $M=1$ and

647: \mbox{$\km\to\infty$}. This single-particle flashing ratchet (SPR) has

648: been studied extensively \citep{reimann,astumian97} in different

649: versions, and its properties are qualitatively different from those of

650: the interacting model in several interesting respects. For an SPR in

651: this simple version, the mechanism to pull the particle to the next

652: potential well is not present. Instead, forward motion relies on

653: thermal noise to make the particle diffuse forward while in the

654: unbound state. This means that the probability for a forward step can

655: never exceed $1/2$ for a single chemical cycle even without applied

656: force. With nonzero applied force, the free diffusion is superimposed

657: on a backward motion with velocity

658: \mbox{$v_{\text{drift}}=-F/\gamma$}; hence, the velocity is

659: substantially reduced even at very low forces.  These features

660: conspire to make both the maximal velocity and the stall force depend

661: strongly on the friction constant, the reaction rate, and how much

662: time the particle spends unbound during a reaction cycle

663: \citep{astumian97,astumian94}.  Two examples of force-velocity

664: relations for SPR are shown in Fig.\ \ref{VFfit}, with 4 nm

665: periodicity, $\dE=160$ pN nm, $\lambda=2000$ $s^{-1}$, and damping as

666: above. The transitions between the bound and unbound states are

667: deterministic in these cases, and the time $t_{\text{off}}$ spent

668: unbound during each cycle was 2.5 ms and 25 ms.

669: \begin{figure}[t]\begin{center}

670: \includegraphics[width=8.7cm]{fig4.eps}

671: \caption{(Color online) Force-velocity characteristics from simulation

672: and experiment.  Triangles: average experimental velocity from Fig.\

673: 3(b) of Ref.\ \citep{maier04}.  Thick solid curve: $M=5$ model with

674: $\dE=200$ pN nm, $\km=25$ pN/nm, $\Fst=123$ pN, and $\lambda=1180$

675: s$^{-1}$.  Diamonds: a single retraction event from Ref.\

676: \citep{maier02}. Dashed curve: $M=5$ model with $\dE=200$ pN nm,

677: $\km=60$ pN/nm, $\Fst=160$ pN, and $\lambda=1800$ s$^{-1}$.  Isolated

678: motor subunits -- i.e., simple flashing ratchet models (black symbols)

679: are qualitatively different, as illustrated for $M=1$, $\dE=200$ pN

680: nm, $\lambda=2000$ s$^{-1}$, and $t_{\text{off}}=25$ ms ($\star$) or

681: $2.5$ ms ($\ast$).  }\label{VFfit}

682: \end{center}\end{figure}

683: \subsection{Application to a real motor system}

684: In Fig.\ \ref{VFfit} we compare model results with experimental data

685: on \bakterie{N. gonorrhoeae}. Several force-velocity relations

686: averaged over many retraction events, as well as two single events,

687: are preesented in Ref.\ \citep{maier02,maier04}.  The maximum forces

688: produced in the different events are distributed between 40 and 160

689: pN. There are several experimental factors that can make the maximum

690: measured force in a particular retraction event lower than the

691: intrinsic stall force \citep{maier02,maier04}. These factors include,

692: for example, breakage of the filament and are not included in the

693: model. Accordingly, the model should be compared to the data that

694: reach the highest forces. This leaves one single trajectory

695: \citep{maier02}, shown in Fig. \ref{VFfit}, and several average

696: curves. The average force-velocity data in Fig.\ \ref{VFfit} represent

697: an average of the data in Fig.\ 3 of \Ref\ \citep{maier04}.  Both

698: experimental curves give positive velocities up to about 160 pN, which

699: is in the upper tail of the maximum force distribution

700: \citep{maier04}, but the velocity at lower forces differs considerably

701: between the two curves.

702: 

703: To describe the PilT system, there are two natural choices for $M$:

704: namely, $M=5$, reflecting the five-fold symmetry of the filament, and

705: $M=6$, reflecting the six-fold symmetry of the PilT molecule. We

706: present simulation results for $M=5$, but $M=6$ is qualitatively no

707: different.

708: 

709: The model can describe both experimental curves to some extent. The

710: parameter values that describe the single event and the average data

711: (see Fig.\ \ref{VFfit}) differ in reaction $\lambda$ rate and

712: stiffness $\km$. There are two important deviations, which we discuss

713: next.

714: 

715: The average velocity falls off exponentially at high forces

716: \citep{maier04}, while the results of the model decay faster.  The

717: exponential decay can be described by an Arrhenius law for the rate

718: limiting step \citep{maier04}. The present model does not account for

719: this behavior.  

720: 

721: The single retraction event has a different decay at high forces and

722: agrees better with the characteristics of the model. However, the

723: single event shows an initial increase in velocity at low forces,

724: rather than a plateau. Such initial increases in velocity are also

725: obtained in some individual simulation runs, but disappear when the

726: average is computed. All the experimental data also suffer from a

727: possible systematic underestimation of the velocity near $F$=0

728: \cite{maier02,maier04}. Within the experimental accuracy, the general

729: trend is a constant velocity up to about $50$ pN

730: \cite{maier02,maier04}, which is consistent with the results of the

731: model.

732: 

733: \subsection{Deformations and elastic approximation}

734: From the results presented in Figs.\ \ref{scaleFig} and\ref{VFfit} and

735: Eq.\ \eqref{scaleEq}, it is possible to estimate the magnitude of the

736: deformations $y_i$ of the motor complex near the stall force. This is

737: useful, since the harmonic approximation for the confining force on

738: the motor subunits is questionable for large displacements.  At the

739: stall force, with $M-1$ subunits bound to the filament, the mean

740: displacement of a subunit can be estimated to $\Fst/\km(M-1)$. As seen

741: earlier, the stall force is less than $(M-1)\dE/d(1-a)$, which gives a

742: displacement less than $U/\km d(1-a)$. From Fig.\ \ref{scaleFig}, the

743: interesting and relevant regime with high normalized stall force has

744: $kd/U\gtrsim 1$ nm$^{-1}$. This gives a displacement less than $1$ nm

745: near the stall force. This is not excessively large compared to the

746: dimensions of the PilT ring, whose diameter is $11.5$ nm

747: \citep{forest04}.

748: \subsection{Role of order in the binding process}

749: We next examine the effect of the ad hoc assumption of a sequential

750: binding process on the results of the model. We compared the ordered

751: sequence with two less correlated binding schemes. For random order

752: with only one subunit free at the same time, the stall force is

753: essentially unchanged, but the velocity decreases with about

754: 50\%. Alternatively, if the binding and unbinding events are assumed

755: to be independent for the different subunits, all subunits might

756: occasionally become unbound simultaneously, which releases the

757: filament from the motor. The resulting retraction events become highly

758: irregular and have low stall force and mean velocity. We conclude that

759: some degree of correlation between the motor subunits is necessary in

760: order for the model to simulate the experiments.

761: \section{Discussion and Conclusion}

762: We investigate a model for force generation in finite ensembles of

763: motor subunits interacting through an elastic backbone, which is

764: inspired by the pilus retraction machinery, the strongest molecular

765: motor reported so far.  The model is prototypical, rather than

766: realistic in detail, and offers a possible mechanism for generation of

767: large retraction forces.  It includes a ring of motor subunits

768: surrounding the pilus filament, following a suggestion in Ref.\

769: \citep{kaiser00}. Some parameters in the model can be roughly

770: estimated, and we explore parts of the remaining parameter space and

771: focus on generation of large forces.

772: 

773: We find that the stall force depends on the binding strength $\dE$

774: between motor and filament, the stiffness $\km$ of the motor complex,

775: the number of motor subunits $M$, and an asymmetry parameter $a$.  For

776: high enough stiffness we find qualitatively different properties

777: compared to the well-studied model of a single particle in a flashing

778: ratchet potential, which is the basic building block of our

779: model. This is not surprising, since the mechanisms that generate

780: motion are different in the two cases. The motion in the flashing

781: ratchet model is dependent on diffusion \citep{reimann}. Our model

782: also contains diffusive motion of the motor subunits, but diffusion is

783: not necessary for the motor to work \citep{dan03,klumpp01}.  The

784: dependence of the stall force on $\dE$, $\km$, and $M$ is well

785: parametrized in empirical scaling plots (Fig.\ \ref{scaleFig}). The

786: scaling ansatz in Eq.\ \eqref{scaleEq} relies on the presence of an

787: interaction between the motor subunits, and the stall force depends

788: nonlinearly on the number of motor subunits.

789: 

790: Low stiffness compared to binding strength has a strong destructive

791: effect on the force production. For a small number of motor subunits,

792: the stall force increases monotonically with increasing stiffness, but

793: when the motor subunits become dense enough there is a crossover to a

794: different regime, where a high stiffness instead has a destructive

795: effect on the stall force. For the asymmetry studied here, the

796: crossover occurs around $M=1/a$.  For $M=$ 5 or 6, corresponding to

797: the number of filament strands or PilT subunits, a system with $a>0.2$

798: would have the interesting property that the maximum stall force is

799: obtained for finite stiffness. If the binding strength between the

800: PilT subunits can be genetically engineered, it might be possible to

801: observe this effect in future experiments.

802: 

803: We compare results from the motor model for the filament retraction

804: force-velocity characteristics with experiments on \bakterie{N.\

805: gonorrhoeae} \citep{maier02,maier04}. Since the molecular details of

806: the retraction mechanism are unknown, it is unclear to what extent the

807: agreement we see reflects actual similarities between the model and

808: the real system.  More information would be useful for the

809: construction of more detailed models. Nevertheless, the model can

810: describe the general features of the experimental results -- i.e., the

811: plateau in the force-velocity relation at low applied forces and the

812: high stall force that is independent of reaction rate

813: \citep{maier02,maier04}.

814: 

815: For a quantitative comparison we select two different experimental

816: force-velocity relations that show large forces, one single event, and

817: one averaged curve, as shown in Fig.\ \ref{VFfit}. With different

818: parameters, the model gives a reasonable description of the average

819: data, as well as what looks like an atypical single event. The model

820: deviates from the average data at high forces, which indicates that

821: something is missing from the description.  However, the data might

822: include variations in cellular conditions that affect the average at

823: high forces. This is not accounted for in the simulations, where the

824: average is taken over thermal fluctuations with fixed parameters. The

825: single event is described better and does not suffer from such a

826: complication. In this case, the model is limited at high forces mainly

827: by thermally assisted transitions of motor subunits between potential

828: minima. Given the simplicity of the model, we find the agreement with

829: experiments encouaging.

830: 

831: Small ensembles of interaction motor systems have previously been

832: found to possess rich behavior without applied force

833: \citep{klumpp01,dan03,kolomeisky05}. As we have shown, this is true

834: also in the limit of high forces, which is a realistic experimental

835: situation.  Generation of strong forces in nanoscale devices is also

836: of technological interest, and it is tempting to speculate about the

837: possibility to realize a setup of interacting motors units pulling on

838: an artificial filament such as a carbon nano tube.

839: \begin{acknowledgments}

840: The authors thank Berenike Maier for valuable discussions and

841: comments. This work was supported by the Swedish Research Council

842: [Grants Nos.\ 2003-5001 (M.W.) 2001-6456, 2002-6240, and 2004-4831

843: (A.B.J.)], the Royal Institute of Technology, the G\"oran Gustafsson

844: Foundation, the Swedish Cancer Society, and Uppsala University.

845: \end{acknowledgments}

846: 

847: \bibliography{motorRefs}

848: 

849: \end{document}

850: 

851: