1: %\documentclass[aps,preprint,prl]{revtex4}
2: \documentclass[aps,prl,twocolumn]{revtex4}
3: \usepackage{graphicx}
4: \usepackage{latexsym}
5: \usepackage{amsmath}
6: \usepackage{amsfonts}
7: \usepackage{amssymb}
8: \newcommand{\be}{\begin{equation}}
9: \newcommand{\ee}{\end{equation}}
10: \newcommand{\bea}{\begin{eqnarray}}
11: \newcommand{\eea}{\end{eqnarray}}
12: \pagestyle{myheadings} \markboth{\today}{\today}
13: \begin{document}
14: \title{Worm-like Polymer Loops and Fourier Knots}
15: \author{S.M. Rapaport$^{\dag}$, Y. Rabin$^{\dag}$ and A. Yu. Grosberg$^{\ddag}$}
16: \affiliation{$^{\dag}$Department of Physics, Bar--Ilan University, Ramat-Gan 52900, Israel }
17: \affiliation{$^{\ddag}$Department of Physics, University of Minnesota, 116 Church Street
18: SE, Minneapolis, MN 55455, USA}
19:
20: \begin{abstract}
21: Every smooth closed curve can be represented by a suitable Fourier
22: sum. We show that the ensemble of curves generated by randomly
23: chosen Fourier coefficients with amplitudes inversely proportional
24: to spatial frequency (with a smooth exponential cutoff), can be
25: accurately mapped on the physical ensemble of worm-like polymer
26: loops. We find that measures of correlation on the scale of the
27: entire loop yield a larger persistence length than that calculated
28: from the tangent-tangent correlation function at small length
29: scales. The conjecture that physical loops exhibit additional
30: rigidity on scales comparable to the entire loop due to the
31: contribution of twist rigidity, can be tested experimentally by
32: determining the persistence length from the local curvature and
33: comparing it with that obtained by measuring the radius of
34: gyration of dsDNA plasmids. The topological properties of the
35: ensemble randomly generated worm-like loops are shown to be
36: similar to that of other polymer models.
37:
38: \end{abstract}
39: \maketitle
40:
41:
42: Classical polymer theories, from Debye to Flory to De Gennes
43: \cite{DeGennes} and beyond, make extensive use of the analogy
44: between conformation of a chain molecule and a Brownian random
45: walk (BRW). In mathematically idealized form, such a random walk
46: is thought of as a Wiener trajectory $\vec{r}(\tau)$ generated by
47: the measure $P\left\{ \vec{r}(\tau)\right\} \propto\exp\left[
48: -\mathrm{const} {\displaystyle\int} \left(
49: \partial\vec{r}/\partial\tau\right) ^{2}d\tau\right]$, where
50: $\tau$ is a parameter running along the trajectory. The BRW model
51: brings powerful mathematical techniques, such as the diffusion
52: equation, to bear on polymers but, however fruitful, this method
53: has its limitations. For example, it is fundamentally unable to
54: reproduce the fact of finite extensibility of polymer chains
55: stretched by a strong force\cite{MarcoSiggiaBustamante}. Another
56: field in which the Wiener trajectory model fails miserably is the
57: study of polymers with knots. Indeed, simulations of discrete
58: polymer models, or random polygons with $N$ steps, show
59: \cite{N01,N02,N03}, that the probability of trivial knot
60: configuration decays exponentially with $N$ as
61: $P_{0}(N)\sim\exp\left( -N/N^{\ast}\right) $, where $N^{\ast}$
62: defines the crossover from an unknotted to a knotted regime. It
63: was argued that Wiener trajectory polymer models can not be used
64: to calculate this probability \cite{grosberg}, since they
65: correspond to the limit $N\rightarrow\infty$, $l\rightarrow0$ ($l$
66: is a step length); $Nl^{2}=\mathrm{const}$, in which the
67: probability of a trivial knot vanishes. Notice that the contour
68: length of Wiener trajectory diverges, $Nl\rightarrow\infty$, so it
69: is not surprising that these models do not exhibit finite
70: extensibility. A better continuum model of polymers which can
71: handle the constraints imposed both by finite extensibility and by
72: the presence of knots, and can describe the elastic response of
73: such objects, is that of a worm-like chain. However, while the
74: statistical physics of linear worm-like polymers is well
75: understood, little is known about the conformations and the
76: topology of worm-like loops (WLL). Our goal in this article is to
77: work out a method allowing computational generation of an ensemble
78: of smoothly curved conformations of such polymers and to carry out
79: the statistical analysis of their geometric and topological
80: properties.
81:
82: The conformations of worm-like polymers are generated by the
83: measure $P\left\{ \vec{r}(s)\right\} \propto\exp\left[
84: -\mathrm{const} {\displaystyle\int} \left(
85: \partial^{2}\vec{r}/\partial s^{2}\right) ^{2}ds\right]$,
86: subject to the constraint of non-extensibility, $\left\vert
87: \partial\vec{r}/\partial s\right\vert =1$. That simply means that
88: conformations are Boltzmann weighted by the bending energy
89: proportional to the squared curvature. In more general elastic
90: models \cite{RabinPanyukov}, the three generalized curvatures
91: describing the object could be treated as independent Gaussian
92: variables. However, this is true only for linear (with open ends)
93: filaments and attempts to generalize these methods to the case of
94: closed loop failed because the loop closure conditions introduce a
95: non-local coupling between the curvatures that makes the problem
96: practically intractable (except for the case of small fluctuations
97: of a planar ring in which this coupling can be explicitly taken
98: into account \cite{ring}). One natural way to generate the
99: conformations of a closed loop is to expand each component
100: $r^{i}\left( \tau\right) $ in a Fourier series:
101: \begin{equation}
102: r^{i}\left( \tau\right) =\sum\limits_{n=1}\left[
103: A_{n}^{i}\cos\left( \frac{2\pi n\tau}{T}\right)
104: +B_{n}^{i}\sin\left( \frac{2\pi n\tau} {T}\right) \right] \ .
105: \label{eq:fourier}
106: \end{equation}
107: Here, $\tau$ is a parameter along the curve $\mathbf{r}(\tau)$
108: ($0\leq\tau\leq T$) and the summation goes up to some cutoff
109: frequency $n_{\max}$ (if the series converges sufficiently
110: rapidly, the cutoff can be replaced by infinity). Any
111: conformation of a sufficiently smooth closed loop can be fully
112: described by the set of Fourier coefficients $\left\{ A_{n}^{i},
113: B_{n}^{i}\right\} $, giving rise to the concept of Fourier knots
114: introduced in \cite{Trautwein,Kauffman}. Notice that a Fourier
115: knot is a much more general concept than the more traditional
116: Lissajou knot, for which only one frequency is present for each
117: coordinate direction.
118:
119: Using the above prescription one can generate an ensemble of loops of
120: different shapes by taking the Fourier coefficients from some random
121: distribution. However, even though all the loops have the same period $T$,
122: their contour lengths $L_{T}=\int_{0}^{T}d\tau^{\prime}\left\vert
123: d\mathbf{r/}d\tau^{\prime}\right\vert $ are different for each realization of
124: the Fourier coefficients. Even more importantly, since, in general,
125: $\left\vert \partial\vec{r}/\partial\tau\right\vert \neq1$, loops generated by
126: (\ref{eq:fourier}) do not obey the inextensibility condition and can not be
127: used to model worm-like polymers. In order to generate an ensemble of
128: different conformations of an inextensible loop of some well-defined contour
129: length, one has to transform to a representation in which trajectories are
130: parametrized by the arclength $s$
131: \begin{equation}
132: s\left( \tau\right) =\int_{0}^{\tau}d\tau^{\prime}\left\vert
133: d\mathbf{r/} d\tau^{\prime}\right\vert
134: ,\quad\text{with\quad}\left\vert \partial\vec {r}/\partial
135: s\right\vert =1\label{eq:s}
136: \end{equation}
137: and then bring all the generated conformations to the same contour length by a
138: suitable affine transformation of all lengths and coordinates (this
139: transformation does not affect the topology since the latter is independent of
140: the parametrization). Thus, we can replace $L_{T}=s(T)$ by any standard length
141: $L$, provided that we rescale all lengths using the affine transformation:
142: $\mathbf{r}\rightarrow\mathbf{r}L/L_{T}$. Therefore, in three steps
143: (generation of random coefficients and Fourier summation (\ref{eq:fourier}),
144: reparametrization (\ref{eq:s}), and affine transformation $s\rightarrow
145: sL/L_{T}$), of which the latter two couple all Fourier harmonics together in a
146: complex way, we obtain a statistical ensemble of smoothly bent conformations
147: of an inextensible loop of length $L$. Let us now check if this ensemble is
148: representative of physical conformations of a worm-like loop which possesses
149: some characteristic bending and (possibly) twist rigidities.
150:
151: The first step is to recognize that on length scales much larger
152: than some microscopic cutoff, the conformations of a worm-like
153: polymer are well represented by those of a BRW. For the latter,
154: the amplitudes of the Fourier components can be readily shown to
155: be of the form $\left\{ A_{n}^{i} ,B_{n}^{i}\right\}
156: _{BRW}=\lambda/n$ for $n\leq n_{\max}$ and zero otherwise, where
157: $\lambda$ is a random number (say, uniformly distributed between
158: $-1$ and $1$; the choice of another symmetric interval is
159: equivalent to rescaling the contour length, see below). As will be
160: demonstrated in the following, short scale behavior characteristic
161: of WLL can be obtained by replacing the abrupt cutoff $n_{\max}$
162: by Fourier coefficients that decrease smoothly with a
163: characteristic decay frequency $n_{0}$,
164: \begin{equation}
165: \left\{ A_{n}^{i},B_{n}^{i}\right\} _{WLL}=\lambda
166: e^{-n/n_{0}}/n, \label{WLC coeff}
167: \end{equation}
168: (notice that the WLL and the BRW expressions for the coefficients coincide for
169: $n\ll n_{0}$).
170:
171: The next step is to notice that the conformations of linear
172: worm-like polymer are governed by the persistence length, which
173: marks the cross-over between straight line conformations at small
174: length scales and random walk type behavior at large length scales
175: \cite{DeGennes}. We expect that on small scales, the conformations
176: of linear worm-like polymers and WLL are quite similar and that
177: for the latter this cross-over is represented in Fourier space in
178: terms of $n_{0}$. Of course, the analogy is meaningful only as
179: long as $n_{0}\gg1$, when there is a sufficiently broad range of
180: length scales between the persistence length and the contour
181: length of the knot. The characteristic property of the worm-like
182: chain model is exponential decay (as $\exp(-s/l_{1})$) of the
183: tangent-tangent correlation function, which defines the
184: persistence length $l_{1}$; similarly, for WLL, on length scales
185: sufficiently small compared to the entire loop, one expects that
186: $\left\langle \left\langle t(s)\cdot t(0)\right\rangle
187: \right\rangle \simeq\exp(-s/l_{1})$, where $\left\langle
188: \left\langle {}\right\rangle \right\rangle $ denotes averaging
189: both over the contour of a given loop and over the ensemble of
190: loops. This expectation is confirmed in Figure
191: \ref{fig:correlations}a, where the logarithm of the correlation
192: function is plotted against the dimensionless arclength $s/l_{1}$.
193: The choice $l_{1}=0.43/n_{0}$ allows us to superimpose data for
194: different values of $n_{0}$ in the range $30\leq n_{0}\leq270$
195: (the small shift of the exponent is the result of the finite
196: discretization of the contour length $s$).
197:
198:
199: \begin{figure*}
200: \centerline{\scalebox{0.9}{\includegraphics{fig1.eps}}}
201: \caption{(Color online) (a) Universal semi-log plot of the short
202: length scale behavior of the tangent-tangent correlation function
203: vs the scaled distance, $s/l_{1}$; the best fit
204: $exp(-(s/l_{1}-0.07))$ is given by the solid line. (b) Large scale
205: behavior of the correlation function, for different values of
206: $n_{0}$ (the height of the curves increases with $n_{0}$). (c)
207: Scaled (by $l_{3}$) mean-squared distance between points on the
208: loop. The BRW expression is plotted by the solid line. Different
209: values of $n_{0}$ are represented by black ($n_{0}=30$), red
210: ($n_{0}=50$), green ($n_{0}=70$), blue ($n_{0}=100$), cyan
211: ($n_{0}=150 $), magenta ($n_{0}=200$), yellow ($n_{0}=220$), dark
212: yellow ($n_{0}=250$) and navy ($n_{0}=270$). }
213: \label{fig:correlations}
214: \end{figure*}
215:
216:
217:
218: At first sight, analogy with worm-like chain models of linear
219: polymers suggests that on length scales much larger than $l_{1}$
220: the conformations of the loops are those of BRW with step size
221: (Kuhn length) given by twice the persistence length, $2l_{1}$.
222: However, this analogy is open to question because the large scale
223: behavior of a loop is strongly affected by the loop closure
224: constraint and it is not clear, a'priori, whether the same
225: persistence length controls both the small and the large scale
226: behavior. We therefore decided to apply our method to generate a
227: representative ensemble of configurations of WLL and use it to
228: compute the tangent-tangent correlation function and the mean
229: squared distance between two well-separated points $s$ and
230: $s^{\prime}$ ($\left\vert s-s^{\prime}\right\vert \gg l_{1}$)
231: along the contour.
232:
233: A simple estimate shows that on length scales much larger than the
234: persistence length, the tangent-tangent correlation function of a
235: WLL approaches a constant negative value $\left\langle
236: \left\langle t(s)\cdot t(0)\right\rangle \right\rangle
237: \simeq-2l_{2}/L$ where $l_{2}$ is, in general,different from
238: $l_{1}$ (the negative value of the correlation follows from the
239: fact that the tangent has to turn back on itself in order to come
240: back to its initial direction upon traversing the contour of the
241: loop). In Fig. \ref{fig:correlations}b this correlation function
242: is plotted in the interval $0.1<s<0.5$ (here and in the following
243: we take $L=1$) for different values of $n_{0}$. Upon averaging the
244: correlation functions over the oscillations, all the results for
245: different values of $n_{0}$ can be collapsed to a single
246: horizontal line by dividing them by $l_{2}=0.575/n_{0}$. We
247: therefore establish that unlike the case of linear worm-like
248: chains in which only a single persistence length exists, WLL have
249: at least two distinct persistence lengths, with
250: $l_{1}/l_{2}\simeq3/4$.
251:
252: We now proceed to compute the mean squared distance between two
253: points on the loop separated by a contour distance $s$,
254: $\left\langle \langle R^{2}\left( s\right) \rangle\right\rangle
255: =(1/L)\left\langle \int_{0}^{L}\left\vert \mathbf{r}\left(
256: s^{\prime}+s\right) -\mathbf{r}\left( s^{\prime}\right)
257: \right\vert ^{2}ds^{\prime} \right\rangle $. Since we expect WLL
258: to behave like BRW on scales much larger than some persistence
259: length $l_{3}$, the probability distribution of $R^{2}(s)$ can be
260: easily written down assuming that both $s\gg l_{3}$ and $L-s\gg
261: l_{3}$. Apart from a normalization factor, this probability is
262: equal to the product of two Gaussian functions: $\exp\left[
263: -R^{2}/(2s\ell_{3})\right] \exp\left[ -R^{2}/(2(L-s)\ell
264: _{3})\right] $. Averaging with this distribution yields the well
265: known relation (see, e.g., \cite{RedBook}), $\left\langle
266: R^{2}\left( s\right) \right\rangle _{\mathrm{BRW}}=2l_{3}s\left(
267: 1-s/L\right) .$ This result indicates that the plot of $\left.
268: \langle\langle R^{2}(s)\rangle \rangle\right/ l_{3}L$ against
269: $\sigma=s/L$ is universal, i.e., it is the unit height parabola
270: $4\sigma\left( 1-\sigma\right)$, independent of either
271: persistence length $l_{3}$ or total contour length $L$. Figure
272: \ref{fig:correlations}c shows that the data averaged over $1000$
273: different configurations collapse quite accurately on the expected
274: parabolic master curve. Furthermore, by looking at our
275: computational data for $\left\langle \left\langle R^{2}\left(
276: s=L/2\right) \right\rangle \right\rangle $ averaged over all pairs
277: of diametrically opposite points on the loop ($s=L/2=0.5$) we were
278: able to relate the persistence length to the cutoff $n_{0}$.
279: Within the accuracy of our simulation $l_{3}$ coincides with
280: $l_{2}$ (see Fig. \ref{fig:persistence_lengths}).
281:
282: \begin{figure}
283: \centerline{\scalebox{0.6}{\includegraphics{fig2.eps}}}
284: \caption{(Color online). Log-log plot of persistence lengths
285: $l_{1},l_{2},l_{3}$ (black squares, green triangles and red
286: circles respectively) vs $n_{0}$. The expressions $0.435/n_{0}$
287: and $0.57/n_{0}$ are shown as solid lines. The standard deviations
288: are represented by error bars.} \label{fig:persistence_lengths}
289: \end{figure}
290:
291: The conclusion that local and global statistical properties of
292: worm-like loops are characterized by two different persistence
293: lengths, $l_{1}$ and $l_{2}(=l_{3}),$ respectively, is a new
294: result of this work. Although analytical theory of worm-like loops
295: does not exist at present, we can offer some tentative
296: considerations about the origin of the two persistence lengths,
297: based on the study of small fluctuations of elastic rings
298: \cite{ring}. For rings of zero linking number that possess both
299: bending and twist persistence lengths, it was shown that while on
300: small scales writhe fluctuations depend only on the bending
301: persistence length (on such scales the ring is essentially
302: straight and bending and twist decouple), both bending and twist
303: persistence lengths contribute to the writhe on larger scales, for
304: which the global geometry of the closed ring becomes important.
305: Extrapolating this result to the case of a strongly fluctuating
306: loop considered in this work, suggests that while $l_{1}$
307: represents the pure bending contribution to the persistence
308: length, $l_{2}$ contains both the bending and the twist
309: contributions and is therefore larger than $l_{1}$ \cite{Mezard}.
310:
311: We now turn to the topological properties of WLL and examine the dependence of
312: the trivial knot probability $P_{0}$ on the cutoff frequency $n_{0}$. Since
313: $n_{0}$ determines the Kuhn length defined as either $2l_{1}$ or the $2l_{2}$,
314: we can bring our knotting probability data to a form comparable to that for
315: discrete polymer models, where trivial knot probability depends on the number
316: of segments $N$. In our case, we generate Fourier knots as completely smooth
317: curves, but we can define the number of effective segments as the ratio of the
318: contour length to the Kuhn length. For long random--walk-like polymers, this
319: definition coincides with the standard one accepted in polymer physics for the
320: Kuhn segments \cite{RedBook} but there remains an ambiguity associated with
321: the choice of $N_{1}=L/2l_{1}$ or $N_{2}=L/2l_{2}$. In order to resolve this
322: ambiguity we will measure the probability to obtain a trivial knot as a
323: function of the cutoff frequency $n_{0}.$
324:
325: To address the topology of the loops computationally, we employ the knot
326: analysis routine due to R. Lua \cite{lua} which identifies knots by computing
327: the Alexander polynomial invariant $\Delta(t)$ at one value of argument,
328: $t=-1$, and Vassiliev invariants of degrees two, $v_{2}$, and three, $v_{3}$.
329: This set of invariants is widely considered as powerful enough for reliable
330: identification of the trivial knot for all lengths achievable in practical
331: computations. The details of this topological routine are described in
332: \cite{lua}.
333:
334: The data on the trivial knot probability are shown in Figure
335: \ref{fig:trivial_knot_probability}. As this figure indicates, the
336: trivial knot probability fits well to the exponential
337: \begin{equation}
338: P_{0}\left( n_{0}\right) \sim\exp\left[ -n_{0}/242\right] \ .
339: \label{eq:exponential_fit}
340: \end{equation}
341: Notice that since this relation involves only the cutoff on the Fourier
342: series, formula (\ref{eq:exponential_fit}) can be re-interpreted in purely
343: mathematical form, not involving any references to polymers, or, for that
344: matter, to any physics. While there is no fundamental understanding of the
345: origin of this large cutoff ($n_{0}^{\ast}=242$) at present, our formulation
346: hints at the existence of a hitherto unexplored connection between Fourier
347: analysis and topology of space curves and will hopefully stimulate new work on
348: this fundamental problem.
349:
350: \begin{figure}
351: \centerline{\scalebox{0.6}{\includegraphics{fig3.eps}}}
352: \caption{(Color online). Semi-log plot of the trivial knot
353: probability as a function of $n_{0}$.}
354: \label{fig:trivial_knot_probability}
355: \end{figure}
356:
357: In summary, we have shown that the choice of random Fourier
358: coefficients with amplitudes that decay with frequency as
359: $n^{-1}\exp(-n/n_{0})$, generates a statistical ensemble of
360: Fourier knots whose local properties coincide with those of
361: worm-like polymers with persistence length that scales as
362: $1/n_{0}$. We found that even though WLL behave on large scales as
363: BRW as expected, the effective step size of this BRW is larger
364: than that calculated from the local persistence length of the
365: loop. This new prediction can be directly tested by experiments on
366: double stranded DNA loops that can monitor both local (persistence
367: length) and global (e.g., radius of gyration) properties of the
368: polymers (see, e.g., \cite{AFM1,AFM2,SFM}). We also demonstrated
369: that similar to discrete models of polymers or random polygons,
370: our Fourier knots exhibit exponential decay of the unknotting
371: probability with the number of \emph{effectively} straight
372: segments, or, equivalently, with the maximal spatial frequency
373: included in Fourier expansion. The characteristic cross-over
374: determined by this exponential decay represents a large number,
375: which, although in the same ballpark as for other known models,
376: remains an unexplained puzzle.
377:
378: \acknowledgements We acknowledge Ronald Lua's help in the use of his
379: computational knot analysis routine. The work of AG was supported in part by
380: the MRSEC Program of the National Science Foundation under Award Number
381: DMR-0212302. This research was supported in part by a grant from the US-Israel
382: Binational Science Foundation. YR would like to acknowledge the hospitality of
383: Institute for Mathematics and its Applications of the University of Minnesota.
384:
385: \begin{thebibliography}{99}
386:
387:
388: \bibitem{DeGennes} P.-G. De Gennes, \emph{Scaling Concepts in Polymer Physics}
389: (Cornell University Press, Ithaca, NY, 1979.
390:
391: \bibitem{MarcoSiggiaBustamante} C. Bustamante, J. F. Marko, E. D. Siggia, and
392: S. Smith, \emph{Science} \textbf{265}, 1599 (1994).
393:
394: \bibitem{N01} K. Koniaris and M. Muthukumar, Phys. Rev. Lett. \textbf{66},
395: 2211-2214 (1991).
396:
397: \bibitem{N02} T. Deguchi, and K. Tsurusaki, \emph{Phys. Rev. E.} \textbf{55},
398: 6245-6248 (1997).
399:
400: \bibitem{N03} N.T. Moore, R. Lua, A.Y. Grosberg, \emph{Proc. Nat. Ac. Sci.}
401: \textbf{101}, 13431 (2004).
402:
403: \bibitem{grosberg} A. Yu. Grosberg, \emph{Phys. Rev. Lett.} \textbf{85},
404: 3858-3861 (2000).
405:
406: \bibitem{RabinPanyukov} S. Panyukov and Y. Rabin, \emph{Phys. Rev. Lett.}
407: \textbf{85}, 2404 (2000); \emph{Phys. Rev. E} \textbf{62}, 7135 (2000).
408:
409: \bibitem{ring} S. Panyukov and Y. Rabin, \emph{Phys. Rev. E} \textbf{64},
410: 011909 (2001).
411:
412: \bibitem{Trautwein} A.Trautwein, \emph{Harmonic Knots} Ph. D. Thesis,
413: University of Iowa, 1995; Web Page: http://www2.carthage.edu/\symbol{126}trautwn/
414:
415: \bibitem{Kauffman} L.H. Kauffman, Chapter 19 in: \emph{Ideal knots},
416: A.Stasiak, V.Katrich, L.H.Kaufman - Editors (World Scientific, Singapore, 1998).
417:
418: \bibitem{RedBook} A. Yu. Grosberg and A. R. Khokhlov, \emph{Statistical
419: Physics of Macromolecules }(AIP Press, NY, 1994).
420:
421: \bibitem{lua} R. Lua, A. Borovinskiy and A. Yu. Grosberg, \emph{Polymer}
422: \textbf{45}, 717 (2004).
423:
424: \bibitem{Mezard} This effect is expected only for worm-like loops; for linear
425: worm-like chains it can be shown that twist degrees of freedom can be
426: integrated over and that the spatial conformations of the polymer depend on
427: the bending rigidity only: C. Bouchiat and M. Mezard, \emph{Eur. Phys. J.} E
428: \textbf{2}, 377 (2000).
429:
430: \bibitem{AFM1} Y.L. Lyubchenko and L.S. Shlyakhtenko, \emph{Proc. Natl. Acad.
431: Sci. USA }\textbf{94}, 496 (1997).
432:
433: \bibitem{AFM2} A. Scipioni et al., \emph{Chemistry \& Biology} \textbf{9}, 1315 (2002)
434:
435: \bibitem{SFM} M. Bussiek, N. M\"ucke and J. Langowski,
436: \emph{Nucleic Acids Research }\textbf{31}, e137 (2003).
437: \end{thebibliography}
438:
439:
440:
441:
442:
443:
444:
445:
446:
447:
448: \end{document}
449: